CN116528358B - Wi-Fi-based fall positioning joint detection method - Google Patents
Wi-Fi-based fall positioning joint detection method Download PDFInfo
- Publication number
- CN116528358B CN116528358B CN202310785669.5A CN202310785669A CN116528358B CN 116528358 B CN116528358 B CN 116528358B CN 202310785669 A CN202310785669 A CN 202310785669A CN 116528358 B CN116528358 B CN 116528358B
- Authority
- CN
- China
- Prior art keywords
- falling
- data
- positioning
- csi
- feature vector
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 101
- 239000013598 vector Substances 0.000 claims abstract description 252
- 239000011159 matrix material Substances 0.000 claims description 127
- 238000004364 calculation method Methods 0.000 claims description 80
- 238000000034 method Methods 0.000 claims description 78
- 230000004044 response Effects 0.000 claims description 76
- 238000012360 testing method Methods 0.000 claims description 56
- 238000012549 training Methods 0.000 claims description 39
- 238000000605 extraction Methods 0.000 claims description 33
- 238000013507 mapping Methods 0.000 claims description 23
- 230000008859 change Effects 0.000 claims description 21
- 238000007781 pre-processing Methods 0.000 claims description 18
- 238000001914 filtration Methods 0.000 claims description 13
- 238000006467 substitution reaction Methods 0.000 claims description 11
- 238000006073 displacement reaction Methods 0.000 claims description 8
- 238000005070 sampling Methods 0.000 claims description 8
- 239000000969 carrier Substances 0.000 claims description 6
- 230000009466 transformation Effects 0.000 claims description 6
- 230000004807 localization Effects 0.000 claims description 4
- 230000000694 effects Effects 0.000 abstract description 5
- 238000010586 diagram Methods 0.000 description 14
- 230000006870 function Effects 0.000 description 11
- 230000002159 abnormal effect Effects 0.000 description 6
- 108010076504 Protein Sorting Signals Proteins 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 230000009471 action Effects 0.000 description 4
- 230000005540 biological transmission Effects 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 3
- 238000004590 computer program Methods 0.000 description 3
- 230000009977 dual effect Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000005484 gravity Effects 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 230000001133 acceleration Effects 0.000 description 2
- 238000012850 discrimination method Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000035772 mutation Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000001228 spectrum Methods 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 230000001627 detrimental effect Effects 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000009323 psychological health Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W64/00—Locating users or terminals or network equipment for network management purposes, e.g. mobility management
- H04W64/006—Locating users or terminals or network equipment for network management purposes, e.g. mobility management with additional information processing, e.g. for direction or speed determination
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S5/00—Position-fixing by co-ordinating two or more direction or position line determinations; Position-fixing by co-ordinating two or more distance determinations
- G01S5/02—Position-fixing by co-ordinating two or more direction or position line determinations; Position-fixing by co-ordinating two or more distance determinations using radio waves
- G01S5/0294—Trajectory determination or predictive filtering, e.g. target tracking or Kalman filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
- H04W4/30—Services specially adapted for particular environments, situations or purposes
- H04W4/33—Services specially adapted for particular environments, situations or purposes for indoor environments, e.g. buildings
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W84/00—Network topologies
- H04W84/02—Hierarchically pre-organised networks, e.g. paging networks, cellular networks, WLAN [Wireless Local Area Network] or WLL [Wireless Local Loop]
- H04W84/10—Small scale networks; Flat hierarchical networks
- H04W84/12—WLAN [Wireless Local Area Networks]
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Remote Sensing (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Radar, Positioning & Navigation (AREA)
- Position Fixing By Use Of Radio Waves (AREA)
Abstract
The invention relates to a Wi-Fi-based fall positioning joint detection method, which comprises the following steps of: acquiring CSI data and extracting features of the CSI data to obtain a falling feature vector and a positioning feature vector; identifying a falling type label corresponding to the falling feature vector through a falling detection model; judging whether falling happens or not according to the falling type label; if the falling type label is 0, judging that falling does not occur; if the falling type label is 1, judging that falling occurs, matching the positioning feature vector with a preset area through a positioning model, and taking the matched preset area as a positioning result; the preset area refers to an area divided by the CSI data sample adopted for constructing the positioning model. The falling positioning joint detection method can realize falling detection without any wearing equipment, and can detect the falling position, so that a more accurate falling positioning joint detection effect can be realized with lower cost.
Description
Technical Field
The invention relates to the technical field of data processing, in particular to a Wi-Fi-based fall positioning joint detection method.
Background
The fall event is detrimental to both the physiological and psychological health of the elderly. Studies have shown that the medical outcome after a fall is largely dependent on the time of reaction and rescue, and that delaying treatment after a fall greatly increases the risk of death. It is thus possible to save the life of the elderly to detect in a timely and automatic manner whether a fall event has occurred.
In the prior art, a conventional means for detecting whether a fall event occurs is to let the elderly wear a wearable intelligent device; be equipped with acceleration sensor on the wearable smart machine, judge the motion state of old man through acceleration sensor to detect whether the event of falling takes place.
Disclosure of Invention
Based on the above, the invention aims to provide a Wi-Fi-based fall positioning joint detection method and device, which have the advantages that fall detection can be realized without wearing any equipment, and the position where the fall occurs can be detected.
A Wi-Fi-based fall positioning joint detection method comprises the following steps:
s1, acquiring CSI data, and carrying out feature extraction on the CSI data to obtain a falling feature vector and a positioning feature vector;
s2, identifying a falling type label corresponding to the falling feature vector through an SVM falling detection model; the falling type label is 0 or 1;
S3, judging whether falling occurs or not according to the falling type label; if the falling type label is 0, judging that falling does not occur; if the falling type label is 1, judging that falling occurs, and executing step S4;
s4, matching the positioning feature vector with a preset area through an SVM positioning model, and taking the matched preset area as a positioning result; the preset area is an area divided by CSI data samples adopted for constructing an SVM positioning model.
The Wi-Fi-based falling positioning joint detection method can realize falling detection without any wearing equipment, and can detect the position of falling, so that a more accurate falling positioning joint detection effect can be realized at lower cost.
Further, identifying the falling type label corresponding to the falling feature vector through an SVM falling detection model, specifically comprising the following steps:
recording the falling characteristic vector asThe dimension is +.>;
The SVM fall detection model comprises a first hyperplane dividing a first feature space into two areas; the first hyperplane equation is, wherein />Is->Vector of dimensions>To determine the normal vector of the first hyperplane direction,is->Transposed matrix of >Is a displacement term;
will fall feature vectorSubstituting into the first hyperplane equation, and judging the area of the falling characteristic vector in the first characteristic space according to the value obtained by substitution calculation, namely +.>, in the formula />Representing a fall feature vector +.>In the area of (2),>is a sign function;
and taking the area of the falling feature vector in the first feature space as a falling type label corresponding to the falling feature vector.
Further, the positioning feature vector is matched with a preset area through an SVM positioning model, and the method specifically comprises the following steps:
recording the positioning characteristic vector asThe dimension is +.>The method comprises the steps of carrying out a first treatment on the surface of the The number of the preset areas is recorded as k;
the SVM positioning model comprises k-1 second hyperplanes which divide the second characteristic space into k areas; the second hyperplane equation is, wherein />Is->Vector of dimensions>To determine the normal vector of the second hyperplane direction,is->Transposed matrix of>Is a displacement term; />An integer between 1 and k-1, representing an ith equation of the k-1 second hyperplane equations;
will locate feature vectorsSubstituting the locating feature vector into each second hyperplane equation respectively, and judging the area of the locating feature vector in the second feature space according to the value obtained by substitution calculation; i.e. < - >, in the formula />Representing a localization feature vector +.>Relative to (I)>The region of the second hyperplane, < >>Is a sign function; comprehensive positioning feature vector->Substituted into each second hyperplane equation>Judging the area of the positioning feature vector in the second feature space;
taking the area of the positioning feature vector in the second feature space as a positioning class label corresponding to the positioning feature vector; the positioning class label is an integer between 1 and k, and indicates what preset area the CSI data is matched with.
Further, the SVM fall detection model is constructed by the following steps:
s21, collecting a plurality of CSI data samples; these CSI data samples include both fall samples and non-fall samples;
s22, preprocessing and extracting features of the CSI data samples to obtainCharacteristic values for fall detection, which characteristic values constitute +.>A fall feature vector of dimension; taking 1 as a falling type label corresponding to the falling characteristic vector of the falling sample, and taking 0 as a falling type label corresponding to the falling characteristic vector of the non-falling sample;
s23, randomly dividing the CSI data sample into a training set and a testing set according to a certain proportion;
S24, judging whether the falling feature vectors of the training set are linearly separable; the linear separable means that the falling feature vectors of the training set can be divided into two types by a hyperplane; if yes, taking the falling feature vector of the training set as a mapped falling feature vector; if not, performing first transformation on the falling feature vector of the training set to map the falling feature vector to another high-dimensional space, and converting the falling feature vector into mapped falling feature vectors which can be divided into two types by a hyperplane;
s25, constructing a hyperplane equation according to the mapped falling feature vector and the falling type label thereof, so that the hyperplane equation can divide the first feature space into two areas, and the falling classification accuracy of the mapped falling feature vector is within a certain allowable range;
the fall classification accuracy refers to that a fall feature vector is substituted into the hyperplane equation, the region of the fall feature vector in the first feature space is judged according to a value obtained by substitution calculation, and the obtained predicted fall label is compared with the true fall label in accuracy;
s26, substituting the falling feature vector of the test set into the hyperplane equation constructed in the S25, and judging the area of the falling feature vector of the test set in the feature space according to the value calculated by the substituted equation to obtain a predicted falling type label of the test set;
Comparing the predicted falling type label of the test set with the real falling type label to obtain the falling classification accuracy of the test set of the hyperplane equation constructed at this time;
s27, returning to the step S23, reselecting the training set and the testing set to construct a hyperplane equation, and repeating for a plurality of times;
and comparing the test set falling classification accuracy of the hyperplane equations generated each time, and selecting the hyperplane equation with the highest test set falling classification accuracy as the first hyperplane equation.
