CN109522785A - A kind of gesture identification method and device based on big data and wireless signal model - Google Patents
A kind of gesture identification method and device based on big data and wireless signal model Download PDFInfo
- Publication number
- CN109522785A CN109522785A CN201811123809.8A CN201811123809A CN109522785A CN 109522785 A CN109522785 A CN 109522785A CN 201811123809 A CN201811123809 A CN 201811123809A CN 109522785 A CN109522785 A CN 109522785A
- Authority
- CN
- China
- Prior art keywords
- csi data
- csi
- data packet
- data flow
- training
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/107—Static hand or arm
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Human Computer Interaction (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
The present invention relates to a kind of gesture identification method based on big data and wireless signal model, comprising: obtain the first CSI data flow, wherein the first CSI data flow includes multiple first CSI data packets；The first CSI data flow is intercepted, dimensionality reduction and denoising, with the first CSI data packet that obtains that treated from the first CSI data flow；The first CSI data packet training Random Forest model using treated；The 2nd CSI data flow is identified using the Random Forest model after training.The Random Forest model that the present invention obtains has high accuracy rate, is capable of handling very high-dimensional data, does not need to do feature selecting, and training speed is fast and implementation is simple.
Description
Technical field
The invention belongs to technical field of hand gesture recognition, and in particular to a kind of gesture based on big data and wireless signal model
Recognition methods and device.
Background technique
With the fast development of computer information technology, human-computer interaction technology plays in people's daily life
More and more important role.Gesture is a kind of people and extraneous exchange way most intuitive when linking up, people can by body or
Gesture is intuitive, succinct, natural terrain reaches oneself idea, therefore the human-computer interaction technology based on gesture becomes the heat studied at present
Point, i.e. Gesture Recognition.
Gesture Recognition developing direction is broadly divided into two aspects at present: on the one hand for target carry special sensor or
Person's equipment, i.e., active Gesture Recognition, active Gesture Recognition mainly carry 3-axis acceleration sensing by target
The sensor devices such as device, gyroscope, electronic compass acquire hand-type or tracking hand spatial movement data, current active gesture
Data glove is most widely used in identification, but due to needing user to carry special equipment, it has not been convenient to operate, be not suitable for remote behaviour
Make, active Gesture Recognition application scenarios are limited by very large.It on the other hand is that target does not need to carry any biography
The Gesture Recognition of sensor or equipment, i.e. passive type Gesture Recognition, passive type Gesture Recognition mainly pass through nothing
Radio channel status information (the Channel State transmitted between wireless router and network interface card is utilized in line communication signal
Information, abbreviation CSI) acquisition gesture motion data, it is easy to operate since its is at low cost, meet user's habit, becomes
The hot spot studied both at home and abroad.
However, still remaining that recognition accuracy is not high to ask in such a way that wireless communication signal carries out gesture identification
Topic.
Summary of the invention
In order to solve the above-mentioned problems in the prior art, the present invention provides one kind to be based on big data and wireless signal
The gesture identification method and device of model.
The embodiment of the invention provides a kind of gesture identification method based on big data and wireless signal model, comprising:
Obtain the first CSI data flow, wherein the first CSI data flow includes multiple first CSI data packets；
The first CSI data flow is successively intercepted, dimensionality reduction and denoising, with from the first CSI data flow
It is middle to obtain treated the first CSI data packet；
The first CSI data packet training Random Forest model using treated, and utilize described random after training
Forest model identifies the 2nd CSI data flow.
In one embodiment of the invention, the first CSI data flow is successively intercepted, at dimensionality reduction and denoising
Reason, comprising:
The first CSI data packet is intercepted from the first CSI data flow using Pearson correlation coefficients；
Dimension-reduction treatment is carried out to the first CSI data packet after interception；
Denoising is carried out to the first CSI data packet after dimension-reduction treatment.
In one embodiment of the invention, institute is intercepted from the first CSI data flow using Pearson correlation coefficients
State the first CSI data packet, comprising:
The point of cut-off of the first CSI data flow is determined using Pearson correlation coefficients；
The first CSI number before the point of cut-off of the first CSI data flow with preset quantity after point of cut-off is intercepted respectively
According to packet.
In one embodiment of the invention, the truncation of the first CSI data flow is determined using Pearson correlation coefficients
Point, comprising:
N number of continuous first CSI data packet is obtained from the first CSI data flow, calculates the continuous institute of group
State the summing value or average value of the first CSI data packet, wherein N is the natural number greater than 1；
Using the summing value or the average value as the input data of the Pearson correlation coefficients, institute is utilized later
It states Pearson correlation coefficients and traverses the first CSI data flow, determine the point of cut-off of the first CSI data flow.
In one embodiment of the invention, the point of cut-off of the first CSI data flow is the Pearson correlation coefficients
Correlation minimum point.
