CN109768838B - Interference detection and gesture recognition method based on WiFi signal - Google Patents

Interference detection and gesture recognition method based on WiFi signal Download PDF

Info

Publication number
CN109768838B
CN109768838B CN201811630072.9A CN201811630072A CN109768838B CN 109768838 B CN109768838 B CN 109768838B CN 201811630072 A CN201811630072 A CN 201811630072A CN 109768838 B CN109768838 B CN 109768838B
Authority
CN
China
Prior art keywords
csi data
detected
gesture
sample
data
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201811630072.9A
Other languages
Chinese (zh)
Other versions
CN109768838A (en
Inventor
陈晓江
刘雨田
牛进平
颉麦杰
郭艺
马跃
房鼎益
陈�峰
张涛
刘宝英
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Northwestern University
Original Assignee
Northwestern University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Northwestern University filed Critical Northwestern University
Priority to CN201811630072.9A priority Critical patent/CN109768838B/en
Publication of CN109768838A publication Critical patent/CN109768838A/en
Application granted granted Critical
Publication of CN109768838B publication Critical patent/CN109768838B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Landscapes

  • Image Analysis (AREA)

Abstract

The invention belongs to the technical field of gesture recognition, and particularly relates to an interference detection and gesture recognition method based on WiFi signals, which comprises the steps of firstly judging whether interference exists in to-be-detected CSI data containing gesture signals, calculating time-frequency domain statistical characteristics of the to-be-detected CSI data and sample CSI data when the interference does not exist, respectively comparing the similarity of the time-frequency domain characteristics of each gesture of the to-be-detected CSI data and the sample CSI data, and recognizing a gesture corresponding to the to-be-detected CSI data; when interference exists, calculating the time-frequency domain statistical characteristics of the to-be-detected CSI data and the sample CSI data, wherein the to-be-detected CSI data and the sample CSI data contain cross correlation coefficient characteristics, respectively comparing the similarity of the time-frequency domain statistical characteristics of each gesture of the to-be-detected CSI data and the sample CSI data, wherein the similarity of the time-frequency domain statistical characteristics contains the cross correlation coefficient characteristics, and identifying the gesture corresponding to the to-be-detected CSI data. The invention provides a gesture recognition method based on weighted cross-correlation coefficient characteristics, which is used for extracting values for different characteristics of gesture action waveforms to match with corresponding weights so as to further improve the classification performance.

