CN107884744A - Passive passive type indoor orientation method and device - Google Patents
Passive passive type indoor orientation method and device Download PDFInfo
- Publication number
- CN107884744A CN107884744A CN201710951127.5A CN201710951127A CN107884744A CN 107884744 A CN107884744 A CN 107884744A CN 201710951127 A CN201710951127 A CN 201710951127A CN 107884744 A CN107884744 A CN 107884744A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msub
- sample
- coefficient
- sub
- 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.)
- Granted
Links
Classifications
-
- 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
- G01S5/0252—Radio frequency fingerprinting
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Position Fixing By Use Of Radio Waves (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
A kind of passive passive type indoor orientation method, including：Using the radio frequency network link RSS values that target has gathered at non-spacious indoor each coordinate as training sample, sample label is used as using coordinate numbering；Two-dimentional two-phase is carried out to training sample and closes distributed wavelet filtering processing, determines filtered training sample；The Adaboost.M2 integrated study models based on Geordie decision tree are established, are trained using filtered training sample and sample label, it is determined that the model after training；RSS value of the target in monitored area during any movement is gathered, as test sample, test sample is subjected to two-dimentional two-phase and closes distributed wavelet filtering processing, then is inputted in the model to the training, determines positioning result.RSS sample noises and random disturbances can be differentiated and filtered out to this method, retain normal saltus step data, while position fixing process has very strong generalization ability, can generally improve locating accuracy and stability.Present invention also offers a kind of corresponding device.
Description
Technical field
The present invention relates to radio sensing network indoor occupant technical field of target location, more particularly to it is a kind of non-spacious
Compared with the passive passive type indoor objects field of locating technology based on radio frequency receiving signal intensity under complex indoor environment.
Background technology
Personnel targets indoor orientation method based on less radio-frequency sensing network, it is not necessary to be positioned target cooperate with one's own initiative and
Electronic tag is carried, its principle is mainly：Human body target can produce when being in monitoring area to associated radio frequency link in network
Block or disturb, generation is blocked, namely generates shadow effect, so as to influence network sensor node received signal strength value
(RSS), then target position can be judged by gathering and analyzing RSSI samples situation., should due to the characteristic of radiofrequency signal
Method has, the advantages such as networking facilitate cost cheap insensitive to environment temperature, humidity, light and non-metal barriers, therefore
It is with a wide range of applications in fields such as intelligent building monitoring, military security monitor, old monitoring, patient monitorings.
The localization method mainly has two class technic relization schemes at present：The first kind is the technical scheme based on machine learning.
Such as realized using artificial neural network, fingerprint matching, SVMs scheduling algorithm, it is necessary to artificially be demarcated in monitored area
Coordinate, collection sensing network RSS data are carried out as training sample training machine learning model, on-line stage by test data
Positioning.Second class is to influence the targeting scheme of model on radiofrequency signal link based on human body.The program does not need off-line learning, main
If influence model of the preferable human body to respective link received signal strength is found, such as model of ellipse, saddle model etc..It is fixed
Position process is to infer relevant parameter by influenceing model and RSS values, then coordinates particle filter, Bayesian probability, radio frequency chromatography
The technologies such as imaging realize the positioning to target.But under non-spacious environment, such as in general room environment, its regional structure becomes
Change various, article of furniture material is changeable and shape differs, and thus to various construction caused by different link radio frequency signals or non-builds
If the influences such as property decline, reflection, scattering have a huge difference, and current localization method will exist in this case it is following
Problem：
(1) RSS samples filter effect is bad.When monitored area is relative complex, RSS data can by various noises and with
The influence of machine interference, and intensity is different, some can even make sample occur to distort strongly in itself.And the translational speed of measured target
It is and non-constant so that some signals and noise time-frequency characteristic mix, it is difficult to differentiate.Traditional filtering method effect is unstable, filter
Ripple condition can strictly influence data self character, or even pollution valid data, condition loosely can then leave noise and distortion point, this
It can all influence locating accuracy a bit.
(2) machine learning positioning method accuracy rate and generalization ability have much room for improvement.When environment is complicated, fixed link influences
Model is difficult correct expression RSS sample changeds and the relation of target location change, and machine learning mode is to select well.It is but multiple
Under heterocycle border, link change is complicated with target location relation, and each bar Link Significance is difficult to determine, node weights distribution is not solid
Fixed, the training sample amount gathered is very limited, and specimen types are not comprehensive.And because the single machine learning model after training is extensive
It is indifferent, bad adaptability, when the position between target is in some training sample coordinates be target at nonstandard position fixing,
It is likely to navigate to from actual position with respect to the coordinate of coordinate farther out around it up, so as to influence locating effect.Such as target
During position between coordinate 3 and 4, generalization ability and the not strong model of adaptability are likely to be located to coordinate farther out
10 or 13, it is actual it is expected it is then that the positioning result that model export is coordinate 3 or 4.
The content of the invention
(1) technical problems to be solved
It is an object of the invention to provide a kind of passive passive type indoor orientation method and device, to solve existing be based on
The passive passive type indoor positioning technologies of RSS are compared with complex environment, i.e., non-spacious environment, such as under general room environment, to RSS
Sample filter effect is bad, and machine learning method locating accuracy is not high, at least one problem of generalization ability not persistent erection.
(2) technical scheme
The invention provides a kind of passive passive type indoor orientation method, including：
Step A, using non-spacious indoor environment as monitored area, by target in monitored area at each coordinate when,
All radio frequency network link RSS values of collection are used as sample label as training sample, and using coordinate numbering；
Step B, two-dimentional two-phase is carried out to training sample and closes distributed wavelet filtering processing, determines filtered training sample
This；
Step C, the Adaboost.M2 integrated study models based on Geordie decision tree are established, utilize filtered training sample
This and sample label are trained, it is determined that the integrated study model after training；And
Step D, all radio frequency network link RSS values when collection target is any mobile in the monitored area, as
Test sample, test sample is subjected to two-dimentional two-phase and closes distributed wavelet filtering processing, determines filtered test sample, and will
It is input in the integrated study model after training, determines positioning result.