Further, the SVM positioning model is constructed by the following method:
s41, dividing an indoor space into k preset areas, and collecting a plurality of CSI positioning data samples in each preset area;
s42, preprocessing and extracting features of the CSI positioning data sample to obtainA plurality of feature values for positioning; these characteristic values constitute->A dimensional positioning feature vector; taking a preset area corresponding to the CSI positioning data sample as a positioning class label of the positioning feature vector;
s43, randomly dividing the CSI positioning data sample into a training set and a testing set according to a certain proportion;
s44, judging whether the positioning feature vectors of the training set are linearly separable; the linear separable means that the positioning feature vector of the training set can be divided into k classes by k-1 hyperplanes; if yes, the positioning feature vector of the training set is used as a mapping positioning feature vector; if not, performing second transformation on the positioning feature vector of the training set to map the positioning feature vector to another high-dimensional space, and converting the positioning feature vector into a mapping positioning feature vector which can be divided into k classes by k-1 hyperplanes;
S45, constructing k-1 hyperplane equations according to the mapping positioning feature vector and the positioning class labels thereof, so that the k-1 hyperplanes can divide the second feature space into k areas, and the positioning classification accuracy of the mapping positioning feature vector is within a certain allowable range;
the positioning classification accuracy refers to the accuracy of the obtained predicted positioning class label compared with the true positioning class label by substituting the positioning feature vector into k-1 hyperplane equations and judging the area of the positioning feature vector in the second feature space according to the value obtained by substitution calculation;
s46, substituting the positioning feature vector of the test set into the k-1 hyperplane equations constructed in the S45, and judging the area of the positioning feature vector of the test set in the second feature space according to the substituted and calculated value to obtain a predicted positioning class label of the test set;
comparing the predicted positioning class label of the test set with the real positioning class label to obtain the positioning classification accuracy of the test set of the hyperplane equation constructed at this time;
s47, returning to the step S43, reselecting the training set and the testing set to construct a hyperplane equation, and repeating for a plurality of times;
and comparing the test set positioning classification accuracy of the hyperplane equations generated each time, and selecting the hyperplane equation group with the highest test set positioning classification accuracy as a second hyperplane equation.
Further, the step S1 specifically includes the following steps:
s11, obtaining CSI data;
s12, carrying out feature extraction on the CSI data to obtain a plurality of feature values;
s13, dividing a plurality of characteristic values into characteristic values for fall detection and characteristic values for positioning; the characteristic values used for falling detection form falling characteristic vectors, and the characteristic values used for positioning form positioning characteristic vectors;
further, the step S12 specifically includes the following steps:
s121, performing time domain feature extraction on the CSI data to obtain a time domain feature value; the time domain characteristic value comprises a time domain amplitude value mean valueEnergy->Amplitude margin interval->Amplitude difference->Margin factor->Signal rate of change->Mean value of phase->Standard deviation of phase->Phase margin interval->And phase difference->;
S122, carrying out frequency domain feature extraction on the CSI data to obtain a frequency domain feature value; the frequency domain characteristic value comprises a frequency domain amplitude average valueFrequency domain standard deviation->Maximum value of frequency domain->And frequency domain skewness coefficient->;
The CSI data are a plurality of CSI data packets; the content of a CSI packet is a p×n matrix:
wherein ,indicate->No. H of the link>Subcarrier data; the subcarrier data is a complex number, and comprises amplitude data and phase data; the amplitude data of all the subcarriers form CSI amplitude data; the phase data of all sub-carriers form CSI phase data;
The step S121 further includes the steps of:
s121a, calculating a time domain amplitude mean value of the CSI amplitude dataThe method comprises the steps of carrying out a first treatment on the surface of the The time domain amplitude mean->The calculation formula of (2) is as follows:
, wherein />Is->CSI amplitude data matrix of individual data packets +.>Line->Subcarrier amplitude data of column,/>The number of the data packets;
s121b, calculating the energy of the CSI amplitude dataThe method comprises the steps of carrying out a first treatment on the surface of the Said energy->The calculation formula of (2) is as follows: />, wherein />Is->CSI amplitude data matrix of individual data packets +.>Line->Subcarrier amplitude data of column,/>The number of the data packets;
s121c, calculating amplitude range interval of the CSI amplitude dataThe method comprises the steps of carrying out a first treatment on the surface of the The amplitude margin interval +.>The calculation formula of (2) is as follows:
, in the formula />The number of data packets corresponding to the maximum amplitude data,/-for the maximum amplitude data>The data packet number corresponding to the minimum amplitude data;
s121d, calculating the amplitude difference of the CSI amplitude dataThe method comprises the steps of carrying out a first treatment on the surface of the The amplitude difference->The difference of subcarrier amplitude data of different links is indicated, and the calculation formula is as follows:
, in the formula />Representing the link number>Is->CSI amplitude data matrix of individual data packets +.>Line->Subcarrier amplitude data of the columns;
s121e, calculating margin factor of CSI amplitude dataThe method comprises the steps of carrying out a first treatment on the surface of the Said margin factor->The calculation formula of (2) is as follows:
Middle square root amplitude +.>; wherein />Is the firstCSI amplitude data matrix of individual data packets +.>Line->Subcarrier amplitude data of column,/>The number of the data packets;
s121f: calculating the signal speed change rate of the CSI amplitude dataThe method comprises the steps of carrying out a first treatment on the surface of the The rate of change of the signal speed ∈ ->The calculation formula of (2) is as follows:
, in the formula />Is->CSI amplitude data matrix of individual data packets +.>Line->Subcarrier amplitude data of column,/>;/>Is a data packetA number of; />The sampling period is the interval of sampling time points corresponding to adjacent data packets;
s121g: calculating the phase mean of CSI phase dataThe method comprises the steps of carrying out a first treatment on the surface of the The phase mean->The calculation formula of (2) is as follows:
, wherein />Is->CSI phase data matrix of individual data packets +.>Line->Subcarrier phase data of column,/->The number of the data packets;
s121h: calculating the phase standard deviation of the CSI phase dataThe method comprises the steps of carrying out a first treatment on the surface of the The phase standard deviation->The calculation formula of (2) is as follows:
, wherein />Is->CSI phase data matrix of individual data packets +.>Line->Subcarrier phase data of column,/->For the number of data packets, < > for>Is the phase average value;
s121i, calculating the phase margin interval of the CSI phase dataThe method comprises the steps of carrying out a first treatment on the surface of the The phase margin interval->The calculation formula of (2) is as follows:
, wherein />For the maximum number of packets corresponding to the phase data,/-for the phase data >The data packet number corresponding to the minimum phase data;
s121j, calculating the phase difference of the CSI amplitude dataThe method comprises the steps of carrying out a first treatment on the surface of the Said phase difference->Refers to the difference in subcarrier phase data for different links, whichThe calculation formula is as follows:
, in the formula />Representing the link number>Is->CSI phase data matrix of individual data packets +.>Line->Subcarrier phase data of the columns;
the step S122 further includes the steps of:
s112a, counting the CSI dataIs fast Fourier transformed to give +.>A frequency response matrix; then extract +.>The frequency response matrix representing the low-frequency component forms a low-frequency response matrix; sign->Indicate->A low frequency response matrix +.>Line->Frequency response amplitude data for the columns; k is 1 to->An integer therebetween; />Less than or equal to->;
S112b, calculating frequency domain amplitude average value according to the low frequency response matrixThe method comprises the steps of carrying out a first treatment on the surface of the The frequency domain amplitude average value +.>The calculation formula of (2) is as follows: />, in the formula />Represents the kth low frequency response matrix +.>Line->Frequency response amplitude data of column,/>The number of the low-frequency response matrixes;
s112c, calculating frequency domain standard deviation according to the low-frequency response matrixThe method comprises the steps of carrying out a first treatment on the surface of the The frequency domain standard deviation->The calculation formula of (2) is as follows:
, in the formula />Represents the kth low frequency response matrix +.>Line->Frequency response amplitude data of column,/>For the number of low frequency response matrices, < +.>Is the average value of the frequency domain amplitude;
s112d, calculating the frequency domain maximum according to the low-frequency response matrixThe method comprises the steps of carrying out a first treatment on the surface of the The calculation formula of the frequency domain maximum value is as follows:
the method comprises the steps of carrying out a first treatment on the surface of the The first>Line->Column element->Equal to->Line->Column element->At->Maximum value in the low frequency response matrix;
s112e, calculating frequency domain skewness coefficient according to the low-frequency response matrixThe method comprises the steps of carrying out a first treatment on the surface of the The skewness coefficient->The calculation formula of (2) is as follows:
, wherein />Represents the kth low frequency response matrix +.>Line->Frequency response amplitude of column, +.>Maximum in frequency domain>Is the average value of the amplitude values in the frequency domain,is the standard deviation of frequency domain>Is the number of low frequency response matrices.
Further, the Wi-Fi-based fall positioning joint detection method further comprises the following step S14: filtering and denoising the CSI data to obtain filtered CSI data; the step S14 is located after the step S11 and before the step S12; step S12 becomes: extracting features of the filtered CSI data to obtain a plurality of feature values;
the step S14 specifically includes the following steps:
S14a, performing linear denoising on the CSI phase data to obtain a linear denoising CSI phase data matrix;
s14b, performing outlier removal on the CSI amplitude data and the linear denoising CSI phase data to obtain outlier removed CSI data;
and S14c, performing Butterworth low-pass filtering on the CSI data with the outlier removed to obtain filtered CSI data.
A Wi-Fi-based fall positioning joint detection device comprises a preprocessing module, an SVM fall detection model, a judgment module and an SVM positioning model.
The preprocessing module is used for acquiring the CSI data and extracting the characteristics of the CSI data to obtain a falling characteristic vector and a positioning characteristic vector;
the SVM fall detection model is used for identifying a fall type label corresponding to the fall feature vector; the falling type label is 0 or 1;
the judging module is used for judging whether falling happens according to the falling type label; if the falling type label is 0, judging that falling does not occur; if the falling type label is 1, judging that falling occurs, and calling the SVM positioning model;
the SVM positioning model is used for matching the positioning feature vector with a preset area, and taking the matched preset area as a positioning result; the preset area is an area divided by CSI data samples adopted for constructing an SVM positioning model.
Further, the preprocessing module comprises a data acquisition unit, a feature extraction unit and a feature vector extraction unit.
The data acquisition unit is used for acquiring the CSI data;
the feature extraction unit is used for carrying out feature extraction on the CSI data to obtain a plurality of feature values;
the characteristic vector extraction unit is used for dividing a plurality of characteristic values into characteristic values for fall detection and characteristic values for positioning; wherein the feature values for fall detection constitute a fall feature vector, and the feature values for positioning constitute a positioning feature vector.
A fall positioning joint detection system based on Wi-Fi comprises Wi-Fi equipment, a wireless network card and a fall positioning joint detection device; the Wi-Fi equipment is fixedly arranged at a certain place in the room and transmits Wi-Fi signals; the wireless network card is fixedly arranged at a certain indoor position, receives Wi-Fi signals, demodulates the Wi-Fi signals, and sends the CSI data packet obtained through demodulation to any Wi-Fi-based falling positioning combined detection device.