Description

Interference detection and gesture recognition method based on WiFi signal
Technical Field
The invention belongs to the technical field of gesture recognition, and particularly relates to an interference detection and gesture recognition method based on WiFi signals.
Background
WiFi equipment is common indoor deployment equipment in modern society, need not the user and carry on just can gather demand data, has low cost, passive and user-friendly's characteristics. The basic idea of WiFi sensing is to capture the change of the surrounding environment according to the change of the WiFi signal, and two features of the WiFi signal participating in the sensing process are: received Signal Strength (RSS) and Channel State Information (CSI). Since the CSI data includes large-scale fading and small-scale fading channel information of each subcarrier of the WiFi system, and has finer-grained channel characteristics, we often use the CSI data to represent the characteristics of the WiFi signal. In the field of gesture recognition, different types of gesture information are contained in the CSI data, and gestures sensed by WiFi can be recognized and classified through processing of the CSI data.
The existing research of the gesture recognition method based on WiFi mainly focuses on distinguishing target gestures, but neglects the interference behavior when the gestures are made, but the interference behavior may exist all the time, especially the dynamic interference caused by some indoor daily behaviors, when the problem of large interference exists, the recognition accuracy of the algorithm is reduced when the existing algorithm is used for recognizing the individual gestures by using the detection of low-cost commercial WiFi equipment. While some existing methods involve interference problems in gesture recognition, these methods do not use commercial WiFi devices but rather specialized devices, such as wireless Adhoc positioning systems (WASPs) with wider bandwidths, are costly.
Disclosure of Invention
Aiming at the problem that the identification precision of the prior art for the target individual behavior is reduced when other people perform daily behaviors in an indoor environment in the prior art, the invention provides an interference detection and gesture identification method based on WiFi signals, which is realized by adopting the following technical scheme:
an interference detection method based on WiFi signals comprises the following steps:
step 1: acquiring CSI data by using WiFi equipment, and preprocessing amplitude information of the CSI data to obtain the CSI data containing gesture information;
step 2: performing signal analysis on the CSI data containing the gesture information obtained in the step 1 to obtain time domain subcarrier data;
and step 3: converting the subcarrier data of the time domain obtained in the step (2) into a frequency domain, and carrying out phase calibration;
and 4, step 4: performing normalization processing on the phase calibration result of the CSI data obtained in the step (3), calculating the variance of the phase sample, and if the variance of the phase sample is greater than or equal to a threshold value, judging that the CSI data has interference; otherwise, the CSI data is interference free.
A gesture recognition method based on WiFi signals comprises the following steps:
step 1: setting a group of non-interference sample CSI data, and respectively calculating time-frequency domain characteristics of the sample CSI data corresponding to different gestures;
step 2: by adopting the interference detection method based on the WiFi signal, when the CSI data to be detected has no interference, the step 3 is executed, and when the CSI data to be detected has the interference, the step 4 is executed;
and step 3: calculating time-frequency domain statistical characteristics of the CSI data to be detected and the sample CSI data, respectively comparing the similarity of the time-frequency domain characteristics of each gesture of the CSI data to be detected and the sample CSI data based on a k-NN method, and identifying the gesture corresponding to the CSI data to be detected;
and 4, step 4: and calculating time-frequency domain statistical characteristics of the to-be-detected CSI data and the sample CSI data, wherein the to-be-detected CSI data and the sample CSI data contain cross correlation coefficient characteristics, respectively comparing the similarity of the time-frequency domain statistical characteristics of each gesture of the to-be-detected CSI data and the sample CSI data, which contain the cross correlation coefficient characteristics, based on a k-NN method, and identifying the gesture corresponding to the to-be-detected CSI data.
Further, step 3 of the gesture recognition method based on the WiFi signal includes the following sub-steps:
when the CSI data to be detected has no interference, calculating the time-frequency domain statistical characteristics of the CSI data s1 to be detected to form a characteristic vector: [ lambda ]1(s1),λ2(s1),λ3(s1),…λk(s1)…,λK(s1)]，λk(s1) denotes the kth time-frequency domain statistic feature of s1, K ∈ [1, K ∈]And K is a positive integer, and the similarity of the time-frequency domain characteristics of the CSI data to be detected and the sample CSI data of each gesture is calculated by using the formula I