In some embodiments of the invention, step B includes sub-step：
Sub-step B1, training sample is formed into sample matrix, each row represent a training sample, and columns is sample
Number, the RSS values with dimension in different samples are represented per a line, the sample matrix is subjected to each dimension data by dimension
Wavelet decomposition, wavelet function choice db1 small echos, obtain low-frequency wavelet coefficients ca and h layer high-frequency wavelet coefficient cdh, wherein,
One dimension refers to a radio frequency link, and 1≤h≤s, s are the number of plies of wavelet decomposition, s >=2；
Sub-step B2, retain low-frequency wavelet coefficients ca, ask two-dimentional double using the high-frequency wavelet coefficient of h layers and h+1 layers
Related longitudinal coefficient correlation corrh, determine high-frequency wavelet coefficient energy PcdhWith longitudinal coefficient correlation energy Pcorrh, it is determined that returning
One changes longitudinal coefficient correlation corrnh；
Sub-step B3, compare the normalization longitudinal direction coefficient correlation of high-frequency wavelet coefficient cdh and respective layer, will be greater than normalizing
The high-frequency wavelet coefficient of longitudinal coefficient correlation is arranged to zero, is retained less than being equal to the high frequency wavelet system for normalizing longitudinal coefficient correlation
Number, last layer of high-frequency wavelet coefficient all retains, final to determine reserved high-frequency wavelet coefficient cdi；
Sub-step B4, reserved high-frequency wavelet coefficient is divided into U sections, U >=2, d data of every section of selection, d is positive integer, is pressed
Layer is utilized respectively the reserved high-frequency wavelet coefficient of+1 section of jth section and jth to calculate the laterally non-time shift phase relation of the two-dimentional two-phase Central Shanxi Plain
Number R1ij, wherein, 1≤j≤U-1 ,+1 section of reserved high-frequency wavelet coefficient of jth is subjected to time shift, wherein, time shift amount be a, a for less than
Positive integer equal to d/2, calculate horizontal time shift coefficient R 2 respectively by layerij, determine that time shift coefficient correlation is related to non-time shift
Coefficient differentials Rmij, and select each layer of R1ijThe minimum R1 of middle absolute valueijCorresponding data segment cdriWith each layer of RmijIn absolutely
To the Rm that value is minimumijCorresponding data segment cdrmi, it is merged into filter threshold parameter Estimation data cdmi；
Sub-step B5, according to cdmiFilter threshold parameter σ is determined, then determines filter threshold thr, to reserved high-frequency wavelet systems
Number carries out Distributed filtering, obtains filtered high-frequency wavelet coefficient cdfi；And
Sub-step B6, wavelet reconstruction is carried out using low-frequency wavelet coefficients ca and filtered high-frequency wavelet coefficient cdfi, is obtained
Obtain filtered training sample.
In some embodiments of the invention, determine to normalize longitudinal coefficient correlation according to below equation in sub-step B2：
corrh=cdh·cdh+1
Wherein, n is the length of high-frequency wavelet coefficient cdh sequences, and 1≤g≤n, g are number in high-frequency wavelet coefficient cdh sequences
According to sequence number.
In some embodiments of the invention, in sub-step B4, R1 is determined according to below equationij、R2ij、Rmij, and select
cdri、cdrmi、cdmi：
Rmij=| R1ij-R2ij|
cdrmi=cdijIf
cdri=cdijIf
cdmi={ cdrmi, cdri}
Wherein, cdij+1It is cdijNext adjacent non-time shift segment, cdij+1It is cdij+1Move to right that a time shift amount formed when
Move segment, Cov cdij+1With cdijBetween covariance, Var cdij+1With cdijBetween variance.
In some embodiments of the invention, filter threshold parameter σ and thresholding are determined according to below equation in sub-step B5
Thr, and carry out distributed wavelet filtering and obtain filtered high-frequency wavelet coefficient cdfi：
Wherein, 2≤q≤s, L are each layer of reserved high-frequency wavelet coefficient cdiLength, L=U × d, median is in taking
Between be worth, y is the sequence number of each data in each layer of reserved high-frequency wavelet coefficient, 1≤y≤L.
In some embodiments of the invention, step C includes sub-step：Sub-step C1, determines each in training sample
The Gini coefficients of attribute, attribute A corresponding to Gini coefficient maximums is selected to carry out the growth of decision tree as best attributes, its
In, Gini coefficient Gini (A) formula is：
Wherein, V is the branch amount of the decision tree, and N is the number of coordinate, S 'bcThe sample separated when being using A as Split Attribute
Subset S 'cIn belong to the number of samples of b classes, E is the number of population sample at split point, 1≤b≤N；And sub-step C2, base
Decision tree in sub-step C1, it is determined that the integrated study model after training：
Iteration updates sample weights：
Determine the error rate of decision tree：
Determine the weight of decision tree：
Wherein, W0(f) be f-th of sample initial weight, N is the number of coordinate, and M is the training included under each coordinate
The number of sample, Wk(f, z) is the possibility sum that sample f is divided into all error category z in kth wheel iteration, and k is to change
Generation number, hk(xf, zf) it is that kth decision tree is its correct classification z to f-th of sample classification resultfPossibility, hk(xf, z)
Kth decision tree is except its correct classification z to f-th of sample classification resultfThe possibility of every other incorrect classification z in addition
Property, εkIt is the error rate of kth decision tree, akIt is the weight of kth decision tree, Sum is summation.
In some embodiments of the invention, sub-step B1 also includes step：Monitored area is not had what is gathered during target
All radio frequency network link RSS values carry out difference processing as reference sample, by the training sample and reference sample, it is determined that
Difference signal, as the training sample.
Based on same inventive concept, the present invention also provides a kind of passive passive type indoor positioning device, including：
Memory, for store instruction；And
Processor, for according to the instruction, performing foregoing passive passive type indoor orientation method.
(3) beneficial effect
Passive passive type indoor orientation method and device provided by the invention, compared to prior art, at least with following
Advantage：
1st, RSS data sample is subjected to vertical and horizontal two dimension two-phase and closes distributed wavelet filtering processing, targetedly
RSS sample noises are differentiated and filtered out, retains be mingled in high-frequency noise due to normal number of transitions caused by target movement as far as possible
According to, while there is good filtration result to data exception trip point, high-frequency noise, and have two-dimentional two-phase and close adaptive-filtering
Threshold value.
2nd, the Adaboost.M2 integrated study models of the invention based on Geordie decision tree, solves individual machine study mould
For type because generalization ability is not strong, target caused by bad adaptability punishes the problem of class is ineffective, Neng Gou in nonstandard position fixing
Link change is complicated with target location relation, and in the case that each bar Link Significance is difficult determination, sample characteristics and attribute are entered
The rational weight distribution of row and selection, generalization ability and correct localization are improved so as to train multigroup learning model.
Brief description of the drawings
Fig. 1 is the step schematic diagram of the passive passive type indoor orientation method of the embodiment of the present invention.
Fig. 2A is the monitoring of environmental overall schematic after the coordinate division of embodiments of the invention 1.