In one embodiment of the invention, dimension-reduction treatment is carried out to the first CSI data packet after interception, comprising:
Dimension-reduction treatment is carried out to the first CSI data packet after interception using Principal Component Analysis.
In one embodiment of the invention, denoising, packet are carried out to the first CSI data packet after dimension-reduction treatment
It includes:
Denoising is carried out to the first CSI data packet after dimension-reduction treatment using Butterworth low pass wave algorithm.
In one embodiment of the invention, the first CSI data packet training Random Forest model using treated,
Include:
Training sample is extracted from the first CSI data packet using Bootstraping algorithm；
Utilize training sample training Random Forest model.
In one embodiment of the invention, the 2nd CSI data flow is identified using the Random Forest model after training
Before, further includes:
The 2nd CSI data flow is successively intercepted, dimensionality reduction, denoising, from the 2nd CSI data flow
Obtain treated the 2nd CSI data packet；
To treated, the 2nd CSI data packet carries out first-order difference processing, and by first-order difference, treated described the
The Random Forest model after the input training of two CSI data packets.
The embodiment of the invention also provides a kind of gesture identifying device based on big data and wireless signal model, including
CSI data acquisition module, CSI data processing module and neural metwork training module；
The CSI data acquisition module, for obtaining the first CSI data flow, wherein the first CSI data flow includes
Multiple first CSI data packets；
The CSI data processing module successively intercepts the first CSI data flow, dimensionality reduction and denoising, with
Treated the first CSI data packet is obtained from the first CSI data flow；
The neural metwork training module, the first CSI data packet training Random Forest model using treated, and
The 2nd CSI data flow is identified using the Random Forest model after training.
Compared with prior art, beneficial effects of the present invention:
1, the Random Forest model that the present invention obtains has high accuracy rate, is capable of handling very high-dimensional data, no
Need to do feature selecting, and training speed is fast and implementation is simple.
2, the present invention determines that point occurs for gesture motion using Pearson correlation coefficients in the first CSI Data Stream Processing, right
First CSI data flow is intercepted, and only retains before point occurs for gesture motion and the first CSI data after point occur with gesture motion
Packet, avoids the subsequent calculating to useless first CSI data packet, reduces calculation amount.
3, the present invention is due to having used Butterworth low pass wave algorithm in the first CSI data packet treatment process, to the
One CSI data packet has carried out denoising, alleviates adverse effect of the environmental noise for the first CSI data packet, helps to mention
Rise gesture identification accuracy rate.
Detailed description of the invention
Fig. 1 is a kind of gesture identification method process based on big data and wireless signal model provided in an embodiment of the present invention
Schematic diagram；
Fig. 2 is the positional diagram of wireless router provided in an embodiment of the present invention and receiving end；
Fig. 3 is the first CSI data flow diagram provided in an embodiment of the present invention；
Fig. 4 is the first CSI data flow provided in an embodiment of the present invention using the knot after improved Pearson correlation coefficients
Fruit schematic diagram；
Fig. 5 is the first CSI data packet schematic diagram provided in an embodiment of the present invention after PCA dimensionality reduction；
Fig. 6 shows through the filtered first CSI data packet of Butterworth after PCA dimensionality reduction again to be provided in an embodiment of the present invention
It is intended to；
Fig. 7 is the accuracy rate schematic diagram provided in an embodiment of the present invention that gesture identification is carried out using Random Forest model；
Fig. 8 is provided in an embodiment of the present invention a kind of based on the signal of the gesture identifying device of big data and wireless signal model
Figure.
Description of symbols:
CSI data acquisition module 21；CSI data processing module 22；Neural metwork training module 23.
Specific embodiment
Further detailed description is done to the present invention combined with specific embodiments below, but embodiments of the present invention are not limited to
This.
Embodiment one
Referring to Figure 1, Fig. 1 is provided in an embodiment of the present invention a kind of based on big data and the knowledge of the gesture of wireless signal model
Other method flow schematic diagram.The embodiment of the invention provides a kind of gesture identification side based on big data and wireless signal model
Method, the recognition methods include the following steps:
Step 1 obtains the first CSI data flow；
Specifically, device obtains the first CSI data flow by wireless communication, wherein the first CSI data flow includes multiple the
One CSI data packet.
Further, Fig. 2 is referred to, Fig. 2 is the position pass of wireless router provided in an embodiment of the present invention and receiving end
It is schematic diagram.Fig. 2 describes sender unit (wireless router) and signal receiving device (receiving end, such as wireless network card)
Orientation.Under different indoor scenes, arrangement wireless communication device, including wireless router, multiple transmittings and receiving antenna,
Receiving end configured with wireless network card (such as 5300 wireless network card of Intel) is to form monitoring area.It is read from wireless network card
The first CSI data flow of channel, wherein the first CSI data flow includes the phase information of the wireless signal transmitted in wireless channel
And amplitude information.