Figure GDA0002878160840000037
Figure GDA0002878160840000031
Wherein S1 represents the CSI data to be detected without interference, SjSample CSI data representing the jth gesture,
Figure GDA0002878160840000032
denotes the similarity of s1 with the j gesture, λk(Sj) Denotes SjThe kth time-frequency domain statistical characteristic of (1);
similarity between the CSI data to be detected and the sample CSI data of the jth gesture if no interference exists
Figure GDA0002878160840000038
And if the detected CSI data is the smallest, the CSI data to be detected is identified as the jth gesture.
Further, step 4 of the gesture recognition method based on the WiFi signal includes the following sub-steps:
when the CSI data to be detected have interference, calculating the time-frequency domain statistical characteristics of the CSI data s2 to be detected to form a characteristic vector:
Figure GDA0002878160840000033
λk(s2) denotes the kth time-frequency domain statistic feature of s2, K ∈ [1, K ∈]And K is a positive integer and is a positive integer,
Figure GDA0002878160840000034
expressing the cross-correlation coefficient characteristics between the s2 and the sample CSI data, and calculating the similarity of the time-frequency domain characteristics of the CSI data s2 to be detected and the sample CSI data of each gesture by using a formula II
Figure GDA0002878160840000035
Figure GDA0002878160840000036
Where S2 denotes the interference CSI data to be detected, SjSample CSI data representing the jth gesture,
Figure GDA0002878160840000041
denotes the similarity of s2 with the j gesture, λk(Sj) Denotes SjThe kth time-frequency domain statistical characteristic of (1), wherein eta and gamma are weighting coefficients and are both larger than zero, gamma is the weighting coefficient for the cross-correlation coefficient characteristic, and eta is the weighting coefficient for the other time-frequency domain characteristics;
similarity between interference CSI data to be detected and sample CSI data of jth gesture
Figure GDA0002878160840000042
And if the detected CSI data is the smallest, the CSI data to be detected is identified as the jth gesture.
The invention has the following beneficial effects:
1. when the indoor gesture recognition based on WiFi is processed, interference generated by actions of other people is considered, the CSI amplitude information is converted into frequency domain information by using the Fourier transform theory, and then a phase-based detection scheme is provided to determine whether interference behaviors exist.
2. When interference exists, the invention provides a gesture recognition method based on weighted cross-correlation coefficient characteristics, which utilizes a k-NN classification method, takes the cross-correlation coefficient as one characteristic, and extracts values for different characteristics of gesture action waveforms to match corresponding weights so as to further improve the classification performance.
3. The gesture recognition system provided by the invention can obtain good detection performance no matter whether interference exists or not, and compared with the existing classification scheme, the gesture recognition system provided by the invention has the advantage that the recognition accuracy can be improved to more than 78% under the condition of the interference.
Drawings
FIG. 1 is a flow diagram of gesture recognition of the present invention;
FIG. 2 is a schematic diagram of the rate of gesture recognition for pushing and pulling a target while interfering with activities of daily activities;
FIG. 3 is a PDF curve of sample variance with/without interference;
FIG. 4 is a schematic view of the apparatus setup;
FIG. 5 is a diagram illustrating the recognition rate of target behavior under the influence of interference at different locations (near the transmitting end; near the receiving end; on the intermediate link);
FIG. 6 is an interference detection rate;
FIG. 7 is a graph of recognition rate versus feature weighting;
FIGS. 8(a) and (b) are graphs comparing performance of k-NN-based algorithm classification method and reference method against target behavior recognition rate;
FIG. 9 is a graph comparing the recognition rate of the algorithm of the present patent with that of a conventional algorithm in the presence of interference;
FIG. 10 is a graph of the impact of interfering activity intensity on activity recognition performance;
FIGS. 11(a) and (b) are graphs comparing the performance of the patented method and the referenced method in identifying the rate of target behavior when the interferer is doing random daily behavior and the target is doing a specific action;
fig. 12 is a CSI differential amplitude waveform diagram acquired in a real scene.
Detailed Description
An interference detection method based on WiFi signals comprises the following steps:
step 1: acquiring CSI data by using WiFi equipment, and preprocessing amplitude information of the CSI data to obtain the CSI data containing gesture information;