Fig. 2 B are the monitoring of environmental overall schematic after the coordinate division of embodiments of the invention 2.
Fig. 3 A are the positioning of the Adaboost.M2 integrated study models without filtering and based on Geordie decision tree of embodiment 1
Design sketch.
Fig. 3 B are embodiment 1 without filtering and based on the locating effect figure of deep neural network DNN models.
Fig. 3 C are the locating effect figure without filtering and based on fingerprint recognition model of embodiment 1.
Fig. 3 D are the locating effect figure without filtering and based on SVM models of embodiment 1.
Fig. 3 E are the positioning of the Adaboost.M2 integrated study models without filtering and based on Geordie decision tree of embodiment 2
Design sketch.
Fig. 3 F are embodiment 2 without filtering and based on the locating effect figure of deep neural network DNN models.
Fig. 3 G are the locating effect figure without filtering and based on fingerprint recognition model of embodiment 2.
Fig. 3 H are the locating effect figure without filtering and based on SVM models of embodiment 2.
Fig. 4 A are that the two-dimentional two-phase of embodiment 1 closes distributed wavelet filtering and based on the Adaboost.M2 of Geordie decision tree
The locating effect figure of integrated study model.
Fig. 4 B are the simple correlation threshold value wavelet filtering of embodiment 1 and integrated of Adaboost.M2 based on Geordie decision tree
Practise the locating effect figure of model.
Fig. 4 C are the correlation Wavelet Entropy filtering of embodiment 1 and the Adaboost.M2 integrated studies based on Geordie decision tree
The locating effect figure of model.
Fig. 4 D are complete zero wavelet filtering of high frequency coefficient of embodiment 1 and the Adaboost.M2 based on Geordie decision tree is integrated
The locating effect figure of learning model.
Fig. 4 E are that the two-dimentional two-phase of embodiment 2 closes distributed wavelet filtering and based on the Adaboost.M2 of Geordie decision tree
The locating effect figure of integrated study model.
Fig. 4 F are the simple correlation threshold value wavelet filtering of embodiment 2 and integrated of Adaboost.M2 based on Geordie decision tree
Practise the locating effect figure of model.
Fig. 4 G are the correlation Wavelet Entropy filtering of embodiment 2 and the Adaboost.M2 integrated studies based on Geordie decision tree
The locating effect figure of model.
Fig. 4 H are complete zero wavelet filtering of high frequency coefficient of embodiment 2 and the Adaboost.M2 based on Geordie decision tree is integrated
The locating effect figure of learning model.
Fig. 5 A are that the two-dimentional two-phase of embodiment 1 closes distributed wavelet filtering and based on the Adaboost.M2 of Geordie decision tree
The locating effect figure of integrated study model.
Fig. 5 B are that the two-dimentional two-phase of embodiment 1 is closed distributed wavelet filtering and determined based on deep neural network DNN models
Position design sketch.
Fig. 5 C are that the two-dimentional two-phase of embodiment 1 closes distributed wavelet filtering and based on the locating effect of fingerprint recognition model
Figure.
Fig. 5 D are that the two-dimentional two-phase of embodiment 1 closes distributed wavelet filtering and the locating effect figure based on SVM models.
Fig. 5 E are that the two-dimentional two-phase of embodiment 2 closes distributed wavelet filtering and based on the Adaboost.M2 of Geordie decision tree
The locating effect figure of integrated study model.
Fig. 5 F are that the two-dimentional two-phase of embodiment 2 is closed distributed wavelet filtering and determined based on deep neural network DNN models
Position design sketch.
Fig. 5 G are that the two-dimentional two-phase of embodiment 2 closes distributed wavelet filtering and based on the locating effect of fingerprint recognition model
Figure.
Fig. 5 H are that the two-dimentional two-phase of embodiment 2 closes distributed wavelet filtering and the locating effect figure based on SVM models.
Fig. 6 is the structural representation of the passive passive type indoor positioning device of the embodiment of the present invention.
Embodiment
In order to solve existing localization method, RSS sample filter effects to be present bad, and machine learning positioning method is accurate
The defects of true rate and generalization ability have much room for improvement, it is main to be applicable the invention provides a kind of passive passive type indoor orientation method
In more complicated interior, i.e., non-spacious environment, such as under general room environment, using two-dimentional double correlated wavelets filtering methods to adopting
The RSS sample datas of collection are filtered, using filtered data to the Adaboost.M2 integrated studies based on Geordie decision tree
Model is trained, with the model after being trained.Finally, RSS value of the target in monitored area during any movement is gathered, is made
For test sample, test sample is subjected to two-dimentional two-phase and closes distributed wavelet filtering processing, then the model inputted to the training
In, determine positioning result.In general, in passive passive type localization method, target need not carry electronic tag and signal
R-T unit, it becomes possible to be localized indirectly.
For the object, technical solutions and advantages of the present invention are more clearly understood, below in conjunction with specific embodiment, and reference
Accompanying drawing, the present invention is described in more detail.
The embodiments of the invention provide a kind of passive passive type indoor orientation method, as shown in figure 1, this method includes step
Suddenly：
Step A, using non-spacious general room environment as monitored area, by target in monitored area at each coordinate
When, all radio frequency network link RSS values gathered are used as sample label as training sample, and using coordinate numbering.Wherein, institute
Stating the determination of coordinate includes step：The ground of clear area in the non-spacious indoor environment, choose any position conduct
Origin coordinates, along the vertical and horizontal of the origin coordinates, one coordinate is set every a preset distance from the origin coordinates, its
In, the coordinate includes the origin coordinates.In general, can be by setting multiple radio frequencies to pass around the monitored area
Sensor, obtains the training sample, the height of the radio frequency sensor is generally 1.5m~2m (rice), preferably radio frequency sensor
It is highly identical.
If the number of coordinate is N, N=18 in embodiment 1, N=12 in embodiment 2.M training sample of collection under each coordinate
This, M=480 in two embodiments, the training sample gathered at I coordinates is RIJ={ rssIJ1, rssIJ2..., rssIJH}(I
=1,2...N, J=1,2...M), H is link number, in the present embodiment H=48, then I points training sample set is combined into RI=
{RI1, R12... RIM, RICorresponding label collection is combined into YI={ I1, I2... IM, because being all at same coordinate I points
Collection, so I1=I2=...=IM=I.Then training sample is totally represented by R={ R1, R2 ... RN, it is corresponding
Label Y={ Y1, Y2... YN}。
Then, in conjunction with the embodiments 1 and determination of the embodiment 2 to coordinate be illustrated, Fig. 2 Fig. 2A be the present invention reality
The monitoring of environmental overall schematic after the coordinate division of example 1 is applied, Fig. 2 B are the monitoring after the coordinate division of embodiments of the invention 2
Environment overall schematic, as shown in Fig. 2A to Fig. 2 B, the monitored area of the two embodiments is all the general room for having barrier
Environment, the furniture size of two embodiments, put and the equal difference such as material.