For a better understanding and implementation, the present invention is described in detail below with reference to the drawings.
Drawings
Fig. 1 is a schematic diagram of a Wi-Fi-based fall positioning joint detection system according to an embodiment of the present invention;
Fig. 2 is a schematic diagram of a Wi-Fi-based fall positioning joint detection device according to an embodiment of the present invention;
fig. 3 is a flow chart of a Wi-Fi-based fall positioning joint detection method according to an embodiment of the present invention;
fig. 4 is a schematic diagram of CSI amplitude data of 30 subcarriers of a link according to an embodiment of the present invention;
fig. 5 is a schematic diagram of CSI phase data before linear denoising according to an embodiment of the present invention;
fig. 6 is a schematic diagram of CSI phase data with linear denoising according to an embodiment of the present invention;
fig. 7 is a schematic diagram of CSI data and filtered CSI data with outlier removal in an embodiment of the present invention;
FIG. 8 is a schematic view illustrating indoor space region division according to an embodiment of the present invention;
FIG. 9 is a schematic diagram of classification accuracy of SVM positioning models according to an embodiment of the present invention;
fig. 10 is a schematic diagram of prediction accuracy of a fall positioning joint detection method according to an embodiment of the present invention.
Detailed Description
The channel state information CSI (Channel State Information) for Wi-Fi signals carries rich features. And filtering and extracting the CSI data to obtain characteristic values which can be used for identifying the change of the indoor environment. In the field of wireless communications, CSI refers to the channel properties of a communications link and describes the attenuation factors of signals on each transmission path, such as signal scattering, environmental attenuation, and distance attenuation. Wi-Fi technology based on 802.11 n protocol adopts MIMO-OFDM system, and 30 sub-carriers in a wireless communication channel can be obtained by using tool software provided by Daniel Halperin of Washington university; finally, a normalized CSI data matrix can be obtained, wherein each row of the CSI data matrix has 30 elements, and each element represents one subcarrier data; different columns of the CSI data matrix represent different links.
Since Wi-Fi devices have become popular in various places in the life of residents, the scheme of the present invention contemplates using Wi-Fi signals for fall detection without introducing additional devices. According to the Wi-Fi fall detection method, only one Wi-Fi device for transmitting Wi-Fi signals and a wireless network card for receiving the Wi-Fi signals are arranged indoors, fall detection can be completed by analyzing signal data received by the wireless network card, and the old people do not need to be forced to wear wearable intelligent equipment. According to the invention, falling detection is realized through the CSI data of Wi-Fi signals, firstly, the CSI data is acquired, then, the CSI data is preprocessed and extracted in characteristics, and whether falling occurs or not is judged according to the extracted characteristic value; if the falling happens, the falling position is further judged. Based on the technical conception, the fall positioning joint detection method comprises two large modules of data acquisition and preprocessing, fall detection and indoor positioning, wherein the fall detection and the indoor positioning also relate to the construction problem of a classification model.
Referring to fig. 1, fig. 2 and fig. 3, fig. 1 is a schematic diagram of a Wi-Fi-based fall positioning joint detection system according to an embodiment of the present invention; fig. 2 is a schematic diagram of a fall positioning joint detection device according to an embodiment of the present invention; fig. 3 is a flow chart of a Wi-Fi-based fall positioning joint detection method according to an embodiment of the present invention.
The invention discloses a Wi-Fi-based falling positioning joint detection system, which comprises Wi-Fi equipment, a wireless network card and a falling positioning joint detection device; the Wi-Fi equipment is fixedly arranged at a certain place in the room and transmits Wi-Fi signals; the wireless network card is fixedly arranged at a certain indoor position, receives Wi-Fi signals, demodulates the Wi-Fi signals, and sends the CSI data packet obtained through demodulation to the falling positioning combined detection device; the falling positioning joint detection device receives the CSI data packet, judges whether falling occurs according to the CSI data packet, and continuously judges the falling position if the falling occurs.
The fall positioning joint detection device comprises a preprocessing module 1, an SVM fall detection model 2, a judging module 3 and an SVM positioning model 4.
The preprocessing module 1 is configured to execute step S1: acquiring CSI data and extracting features of the CSI data to obtain a falling feature vector and a positioning feature vector;
the SVM fall detection model 2 is configured to perform step S2: identifying a falling type label corresponding to the falling feature vector; the fall-type tag is 0 or 1.
The judging module 3 is configured to execute step S3: judging whether falling happens or not according to the falling type label; if the falling type label is 0, judging that falling does not occur; if the falling type label is 1, the falling is judged to occur, and the SVM positioning model 4 is called.
The SVM positioning model 4 is configured to execute step S4: matching the positioning feature vector with a preset area, and taking the matched preset area as a positioning result; the preset area is an area divided by CSI data samples adopted for constructing an SVM positioning model.
(1) Pretreatment part
The preprocessing module 1 includes a data acquisition unit 11, a feature extraction unit 12, and a feature vector extraction unit 13.
The data acquisition unit 11 is configured to execute step S11: and obtaining the CSI data.
The feature extraction unit 12 is configured to perform step S12: and carrying out feature extraction on the CSI data to obtain a plurality of feature values.
The feature vector extraction unit 13 is configured to perform step S13: dividing a plurality of characteristic values into characteristic values for fall detection and characteristic values for positioning; wherein the feature values for fall detection constitute a fall feature vector, and the feature values for positioning constitute a positioning feature vector.
Specifically, the CSI data in S11 refers to a CSI data packet sent by the wireless network card; the content of a CSI packet is a p×n matrix:
wherein ,indicate->No. H of the link>Subcarrier data; the subcarrier data is a complex number, and comprises amplitude data and phase data; all amplitude data form CSI amplitude data; all phase data constitutes CSI phase data. Referring to fig. 4, fig. 4 is a schematic diagram showing CSI amplitude data of 30 subcarriers of one link according to an embodiment of the present invention.
Each row of the p x N matrix represents different subcarrier data on one link, and different rows represent different links; there are p links in total, each link having N subcarriers. The link refers to a connection between one transmitting antenna and one receiving antenna, in this embodiment, the wireless network card has three receiving antennas, and the Wi-Fi device has one transmitting antenna, so the number of links p=3. The number of subcarriers is determined by the modulation method, and in the present embodiment, the number of subcarriers n=30. The content of one CSI packet in this embodiment is a 3×30 matrix. In the embodiment, the once acquisition duration is set to be 6s, the wireless network card packet sending frequency is set to be 100 packets/s, and 600 CSI data packets are acquired once.
Specifically, the feature extraction unit 12 further includes a time domain feature extraction unit 121 and a frequency domain feature extraction unit 122.
The time domain feature extraction unit 121 is configured to perform step S121: performing time domain feature extraction on the CSI data to obtain a time domain feature value; the time domain characteristic value comprises a time domain amplitude value mean valueEnergy->Amplitude margin interval->Amplitude difference->Margin factor->Signal rate of change->Mean value of phase->Standard deviation of phase- >Phase margin interval->And phase difference->。
The time domain feature extraction unit 121 further includes a time domain amplitude mean value calculation unit 121a, an energy calculation unit 121b, an amplitude range interval calculation unit 121c, an amplitude range calculation unit 121d, a margin factor calculation unit 121e, a signal speed change rate calculation unit 121f, a phase mean value calculation unit 121g, a phase standard deviation calculation unit 121h, a phase range interval calculation unit 121i, and a phase difference calculation unit 121j.
The time domainThe amplitude average value calculation unit 121a is configured to perform step S121a: calculating the time domain amplitude mean of the CSI amplitude dataThe method comprises the steps of carrying out a first treatment on the surface of the The time domain amplitude mean->The calculation formula of (2) is as follows:
, wherein />Is->CSI amplitude data matrix of individual data packets +.>Line->Amplitude data of subcarriers of column, +.>Is the number of data packets. In this embodiment, the CSI amplitude data matrix of one data packet is a 3×30 matrix, and the time domain amplitude average value of each subcarrier amplitude data in the CSI amplitude data matrix of 600 data packets is calculated, so as to obtain a 3×30 time domain amplitude average value matrix.
The energy calculating unit 121b is configured to perform step S121b: calculating the energy of the CSI amplitude data The method comprises the steps of carrying out a first treatment on the surface of the Said energyThe calculation formula of (2) is as follows:
, wherein />Is->CSI amplitude data matrix of individual data packets +.>Line->Subcarrier amplitude data of column,/>Is the number of data packets. In this embodiment, the CSI amplitude data matrix of one data packet is a 3×30 matrix, and energy is calculated for each subcarrier amplitude data in the CSI amplitude data matrix of 600 data packets, so as to obtain a 3×30 energy matrix.
The amplitude range interval calculation unit 121c is configured to perform step S121c: calculating amplitude range interval of CSI amplitude dataThe method comprises the steps of carrying out a first treatment on the surface of the The amplitude margin interval +.>The calculation formula of (2) is as follows:
, in the formula />The number of data packets corresponding to the maximum amplitude data,/-for the maximum amplitude data>The data packet number corresponding to the minimum amplitude data. For example, assume that for the 20 th subcarrier data of the 1 st link, the maximum value of the amplitude occurs in the 500 th data packet and the minimum value of the amplitude occurs in the 100 th data packetThe range interval t=400 and the value of the amplitude range interval matrix row 1 and column 20 is 400. In this embodiment, the CSI amplitude data matrix of one data packet is a 3×30 matrix, and the range interval T is calculated for each subcarrier amplitude data in the CSI amplitude data matrix of 600 data packets, so as to obtain a range interval matrix of 3×30.
The amplitude difference calculating unit 121d is configured to perform step S121d to calculate an amplitude difference of the CSI amplitude dataThe method comprises the steps of carrying out a first treatment on the surface of the The amplitude difference->The difference of subcarrier amplitude data of different links is indicated, and the calculation formula is as follows:
, in the formula />Representing the link number>Is->CSI amplitude data matrix of individual data packets +.>Line->Subcarrier amplitude data of the columns.
Since the CSI data of different links come from receiving antennas at different positions, the amplitude difference between the links can represent the position difference of the receiving antennas, and the extracted amplitude difference can be effectively used for positioning.
In this embodiment, the number of links p=3, and the number of links p=3 are combined two by two to obtain 3 sets of amplitude differences, including the amplitude difference between the 1 st link and the 2 nd link, the amplitude difference between the 1 st link and the 3 rd link, and the amplitude difference between the 2 nd link and the 3 rd link.