step 2: performing signal analysis on the CSI data containing the gesture information obtained in the step 1 to obtain time domain subcarrier data;
and step 3: converting the subcarrier data of the time domain obtained in the step (2) into a frequency domain, and carrying out phase calibration;
and 4, step 4: performing normalization processing on the phase calibration result of the CSI data obtained in the step (3), calculating the variance of the phase sample, and if the variance of the phase sample is greater than or equal to a threshold value, judging that the CSI data has interference; otherwise, the CSI data is interference free.
Specifically, step 1 specifically includes the following substeps:
step 1.1: acquiring CSI data by utilizing WiFi equipment, and preprocessing amplitude information of the acquired CSI data, wherein the preprocessing comprises data denoising processing and signal extraction processing, and since the data frequency containing gesture information is mainly in a low frequency band, denoising is performed by utilizing a Low Pass Filter (LPF);
step 1.2: and (3) performing signal extraction processing on the de-noised CSI data in the step 1.1, and keeping the data containing the gesture signals and deleting the residual data through the signal extraction processing because the collected CSI comprises data samples containing the gesture signals or data samples without the gesture signals.
Preferably, in step 1.1, the Low Pass Filter (LPF) is a Butterworth low pass filter with a sampling rate set to 1ms and passband and stopband edge frequencies set to 0.008 π and 0.04 π rad/s, respectively.
Preferably, in step 1.2, performing signal extraction processing on the de-noised CSI data in step 1.1 specifically means determining a starting point M and an end point M of an area range where a signal data sample is located according to formulas 1 and 2 to obtain data affected by target activity:
Figure GDA0002878160840000061
Figure GDA0002878160840000062
where Nw + is represented as the length of the time window used to determine the start and end regions of the data time sample and Nw-is represented as the time window used to determine the particular start and end. Wherein the minimum value of k satisfying formula 3 is counted as ksSatisfy formula 4The maximum value of k is denoted as keFormula 3 is represented as:
Figure GDA0002878160840000063
formula 4 is represented as:
Figure GDA0002878160840000064
representing a threshold value.
Specifically, step 2 includes the following substeps:
performing signal analysis expression on the preprocessed CSI data obtained in the step 1 by using a formula 5 to obtain time domain subcarrier data d (t):
d(t)＝s(t)+sgn(t)i(t)+z(t)
Figure GDA0002878160840000065
Figure GDA0002878160840000071
where z (t) is random white Gaussian noise with a mean value of zero, sgn (t) is a sign function which can be written as i (t) is dynamic interference caused by daily behaviors of other people in the same environment, blAnd vlRespectively the amplitude and angular frequency, phi, of the ith sinusoidal componentlIs the initial phase information of the ith sinusoidal component, assuming it follows a uniform distribution between 0 and 2 pi, phil～U[0，2π)。
Specifically, in step 3, transforming the time domain subcarrier data obtained in step 2 to the frequency domain, and performing phase calibration specifically includes:
performing M-point Discrete Fourier Transform (DFT), θ, on subcarrier data in time domain(m,n)Indicating the phase of the mth frequency component of the nth subcarrier after the subcarrier data has undergone DFT, to reduce the negative effect of the uncertainty offset, on θ(m,n)Performing phase calibration by using formula 6 to obtain
Figure GDA0002878160840000072
Figure GDA0002878160840000073
Wherein alpha ismAn average expression representing the phase of the mth frequency component of the CSI amplitude signal obtained on the first subcarrier and the phase of the mth frequency component of the CSI amplitude signal obtained on the thirtieth subcarrier is:
Figure GDA0002878160840000074
βmthe mean value of the mth frequency phase of the CSI amplitude signal obtained over thirty subcarriers after DFT conversion is represented by the following expression:
Figure GDA0002878160840000075
representing the individual subcarriers defined by the 802.11 protocol in a WiFi wireless network.
Specifically, step 4 includes the following substeps:
step 4.1: in order to reduce the phase difference of the measurement data under different environments, the normalized phase obtained by normalizing the phase after phase calibration by using formula 7 is obtained
Figure GDA0002878160840000076
Figure GDA0002878160840000077
Step 4.2: calculating sampling phase variance sigma of WiFi subcarrier signal by formula 82：
Figure GDA0002878160840000081
Wherein the content of the first and second substances,
Figure GDA0002878160840000082
representing the nth subcarrierThe phase variance expression of the wave measurement data is:
Figure GDA0002878160840000083
Figure GDA0002878160840000084
representing the phase samples carrying measurement data for the nth subcarrier, the expression is:
Figure GDA0002878160840000085