The number of coordinates of embodiment 1 is 18, and numbering is 1~18 respectively.It should be noted that in order to reduce interference fringe
Influence, monitored area can be there is no to all radio frequency network link RSS values gathered during target as reference sample, by the instruction
Practice sample and carry out difference processing with reference sample, difference signal is determined, as the training sample.Therefore in the present embodiment
In, situation during without target is also needed plus monitored area, altogether 19 groups of training data samples.Similarly, the number of coordinates of embodiment 2
For 12, numbering is 1~12 respectively, and situation during plus monitored area without target, one shares 13 groups of training samples.Collection training
During sample, target is stood at each coordinate respectively, and each coordinate points gather 480 samples, and each sample is (mutual by 48 dimensions
The number of link between the sensor of communication) composition.
Step B, two-dimentional two-phase is carried out to training sample and closes distributed wavelet filtering processing, determines filtered training sample
This.
Wherein, step B mainly includes following sub-step：
Sub-step B1, training sample is formed into sample matrix, each row represent a training sample, and columns is sample
Number, the RSS values with dimension in different samples are represented per a line, the sample matrix is subjected to each dimension data by dimension
Wavelet decomposition, wavelet function choice db1 small echos, obtain low-frequency wavelet coefficients ca and h layer high-frequency wavelet coefficient cdh, wherein,
One dimension refers to a radio frequency link, and 1≤h≤s, s are the number of plies of wavelet decomposition, s >=2；Wherein, the number of plies of wavelet decomposition can be with
According to data length optimum selecting of the training sample matrix per dimension.
Sub-step B2, retain low-frequency wavelet coefficients ca, ask two-dimentional double using the high-frequency wavelet coefficient of h layers and h+1 layers
Related longitudinal coefficient correlation corrh, determine high-frequency wavelet coefficient energy PcdhWith longitudinal coefficient correlation energy Pcorrh, it is determined that returning
One changes longitudinal coefficient correlation corrnh；
corrh=cdh·cdh+1
Wherein, n is high-frequency wavelet coefficient cdhThe length of sequence, 1≤g≤n, g are high-frequency wavelet coefficient cdhData in sequence
Sequence number.
Sub-step B3, compare high-frequency wavelet coefficient cdhWith the normalization longitudinal direction coefficient correlation of respective layer, will be greater than normalizing
The high-frequency wavelet coefficient of longitudinal coefficient correlation is arranged to zero, is retained less than being equal to the high frequency wavelet system for normalizing longitudinal coefficient correlation
Number, last layer of high-frequency wavelet coefficient all retains, final to determine reserved high-frequency wavelet coefficient cdi；
Sub-step B4, reserved high-frequency wavelet coefficient is divided into U sections, U >=2, d data of every section of selection, d is positive integer, excellent
Selection of land, 10≤d≤20.The reserved high-frequency wavelet coefficient for being utilized respectively+1 section of jth section and jth by layer again closes to calculate two-dimentional two-phase
The middle non-time shift coefficient R 1 of transverse directionij, wherein, 1≤j≤U-1 ,+1 section of reserved high-frequency wavelet coefficient of jth is subjected to time shift, its
In, time shift amount is a, and a is the positive integer less than or equal to d/2, calculates horizontal time shift coefficient R 2 respectively by layerij, determine time shift
Coefficient correlation and non-time shift coefficient correlation difference Rmij, and select each layer of R1ijThe minimum R1 of middle absolute valueijCorresponding data segment
cdriWith each layer of RmijThe minimum Rm of middle absolute valueijCorresponding data segment cdrmi, it is merged into filter threshold parameter Estimation data
cdmi.R1 is determined according to below equationij、R2ij、Rmij, and selected cdri、cdrmi、cdmi：
Rmij=| R1ij-R2ij|
cdrmi=cdijIf
cdri=cdijIf
cdmi={ cdrmi, cdri}
Wherein, cdij+1It is cdijNext adjacent non-time shift segment, cdij+1It is cdij+1Move to right that a time shift amount formed when
Move segment, Cov cdij+1With cdijBetween covariance, Var cdij+1With cdijBetween variance.
Sub-step B5, according to cdmiFilter threshold parameter σ is determined, then determines filter threshold thr, to reserved high-frequency wavelet systems
Number carries out Distributed filtering, obtains filtered high-frequency wavelet coefficient cdfi。
Filter threshold parameter σ and thresholding thr are determined according to below equation in sub-step B5, and carries out distributed wavelet filtering
Obtain filtered high-frequency wavelet coefficient cdfi：
Wherein, 2≤q≤s, L are each layer of reserved high-frequency wavelet coefficient cdiLength, L=U × d, median is in taking
Between be worth, y is the sequence number of each data in each layer of reserved high-frequency wavelet coefficient, 1≤y≤L.
Sub-step B6, wavelet reconstruction is carried out using low-frequency wavelet coefficients ca and filtered high-frequency wavelet coefficient cdfi, is obtained
Obtain filtered training sample.
Step C, the Adaboost.M2 integrated study models based on Geordie decision tree are established, utilize filtered training sample
This and sample label are trained, it is determined that the integrated study model after training.
Including following sub-step：
Sub-step C1, the Gini coefficients of each attribute in training sample are determined, selected corresponding to Gini coefficient maximums
Attribute A carries out the growth of decision tree as best attributes, wherein, Gini coefficient Gini (A) formula is：
Wherein, V is the branch amount of the decision tree, and N is the number of coordinate, S 'bcThe sample separated when being using A as Split Attribute
Subset S 'cIn belong to the number of samples of b classes, E is the number of population sample at split point, 1≤b≤N；And
Sub-step C2, based on the decision tree in sub-step C1, it is determined that the integrated study model after training：
Iteration updates sample weights：
Determine the error rate of decision tree：
Determine the weight of decision tree：
Wherein, W0(f) be f-th of sample initial weight, N is the number of coordinate, and M is the training included under each coordinate
The number of sample, Wk(f, z) is the possibility sum that sample f is divided into all error category z in kth wheel iteration, and k is to change
Generation number, hk(xf, zf) it is that kth decision tree is its correct classification z to f-th of sample classification resultfPossibility, hk(xf, z)
Kth decision tree is except its correct classification z to f-th of sample classification resultfThe possibility of every other incorrect classification z in addition
Property, εkIt is the error rate of kth decision tree, akIt is the weight of kth decision tree, Sum is summation.In addition, in general, which
Individual decision tree also represents has carried out for which time iteration.