The margin factor calculating unit 121e is configured to perform step S121e: calculating margin factors for CSI amplitude dataThe method comprises the steps of carrying out a first treatment on the surface of the Said margin factor->The calculation formula of (2) is as follows:
middle square root amplitude +.>; wherein />Is the firstCSI amplitude data matrix of individual data packets +.>Line->Subcarrier amplitude data of column,/>Is the number of data packets.
The margin factor is the ratio of the peak value to the square root amplitude of the signal, and represents the stability and reliability of the signal in the transmission process. Compared with a static environment, the environment when the human body motion changes has obvious difference between margin factors, and can be used as one of indexes of the human body motion change. In this embodiment, the CSI amplitude data matrix of one data packet is a 3×30 matrix, and the margin factor is calculated for each subcarrier amplitude data in the CSI amplitude data matrix of 600 data packets, so as to obtain a 3×30 margin factor matrix.
The signal speed change rate calculation unit 121f is configured to perform step S121f: calculating the signal speed change rate of the CSI amplitude dataThe method comprises the steps of carrying out a first treatment on the surface of the The rate of change of the signal speed ∈ ->The calculation formula of (2) is as follows:
, in the formula />Is->CSI amplitude data matrix of individual data packets +.>Line->Subcarrier amplitude data of column,/>;/>The number of the data packets; />The sampling period is the interval between sampling time points corresponding to adjacent data packets.
Because the falling signal is discrete and irregular, the method introduces a brand new variable for exactly judging that the waveform mutation exactly corresponds to the change of the actual environment: rate of change of signal speed. The signal speed change rate refers to dividing subcarrier data into a plurality of groups according to time, each time period comprises a plurality of data, calculating the amplitude change rate of the CSI data in each time period, carrying out iterative comparison on the amplitude change rate of each time period to extract the maximum value, wherein the obtained maximum amplitude change rate is the intensity of the most intense signal mutation, the actual physical meaning is the maximum value of the signal fluctuation speed, and the time point corresponding to the maximum value reflects the time point of occurrence of a falling event.
In this embodiment, the number of data packets =600, a sampling frequency of 100 packets/s, i.e. sampling period +.>=0.01 s; the signal rate of change is calculated for each subcarrier amplitude data in the CSI amplitude data matrix of 600 data packets, resulting in 599 3 x 30 signal rate of change matrices.
The phase average value calculation unit 121g is configured to perform step S121g: calculating the phase mean of CSI phase dataThe method comprises the steps of carrying out a first treatment on the surface of the The phase mean->The calculation formula of (2) is as follows:
, wherein />Is->CSI phase data matrix of individual data packets +.>Line->Subcarrier phase data of column,/->Is the number of data packets. In this embodiment, the CSI phase data matrix of one data packet is a 3×30 matrix, and the phase average value of each subcarrier phase data in the CSI phase data matrix of 600 data packets is calculated, so as to obtain a 3×30 phase average value matrix.
The phase standard deviation calculation unit 121h is configured to perform step S121h: calculating the phase standard deviation of the CSI phase dataThe method comprises the steps of carrying out a first treatment on the surface of the The phase standard deviation->The calculation formula of (2) is as follows:
, wherein />Is->CSI phase data matrix of individual data packets +.>Line->Subcarrier phase data of column,/->For the number of data packets, < > for>Is the phase average. In this embodiment, the CSI phase data matrix of one data packet is a 3×30 matrix, and the phase standard deviation of each subcarrier phase data in the CSI phase data matrix of 600 data packets is calculated, so as to obtain a 3×30 phase standard deviation matrix.
The phase margin interval calculation unit 121i is configured to perform step S121i: calculating CSI phasePhase margin interval of bit dataThe method comprises the steps of carrying out a first treatment on the surface of the The phase margin interval->The calculation formula of (2) is as follows:
, wherein />For the maximum number of packets corresponding to the phase data,/-for the phase data>The data packet number corresponding to the minimum phase data. For example, assuming that for the 20 th subcarrier data of the 1 st link, the maximum value of the phase appears in the 200 th data packet and the minimum value of the phase appears in the 500 th data packet, the phase margin interval n=300, and the value of the 20 th column of the phase margin interval matrix 1 is 300. In this embodiment, the CSI phase data matrix of one data packet is a 3×30 matrix, and the phase margin interval T is calculated for each subcarrier phase data in the CSI phase data matrix of 600 data packets, so as to obtain a 3×30 phase margin interval matrix.
The phase difference calculation unit 121j is configured to perform step S121j: calculating the phase difference of the CSI amplitude dataThe method comprises the steps of carrying out a first treatment on the surface of the Said phase difference->Refers to the difference of subcarrier phase data of different links, and the calculation formula is as follows:
, in the formula />Representing the link number>Is->CSI phase data matrix of individual data packets +. >Line->Subcarrier phase data of the columns.
Since the CSI data of different links come from receiving antennas at different positions, the phase difference between the links can reflect the position difference of the receiving antennas, and the extracted phase difference can be effectively used for positioning.
In this embodiment, the number of links p=3, 3 links are combined two by two to obtain 3 sets of phase differences, including a phase difference between the 1 st link and the 2 nd link, a phase difference between the 1 st link and the 3 rd link, and a phase difference between the 2 nd link and the 3 rd link.
The frequency domain feature extraction unit 122 is configured to perform step S122: extracting frequency domain characteristics of the CSI data to obtain frequency domain characteristic values; the frequency domain characteristic value comprises a frequency domain amplitude average valueFrequency domain standard deviation->Maximum value of frequency domain->And frequency domain skewness coefficient->。
The frequency domain feature extraction unit 122 further includes a frequency response calculation unit 112a, a frequency domain amplitude average calculation unit 112b, a frequency domain standard deviation calculation unit 112c, a frequency domain maximum calculation unit 112d, and a frequency domain skewness coefficient calculation unit 112e.
The frequency response calculation unit 112a is configured to perform step S112a: counting the CSI dataIs fast Fourier transformed to give +.>A frequency response matrix; then extract +. >The frequency response matrix representing the low-frequency component forms a low-frequency response matrix; sign->Indicate->A low frequency response matrix +.>Line->Frequency response amplitude data for the columns; k is 1 to->An integer therebetween; />Less than or equal to->。
Since different data packets represent different points in time, for the firstThe p×N CSI data matrix of the individual data packets>Line->Column element->For example, a +>=1 to->Corresponding +.>Form a length of->Is a time domain signal sequence of (a); the time domain signal sequence is counted as +.>Is a fast Fourier transform of (2) to obtain a length of +.>Frequency domain signal sequence>。
For a pair ofElements of each column of each row in p×n CSI data matrix of each packet +.>The length of the composition is->Is carried out with the above-mentioned point number +.>Fast fourier transform, obtaining p×n lengths +.>Frequency domain signal sequence>These frequency signal sequences are then combined according to the corresponding +.>Arranging at the arrangement position of the original p×N CSI data matrix to obtain +.>P x N frequency response matrices. In this embodiment, the number of data packets +.>=600, fast fourier transform points +.>=64, link number p=3, subcarrier number n=30; the filtered CSI data of 600 packets is subjected to a fast fourier transform with a number of points of 64, resulting in 64 frequency response matrices of 3×30.
Since the human motion signal frequency is concentrated at low frequency and the high frequency part is mostly noise irrelevant to useful signals, the human motion signal frequency is also needed to be controlled fromExtracting +.>A frequency response matrix representing the low frequency content. In this embodiment, <' > a->=25; extracting 2 nd to 26 th frequency response matrixes from 64 3×30 frequency response matrixes, and forming a low-frequency response matrix by 25 total frequency response matrixes, so as to perform subsequent frequency characteristic extraction; these 25 low frequency response matrices represent low frequency components of the CSI signal with frequencies between 0.2Hz and 10 Hz; sign->Frequency response amplitude data representing the kth low frequency response matrix, the ith row and the jth column.
The frequency domain amplitude average value calculation unit 112b is configured to perform step S112b: calculating frequency domain amplitude average value according to low-frequency response matrixThe method comprises the steps of carrying out a first treatment on the surface of the The frequency domain amplitude average value +.>The calculation formula of (2) is as follows: />, in the formula Represents the kth low frequency response matrix +.>Line->Frequency response amplitude data of column,/>Is the number of low frequency response matrices.
In the present embodiment, the number of low-frequency response matricesThe frequency domain amplitude average value is calculated by using the frequency domain amplitude response matrix of 25, 25 3×30 low frequencies to obtain a frequency domain amplitude average value matrix of 3×30.
The frequency domain standard deviation calculation unit 112c is configured to perform step S112c: calculating frequency domain standard deviation according to low frequency response matrixThe method comprises the steps of carrying out a first treatment on the surface of the The frequency domain standard deviation->The calculation formula of (2) is as follows:
, in the formula />Represents the kth low frequency response matrix +.>Line->Frequency response amplitude data of column,/>For the number of low frequency response matrices, < +.>Is the average value of the frequency domain amplitude.
The frequency domain standard deviation is the radius of inertia centered on the center of gravity frequency. If the frequency spectrum amplitude near the center of gravity is larger, the frequency standard deviation is smaller; if the frequency spectrum around the center of gravity is small, the standard deviation of the frequency is large. The frequency domain standard deviation describes the degree of dispersion of the power spectral energy distribution.
In the present embodiment, the number of low-frequency response matricesAfter calculating the frequency domain standard deviation for the frequency response matrix of 25, 25 frequency response matrices of 3×30, a frequency domain standard deviation matrix of 3×30 is obtained.
The frequency domain maximum value calculation unit 112d is configured to perform step S112d: calculating the frequency domain maximum from the low frequency response matrixThe method comprises the steps of carrying out a first treatment on the surface of the The calculation formula of the frequency domain maximum value is as follows: />
The method comprises the steps of carrying out a first treatment on the surface of the The first>Line->Column element->Equal to->Line->Column element->At->The maximum value in the low frequency response matrix. In this embodiment, the frequency domain maximum value is calculated by 25 3×30 low frequency response matrices, and then a 3×30 frequency domain maximum value matrix is obtained.
The frequency domain bias factor calculating unit 112e is configured to perform step S112e: calculating frequency domain skewness coefficient according to low-frequency response matrixThe method comprises the steps of carrying out a first treatment on the surface of the The skewness coefficient->The calculation formula of (2) is as follows:
, wherein />Represents the kth low frequency response matrix +.>Line->Frequency response amplitude of column, +.>Maximum in frequency domain>Is the average value of the amplitude values in the frequency domain,is the standard deviation of frequency domain>Is the number of low frequency response matrices.
The skewness factor measures the degree of symmetry of the dataset. The closer the skewness coefficient is to 0, the more symmetrical the data set is; the further the coefficient is from 0, the more asymmetric the data set is explained. If the skewness factor is positive, this indicates that the data is more spread on the right, and if negative, this indicates that the data set is more spread on the left. The difference of the degree of symmetry of the data in the frequency domain can be caused by the difference of the human actions, and can be used as the basis of human action classification.