Fig. 3 is referred to, Fig. 3 is the first CSI data flow diagram provided in an embodiment of the present invention.For example, will be equipped with
The WIFI network of the host computer of Ubuntu system is connected on ready wireless router, in the distal end command window ping
Wireless router is to obtain ACK (Acknowledgement, abridge ACK) data packet, by the ack msg packet of acquisition come really
Recognize network link status, if network-like link state is normal, setting sends the frequency of CSI data packet, starts to carry out CSI data flow
Collection.The present embodiment wireless router transmitting terminal is provided with two antennas, and wireless network card receiving end is provided with three antennas, and
And have chosen 30 subcarriers that channel state information can be indicated in 56 subcarriers, then the CSI data packet of each transmission
It will receive 2*3*30=180 CSI data value, the CSI matrix of each pair of transmitting terminal and the 1*1*30 of receiving end is just in these data
It is a CSI data flow.The study found that not having in this six pairs of CSI data flows for the expression of gesture action message in experimentation
There is significant difference, therefore chooses one pair of them CSI data flow, therefore can once collect a transmitting terminal and a piece-root grafting
First CSI data flow of receiving end finally obtains the CSI dfd matrix of 30* time series, this is the first final CSI number
According to stream, wherein the first CSI data flow is the CSI data flow for including all gesture motions in monitoring area.
Preferably, the frequency for sending CSI data packet is 500HZ, i.e., the 500 CSI data packets of transmission in one second.
Step 2 successively intercepts the first CSI data flow, dimensionality reduction and denoising, from the first CSI data flow
Obtain treated the first CSI data packet；
Step 2.1 intercepts the first CSI data packet using Pearson correlation coefficients from the first CSI data flow；
Specifically, the point of cut-off of the first CSI data flow is determined using Pearson correlation coefficients.
Further, it obtains N number of continuous first CSI data packet from the first CSI data flow, calculates N number of continuous the
The summing value or average value of one CSI data packet, wherein N is the natural number greater than 1.
Wherein, summing value is the data of the cumulative summation of N number of continuous first CSI data packet, and average value is that a group is continuous
The average data taken after the cumulative summation of first CSI data packet divided by N.
Preferably, N value is 3.
Further, using summing value or average value as the input data of Pearson correlation coefficients, Pierre is utilized later
Gloomy related coefficient traverses the first CSI data flow, determines the point of cut-off of the first CSI data flow.
Preferably, the point of cut-off of the first CSI data flow is Pearson correlation coefficients correlation minimum point.
Fig. 4 is referred to, Fig. 4 is that the first CSI data flow provided in an embodiment of the present invention uses improved pearson correlation
Result schematic diagram after coefficient.Firstly, determining the point of cut-off of the first CSI data flow, i.e. the generation point of gesture motion.First CSI
The determination of the point of cut-off of data flow uses Pearson correlation coefficients.Pearson correlation coefficients be mainly used to two variables of measurement it
Between correlation, the related coefficient between two variables is defined as the quotient of covariance and standard deviation between two variables.This
The case where a coefficient undulating value, can reflect the point of cut-off of the first CSI data flow, that minimum point of related coefficient is the
The point of cut-off of one CSI data flow.
Further, Pearson correlation coefficients originally have only used a first CSI data packet, the table when data are preferable
It is existing excellent, but when unstable variation occurs for the power of wireless router transmission antenna or wireless network card receiving antenna, it will lead to
The fluctuation of first CSI data flow, especially when the first CSI data flow is there are when noise, using only the skin of a first CSI data packet
Not when the first CSI data flow is poor Ademilson related coefficient can not find that minimum activity well and point occurs, i.e.,
The point of cut-off of the first CSI data flow can be determined well.In order to keep the undulating value of the first CSI data flow more obvious, the present invention
Embodiment improves Pearson correlation coefficients, and the summing value or average value for having used N number of first CSI data packet are as Pearson
The input data of related coefficient traverses the first CSI data flow, finds the correlation minimum point of Pearson correlation coefficients, that is, determines
The point of cut-off of first CSI data flow.Wherein, N is the natural number greater than 1.Because Pearson correlation coefficients have used N number of first
CSI data packet determines that the effect of the point of cut-off of the first CSI data flow can be more preferable.
Further, the first CSI number with preset quantity after point of cut-off before the point of cut-off of the first CSI data flow is intercepted respectively
According to packet；
On the basis of the point of cut-off of the first CSI data flow determined, respectively to before the point of cut-off of the first CSI data flow and
The first CSI data packet of preset quantity is intercepted after point of cut-off, this mode can guarantee certain in the first CSI data packet of interception
It can include complete gesture motion.
Preferably, preset quantity 1500.
For example, referring again to Fig. 4, in the position that the first CSI data packet is 2500, the correlation of Pearson correlation coefficients
Minimum, then the point of cut-off of the first CSI data flow of the invention intercepts respectively just in the position that the first CSI data packet is 2500
1500 the first CSI data packets, totally 3000 the first CSI data packets, in this 3000 the first CSI data before 2500 and after 2500
Packet centainly contains complete gesture motion.