Finally progress step D, all radio frequency network link RSS values when collection target is any mobile in the monitored area,
As test sample, test sample is subjected to two-dimentional two-phase and closes distributed wavelet filtering processing, determines filtered test
Sample, and be entered into the integrated study model after training, determine positioning result.
Next, test checking is carried out to method provided by the invention：
In verification process, test sample is personnel targets from monitored area any point continuous moving to any point
At the end of gathered, each test sample can be made up of 48 dimensions (can select different dimensions in other embodiments
Number of degrees mesh).Because the change of general room environmental structure is various, article of furniture material composition is changeable and not of uniform size, thus to penetrating
Phenomena such as various construction caused by frequency signal or non-constructive decline, reflection, scattering, varies, and is made an uproar in environment there is also various
Sound and random disturbances, therefore the data sample gathered contains various noises and data mutation.
In target moving process, because target translational speed may change at any time, so the sample size gathered everywhere is not to the utmost
It is identical, and target is except the region that be able to will also pass through by having demarcated the place of coordinate between some coordinate points, and target
The sample that region between these demarcation coordinates is gathered is exactly to verify location model generalization ability, localization method accuracy rate
Key point.For example, when target is in the position between coordinate 3 and 4, affected by various factors, other localization methods may
Be located to farther out on coordinate (such as coordinate 10 or 13) so that positioning result has relatively large deviation, and is actually contemplated to be
The positioning result that model can export is coordinate 3 or 4.
In embodiment 1 and embodiment 2, under general room environment, using the present invention based on Geordie decision tree
The localization method of Adaboost.M2 integrated study models, different assignment tests is carried out, and it is related to employing the field fixed
3 kinds of different localization methods of Fingerprint Model, SVM models, deep neural network DNN models in bit model compare.
Compared for respectively, without filtering process when locating effect；Locating effect under different wavelet filtering modes；And employ this
When inventing described two dimension pair correlated wavelets filtering method, the locating effect of different location models.Wherein, deep neural network mould
Type have selected 5 layers of connection DNN structures, and used regularization and dropout layers entirely, can avoid part as far as possible by debugging
Minimum and over-fitting problem.
In two embodiments, compared to embodiment 2, the ambient noise of embodiment 1 and data exception saltus step are less, data matter
Amount is higher, therefore embodiment 2 clearer can embody under the non-spacious general room environment of the present invention and determine in passive passive type room
The filtering of position method and generalization ability advantage.The coordinate precision of two embodiments is all 0.5 meter, therefore experimental result will be examined and divided
Locating accuracy and stability when resolution is 0.5 meter.
When Fig. 3 A to Fig. 3 H are no filtering process, the Adaboost.M2 integrated studies of the invention based on Geordie decision tree
The localization method of model and positioning result figure of the localization method of other models under test sample, table 2 are Fig. 3 A's to Fig. 3 H
The comparison sheet of positioning result.Abscissa represents the test sample numbering arranged in chronological order, and ordinate represents the coordinate of positioning
As a result.Wherein, the test sample of embodiment 1 and embodiment 2 is target respectively by demarcation coordinate 16-13-7-4-3-2-9-11
With gather during 1-6-7-8-9-10-11.It can be seen that the Adaboost.M2 based on Geordie decision tree in the present invention
Integrated study model has more preferable stability than other models, the accuracy rate (being shown in Table 1) under 0.5 meter of resolution ratio apparently higher than
Employ the localization method of other models.
Table 1
Fig. 4 A to Fig. 4 H, it is to use using the integrated study model of the present invention under different wavelet filtering modes, what is drawn determines
Position design sketch, table 2 are Fig. 4 A to Fig. 4 H locating effect comparison sheet.The test sample that abscissa represents to arrange in chronological order is compiled
Number, ordinate represents the coordinate result of positioning.Different wavelet filtering modes are that the two-dimentional two-phase pass of the present invention is distributed small respectively
Ripple filtering, the filtering of simple correlation threshold value wavelet filtering, correlation Wavelet Entropy, complete zero wavelet filtering of high frequency coefficient, the number of plies are both configured to 3
Layer.It can be seen that use two-dimentional two-phase in the present invention to close the localization method that distributed wavelet filteration method is filtered,
Locating accuracy (being shown in Table 2) under 0.5 meter of resolution ratio is higher than the localization method for employing other wavelet filtering modes, and surely
It is qualitative also more preferable.
Table 2
Fig. 5 A to Fig. 5 H, when closing distributed wavelet filteration method all to employ the two-dimentional two-phase of the present invention, different determines
The positioning result figure of bit model, table 3 are Fig. 5 A to Fig. 5 H positioning result comparison sheet.Abscissa represents what is arranged in chronological order
Test sample is numbered, and ordinate represents the coordinate result of positioning.As can be seen from the figure it is of the invention based on Geordie decision tree
Localization method of the localization method than employing other location models of Adaboost.M2 integrated study models, in 0.5 meter of resolution
Locating accuracy under rate is higher (being shown in Table 3), and stability is more preferable.Contrast table 1 is it can be found that the localization method of model of the same race simultaneously
After the two-dimentional two-phase of the present invention closes distributed wavelet filtering progress data prediction, its positional accuracy has also obtained different journeys
The raising of degree.
Table 3
The another aspect of the embodiment of the present invention, additionally provides a kind of passive passive type indoor positioning device, and Fig. 6 is the present invention
The structural representation of the passive passive type indoor positioning device of embodiment, as shown in fig. 6, the device includes：Memory 61, is used for
Store instruction；And processor 62, for the instruction in memory 61, perform foregoing passive passive type indoor positioning side
Method.The present invention can differentiate and filter out RSS sample noises and random disturbances, retain normal saltus step data, while position fixing process has
There is very strong generalization ability, can generally improve locating accuracy and stability.
Particular embodiments described above, the purpose of the present invention, technical scheme and beneficial effect are carried out further in detail
Describe in detail it is bright, should be understood that the foregoing is only the present invention specific embodiment, be not intended to limit the invention, it is all
Within the spirit and principles in the present invention, any modification, equivalent substitution and improvements done etc., it should be included in the guarantor of the present invention
Within the scope of shield.