In this embodiment, 25 3×30 low frequency response matrices calculate the skewness coefficient to obtain a 3×30 skewness coefficient matrix.
The feature vector extraction unit 13 is configured to perform step S13: dividing a plurality of characteristic values into characteristic values for fall detection and characteristic values for positioning; the characteristic values used for falling detection form falling characteristic vectors, and the characteristic values used for positioning form positioning characteristic vectors; note the characteristic vector of falling as Dimension is +.>The method comprises the steps of carrying out a first treatment on the surface of the Recording the locating feature vector as +.>Dimension of。
In addition, in order to improve the effect of feature extraction, the preprocessing module 1 further includes a filtering unit; the filtering unit is configured to perform the steps before the feature extraction unit 12 performs the step S12: filtering and denoising the CSI data to obtain filtered CSI data; the feature extraction unit 12 performs step S12 of: and extracting the characteristics of the filtered CSI data to obtain a plurality of characteristic values.
Specifically, the filtering unit further includes a linear denoising unit, an outlier removing unit, and a butterworth filtering unit.
The linear denoising unit is used for executing the following steps: and carrying out linear denoising on the CSI phase data to obtain a linear denoising CSI phase data matrix.
Record the first of a certain link of the CSI phase data matrixThe phase of the sub-carrier is +.>Then there is, in the formula />For measuring phase +.>Is the true phase; />For the time offset between receiver and transmitter, < >>Z is the noise of the measurement process for unknown phase offset; />Is->Sub-carrier waveSubcarrier number values of (a); the subcarrier numbers corresponding to the 30 subcarriers are-28, -26, -24, -4, -2, -1, 3,5, -25, 27, 28, respectively, in the case of 2.4G, i.e. standard bandwidth of 20 MHz; in the case of 5G, i.e., 40MHz standard bandwidth, the subcarrier numbers corresponding to 30 subcarriers are-58, -54, -50, -46, 46, 50, 54, 58, respectively; / >Points are fast fourier transforms. In the present embodiment, the number of points of the fast fourier transform +.>64.
To eliminate (And->Defining two quantitiesaAnd (3) withb:
/>
Assuming that the center frequencies of the sub-carriers are symmetrical and that the phase difference between adjacent sub-carriers is small, the sum of all sub-carriers in the whole frequency band is considered to be 0, namelyThen->Can be expressed as
From the measured phaseSubtracting the linear error term +.>Remove and eliminate the error from clock synchronization +.>And the random phase offset of the unknown constant beta, further obtaining a final more reliable linear combination form of the real phase, namely
The saidNamely, one subcarrier phase data in the CSI phase data matrix>And (5) linearly denoising the obtained phase data. And carrying out the linear denoising on each subcarrier phase data in the CSI phase data matrix to obtain the linear denoising CSI phase data matrix. Referring to fig. 5 and 6, fig. 5 is a schematic diagram of CSI phase data before linear denoising according to an embodiment of the present invention, and fig. 6 is a schematic diagram of CSI phase data after linear denoising according to an embodiment of the present invention.
The outlier removing unit is configured to perform the steps of: performing outlier removal on the CSI amplitude data and the linear denoising CSI phase data to obtain outlier removed CSI data; the abnormal value removed CSI data comprises abnormal value removed CSI amplitude data and abnormal value removed CSI phase data. The abnormal value is removed, and a Hampel abnormal discrimination method based on decision is adopted to fall into an effective interval Data values other than these are considered abnormal disturbances. Wherein (1)>Indicated is the median of the measured data, +.>The absolute median deviation of the measured data is shown. Hampel anomaly discrimination method slides windows in a data sequence according to a set threshold T and a window size w, traverses data points in each window, and calculates absolute deviation between the data points and the median in the window>If the deviation exceeds a threshold T, the data point is considered an outlier. For the data points identified as outliers, the median of the window data is used +.>Instead of them.
The butterworth filtering unit is used for executing the steps of: performing Butterworth low pass filtering on the CSI data with the outlier removed to obtain filtered CSI data; the filtered CSI data includes CSI amplitude data and CSI phase data. This step is to further perform noise reduction processing on the data. The order and cut-off frequency of the butterworth filter are determined according to the actual filtering effect. Referring to fig. 7, fig. 7 is a schematic diagram of CSI data and filtered CSI data with outlier removal according to an embodiment of the invention.
(2) Fall positioning joint detection part
The inventors contemplate that the fall localization joint detection method should fulfil the following functions: judging whether falling occurs or not according to the CSI data of the Wi-Fi signals, if so, further judging the position of falling, and outputting a falling occurrence number and an indoor area number of falling. Therefore, the fall positioning joint detection method includes a fall detection method and an indoor positioning method.
The indoor positioning method should be realized: and predicting which region in the room the CSI data is collected from according to the characteristic value extracted from the CSI data. If the indoor positioning method is regarded as a black box, the input of the black box is the characteristic value extracted from the CSI data, and the output is the class label corresponding to the indoor area. Thus, the indoor positioning problem is converted into the classification problem.
The technical conception of the method is the same as that of the indoor positioning method, and the method for detecting falling predicts whether the CSI data sample belongs to the falling sample according to the characteristic value extracted by the CSI data; if the CSI data sample belongs to a falling sample, the corresponding class label is 1; if the CSI data sample does not belong to a fall sample, its corresponding class label is 0. If the fall detection method is considered as a black box, the input of the black box is the eigenvalue extracted by the CSI data, and the output is a predictive fall-type tag that indicates whether the sample belongs to a fall sample. Thus, the fall detection problem is converted into a two-classification problem.
The SVM algorithm may accomplish the task of classifying based on the feature values. The technical conception of the SVM algorithm is as follows: input devicePersonal->Feature vector of dimension, preset this- >The individual feature vectors should be divided into +.>A class; these->Feature vector of dimension->A point corresponds to the dimensional feature space; if can find +>Personal->Hyperplane of dimension, will->The dimensional space is divided into->The individual regions are then represented by the calculation of the feature vector +.>The locations in the dimensional feature space lie between which hyperplanes can determine which class the feature vector should be divided into. Therefore, the process of searching the hyperplane according to the known feature vector and the class label thereof is the process of constructing the SVM model. The SVM model substitutes the feature vector into the hyperplane equation, and judges which region in the n-dimensional feature space the feature vector is located in, so that the process of predicting the class label corresponding to the feature vector is the classification process.
The SVM fall detection model 2 is configured to perform step S2: identifying a falling type label corresponding to the falling feature vector; the fall-type tag is 0 or 1.
Specifically, the SVM fall detection model identifies a fall-type tag corresponding to the fall feature vector, and specifically includes the following steps:
the SVM fall detection model comprises a first hyperplane dividing a feature space into two areas; the first hyperplane is expressed by an equation as , wherein />Is->Vector of dimensions>To determine the normal vector of the hyperplane direction +.>Is->Transposed matrix of>Is a displacement term;
will fall feature vectorSubstituting into the equation of the first hyperplane, judging the area of the falling characteristic vector in the characteristic space according to the value calculated by the substituted equation, namely +.>, in the formula />Representing a fall feature vector +.>In the area of (2),>is a sign function;
taking the area of the falling feature vector in the feature space as a falling type label corresponding to the falling feature vector; the fall-type tag is 0 or 1.
Specifically, the SVM fall detection model is constructed by:
s21, collecting a plurality of CSI data samples; these CSI data samples include both fall samples and non-fall samples;
s22, preprocessing and extracting features of the CSI data samples to obtainCharacteristic values for fall detection, which characteristic values constitute +.>A fall feature vector of dimension; taking 1 as a falling characteristic vector of a falling sample and taking 0 as a falling of a non-falling sampleA falling type label corresponding to the feature vector;
s23, randomly dividing the CSI data sample into a training set and a testing set according to a certain proportion;
S24, judging whether the falling feature vectors of the training set are linearly separable; the linear separable means that the falling feature vectors of the training set can be divided into two types by a hyperplane; if yes, taking the falling feature vector of the training set as a mapped falling feature vector; if not, performing first transformation on the falling feature vector of the training set to map the falling feature vector to another high-dimensional space, and converting the falling feature vector into mapped falling feature vectors which can be divided into two types by a hyperplane;
s25, constructing a hyperplane equation according to the mapped falling feature vector and the falling type label thereof, so that the hyperplane equation can divide the first feature space into two areas, and the falling classification accuracy of the mapped falling feature vector is within a certain allowable range;
the fall classification accuracy refers to that a fall feature vector is substituted into the hyperplane equation, the region of the fall feature vector in the first feature space is judged according to a value obtained by substitution calculation, and the obtained predicted fall label is compared with the true fall label in accuracy;
s26, substituting the falling feature vector of the test set into the hyperplane equation constructed in the S25, and judging the area of the falling feature vector of the test set in the feature space according to the value calculated by the substituted equation to obtain a predicted falling type label of the test set;
Comparing the predicted falling type label of the test set with the real falling type label to obtain the falling classification accuracy of the test set of the hyperplane equation constructed at this time;
s27, returning to the step S23, reselecting the training set and the testing set to construct a hyperplane equation, and repeating for a plurality of times;
and comparing the test set falling classification accuracy of the hyperplane equations generated each time, and selecting the hyperplane equation with the highest test set falling classification accuracy as the first hyperplane equation.
The judging module 3 is configured to execute step S3: judging whether falling happens or not according to the falling type label; if the falling type label is 0, judging that falling does not occur; if the falling type label is 1, the falling is judged to occur, and the SVM positioning model 4 is called.
The SVM positioning model 4 is configured to execute step S4: matching the positioning feature vector with a preset area, and taking the matched preset area as a positioning result; the preset area is an area divided by a CSI data sample adopted for constructing an SVM positioning model; the number of the preset areas is recorded as k.
Specifically, the positioning feature vector is matched with a preset area through an SVM positioning model, and the method specifically comprises the following steps:
The SVM positioning model comprises k-1 second hyperplanes which divide the feature space into k areas; the second hyperplane is expressed by an equation as, wherein />Is->Vector of dimensions>In order to determine the normal vector of the hyperplane direction,is->Transposed matrix of>Is a displacement term; />An integer between 1 and k-1, representing an ith equation of the k-1 second hyperplane equations;
will locate feature vectorsSubstituting the locating feature vector into an equation of each second hyperplane respectively, and judging the area of the locating feature vector in the second feature space according to the value obtained by substitution calculation; i.e. < ->, in the formula />Representing a localization feature vector +.>Relative to (I)>The region of the second hyperplane, < >>Is a sign function; comprehensive positioning feature vector->Substituted into each second hyperplane equation>Judging the area of the positioning feature vector in the second feature space;
taking the area of the positioning feature vector in the second feature space as a positioning class label corresponding to the positioning feature vector; the positioning class label is an integer between 1 and k, and indicates what preset area the CSI data is matched with.