step 4.3: when the sample variance is greater than or equal to the threshold, interference exists in the CSI data, and when the sample variance is less than the threshold, the interference has less influence on the sample data, and it can be assumed that the CSI data does not have interference at this time.
Preferably, the threshold selected in step 4.3 is 0.1.
A gesture recognition method based on WiFi signals comprises the following steps:
step 1: setting a group of non-interference sample CSI data, and respectively calculating time-frequency domain characteristics of the sample CSI data corresponding to different gestures;
step 2: by adopting the interference detection method based on the WiFi signal in claim 1, when the CSI data to be detected has no interference, the step 3 is executed, and when the CSI data to be detected has interference, the step 4 is executed;
and step 3: calculating time-frequency domain statistical characteristics of the CSI data to be detected and the sample CSI data, respectively comparing the similarity of the time-frequency domain characteristics of each gesture of the CSI data to be detected and the sample CSI data based on a k-NN method, and identifying the gesture corresponding to the CSI data to be detected;
and 4, step 4: and calculating time-frequency domain statistical characteristics of the to-be-detected CSI data and the sample CSI data, wherein the to-be-detected CSI data and the sample CSI data contain cross correlation coefficient characteristics, respectively comparing the similarity of the time-frequency domain statistical characteristics of each gesture of the to-be-detected CSI data and the sample CSI data, which contain the cross correlation coefficient characteristics, based on a k-NN method, and identifying the gesture corresponding to the to-be-detected CSI data.
Specifically, in step 3 of the gesture recognition method based on the WiFi signal, when the gesture recognition method is performedWhen the CSI data to be detected has no interference, calculating the time-frequency domain statistical characteristics of the CSI data s1 to be detected to form a characteristic vector: [ lambda ]1(s1),λ2(s1),λ3(s1),…λk(s1)…,λK(s1)]，λk(s1) denotes the kth time-frequency domain statistic feature of s1, K ∈ [1, K ∈]And K is a positive integer, and the similarity of the time-frequency domain characteristics of the CSI data to be detected and the sample CSI data of each gesture is calculated by using the formula I
Figure GDA0002878160840000099
Figure GDA0002878160840000091
Wherein S1 represents the CSI data to be detected without interference, SjSample CSI data representing the jth gesture,
Figure GDA0002878160840000092
denotes the similarity of s1 with the j gesture, λk(Sj) Denotes SjThe kth time-frequency domain statistical characteristic of (1);
similarity between the CSI data to be detected and the sample CSI data of the jth gesture if no interference exists
Figure GDA00028781608400000910
And if the detected CSI data is the smallest, the CSI data to be detected is identified as the jth gesture.
Specifically, in step 4 of the gesture recognition method based on the WiFi signal, when interference exists in the CSI data to be detected, the time-frequency domain statistical characteristics of the CSI data s2 to be detected are calculated to form a feature vector:
Figure GDA0002878160840000093
λk(s2) denotes the kth time-frequency domain statistic feature of s2, K ∈ [1, K ∈]And K is a positive integer, λj CR(s2) representing the cross-correlation coefficient characteristics between s2 and sample CSI data, and averaging the cross-correlation coefficients
Figure GDA0002878160840000094
As a new characteristic value of the sample CSI data, calculating the similarity of the time-frequency domain characteristics of the to-be-detected CSI data s2 and the sample CSI data of each gesture by using a formula II
Figure GDA0002878160840000095
Figure GDA0002878160840000096
Where S2 denotes the interference CSI data to be detected, SjSample CSI data representing the jth gesture,
Figure GDA0002878160840000097
denotes the similarity of s2 with the j gesture, λk(Sj) Denotes SjThe kth time-frequency domain statistical characteristic of (1), wherein eta and gamma are weighting coefficients which are both larger than zero, gamma is the weighting coefficient for the cross-correlation coefficient characteristic, eta is the weighting coefficient for the other time-frequency domain characteristics, and the setting of the weighting coefficients can ensure that the calculation accuracy is higher;
similarity between interference CSI data to be detected and sample CSI data of jth gesture
Figure GDA0002878160840000098
And if the similarity of the CSI data to be detected and the sample CSI data of a certain gesture is higher, the calculated similarity is smaller, so that the CSI data to be detected is related to the action with the small similarity.
The following embodiments of the present invention are provided, and it should be noted that the present invention is not limited to the following embodiments, and all equivalent changes based on the technical solutions of the present invention are within the protection scope of the present invention.
Example (b):