Claims (8)
1. a kind of passive passive type indoor orientation method, including：
Step A, using non-spacious indoor environment as monitored area, by target in monitored area at each coordinate when, gathered
All radio frequency network link RSS values as training sample, and using coordinate numbering be used as sample label；
Step B, two-dimentional two-phase is carried out to training sample and closes distributed wavelet filtering processing, determines filtered training sample；
Step C, establish the Adaboost.M2 integrated study models based on Geordie decision tree, using filtered training sample and
Sample label is trained, it is determined that the integrated study model after training；And
Step D, all radio frequency network link RSS values when collection target is any mobile in the monitored area, as test
Sample, test sample is subjected to two-dimentional two-phase and closes distributed wavelet filtering processing, determines filtered test sample, and its is defeated
Enter in the integrated study model to after training, determine positioning result.
2. according to the method for claim 1, wherein, step B includes sub-step：
Sub-step B1, training sample is formed into sample matrix, each row represent a training sample, and columns is of sample
Number, the RSS values with dimension in different samples are represented per a line, the sample matrix is subjected to each dimension data by dimension
Wavelet decomposition, wavelet function choice db1 small echos, obtain low-frequency wavelet coefficients ca and h layer high-frequency wavelet coefficient cdh, wherein, one
Individual dimension refers to a radio frequency link, and 1≤h≤s, s are the number of plies of wavelet decomposition, s >=2；
Sub-step B2, retain low-frequency wavelet coefficients ca, ask two-dimentional two-phase to close using the high-frequency wavelet coefficient of h layers and h+1 layers
Longitudinal coefficient correlation corrh, determine high-frequency wavelet coefficient energy PcdhWith longitudinal coefficient correlation energy Pcorrh, it is determined that normalization
Longitudinal coefficient correlation corrnh；
Sub-step B3, compare high-frequency wavelet coefficient cdhWith the normalization longitudinal direction coefficient correlation of respective layer, normalization longitudinal direction will be greater than
The high-frequency wavelet coefficient of coefficient correlation is arranged to zero, is retained less than being equal to the high-frequency wavelet coefficient for normalizing longitudinal coefficient correlation,
Last layer of high-frequency wavelet coefficient all retains, final to determine reserved high-frequency wavelet coefficient cdi；
Sub-step B4, reserved high-frequency wavelet coefficient is divided into U sections, U >=2, d data of every section of selection, d is positive integer, by layer point
The two-dimentional two-phase Central Shanxi Plain laterally non-time shift coefficient correlation is not calculated using the reserved high-frequency wavelet coefficient of+1 section of jth section and jth
R1ij, wherein, 1≤j≤U-1 ,+1 section of reserved high-frequency wavelet coefficient of jth is subjected to time shift, wherein, time shift amount be a, a for less than etc.
In d/2 positive integer, horizontal time shift coefficient R 2 is calculated respectively by layerij, determine time shift coefficient correlation and non-time shift phase relation
Number difference Rmij, and select each layer of R1ijThe minimum R1 of middle absolute valueijCorresponding data segment cdriWith each layer of RmijIn definitely
It is worth minimum RmijCorresponding data segment cdrmi, it is merged into filter threshold parameter Estimation data cdmi；
Sub-step B5, according to cdmiFilter threshold parameter σ is determined, then determines filter threshold thr, reserved high-frequency wavelet coefficient is entered
Row Distributed filtering, obtain filtered high-frequency wavelet coefficient cdfi；And
Sub-step B6, utilize low-frequency wavelet coefficients ca and filtered high-frequency wavelet coefficient cdfiWavelet reconstruction is carried out, is filtered
Training sample afterwards.
3. according to the method for claim 2, wherein, the longitudinal phase relation of normalization is determined according to below equation in sub-step B2
Number：
corrh=cdh·cdh+1
<mrow>
<msub>
<mi>Pcd</mi>
<mi>h</mi>
</msub>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>g</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</munderover>
<msup>
<msub>
<mi>cd</mi>
<mi>h</mi>
</msub>
<mn>2</mn>
</msup>
<mrow>
<mo>(</mo>
<mi>g</mi>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<msub>
<mi>Pcorr</mi>
<mi>h</mi>
</msub>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>g</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</munderover>
<msup>
<msub>
<mi>corr</mi>
<mi>h</mi>
</msub>
<mn>2</mn>
</msup>
<mrow>
<mo>(</mo>
<mi>g</mi>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<msub>
<mi>corrn</mi>
<mi>h</mi>
</msub>
<mo>=</mo>
<msub>
<mi>corr</mi>
<mi>h</mi>
</msub>
<mo>&CenterDot;</mo>
<msqrt>
<mfrac>
<mrow>
<msub>
<mi>Pcd</mi>
<mi>h</mi>
</msub>
</mrow>
<mrow>
<msub>
<mi>Pcorr</mi>
<mi>h</mi>
</msub>
</mrow>
</mfrac>
</msqrt>
</mrow>
Wherein, n is high-frequency wavelet coefficient cdhThe length of sequence, 1≤g≤n, g are high-frequency wavelet coefficient cdhThe sequence of data in sequence
Number.
4. according to the method for claim 2, wherein, in sub-step B4, R1 is determined according to below equationij、R2ij、Rmij, and
Selected cdri、cdrmi、cdmi：
<mrow>
<mi>R</mi>
<msub>
<mn>1</mn>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
</msub>
<mo>=</mo>
<mfrac>
<mrow>
<mi>C</mi>
<mi>o</mi>
<mi>v</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>cd</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
</msub>
<mo>,</mo>
<msub>
<mi>cd</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
<mo>+</mo>
<mn>1</mn>
</mrow>
</msub>
<mo>)</mo>
</mrow>
</mrow>
<msqrt>
<mrow>
<mi>V</mi>
<mi>a</mi>
<mi>r</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>cd</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
</msub>
<mo>)</mo>
</mrow>
<mo>&CenterDot;</mo>
<mi>V</mi>
<mi>a</mi>
<mi>r</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>cd</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
<mo>+</mo>
<mn>1</mn>
</mrow>
</msub>
<mo>)</mo>
</mrow>
</mrow>
</msqrt>
</mfrac>
</mrow>
<mrow>
<mi>R</mi>
<msub>
<mn>2</mn>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
</msub>
<mo>=</mo>
<mfrac>
<mrow>
<mi>C</mi>
<mi>o</mi>
<mi>v</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>cd</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
</msub>
<mo>,</mo>
<msub>
<msup>
<mi>cd</mi>
<mo>&prime;</mo>
</msup>
<mrow>
<mi>i</mi>
<mi>j</mi>
<mo>+</mo>
<mn>1</mn>
</mrow>
</msub>
<mo>)</mo>
</mrow>
</mrow>
<msqrt>
<mrow>
<mi>V</mi>
<mi>a</mi>
<mi>r</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>cd</mi>
<mrow>
<mi>i</mi>
<mi>j</mi>
</mrow>
</msub>
<mo>)</mo>
</mrow>
<mo>&CenterDot;</mo>
<mi>V</mi>
<mi>a</mi>
<mi>r</mi>
<mrow>
<mo>(</mo>
<msub>
<msup>
<mi>cd</mi>
<mo>&prime;</mo>
</msup>
<mrow>
<mi>i</mi>
<mi>j</mi>
<mo>+</mo>
<mn>1</mn>
</mrow>
</msub>
<mo>)</mo>
</mrow>
</mrow>
</msqrt>
</mfrac>
</mrow>
Rmij=| R1ij-R2ij|
cdrmi=cdijIf
cdri=cdijIf
cdmi={ cdrmi, cdri}
Wherein, cdij+1It is cdijNext adjacent non-time shift segment, cdij+1It is cdij+1It is small to move to right the time shift that a time shift amount is formed
Section, Cov cdij+1With cdijBetween covariance, Var cdij+1With cdijBetween variance.
5. according to the method for claim 2, wherein, in sub-step B5 according to below equation determine filter threshold parameter σ and
Thresholding thr, and carry out distributed wavelet filtering and obtain filtered high-frequency wavelet coefficient cdfi：
<mrow>
<mi>&sigma;</mi>
<mo>=</mo>
<mn>2</mn>
<mo>&times;</mo>
<mi>m</mi>
<mi>e</mi>
<mi>d</mi>
<mi>i</mi>
<mi>a</mi>
<mi>n</mi>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>/</mo>
<mi>m</mi>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>m</mi>
</munderover>
<msub>
<mi>cdm</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>/</mo>
<mn>0.6745</mn>
</mrow>
<mrow>
<mi>t</mi>
<mi>h</mi>
<mi>r</mi>
<mo>=</mo>
<mi>&sigma;</mi>
<msqrt>
<mrow>
<mn>2</mn>
<mi>l</mi>
<mi>g</mi>
<mrow>
<mo>(</mo>
<mi>L</mi>
<mo>)</mo>
</mrow>
</mrow>
</msqrt>
</mrow>
Wherein, 2≤q≤s, L are each layer of reserved high-frequency wavelet coefficient cdiLength, L=U × d, median is take median, y
For the sequence number of each data in each layer of reserved high-frequency wavelet coefficient, 1≤y≤L.
6. according to the method for claim 1, wherein, step C includes sub-step：
Sub-step C1, the Gini coefficients of each attribute in training sample are determined, select attribute A corresponding to Gini coefficient maximums
The growth of decision tree is carried out as best attributes, wherein, Gini coefficient Gini (A) formula is：
<mrow>
<mi>G</mi>
<mi>i</mi>
<mi>n</mi>
<mi>i</mi>
<mrow>
<mo>(</mo>
<mi>A</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>c</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>v</mi>
</munderover>
<mfrac>
<msubsup>
<mi>S</mi>
<mi>c</mi>
<mo>&prime;</mo>
</msubsup>
<mi>E</mi>
</mfrac>
<mo>&lsqb;</mo>
<mn>1</mn>
<mo>-</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>b</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>N</mi>
</munderover>
<msup>
<mrow>
<mo>(</mo>
<mfrac>
<msubsup>
<mi>S</mi>
<mrow>
<mi>b</mi>
<mi>c</mi>
</mrow>
<mo>&prime;</mo>
</msubsup>
<msubsup>
<mi>S</mi>
<mi>c</mi>
<mo>&prime;</mo>
</msubsup>
</mfrac>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
<mo>&rsqb;</mo>
</mrow>
Wherein, V is the branch amount of the decision tree, and N is the number of coordinate, S 'bcThe sample set separated when being using A as Split Attribute
S′cIn belong to the number of samples of b classes, E is the number of population sample at split point, 1≤b≤N；And
Sub-step C2, based on the decision tree in sub-step C1, it is determined that the integrated study model after training：
Iteration updates sample weights：
<mrow>
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mrow>
<msub>
<mi>W</mi>
<mn>0</mn>
</msub>
<mrow>
<mo>(</mo>
<mi>f</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<mrow>
<mi>N</mi>
<mi>M</mi>
</mrow>
</mfrac>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<msub>
<mi>W</mi>
<mi>k</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>f</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mrow>
<msub>
<mi>W</mi>
<mi>k</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>f</mi>
<mo>,</mo>
<mi>z</mi>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<munderover>
<mo>&Sigma;</mo>
<mi>f</mi>
<mrow>
<mi>M</mi>
<mi>N</mi>
</mrow>
</munderover>
<msub>
<mi>W</mi>
<mi>k</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>f</mi>
<mo>,</mo>
<mi>z</mi>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<msub>
<mi>W</mi>
<mrow>
<mi>k</mi>
<mo>+</mo>
<mn>1</mn>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<mi>f</mi>
<mo>,</mo>
<mi>z</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<msub>
<mi>W</mi>
<mi>k</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>f</mi>
<mo>,</mo>
<mi>z</mi>
<mo>)</mo>
</mrow>
<mo>&times;</mo>
<msup>
<msub>
<mi>a</mi>
<mi>k</mi>
</msub>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>/</mo>
<mn>2</mn>
<mo>)</mo>
<mo>(</mo>
<mn>1</mn>
<mo>+</mo>
<msub>
<mi>h</mi>
<mi>k</mi>
</msub>
<mo>(</mo>
<mrow>
<msub>
<mi>x</mi>
<mi>f</mi>
</msub>
<mo>,</mo>
<msub>
<mi>z</mi>
<mi>f</mi>
</msub>
</mrow>
<mo>)</mo>
<mo>-</mo>
<msub>
<mi>h</mi>
<mi>k</mi>
</msub>
<mo>(</mo>
<mrow>
<msub>
<mi>x</mi>
<mi>f</mi>
</msub>
<mo>,</mo>
<mi>z</mi>
</mrow>
<mo>)</mo>
<mo>)</mo>
</mrow>
</msup>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
<mo>;</mo>
</mrow>
Determine the error rate of decision tree：
Determine the weight of decision tree：
Wherein, W0(f) be f-th of sample initial weight, N is the number of coordinate, and M is the training sample included under each coordinate
Number, Wk(f, z) is the possibility sum that sample f is divided into all error category z in kth wheel iteration, and k is iteration time
Number, hk(xf, zf) it is that kth decision tree is its correct classification z to f-th of sample classification resultfPossibility, hk(xf, z) and kth
Decision tree is except its correct classification z to f-th of sample classification resultfThe possibility of every other incorrect classification z in addition,
εkIt is the error rate of kth decision tree, akIt is the weight of kth decision tree, Sum is summation.
7. according to the method for claim 1, wherein, sub-step B1 also includes step：Monitored area is not had to adopt during target
Collection all radio frequency network link RSS values be used as reference sample, by the training sample and reference sample progress difference processing,
Difference signal is determined, as the training sample.
8. a kind of passive passive type indoor positioning device, including：
Memory, for store instruction；And
Processor, for according to the instruction, performing the method as described in claim 1 to 7 is any.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710951127.5A CN107884744B (en) | 2017-10-12 | 2017-10-12 | Passive indoor positioning method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710951127.5A CN107884744B (en) | 2017-10-12 | 2017-10-12 | Passive indoor positioning method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107884744A true CN107884744A (en) | 2018-04-06 |
| CN107884744B CN107884744B (en) | 2021-01-01 |
Family
ID=61781631
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710951127.5A Active CN107884744B (en) | 2017-10-12 | 2017-10-12 | Passive indoor positioning method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107884744B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111325261A (en) * | 2020-02-14 | 2020-06-23 | 中国科学院半导体研究所 | Passive indoor positioning algorithm and device for dynamic environment |
| CN117295157A (en) * | 2023-11-24 | 2023-12-26 | 北京国旺盛源智能终端科技有限公司 | Positioning method and system for wearable back splint terminal |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7038617B1 (en) * | 2005-05-26 | 2006-05-02 | Lockheed Martin Corporation | Sensor doppler/wavelet processing |
| CN102186194A (en) * | 2011-05-09 | 2011-09-14 | 松日数码发展(深圳)有限公司 | Method for establishing passive target measurement model based on wireless sensor network |
| CN103023589A (en) * | 2012-12-06 | 2013-04-03 | 中山大学 | Indoor passive motion detection method and device |
| CN104883732A (en) * | 2015-04-14 | 2015-09-02 | 哈尔滨工程大学 | Enhanced indoor passive human body location method |
| CN105116375A (en) * | 2015-07-16 | 2015-12-02 | 北京理工大学 | Robust passive passiveness target positioning method based on multiple frequency points |
-
2017
- 2017-10-12 CN CN201710951127.5A patent/CN107884744B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7038617B1 (en) * | 2005-05-26 | 2006-05-02 | Lockheed Martin Corporation | Sensor doppler/wavelet processing |
| CN102186194A (en) * | 2011-05-09 | 2011-09-14 | 松日数码发展(深圳)有限公司 | Method for establishing passive target measurement model based on wireless sensor network |
| CN103023589A (en) * | 2012-12-06 | 2013-04-03 | 中山大学 | Indoor passive motion detection method and device |
| CN104883732A (en) * | 2015-04-14 | 2015-09-02 | 哈尔滨工程大学 | Enhanced indoor passive human body location method |
| CN105116375A (en) * | 2015-07-16 | 2015-12-02 | 北京理工大学 | Robust passive passiveness target positioning method based on multiple frequency points |
Non-Patent Citations (2)
| Title |
|---|
| 张乐玫: "室内定位特征选择算法研究", 《SOFTWARE》 * |
| 李健伟: "基于RSS的室内无线定位***的设计与实现", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111325261A (en) * | 2020-02-14 | 2020-06-23 | 中国科学院半导体研究所 | Passive indoor positioning algorithm and device for dynamic environment |
| CN111325261B (en) * | 2020-02-14 | 2023-12-05 | 中国科学院半导体研究所 | Passive indoor positioning algorithm and device for dynamic environment |
| CN117295157A (en) * | 2023-11-24 | 2023-12-26 | 北京国旺盛源智能终端科技有限公司 | Positioning method and system for wearable back splint terminal |
| CN117295157B (en) * | 2023-11-24 | 2024-02-13 | 北京国旺盛源智能终端科技有限公司 | Positioning method and system for wearable back splint terminal |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107884744B (en) | 2021-01-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106526693B (en) | Crack identification method and apparatus | |
| CN103810699B (en) | SAR (synthetic aperture radar) image change detection method based on non-supervision depth nerve network | |
| CN104182642B (en) | A kind of fault detection method based on rarefaction representation | |
| CN102721941B (en) | Method for fusing and diagnosing fault information of circuit of electric meter on basis of SOM (self-organized mapping) and D-S (Dempster-Shafer) theories | |
| CN104775811B (en) | A kind of extraction of formation anisotropy information and bearing calibration and system | |
| CN104502997B (en) | A kind of method of utilization fracture spacing curve prediction fracture spacing body | |
| CN105447762B (en) | A kind of calculation method of the low-permeability oil deposit water logging information of fluid replacement | |
| CN109184677A (en) | Reservoir evaluation methods for heterogeneous alternating layers sand body | |
| CN107290782A (en) | Reservoir porosity, water saturation and shale content parameter Simultaneous Inversion new method | |
| CN103886328A (en) | Functional magnetic resonance image data classification method based on brain network modular structure characteristics | |
| CN107862134A (en) | A kind of Wiener process reliability analysis methods for considering autocorrelation measurement error | |
| CN103353923A (en) | Self-adaption spatial interpolation method and system based on spatial feature analysis | |
| CN113534240B (en) | Microseism event detection and positioning method and system | |
| CN107967489A (en) | A kind of method for detecting abnormality and system | |
| CN103454677B (en) | Based on the earthquake data inversion method that population is combined with linear adder device | |
| CN108492298A (en) | Based on the multispectral image change detecting method for generating confrontation network | |
| CN104850867A (en) | Object identification method based on intuitive fuzzy c-means clustering | |
| CN115047531A (en) | Transient electromagnetic data inversion method based on ELM network | |
| CN107230197A (en) | Tropical cyclone based on satellite cloud picture and RVM is objective to determine strong method | |
| CN107884744A (en) | Passive passive type indoor orientation method and device | |
| CN108960334A (en) | A kind of multi-sensor data Weighted Fusion method | |
| CN106650766A (en) | Inherent feature analysis based three-dimensional body waveform classification method | |
| CN110288227A (en) | A method of fracturing effect Dominated Factors are influenced for evaluating | |
| Guo et al. | Prediction of reservoir saturation field in high water cut stage by bore-ground electromagnetic method based on machine learning | |
| CN106842299B (en) | A method of the crack quantification prediction based on seismic properties |
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 |