Specifically, the SVM positioning model is constructed by the following method:
S41, dividing an indoor space into k preset areas, and collecting a plurality of CSI positioning data samples in each preset area;
s42, preprocessing and extracting features of the CSI positioning data sample to obtainA plurality of feature values for positioning; these characteristic values constitute->A dimensional positioning feature vector; taking a preset area corresponding to the CSI positioning data sample as a positioning class label of the positioning feature vector;
s43, randomly dividing the CSI positioning data sample into a training set and a testing set according to a certain proportion;
s44, judging whether the positioning feature vectors of the training set are linearly separable; the linear separable means that the positioning feature vector of the training set can be divided into k classes by k-1 hyperplanes; if yes, the positioning feature vector of the training set is used as a mapping positioning feature vector; if not, performing second transformation on the positioning feature vector of the training set to map the positioning feature vector to another high-dimensional space, and converting the positioning feature vector into a mapping positioning feature vector which can be divided into k classes by k-1 hyperplanes;
s45, constructing k-1 hyperplane equations according to the mapping positioning feature vector and the positioning class labels thereof, so that the k-1 hyperplanes can divide the second feature space into k areas, and the positioning classification accuracy of the mapping positioning feature vector is within a certain allowable range;
The positioning classification accuracy refers to the accuracy of the obtained predicted positioning class label compared with the true positioning class label by substituting the positioning feature vector into k-1 hyperplane equations and judging the area of the positioning feature vector in the second feature space according to the value obtained by substitution calculation;
s46, substituting the positioning feature vector of the test set into the k-1 hyperplane equations constructed in the S45, and judging the area of the positioning feature vector of the test set in the second feature space according to the substituted and calculated value to obtain a predicted positioning class label of the test set;
comparing the predicted positioning class label of the test set with the real positioning class label to obtain the positioning classification accuracy of the test set of the hyperplane equation constructed at this time;
s47, returning to the step S43, reselecting the training set and the testing set to construct a hyperplane equation, and repeating for a plurality of times;
and comparing the test set positioning classification accuracy of the hyperplane equations generated each time, and selecting the hyperplane equation group with the highest test set positioning classification accuracy as a second hyperplane equation.
Further, the characteristic values for fall detection include an amplitude meanEnergy->Amplitude margin interval->Margin factor- >Signal rate of change->Mean value of phase->Standard deviation of phase->Phase margin interval->Frequency domain amplitude averageFrequency domain standard deviation->Maximum value of frequency domain->And frequency domain skewness coefficient->The method comprises the steps of carrying out a first treatment on the surface of the Dimension of Fall feature vector +.>=12。
In this embodiment, 5 testers collect CSI data corresponding to two actions (falling and non-falling) in 9 areas, and each action collects 30 data samples. The one-time acquisition time length is 6s, and the wireless network card packet sending frequency is 100 packets/s. Thus, for each experimenter, a total of 9×2×30=540 samples were collected; 5 experimenters corresponded to 540×5=2700 samples; there are 600 packets per sample, and a total of 600 x 2700 = 1620000 packets. Each data packet includes a 3 x 30 CSI data matrix.
Further, the characteristic values for positioning include amplitude differencesAnd phase difference->The method comprises the steps of carrying out a first treatment on the surface of the Dimension of positioning feature vector +.>=2。
Further, the method for transforming the feature vector to map to another high-dimensional space and converting the feature vector into the mapped feature vector which can be divided into k classes by k-1 hyperplanes, further comprises the following steps:
mapping the positioning feature vector to a high-dimensional space by utilizing an RBF kernel function, and converting the positioning feature vector into a linear separable mapping positioning feature vector; the RBF kernel function is expressed as a formula , in the formula />The width of the kernel function is controlled for the kernel function parameters. />The larger the width representing the kernel function, the smaller the model, but may also lead to overfitting; />The smaller the width representing the kernel function, the simpler the model, but may also result in a lack of fit.
When the SVM positioning model is constructed, the feature vector is a positioning feature vector, and the mapping feature vector is a mapping positioning feature vector; when the SVM fall detection model is constructed, the feature vector is a fall detection vector, the classification number k=2, and the mapping feature vector is a mapping fall feature vector.
Further, the method for constructing the hyperplane equation according to the mapping feature vector and the class label thereof further comprises the following steps:
the programming problem of constructing the hyperplane equations is converted into a dual optimization problem by the Lagrangian multiplier method, and then the dual problem is solved by a sequential minimum optimization algorithm (SMO). The constraint of constructing the hyperplane equations is to minimize the distance of the hyperplane from the edge feature vectors of the different regions, which is a convex Quadratic Programming (QP) problem with linear constraints, so the lagrangian multiplier method can be used to translate the programming problem of constructing the hyperplane equations into a dual optimization problem.
Because the constraint of constructing the hyperplane equations does not require that every feature vector be strictly classified into the correct class, there is a tolerance for classification accuracy, and thus the constraint is constructed to involve penalty parametersThe method comprises the steps of carrying out a first treatment on the surface of the Penalty parameter->The larger the penalty for misclassification, i.e., the smaller the tolerance, the more prone the model to fit the training data, but may also result in overfitting; penalty parameter->Smaller indicates a higher tolerance to misclassification, the more the model tends to generalize, but may also lead to under-fitting.
Condition adjustment parameters using grid search and />. Given parameter->Is in the range of +.>Step size of change is +.>The method comprises the steps of carrying out a first treatment on the surface of the Parameters (parameters)Is in the range of +.>Step size of change is +.>The method comprises the steps of carrying out a first treatment on the surface of the Let->,/>,/>,For each pair of parameters->Training is carried out, and a pair of parameters with the best effect are taken as optimal parameters.
When the SVM positioning model is constructed, the mapping feature vector is a mapping positioning feature vector; and when the SVM fall detection model is constructed, the mapping feature vector is a mapping fall feature vector.
In the present embodiment, the indoor space is divided into 9 areas as shown in fig. 8, and 5 experimenters collect 60 CSI data in each areaSamples, a total of 2700 samples. Each sample includes 600 packets. And extracting the characteristic value of the CSI data of 600 data packets. Since the latter half of 600 data packets includes a falling signal, and the falling signal does not help positioning, only the first 100 data packets are selected for positioning feature value extraction; the characteristic values for positioning include amplitude differences And phase difference->And forming a two-dimensional positioning feature vector.
In this embodiment, the ratio of the training set to the test set in S23 and S43 is 7:3, i.e. 1890 samples are used to build the model and 810 samples are used to test the classification accuracy of the model. The classification accuracy of the SVM positioning model constructed in the embodiment is shown in fig. 9, and the positioning accuracy of each region is more than 90%. The prediction accuracy of the fall positioning joint detection is shown in fig. 10, and the accuracy of the fall positioning joint detection reaches over 94%.
Based on the same inventive concept, the present application also provides an electronic device, which may be a terminal device such as a server, a desktop computing device, or a mobile computing device (e.g., a laptop computing device, a handheld computing device, a tablet computer, a netbook, etc.). The equipment comprises one or more processors and a memory, wherein the processors are used for executing programs to realize the fall positioning joint detection method of the embodiment of the application; the memory is used for storing a computer program executable by the processor.
Based on the same inventive concept, the present application further provides a computer readable storage medium, corresponding to the foregoing embodiment of the fall positioning joint detection method, having stored thereon a computer program, which when executed by a processor, implements the steps of the fall positioning joint detection method described in any of the foregoing embodiments.
The present application may take the form of a computer program product embodied on one or more storage media (including, but not limited to, magnetic disk storage, CD-ROM, optical storage, etc.) having program code embodied therein. Computer-usable storage media include both permanent and non-permanent, removable and non-removable media, and information storage may be implemented by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of storage media for a computer include, but are not limited to: phase change memory (PRAM), static Random Access Memory (SRAM), dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital Versatile Disks (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium, may be used to store information that may be accessed by the computing device.
The above examples illustrate only a few embodiments of the application, which are described in detail and are not to be construed as limiting the scope of the application. It should be noted that modifications and improvements can be made by those skilled in the art without departing from the spirit of the application, and the application is intended to encompass such modifications and improvements.
Claims (9)
1. A Wi-Fi-based fall positioning joint detection method comprises the following steps:
s1, acquiring CSI data, and carrying out feature extraction on the CSI data to obtain a falling feature vector and a positioning feature vector;
s2, identifying a falling type label corresponding to the falling feature vector through an SVM falling detection model; the falling type label is 0 or 1;
s3, judging whether falling occurs or not according to the falling type label; if the falling type label is 0, judging that falling does not occur; if the falling type label is 1, judging that falling occurs, and executing step S4;
s4, matching the positioning feature vector with a preset area through an SVM positioning model, and taking the matched preset area as a positioning result; the preset area is an area divided by the CSI data sample;
the step S2 specifically includes the following steps:
recording the falling characteristic vector asThe dimension is +.>;
The SVM fall detection model comprises a first hyperplane dividing a first feature space into two areas; the first hyperplane equation is, wherein />Is->Vector of dimensions>To determine the normal vector of the first hyperplane direction, < >>Is thatTransposed matrix of>Is a displacement term;
will fall feature vector Substituting into the first hyperplane equation, and judging the area of the falling characteristic vector in the first characteristic space according to the value obtained by substitution calculation, namely +.>, in the formula />Representing a fall feature vector +.>In the area of (2),>is a sign function;
and taking the area of the falling feature vector in the first feature space as a falling type label corresponding to the falling feature vector.
2. The Wi-Fi based fall positioning joint detection method of claim 1, wherein:
the step S4 specifically includes the following steps:
recording the positioning characteristic vector asThe dimension is +.>The method comprises the steps of carrying out a first treatment on the surface of the The number of the preset areas is recorded as k;
the SVM positioning model comprises k-1 second hyperplanes which divide the second characteristic space into k areas; the second hyperplane equation is, wherein />Is->Vector of dimensions>To determine the normal vector of the second hyperplane direction, < >>Is thatTransposed matrix of>Is a displacement term; />An integer between 1 and k-1, representing an ith equation of the k-1 second hyperplane equations;
will locate feature vectorsSubstituting the locating feature vector into each second hyperplane equation respectively, and judging the area of the locating feature vector in the second feature space according to the value obtained by substitution calculation; i.e. < - >, in the formula />Representing a localization feature vector +.>Relative to (I)>The region of the second hyperplane, < >>Is a sign function; comprehensive positioning feature vectorSubstituted into each second hyperplane equation calculationObtained->Judging the area of the positioning feature vector in the second feature space;
taking the area of the positioning feature vector in the second feature space as a positioning class label corresponding to the positioning feature vector; the positioning class label is an integer between 1 and k, and indicates what preset area the CSI data is matched with.