more than 5500 times of measurement data collected from 10 target persons and interference sources are collected and identified by applying the proposed identification method, and the performance of the method is evaluated. Experimental evaluation was performed in both the laboratory and library scenarios, with each target person doing three different actions, with the surrounding others performing random daily actions, and CSI being used as a measure of the wireless channel.
The experiment was performed in a real environment with commercial 5GHz WiFi equipment using a TP-link wireless router with three antennas as the transmitter and a mini PC with an Intel 5300 NIC as the receiver, with the receiver also configured with three antennas. The CSI on the 30 OFDM subcarriers at the receiver is captured using an open source CSI tool, with a packet rate of 1000 per second for the router at the transmitting end. The distance from the transmitter to the receiver is 1.2 meters. Both the transmitter and receiver are located at a height of 0.9 meters as shown in fig. 4. The target is closer to the recognition performance of the communication link than the interferer, as shown in table 1, the recognition rate of the traditional recognition rate VS weighting algorithm for different single gestures (threshold 0.1)
TABLE 1
Push-pull Waving hand Kicking leg
Tradition recognition rate 22％ 83％ 80％
Weighted calculationRate of law recognition 88％ 90％ 82％
In the extreme case where the interference detection threshold is set to 0, as shown in table 2, the recognition accuracy of the method we propose for three types of gestures can reach almost 100%.
Table 2 recognition rate of traditional single-person gesture recognition rate VS weighting algorithm (threshold 0)
Push-pull Waving hand Kicking leg
Tradition recognition rate 22％ 83％ 80％
Weighted algorithm recognition rate 88％ 89％ 97％
Table 3 shows the recognition rates of the conventional recognition rate VS weighting algorithm for different gestures of multiple persons
(number of people: 10, total amount of samples: 1500, threshold 0.1)
Push-pull Waving hand Kicking leg
Tradition recognition rate 42％ 57％ 28％
Weighted algorithm recognition rate 94％ 81％ 95％
Table 4 recognition rate of traditional recognition rate VS weighting algorithm for different gestures under different interferences (threshold 0.1)
Push-pull Waving hand Kicking leg
Tradition recognition rate 22％ 67％ 71％
Weighted algorithm recognition rate 88％ 96.7％ 75％
Firstly, only collecting CSI of a target acting, then collecting the CSI under interference, and finally collecting the CSI of a target person acting under the interference. Fig. 12 shows a plot of the collected CSI amplitude. In the figure, a first portion of the waveform is caused by a gesture of the target, a second portion of the waveform is caused by the disturbance, and a third portion of the waveform is caused by both the target and the disturbance. Tables 1-5 compare the activity recognition performance of our method and previous methods. As can be seen from the figure, the activity recognition rates of the proposed method are 79%, 84% and 89% for gesture push-pull, hand swing and leg kick, respectively. Compared with the reference method, the recognition rate of the three gesture methods is improved by 32%, 8% and 7% respectively, as shown in table 5.
TABLE 5 recognition rates of the traditional recognition rate VS-weighted algorithm for different gestures under multiple disturbances
(threshold 0.1, place: library)
Push-pull Waving hand Kicking leg
Tradition recognition rate 47％ 76％ 81％
Weighted algorithm recognition rate 79％ 84％ 89％
TABLE 6 measured values of interference detection rate
Detection rate Detecting the presence of interference Detection without interference
With interfering data 84％ 16％
Non-interfering data 18％ 82％
We used 50 sets of noisy push-pull data and performed cross-correlations between these push-pull data and the training data set, including push-pull, hand-swing and leg-kicking behaviors. According to the experimental result, the cross correlation coefficient corresponding to 93% of the push-pull training data is above 0.86, and the average cross correlation coefficient is 0.93; the cross correlation coefficient corresponding to only 6% of hand swing training data is above 0.86, and the average cross correlation coefficient is 0.65; the cross correlation coefficient corresponding to 9% of the training data on kicking behaviour was above 0.86, with an average cross correlation coefficient of 0.65. Finally, we use the three average cross-correlation coefficients as correlation coefficient elements of the feature vector. Specifically, the elements of the feature vector corresponding to the cross correlation coefficients of the push-pull training data, the swing training data, and the kick training data are [0.93,0.65,0.65], [0.65,0.93,0.65], and [0.65,0.65,0.93], as shown in table 7, respectively.
TABLE 7 detection rates for different postures
Figure GDA0002878160840000121
Figure GDA0002878160840000131
In summary, the recognition accuracy of the method provided by the invention for the three types of gestures is respectively superior to that of the existing method.