The embodiment of the present invention determines the first CSI data in the first CSI Data Stream Processing, using Pearson correlation coefficients
Point occurs for the point of cut-off of stream, i.e. gesture motion, intercepts to the first CSI data flow, only retains the truncation of the first CSI data flow
With the first CSI data packet after point of cut-off before point, the subsequent calculating to the first useless CSI data packet is avoided, meter is reduced
Calculation amount.
Step 2.2 carries out dimension-reduction treatment to the first CSI data packet after interception；
Specifically, dimension-reduction treatment is carried out to the first CSI data packet after interception using Principal Component Analysis.
Further, Fig. 5 is referred to, Fig. 5 is the first CSI data packet provided in an embodiment of the present invention after PCA dimensionality reduction
Schematic diagram.Dimension-reduction treatment is carried out to the first CSI data packet after interception, used in the embodiment of the present invention is Principal Component Analysis
(Principal Component Analysis, abbreviation PCA).PCA is a kind of statistical method, can be by extracting principal component
First CSI data packet is mapped to lower dimensional space by method.For example, the first CSI data packet dimensionality reduction that the embodiment of the present invention is tieed up 30
The lower dimensional space of the first CSI data packet of this 30 dimension can be extremely indicated with 98%, realization is gone to indicate 30 dimensions with less dimension
First CSI data packet.Dimension-reduction treatment is carried out to the first CSI data packet using the PCA function that MATLAB is carried, by applying PCA
Function obtains the latent parameter for characterizing each dimension contribution rate, according to latent parameter, carries out drafting by MATLAB and allows it
98% ground indicates the first CSI data packet, the first CSI data packet after obtaining dimensionality reduction.
Step 2.3 carries out denoising to the first CSI data packet after dimension-reduction treatment；
Specifically, the first CSI data packet after dimension-reduction treatment is gone using Butterworth low pass wave algorithm
It makes an uproar processing.
Further, Fig. 6 is referred to, Fig. 6 is filtered through Butterworth after PCA dimensionality reduction again to be provided in an embodiment of the present invention
The first CSI data packet schematic diagram afterwards.Denoising, denoising of the embodiment of the present invention are carried out to the first CSI data packet after dimensionality reduction
The method used is handled as Butterworth low pass wave algorithm.The main thought of Butterworth low pass wave algorithm be exactly allow it is low
Frequency signal passes through, and filters out high-frequency signal.Butterworth LPF firstly the need of setting one filter cutoff frequency, if
The first CSI data packet signal frequency domain passed through at this time is higher than cutoff frequency, then the first CSI data packet signal is assigned a value of 0.
Butterworth LPF is designed using butter the and filter function that MATLAB is carried.Wherein, butter
First parameter of function is order n, and order n is higher, and filter effect also can be better, but cost price also can be bigger, consolidates hair
Bright order n takes fixed value 2；The second parameter of butter function is the cutoff frequency W of Butterworth LPFc, this hair
The W of bright embodimentcIt designs as follows:
Wherein, λ represents the wavelength of wireless signal, vrThe spread speed of wireless signal is represented, value is 3 × 108M/s, frGeneration
The frequency of table wireless signal.
Preferably, frThe frequency of wireless signal is 2462MHz.
Wherein, ftRepresent the mobile cutoff frequency of gesture, vsRepresent the mobile speed of gesture.
Preferably, vsThe mobile speed of gesture is 0.5m/s or so.
The then cutoff frequency W of Butterworth LPFcIt calculates as follows:
Wherein, FsRepresent sample frequency.
Preferably, FsSample frequency is 500Hz.
Further, obtained Butterworth LPF parameter order n and cutoff frequency W is utilizedc, progress bar
Special Butterworth low-pass filtering, the first CSI data packet after obtaining noise-removed filtering.
Due to gesture difference, the mobile cutoff frequency f of gesturetThe mobile speed v with gesturesDifference then designs obtained bar
The cutoff frequency of special Butterworth low-pass filter also will be different, i.e., can obtain different Butterworth LPFs.
The embodiment of the present invention due to having used Butterworth low pass wave algorithm in the first CSI data packet treatment process,
Denoising has been carried out to the first CSI data packet, adverse effect of the environmental noise for the first CSI data packet has been alleviated, helps
In promotion gesture identification accuracy rate.
CSI data flow of the first CSI data stream packets containing all gesture motions does the CSI data flow of all gesture motions
Step process as above obtains each gesture motion treated the first CSI data packet, with the first CSI data that constitute that treated
Packet.
For example, gesture motion number of the present invention is 4, treated that the first CSI data packet number is for each gesture motion
3000.Wherein, 4 gesture motions are respectively that a left side is waved, the right side is waved, push, pull.