3. The Wi-Fi based fall positioning joint detection method of claim 2, wherein:
the SVM fall detection model is constructed through the following steps:
s21, collecting a plurality of CSI data samples; these CSI data samples include both fall samples and non-fall samples;
s22, preprocessing and extracting features of the CSI data samples to obtainCharacteristic values for fall detection, which characteristic values constitute +.>A fall feature vector of dimension; taking 1 as a falling type label corresponding to the falling characteristic vector of the falling sample, and taking 0 as a falling type label corresponding to the falling characteristic vector of the non-falling sample;
S23, randomly dividing the CSI data sample into a training set and a testing set according to a certain proportion;
s24, judging whether the falling feature vectors of the training set are linearly separable; the linear separable means that the falling feature vectors of the training set can be divided into two types by a hyperplane; if yes, taking the falling feature vector of the training set as a mapped falling feature vector; if not, performing first transformation on the falling feature vector of the training set to map the falling feature vector to another high-dimensional space, and converting the falling feature vector into mapped falling feature vectors which can be divided into two types by a hyperplane;
s25, constructing a hyperplane equation according to the mapped falling feature vector and the falling type label thereof, so that the hyperplane equation can divide the first feature space into two areas, and the falling classification accuracy of the mapped falling feature vector is within a certain allowable range;
the fall classification accuracy refers to that a fall feature vector is substituted into the hyperplane equation, the region of the fall feature vector in the first feature space is judged according to a value obtained by substitution calculation, and the obtained predicted fall label is compared with the true fall label in accuracy;
s26, substituting the falling feature vector of the test set into the hyperplane equation constructed in the S25, and judging the area of the falling feature vector of the test set in the first feature space according to the value calculated by the substituted equation to obtain a predicted falling type label of the test set;
Comparing the predicted falling type label of the test set with the real falling type label to obtain the falling classification accuracy of the test set of the hyperplane equation constructed at this time;
s27, returning to the step S23, reselecting the training set and the testing set to construct a hyperplane equation, and repeating for a plurality of times;
and comparing the test set falling classification accuracy of the hyperplane equations generated each time, and selecting the hyperplane equation with the highest test set falling classification accuracy as the first hyperplane equation.
4. A Wi-Fi based fall location joint detection method as defined in claim 3, wherein:
the SVM positioning model is constructed by the following method:
s41, dividing an indoor space into k preset areas, and collecting a plurality of CSI positioning data samples in each preset area;
s42, preprocessing and extracting features of the CSI positioning data sample to obtainA plurality of feature values for positioning; these characteristic values constitute->A dimensional positioning feature vector; taking a preset area corresponding to the CSI positioning data sample as a positioning class label of the positioning feature vector;
s43, randomly dividing the CSI positioning data sample into a training set and a testing set according to a certain proportion;
s44, judging whether the positioning feature vectors of the training set are linearly separable; the linear separable means that the positioning feature vector of the training set can be divided into k classes by k-1 hyperplanes; if yes, the positioning feature vector of the training set is used as a mapping positioning feature vector; if not, performing second transformation on the positioning feature vector of the training set to map the positioning feature vector to another high-dimensional space, and converting the positioning feature vector into a mapping positioning feature vector which can be divided into k classes by k-1 hyperplanes;
S45, constructing k-1 hyperplane equations according to the mapping positioning feature vector and the positioning class labels thereof, so that the k-1 hyperplanes can divide the second feature space into k areas, and the positioning classification accuracy of the mapping positioning feature vector is within a certain allowable range;
the positioning classification accuracy refers to the accuracy of the obtained predicted positioning class label compared with the true positioning class label by substituting the positioning feature vector into k-1 hyperplane equations and judging the area of the positioning feature vector in the second feature space according to the value obtained by substitution calculation;
s46, substituting the positioning feature vector of the test set into the k-1 hyperplane equations constructed in the S45, and judging the area of the positioning feature vector of the test set in the second feature space according to the substituted and calculated value to obtain a predicted positioning class label of the test set;
comparing the predicted positioning class label of the test set with the real positioning class label to obtain the positioning classification accuracy of the test set of the hyperplane equation constructed at this time;
s47, returning to the step S43, reselecting the training set and the testing set to construct a hyperplane equation, and repeating for a plurality of times;
and comparing the test set positioning classification accuracy of the hyperplane equations generated each time, and selecting the hyperplane equation group with the highest test set positioning classification accuracy as a second hyperplane equation.
5. The Wi-Fi based fall location joint detection method of claim 4, wherein:
the step S1 specifically comprises the following steps:
s11, obtaining CSI data;
s12, carrying out feature extraction on the CSI data to obtain a plurality of feature values;
s13, dividing a plurality of characteristic values into characteristic values for fall detection and characteristic values for positioning; wherein the feature values for fall detection constitute a fall feature vector, and the feature values for positioning constitute a positioning feature vector.
6. The Wi-Fi based fall location joint detection method of claim 5, wherein:
the step S12 specifically includes the following steps:
s121, performing time domain feature extraction on the CSI data to obtain a time domain feature value; the time domain characteristic value comprises a time domain amplitude value mean valueEnergy->Amplitude margin interval->Amplitude difference->Margin factor->Signal rate of change->Mean value of phase->Standard deviation of phase->Phase margin interval->And phase difference->;
S122, carrying out frequency domain feature extraction on the CSI data to obtain a frequency domain feature value; the frequency domain characteristic value comprises a frequency domain amplitude average valueFrequency domain standard deviation->Maximum value of frequency domain->And frequency domain skewness coefficient- >;
The CSI data are a plurality of CSI data packets; the content of a CSI packet is a p×n matrix:
wherein ,indicate->No. H of the link>Subcarrier data; the subcarrier data is a complex number, and comprises amplitude data and phase data; the amplitude data of all the subcarriers form CSI amplitude data; the phase data of all sub-carriers form CSI phase data;
the step S121 includes the steps of:
s121a, calculating a time domain amplitude mean value of the CSI amplitude dataThe method comprises the steps of carrying out a first treatment on the surface of the The time domain amplitude mean->The calculation formula of (2) is as follows:
, wherein />Is->CSI amplitude data matrix of individual data packets +.>Line->Subcarrier amplitude data of column,/>The number of the data packets;
s121b, calculating the energy of the CSI amplitude dataThe method comprises the steps of carrying out a first treatment on the surface of the Said energy->The calculation formula of (2) is as follows: />, wherein Is->CSI amplitude data matrix of individual data packets +.>Line->Subcarrier amplitude data of column,/>The number of the data packets;
s121c, calculating amplitude range interval of the CSI amplitude dataThe method comprises the steps of carrying out a first treatment on the surface of the The amplitude margin interval +.>The calculation formula of (2) is as follows:
, in the formula />The number of data packets corresponding to the maximum amplitude data,/-for the maximum amplitude data>The data packet number corresponding to the minimum amplitude data;
s121d, calculating the amplitude difference of the CSI amplitude data The method comprises the steps of carrying out a first treatment on the surface of the The amplitude difference->The difference of subcarrier amplitude data of different links is indicated, and the calculation formula is as follows:
, in the formula />Representing the link number>Is->CSI amplitude data matrix of individual data packets +.>Line->Subcarrier amplitude data of the columns;
s121e, calculating margin factor of CSI amplitude dataThe method comprises the steps of carrying out a first treatment on the surface of the Said margin factor->The calculation formula of (2) is as follows:
middle square root amplitude +.>; wherein />Is->CSI amplitude data matrix of individual data packets +.>Line->Subcarrier amplitude data of column,/>The number of the data packets;
s121f: calculating the signal speed change rate of the CSI amplitude dataThe method comprises the steps of carrying out a first treatment on the surface of the The rate of change of the signal speed ∈ ->The calculation formula of (2) is as follows:
, in the formula />Is->CSI amplitude data matrix of individual data packets +.>Line->Subcarrier amplitude data of column,/>;/>The number of the data packets; />The sampling period is the interval of sampling time points corresponding to adjacent data packets;
s121g: calculating the phase mean of CSI phase dataThe method comprises the steps of carrying out a first treatment on the surface of the The phase mean->The calculation formula of (2) is as follows:
, wherein />Is->CSI phase data matrix of individual data packets +.>Line->Subcarrier phase data of column,/->The number of the data packets;
s121h: calculating the phase standard deviation of the CSI phase dataThe method comprises the steps of carrying out a first treatment on the surface of the The phase standard deviation- >The calculation formula of (2) is as follows:
, wherein />Is->CSI phase data matrix of individual data packets +.>Line->Subcarrier phase data of column,/->For the number of data packets, < > for>Is the phase average value;
s121i, calculating the phase margin interval of the CSI phase dataThe method comprises the steps of carrying out a first treatment on the surface of the The phase margin interval->The calculation formula of (2) is as follows:
, wherein />For the maximum number of packets corresponding to the phase data,/-for the phase data>The data packet number corresponding to the minimum phase data;
s121j, calculating the phase difference of the CSI amplitude dataThe method comprises the steps of carrying out a first treatment on the surface of the Said phase difference->Refers to the difference of subcarrier phase data of different links, and the calculation formula is as follows:
, in the formula />Representing the link number>Is->CSI phase data matrix of individual data packets +.>Line->Subcarrier phase data of the columns;
the step S122 includes the steps of:
s112a, counting the CSI dataIs fast Fourier transformed to give +.>A frequency response matrix; then extract outThe frequency response matrix representing the low-frequency component forms a low-frequency response matrix; sign->Indicate->A low frequency response matrix +.>Line->Frequency response amplitude data for the columns; k is 1 to->An integer therebetween; />Less than or equal to->;
S112b, calculating frequency domain amplitude average value according to the low frequency response matrix The method comprises the steps of carrying out a first treatment on the surface of the The frequency domain amplitude average value +.>The calculation formula of (2) is as follows: />, in the formula />Represents the kth low frequency response matrix +.>Line->Frequency response amplitude data of column,/>The number of the low-frequency response matrixes;
s112c, calculating frequency domain standard deviation according to the low-frequency response matrixThe method comprises the steps of carrying out a first treatment on the surface of the The frequency domain standard deviation->The calculation formula of (2) is as follows:
, in the formula />Represents the kth low frequency response matrix +.>Line->Frequency response amplitude data of column,/>For the number of low frequency response matrices, < +.>Is the average value of the frequency domain amplitude;
s112d, calculating the frequency domain maximum according to the low-frequency response matrixThe method comprises the steps of carrying out a first treatment on the surface of the The calculation formula of the frequency domain maximum value is as follows:
the method comprises the steps of carrying out a first treatment on the surface of the The first>Line->Column element->Equal to->Line->Column element->At->Maximum value in the low frequency response matrix;
s112e, calculating frequency domain skewness coefficient according to the low-frequency response matrixThe method comprises the steps of carrying out a first treatment on the surface of the The skewness coefficient->The calculation formula of (2) is as follows:
, wherein />Represents the kth low frequency response matrix +.>Line->Frequency response amplitude of column, +.>Maximum in frequency domain>Is the average value of frequency domain amplitude, < >>Is the standard deviation of frequency domain>Is the number of low frequency response matrices.