Claims (3)

1. A gesture recognition method based on WiFi signals is characterized by comprising the following steps:
step 1: setting a group of non-interference sample CSI data containing gesture information, and respectively calculating time-frequency domain characteristics of the sample CSI data under different gestures;
step 2: judging the CSI data to be detected by adopting an interference detection method based on WiFi signals, executing the step 3 when the interference does not exist in the CSI data to be detected, and executing the step 4 when the interference exists in the CSI data to be detected;
and step 3: calculating time-frequency domain statistical characteristics of the CSI data to be detected and the sample CSI data, respectively comparing the similarity of the time-frequency domain characteristics of each gesture of the CSI data to be detected and the sample CSI data based on a k-NN method, and identifying the gesture corresponding to the CSI data to be detected;
and 4, step 4: calculating time-frequency domain statistical characteristics of the to-be-detected CSI data and the sample CSI data, wherein the to-be-detected CSI data and the sample CSI data contain cross correlation coefficient characteristics, respectively comparing the similarity of the time-frequency domain statistical characteristics of each gesture of the to-be-detected CSI data and the sample CSI data, which contain the cross correlation coefficient characteristics, based on a k-NN method, and identifying the gesture corresponding to the to-be-detected CSI data;
the interference detection method based on the WiFi signal comprises the following steps:
step a: acquiring CSI data by using WiFi equipment, and preprocessing amplitude information of the CSI data to obtain the CSI data containing gesture information;
step b: b, performing signal analysis on the CSI data containing the gesture information obtained in the step a to obtain time domain subcarrier data;
step c: b, converting the subcarrier data of the time domain obtained in the step b into a frequency domain, and carrying out phase calibration;
step d: when dynamic interference exists, performing normalization processing on the phase calibration result of the CSI data obtained in the step c, calculating the variance of the phase sample, and if the variance of the phase sample is larger than or equal to a threshold value, the CSI data has interference; otherwise, the CSI data is interference free.
2. The WiFi signal based gesture recognition method of claim 1 characterized in that step 3 comprises the following sub-steps:
when the CSI data to be detected has no interference, calculating the time-frequency domain statistical characteristics of the CSI data s1 to be detected to form a characteristic vector: [ lambda ]1(s1),λ2(s1),λ3(s1),…λk(s1)…,λK(s1)]，λk(s1) denotes the kth time-frequency domain statistic feature of s1, K ∈ [1, K ∈]And K is a positive integer, and the similarity of the time-frequency domain characteristics of the CSI data to be detected and the sample CSI data of each gesture is calculated by using the formula I
Figure FDA0002850380910000021
Figure FDA0002850380910000022
Wherein S1 represents the CSI data to be detected without interference, SjSample CSI data representing the jth gesture,
Figure FDA0002850380910000023
denotes the similarity of s1 with the j gesture, λk(Sj) Denotes SjThe kth time-frequency domain statistical characteristic of (1);
similarity between the CSI data to be detected and the sample CSI data of the jth gesture if no interference exists
Figure FDA0002850380910000024
And if the detected CSI data is the smallest, the CSI data to be detected is identified as the jth gesture.
3. The WiFi signal based gesture recognition method of claim 1 characterized in that step 4 comprises the following sub-steps:
calculating the time-frequency domain statistical characteristics of the CSI data s2 to be detected to form a characteristic vector:
Figure FDA0002850380910000027
λk(s2) denotes the kth time-frequency domain statistic feature of s2, k ∈ [1 ],K]And K is a positive integer, λj CR(s2) represents the cross-correlation coefficient characteristic, λ, between s2 and sample CSI dataj CR(Sj) Representing the mean value of the cross-correlation coefficients, and calculating the similarity of the time-frequency domain characteristics of the CSI data s2 to be detected and the sample CSI data of each gesture by using a formula II
Figure FDA0002850380910000025
Figure FDA0002850380910000026
Where S2 denotes the interference CSI data to be detected, SjSample CSI data representing the jth gesture,
Figure FDA0002850380910000031
denotes the similarity of s2 with the j gesture, λk(Sj) Denotes SjThe kth time-frequency domain statistical characteristic of (1), wherein eta and gamma are weighting coefficients and are both larger than zero, gamma is the weighting coefficient for the cross-correlation coefficient characteristic, and eta is the weighting coefficient for the other time-frequency domain characteristics;
similarity between interference CSI data to be detected and sample CSI data of jth gesture
Figure FDA0002850380910000032
And if the detected CSI data is the smallest, the CSI data to be detected is identified as the jth gesture.
CN201811630072.9A 2018-12-29 2018-12-29 Interference detection and gesture recognition method based on WiFi signal Active CN109768838B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811630072.9A CN109768838B (en) 2018-12-29 2018-12-29 Interference detection and gesture recognition method based on WiFi signal

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811630072.9A CN109768838B (en) 2018-12-29 2018-12-29 Interference detection and gesture recognition method based on WiFi signal

Publications (2)

Publication Number Publication Date
CN109768838A CN109768838A (en) 2019-05-17
CN109768838B true CN109768838B (en) 2021-04-13

Family

ID=66452403

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811630072.9A Active CN109768838B (en) 2018-12-29 2018-12-29 Interference detection and gesture recognition method based on WiFi signal

Country Status (1)

Country Link
CN (1) CN109768838B (en)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112731286B (en) * 2020-12-29 2023-01-06 浙江毅星科技有限公司 Passive positioning method for WiFi interference source based on AP antenna array direction finding
CN115243308A (en) * 2021-04-23 2022-10-25 成都极米科技股份有限公司 Signal measurement method, device, system, terminal and network equipment
CN113625882B (en) * 2021-10-12 2022-06-14 四川大学 Myoelectric gesture recognition method based on sparse multichannel correlation characteristics
CN114764580A (en) * 2022-06-15 2022-07-19 湖南工商大学 Real-time human body gesture recognition method based on no-wearing equipment

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105979485A (en) * 2016-05-11 2016-09-28 南京邮电大学 Personnel detection method in indoor environment based on channel state information (CSI)
CN106792808A (en) * 2016-12-08 2017-05-31 南京邮电大学 Los path recognition methods under a kind of indoor environment based on channel condition information

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102198003B (en) * 2011-06-07 2014-08-13 嘉兴恒怡科技有限公司 Limb movement detection and evaluation network system and method
CN105933080B (en) * 2016-01-20 2020-11-03 北京大学 Fall detection method and system
CN107994960B (en) * 2017-11-06 2020-11-27 北京大学（天津滨海）新一代信息技术研究院 Indoor activity detection method and system
CN107968689B (en) * 2017-12-06 2020-07-31 北京邮电大学 Perception identification method and device based on wireless communication signals

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105979485A (en) * 2016-05-11 2016-09-28 南京邮电大学 Personnel detection method in indoor environment based on channel state information (CSI)
CN106792808A (en) * 2016-12-08 2017-05-31 南京邮电大学 Los path recognition methods under a kind of indoor environment based on channel condition information

Also Published As

Publication number Publication date
CN109768838A (en) 2019-05-17

Similar Documents

Publication Publication Date Title
CN109768838B (en) Interference detection and gesture recognition method based on WiFi signal
Yousefi et al. A survey on behavior recognition using WiFi channel state information
CN106330385B (en) A kind of interference type recognition methods
CN105844216B (en) Detection and matching mechanism for recognizing handwritten letters by WiFi signals
WO2020103411A1 (en) Fall detection method, device, and system
CN105678273B (en) The starting point detection algorithm of radio-frequency fingerprint identification technology transient signal
JP7150292B2 (en) Action recognition system and action recognition method
CN105429719B (en) Based on power spectrum and multi-scale wavelet transformation analysis high reject signal detection method
CN112560803A (en) Radar signal modulation identification method based on time-frequency analysis and machine learning
CN112333653A (en) Identity intelligent identification method and system based on WiFi channel state information
CN108919203B (en) Radar active interference identification method and system
CN109657572B (en) Wi-Fi-based behind-wall target behavior identification method
CN112861813B (en) Method for identifying human behavior behind wall based on complex value convolution neural network
CN111901028B (en) Human body behavior identification method based on CSI (channel State information) on multiple antennas
CN112381063A (en) Channel state information-based people counting method
CN106100762A (en) A kind of weak signal of communication detection method of cyclo-stationary analysis of spectrum
CN112613443A (en) Robustness communication radiation source intelligent identification method based on deep learning
CN113453180B (en) Intelligent detection method and system for human body tumble and information data processing terminal
Gillespie et al. Optimization of time and frequency resolution for radar transmitter identification
Yousefi et al. A survey of human activity recognition using wifi CSI
CN110289926A (en) Frequency spectrum sensing method based on modulated signal Cyclic Autocorrelation Function asymmetric peak
CN113743374A (en) Personnel identity identification method based on channel state information respiration perception
CN117169346A (en) High-altitude building damage identification method based on wavelet packet energy spectrum analysis
CN110621038B (en) Method and device for realizing multi-user identity recognition based on WiFi signal detection gait
CN108718223B (en) Blind spectrum sensing method for non-cooperative signals

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