Step 3, using treated, the first CSI data packet trains Random Forest model；
To each gesture motion treated the first CSI data packet, following steps processing is carried out:
Step 3.1 extracts training sample from treated the first CSI data packet using Bootstraping algorithm；
Specifically, will treated the first CSI data packet as sample set, first from the first CSI data packet sample set,
Using Bootstraping algorithm, there are n the first CSI data packets of sampling put back to every time, sample m times, establish new samples collection, it will
New samples collection is as training sample.Wherein, new samples integrate number as the first CSI data packet of n*m.
Wherein, Bootstraping algorithm is a kind of methods of sampling for having and putting back to, using limited sample via multiple weight
Multiple sampling, starts against the new samples for being enough to represent original sample distribution.
Preferably, 1000 n, m 3.
Step 3.2 utilizes training sample training Random Forest model；
To above-mentioned n*m the first CSI data packet groups at training sample carry out m nothing and put back to sampling, by what is sampled every time
First CSI data packet sample generates training set, and m sampling generates m training set, wherein the first CSI number of each training set
It is n according to packet number of samples.
Further, to m training set of above-mentioned generation, each training set uses M decision-tree model trained respectively
To M decision tree, m*M decision tree of generation is formed into random forest, later using Bagging strategy to m*M decision tree
Ballot obtains the final classification of the first CSI data packet as a result, which the gesture motion that i.e. the first CSI data packet includes belongs to
Class.Wherein, it to every decision tree, selects optimal mode to be divided according to its information gain or gini index etc., obtains it
Optimal decision tree.
Wherein, information gain indicates that the information of known features makes the degree of the uncertain reduction of the information of categorizing system,
I.e. known features how much information can be brought for categorizing system.Information gain is bigger, it is meant that division institute is carried out using attribute
The purity of acquisition is promoted bigger.
Gini index reflects and randomly selects two samples from sample set, the inconsistent probability of category label, Geordie
Index is smaller, and the purity of sample set is higher.When selection divides attribute, select the smallest attribute of gini index as optimal dividing
Attribute.
Preferably, 5 M.
Preferably, Bagging strategy is most ballot modes.
For example, have 3000 the first CSI data packets after the present invention each gesture motion processing, will treated 3000 the
One CSI data packet as sample set, first to this 3000 the first CSI data packet sample sets using Bootstraping algorithm into
Line sampling extracts 3 times, and 1000 the first CSI data packet samples, re-establish the first new sample of CSI data packet of 3*1000 every time
This collection, using the new samples collection as training sample.Then to the sample in 3000 new samples collection, 3 nothings is carried out and put back to sampling,
1000 samples of sampling every time, every 1000 samples generate 3 training sets as a training set, 3 samplings.This 3 are instructed
Practice collection, each training set generates 5 decision trees using 5 decision-tree models, and 3 training sets generate 15 decision trees, and this 15
Decision tree forms random forest.Then ballot classification is carried out to 15 decision trees that the gesture generates using Bagging strategy, really
Which kind of fixed first CSI data packet belongs to.For example, the gesture motion for having 5 decision trees to think that the first CSI data packet includes is left
It waves, 10 decision trees think that the gesture motion that the first CSI data packet includes is that the right side is waved, then the first CSI data packet includes
Gesture motion belong to the right side and wave.Wherein, 4 gesture motions are divided into 4 classes by the present invention, and respectively promotion, pulling, Zuo Hui, the right side are waved.
After carrying out step process as above to all gesture motions treated the first CSI data packet, i.e. completion random forest
Model training.
Step 4 identifies the 2nd CSI data flow using the Random Forest model after training；
It, can also be to the 2nd CSI data flow before identifying the 2nd CSI data flow using the Random Forest model after training
It is handled, wherein the target gesture that the 2nd CSI data flow as needs to identify.
Specifically, the 2nd CSI data flow is intercepted, dimensionality reduction, denoising, from the 2nd CSI data flow
Obtain treated the 2nd CSI data packet；
The interception of 2nd CSI data flow determines the 2nd CSI data flow point of cut-off using Pearson correlation coefficients, then intercepts the
Before two CSI data flow point of cut-offs and point of cut-off after predetermined quantity the 2nd CSI data packet, to the 2nd CSI data packet after interception
Dimensionality reduction denoises using Butterworth low pass wave method the 2nd CSI data packet after dimensionality reduction using PCA method.Wherein,
Predetermined quantity is 1500.
Further, to treated, the 2nd CSI data packet carries out first-order difference processing；
First-order difference processing refers to two item data of continuous adjacent make difference processing, can show in this way continuous adjacent data it
Between difference.The present invention carries out making poor processing to adjacent two item data in each 2nd CSI data packet, by each 2nd CSI
The difference of two item data adjacent in this way constitutes the 2nd new CSI data packet in data packet.
Further, by first-order difference treated the 2nd CSI data packet inputs random forest mould that above-mentioned training obtains
Type can identify which kind of gesture is the 2nd CSI data flow belong to, that is, realize the identification to target gesture.
Fig. 7 is referred to, Fig. 7 is the accuracy rate provided in an embodiment of the present invention that gesture identification is carried out using Random Forest model
Schematic diagram.The present embodiment is due to, as gesture identification sub-model, having high accuracy rate, table on data set using random forest
It is now good, have great advantage compared with other algorithms, be capable of handling very high-dimensional data, does not need to do feature selecting, and
Training speed is fast and realizes simple.Gesture identification method through the invention can obtain high-precision gesture identification.
The present embodiment the utility model has the advantages that
1, the present embodiment is adopted in training Random Forest model due to obtaining training sample using Bootstraping algorithm
Classified with Bagging strategy, obtained Random Forest model has high accuracy rate, is capable of handling very high-dimensional number
According to not needing to do feature selecting, and training speed is fast and realizes simple.
2, the present embodiment determines that point occurs for gesture motion using Pearson correlation coefficients in the first CSI Data Stream Processing,
First CSI data flow is intercepted, only retains before point occurs for gesture motion and the first CSI data after point occurs with gesture motion
Packet, avoids the subsequent calculating to useless first CSI data packet, reduces calculation amount.
3, the present embodiment is right due to having used Butterworth low pass wave algorithm in the first CSI data packet treatment process
First CSI data packet has carried out denoising, alleviates adverse effect of the environmental noise for the first CSI data packet, facilitates
Promote gesture identification accuracy rate.
4, the present embodiment is a kind of method for carrying out gesture identification by wireless communication signal, and use is common, inexpensive
Wireless telecom equipment constructs monitoring area, and applicable situation is extensive, and is not necessarily to user Portable device.
Embodiment two
Fig. 8 is referred to, Fig. 8 is provided in an embodiment of the present invention a kind of based on big data and the knowledge of the gesture of wireless signal model
Other schematic device.The present embodiment provides a kind of hand based on big data and wireless signal model on the basis of the above embodiments
Gesture identification device, the gesture identifying device include: CSI data acquisition module 21, CSI data processing module 22 and neural network instruction
Practice module 23.
CSI data acquisition module 21；
CSI data acquisition module 21, for obtaining the first CSI data flow, wherein the first CSI data flow includes multiple the
One CSI data packet.
Specifically, the first CSI data flow is obtained for device by wireless communication.Including wireless router, multiple transmittings
With receiving antenna, the receiving end configured with wireless network card (such as 5300 wireless network card of Intel) to form monitoring area；
CSI data acquisition module 21 further includes that the WIFI network of the host computer of Ubuntu system is connected to ready nothing
On line router, the distal end command window ping wireless router to obtain ack msg packet, pass through the ack msg packet of acquisition
After confirming that network link status is normal, setting sends the frequency of CSI data packet, starts to collect the first CSI data flow.
Different types of gesture is identified according to the variation of different sub-carrier amplitude using the first CSI data flow, due to
The amplitude of different sub-carrier is different for the sensibility of different gestures, can be from the hand of multiple a certain types of angle comprehensive description
Gesture has higher resolvability compared to other methods.
CSI data processing module 22；
CSI data processing module 22, for successively being intercepted to the first CSI data flow, dimensionality reduction and denoising,
To obtain treated the first CSI data packet from the first CSI data flow.
Specifically, the first CSI data flow point of cut-off is confirmed including the correlation minimum point using Pearson correlation coefficients, i.e.,
The generation point of gesture motion intercepts the gesture and preceding the first CSI data packet that predetermined number after point occurs with gesture of point occurs, avoids
The subsequent calculating to hash, while external interference do not have much influence the confirmation of gesture action generation point yet.
Butterworth low pass wave algorithm is also used in data processing, not only alleviates environmental noise for
The adverse effect of one CSI data packet also overcomes information redundancy problem brought by influencing each other between different sub-carrier, reduces
Subsequent first CSI data packet computational processing, while helping to be promoted the precision of gesture identification.
Data processing module 22 makes gesture identification method of the present invention have strong robustness, not vulnerable to the advantage of external interference,
It can more accurately identify target gesture.
Neural metwork training module 23.
Neural metwork training module 23, for the first CSI data packet to train Random Forest model, and benefit using treated
The 2nd CSI data flow is identified with the Random Forest model after training.
Specifically, because decision tree to be easy to appear over-fitting, generalization ability weak, the present invention uses Bootstraping
Algorithm generates training sample first, and by training, these training samples generate random forest, and are determined and divided by bagging strategy
Class is as a result, establish Random Forest model.
It inputs the 2nd CSI data to flow to before the Random Forest model, the 2nd CSI data flow is successively intercepted, is dropped
Dimension, denoising and first-order difference processing will treated the 2nd CSI data packets with the 2nd CSI data packet that obtains that treated
Input Random Forest model, so that it may realize the identification to the 2nd CSI data flow, wherein the 2nd CSI data flow is to need to know
Other target gesture.
The present embodiment the utility model has the advantages that
1, the present embodiment is adopted in training Random Forest model due to obtaining training sample using Bootstraping algorithm
Classified with Bagging strategy, obtained Random Forest model has high accuracy rate, is capable of handling very high-dimensional number
According to not needing to do feature selecting, and training speed is fast and realizes simple.
2, the present embodiment determines that point occurs for gesture motion using Pearson correlation coefficients in the first CSI Data Stream Processing,
First CSI data flow is intercepted, only retains before point occurs for gesture motion and the first CSI data after point occurs with gesture motion
Packet, avoids the subsequent calculating to useless first CSI data packet, reduces calculation amount.
3, the present embodiment is right due to having used Butterworth low pass wave algorithm in the first CSI data packet treatment process
First CSI data packet has carried out denoising, alleviates adverse effect of the environmental noise for the first CSI data packet, facilitates
Promote gesture identification accuracy rate.
4, the present embodiment is a kind of method for carrying out gesture identification by wireless communication signal, and use is common, inexpensive
Wireless telecom equipment constructs monitoring area, and applicable situation is extensive, and is not necessarily to user Portable device.
The above content is a further detailed description of the present invention in conjunction with specific preferred embodiments, and it cannot be said that
Specific implementation of the invention is only limited to these instructions.For those of ordinary skill in the art to which the present invention belongs, exist
Under the premise of not departing from present inventive concept, a number of simple deductions or replacements can also be made, all shall be regarded as belonging to of the invention
Protection scope.
Claims (10)
1. a kind of gesture identification method based on big data and wireless signal model characterized by comprising
Obtain the first CSI data flow, wherein the first CSI data flow includes multiple first CSI data packets；
The first CSI data flow is successively intercepted, dimensionality reduction and denoising, to be obtained from the first CSI data flow
The first CSI data packet that takes that treated；
The first CSI data packet training Random Forest model using treated, and utilize the random forest after training
Model identifies the 2nd CSI data flow.
2. the method according to claim 1, wherein successively being intercepted to the first CSI data flow, dimensionality reduction
And denoising, comprising:
The first CSI data packet is intercepted from the first CSI data flow using Pearson correlation coefficients；
Dimension-reduction treatment is carried out to the first CSI data packet after interception；
Denoising is carried out to the first CSI data packet after dimension-reduction treatment.
3. according to the method described in claim 2, it is characterized in that, using Pearson correlation coefficients from the first CSI data
The first CSI data packet is intercepted in stream, comprising:
The point of cut-off of the first CSI data flow is determined using Pearson correlation coefficients；
The first CSI data packet before the point of cut-off of the first CSI data flow with preset quantity after point of cut-off is intercepted respectively.
4. according to the method described in claim 3, it is characterized in that, determining the first CSI number using Pearson correlation coefficients
According to the point of cut-off of stream, comprising:
N number of continuous first CSI data packet is obtained from the first CSI data flow, calculates N number of continuous described first
The summing value or average value of CSI data packet, wherein N is the natural number greater than 1；
Using the summing value or the average value as the input data of the Pearson correlation coefficients, the skin is utilized later
Ademilson related coefficient traverses the first CSI data flow, determines the point of cut-off of the first CSI data flow.
5. according to the method described in claim 4, it is characterized in that, the point of cut-off of the first CSI data flow is the Pierre
Gloomy related coefficient correlation minimum point.
6. according to the method described in claim 2, it is characterized in that, carrying out dimensionality reduction to the first CSI data packet after interception
Processing, comprising:
Dimension-reduction treatment is carried out to the first CSI data packet after interception using Principal Component Analysis.
7. according to the method described in claim 2, it is characterized in that, being carried out to the first CSI data packet after dimension-reduction treatment
Denoising, comprising:
Denoising is carried out to the first CSI data packet after dimension-reduction treatment using Butterworth low pass wave algorithm.
8. the method according to claim 1, wherein using treated the first CSI data packet training with
Machine forest model, comprising:
Training sample is extracted from treated the first CSI data packet using Bootstraping algorithm；
Utilize the training sample training Random Forest model.
9. the method according to claim 1, wherein utilizing the Random Forest model identification second after training
Before CSI data flow, further includes:
The 2nd CSI data flow is successively intercepted, dimensionality reduction, denoising, to be obtained from the 2nd CSI data flow
The 2nd CSI data packet that treated；
To treated, the 2nd CSI data packet carries out first-order difference processing, by first-order difference treated the 2nd CSI
The Random Forest model after data packet input training.
10. a kind of gesture identifying device based on big data and wireless signal model, which is characterized in that including CSI data acquisition
Module, CSI data processing module and neural metwork training module；
The CSI data acquisition module, for obtaining the first CSI data flow, wherein the first CSI data flow includes multiple
First CSI data packet；
The CSI data processing module, for successively being intercepted to the first CSI data flow, dimensionality reduction and denoising, with
Treated the first CSI data packet is obtained from the first CSI data flow；
The neural metwork training module, for the first CSI data packet training Random Forest model described using treated, and
The 2nd CSI data flow is identified using the Random Forest model after training.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811123809.8A CN109522785A (en) | 2018-09-26 | 2018-09-26 | A kind of gesture identification method and device based on big data and wireless signal model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811123809.8A CN109522785A (en) | 2018-09-26 | 2018-09-26 | A kind of gesture identification method and device based on big data and wireless signal model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109522785A true CN109522785A (en) | 2019-03-26 |
Family
ID=65769813
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811123809.8A Pending CN109522785A (en) | 2018-09-26 | 2018-09-26 | A kind of gesture identification method and device based on big data and wireless signal model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109522785A (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104615244A (en) * | 2015-01-23 | 2015-05-13 | 深圳大学 | Automatic gesture recognizing method and system |
| CN107633227A (en) * | 2017-09-15 | 2018-01-26 | 华中科技大学 | A kind of fine granularity gesture identification method and system based on CSI |
| CN107909003A (en) * | 2017-10-16 | 2018-04-13 | 华南理工大学 | A kind of gesture identification method for large vocabulary |
-
2018
- 2018-09-26 CN CN201811123809.8A patent/CN109522785A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104615244A (en) * | 2015-01-23 | 2015-05-13 | 深圳大学 | Automatic gesture recognizing method and system |
| CN107633227A (en) * | 2017-09-15 | 2018-01-26 | 华中科技大学 | A kind of fine granularity gesture identification method and system based on CSI |
| CN107909003A (en) * | 2017-10-16 | 2018-04-13 | 华南理工大学 | A kind of gesture identification method for large vocabulary |
Non-Patent Citations (5)
| Title |
|---|
| SANGJUN O ET AL.: "Real Time Hand Gesture Recognition Using Random Forest and Linear Discriminant Analysis", 《ACM》 * |
| 史白: "基于WiFi的非合作式人体运动检测", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 杨雪: "基于无线信道状态信息的非绑定式人体活动识别与预警算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 绕绍奇: "《中华医学统计百科全书 遗传统计分册》", 31 May 2013, 中国统计出版社 * |
| 陈福山: "基于无线信号CSI和深度学习算法的活动识别技术研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhang et al. | Wi-Fi sensing for joint gesture recognition and human identification from few samples in human-computer interaction | |
| CN104484673B (en) | The Supplementing Data method of real-time stream application of pattern recognition | |
| CN106237604A (en) | Wearable device and the method utilizing its monitoring kinestate | |
| CN106919958B (en) | Human body finger action recognition method based on smart watch | |
| CN113609976B (en) | Direction-sensitive multi-gesture recognition system and method based on WiFi equipment | |
| CN107016347A (en) | A kind of body-sensing action identification method, device and system | |
| CN110502105A (en) | A kind of gesture recognition system and recognition methods based on CSI phase difference | |
| CN105224066A (en) | A kind of gesture identification method based on high in the clouds process | |
| Zinonos et al. | Grape leaf diseases identification system using convolutional neural networks and Lora technology | |
| CN105786189A (en) | Finger independent action recognition method and system based on MMG signal | |
| Al-Naffakh et al. | Continuous user authentication using smartwatch motion sensor data | |
| Shang et al. | LSTM-CNN network for human activity recognition using WiFi CSI data | |
| CN110353693A (en) | A kind of hand-written Letter Identification Method and system based on WiFi | |
| Zhu et al. | Wi-ATCN: Attentional temporal convolutional network for human action prediction using WiFi channel state information | |
| Hao et al. | CSI‐HC: A WiFi‐Based Indoor Complex Human Motion Recognition Method | |
| CN110866468A (en) | Gesture recognition system and method based on passive RFID | |
| CN114423034A (en) | Indoor personnel action identification method, system, medium, equipment and terminal | |
| Barua et al. | Gait signals classification and comparison | |
| CN106375944A (en) | Data acquisition system based on cloud computing | |
| WO2024041053A1 (en) | Indoor passive human behavior recognition method and apparatus | |
| CN109522785A (en) | A kind of gesture identification method and device based on big data and wireless signal model | |
| Pandey et al. | Csi-based joint location and activity monitoring for covid-19 quarantine environments | |
| Lackner et al. | User tracking based on behavioral fingerprints | |
| CN106874739A (en) | A kind of recognition methods of terminal iidentification and device | |
| Yin et al. | Back-guard: Wireless backscattering based user activity recognition and identification with parallel attention model |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190326 |
|
| RJ01 | Rejection of invention patent application after publication |