7. The Wi-Fi based fall location joint detection method of claim 6, wherein:
the method further comprises the step between the step S11 and the step S12; the method comprises the steps of filtering and denoising the CSI data to obtain filtered CSI data, and specifically comprises the following steps:
performing linear denoising on the CSI phase data to obtain a linear denoising CSI phase data matrix;
performing outlier removal on the CSI amplitude data and the linear denoising CSI phase data to obtain outlier removed CSI data;
performing Butterworth low pass filtering on the CSI data with the outlier removed to obtain filtered CSI data;
step S12 is to perform feature extraction on the filtered CSI data to obtain a plurality of feature values.
8. A fall positioning joint detection device based on Wi-Fi comprises a preprocessing module, an SVM fall detection model, a judging module and an SVM positioning model;
the preprocessing module is used for acquiring the CSI data and extracting the characteristics of the CSI data to obtain a falling characteristic vector and a positioning characteristic vector;
the SVM fall detection model is used for identifying a fall type label corresponding to the fall feature vector; the falling type label is 0 or 1;
the judging module is used for judging whether falling happens according to the falling type label; if the falling type label is 0, judging that falling does not occur; if the falling type label is 1, judging that falling occurs, and calling the SVM positioning model;
The SVM positioning model is used for matching the positioning feature vector with a preset area, and taking the matched preset area as a positioning result; the preset area is an area divided by a CSI data sample adopted for constructing an SVM positioning model;
the SVM falling detection model identifies a falling type label corresponding to the falling feature vector, and specifically comprises the following steps:
the SVM fall detection model comprises a first hyperplane dividing a feature space into two areas; the first hyperplane is expressed by an equation as, wherein />Is->Vector of dimensions>To determine the normal vector of the hyperplane direction +.>Is->Transposed matrix of>Is a displacement term;
will fall feature vectorSubstituting into the equation of the first hyperplane, judging the area of the falling characteristic vector in the characteristic space according to the value calculated by the substituted equation, namely +.>, in the formula />Representing a fall feature vector +.>In the area of (2),>is a sign function;
taking the area of the falling feature vector in the feature space as a falling type label corresponding to the falling feature vector; the fall-type tag is 0 or 1.
9. A fall positioning joint detection system based on Wi-Fi comprises Wi-Fi equipment, a wireless network card and a fall positioning joint detection device;
The Wi-Fi equipment is fixedly arranged at a certain place in the room and transmits Wi-Fi signals;
the wireless network card is fixedly arranged at a certain indoor position, receives Wi-Fi signals, demodulates the Wi-Fi signals, and sends the CSI data packet obtained through demodulation to the falling positioning joint detection device;
the falling positioning joint detection device comprises a preprocessing module, an SVM falling detection model, a judging module and an SVM positioning model;
the preprocessing module is used for acquiring the CSI data and extracting the characteristics of the CSI data to obtain a falling characteristic vector and a positioning characteristic vector;
the SVM fall detection model is used for identifying a fall type label corresponding to the fall feature vector; the falling type label is 0 or 1;
the judging module is used for judging whether falling happens according to the falling type label; if the falling type label is 0, judging that falling does not occur; if the falling type label is 1, judging that falling occurs, and calling the SVM positioning model;
the SVM positioning model is used for matching the positioning feature vector with a preset area, and taking the matched preset area as a positioning result; the preset area is an area divided by a CSI data sample adopted for constructing an SVM positioning model;
The SVM falling detection model identifies a falling type label corresponding to the falling feature vector, and specifically comprises the following steps:
the SVM fall detection model comprises a first hyperplane dividing a feature space into two areas; the first hyperplane is expressed by an equation as, wherein />Is->Vector of dimensions>To determine the normal vector of the hyperplane direction +.>Is->Transposed matrix of>Is a displacement term;
will fall feature vectorSubstituting into the equation of the first hyperplane, judging the area of the falling characteristic vector in the characteristic space according to the value calculated by the substituted equation, namely +.>, in the formula />Representing a fall feature vector +.>In the area of (2),>is a sign function;
taking the area of the falling feature vector in the feature space as a falling type label corresponding to the falling feature vector; the fall-type tag is 0 or 1.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310785669.5A CN116528358B (en) | 2023-06-30 | 2023-06-30 | Wi-Fi-based fall positioning joint detection method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310785669.5A CN116528358B (en) | 2023-06-30 | 2023-06-30 | Wi-Fi-based fall positioning joint detection method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN116528358A CN116528358A (en) | 2023-08-01 |
CN116528358B true CN116528358B (en) | 2023-09-12 |
Family
ID=87399731
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310785669.5A Active CN116528358B (en) | 2023-06-30 | 2023-06-30 | Wi-Fi-based fall positioning joint detection method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116528358B (en) |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104637242A (en) * | 2013-11-12 | 2015-05-20 | 广州华久信息科技有限公司 | Elder falling detection method and system based on multiple classifier integration |
WO2017124816A1 (en) * | 2016-01-20 | 2017-07-27 | 北京大学 | Fall detection method and system |
CN109087482A (en) * | 2018-09-18 | 2018-12-25 | 西安交通大学 | A kind of falling detection device and method |
CN109394229A (en) * | 2018-11-22 | 2019-03-01 | 九牧厨卫股份有限公司 | A kind of fall detection method, apparatus and system |
CN112153736A (en) * | 2020-09-14 | 2020-12-29 | 南京邮电大学 | Personnel action identification and position estimation method based on channel state information |
CN114692683A (en) * | 2022-03-23 | 2022-07-01 | 南京邮电大学 | CSI-based fall detection method and device and storage medium |
CN115035686A (en) * | 2022-04-29 | 2022-09-09 | 华南师范大学 | Real-time falling detection method, system and medium based on channel state information |
CN116028885A (en) * | 2023-02-14 | 2023-04-28 | 华南师范大学 | Fall detection method, system, equipment and medium based on channel state information |
-
2023
- 2023-06-30 CN CN202310785669.5A patent/CN116528358B/en active Active
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104637242A (en) * | 2013-11-12 | 2015-05-20 | 广州华久信息科技有限公司 | Elder falling detection method and system based on multiple classifier integration |
WO2017124816A1 (en) * | 2016-01-20 | 2017-07-27 | 北京大学 | Fall detection method and system |
CN109087482A (en) * | 2018-09-18 | 2018-12-25 | 西安交通大学 | A kind of falling detection device and method |
CN109394229A (en) * | 2018-11-22 | 2019-03-01 | 九牧厨卫股份有限公司 | A kind of fall detection method, apparatus and system |
WO2020103411A1 (en) * | 2018-11-22 | 2020-05-28 | 九牧厨卫股份有限公司 | Fall detection method, device, and system |
CN112153736A (en) * | 2020-09-14 | 2020-12-29 | 南京邮电大学 | Personnel action identification and position estimation method based on channel state information |
CN114692683A (en) * | 2022-03-23 | 2022-07-01 | 南京邮电大学 | CSI-based fall detection method and device and storage medium |
CN115035686A (en) * | 2022-04-29 | 2022-09-09 | 华南师范大学 | Real-time falling detection method, system and medium based on channel state information |
CN116028885A (en) * | 2023-02-14 | 2023-04-28 | 华南师范大学 | Fall detection method, system, equipment and medium based on channel state information |
Non-Patent Citations (1)
Title |
---|
一种基于CSI的跌倒检测方法;冉亚鑫;余江;常俊;李晓薇;;云南大学学报(自然科学版)(第02期);全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN116528358A (en) | 2023-08-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Wong et al. | Clustering learned CNN features from raw I/Q data for emitter identification | |
JP5793961B2 (en) | Electromagnetic wave identification device, electromagnetic wave identification method and program | |
CN111475680A (en) | Method, device, equipment and storage medium for detecting abnormal high-density subgraph | |
US10176372B2 (en) | Classification of microwave scattering data | |
CN112633412B (en) | Abnormal electricity utilization detection method, abnormal electricity utilization detection equipment and storage medium | |
US20120116722A1 (en) | Detection of Defects in an Electrochemical Device | |
CN105595990A (en) | Intelligent terminal device for evaluating and distinguishing quality of electrocardiosignal | |
CN109299698B (en) | Wireless channel scene recognition method based on support vector machine | |
CN116684878B (en) | 5G information transmission data safety monitoring system | |
Feng et al. | Evaluation and improvement of activity detection systems with recurrent neural network | |
CN117074965B (en) | Lithium ion battery remaining life prediction method and system | |
Lara-Cueva et al. | On the use of multi-class support vector machines for classification of seismic signals at Cotopaxi volcano | |
Bhattacharyya et al. | Identification of epileptic seizures from scalp EEG signals based on TQWT | |
Tang et al. | Specific emitter identification for IoT devices based on deep residual shrinkage networks | |
CN114465678A (en) | Complex activity WIFI perception method based on deep learning | |
CN109918994B (en) | Commercial Wi-Fi-based violent behavior detection method | |
CN116528358B (en) | Wi-Fi-based fall positioning joint detection method | |
US8918347B2 (en) | Methods and systems for computer-based selection of identifying input for class differentiation | |
Tian et al. | A single accelerometer-based robust human activity recognition via wavelet features and ensemble feature selection | |
CN114692683B (en) | Fall detection method and device based on CSI and storage medium | |
CN116028885A (en) | Fall detection method, system, equipment and medium based on channel state information | |
CN113098640B (en) | Frequency spectrum anomaly detection method based on channel occupancy prediction | |
CN115510949A (en) | Indoor passive human behavior recognition method and device | |
CN108303622A (en) | Delay time estimation method and device | |
KR101418332B1 (en) | Method For Analysis of Epileptic EEG and Electronic Device supporting the same |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |