Gait feature construction method based on wireless signals

Abstract

The invention discloses a gait feature construction method based on wireless signals. The gait feature construction method comprises the steps that frequency modulation continuous wave signals, collected by radar receiving and transmitting equipment, of multiple users are acquired; performing high-pass filtering processing on each frequency-modulated continuous wave signal to obtain a corresponding filtered Doppler spectrogram; gait discretization processing is carried out to obtain a gait signal scatter diagram; sequentially carrying out regression processing and smooth filtering processing on the gait signal scatter diagram to obtain a gait characteristic curve; determining a periodic wave crest in the gait characteristic curve to obtain a segmentation starting point, and performing gait segmentation on the gait characteristic curve by using the segmentation starting point and a fixed time window to obtain a plurality of gait sub-spectrograms segmented from the frequency-modulated continuous wave signal; the gait sub-spectrograms segmented from all the frequency modulated continuous wave signals are used for training preset models, and the preset models comprise an identity recognition model and a gait anomaly recognition model. According to the method, the gait time-varying problem occurring in the Doppler spectrogram construction process can be overcome, and the recognition accuracy and robustness can be improved.

Description

Technical Field

[0001] The invention belongs to the field of gait feature recognition, and in particular relates to a gait feature construction method based on wireless signals. Background Art

[0002] Gait feature construction is a key technical issue in the field of gait recognition. Gait features can be used to achieve identity recognition and abnormal gait diagnosis. At present, identity recognition mainly adopts close-range active recognition methods such as fingerprint recognition, face recognition and voice recognition; abnormal gait diagnosis usually requires the use of wearable devices for data collection, and patients may not be suitable for wearing them for a long time. In comparison, the gait Doppler spectrogram generated based on millimeter wave signals can achieve long-distance, contactless identity recognition without the need for additional cooperation from the user. However, as the distance between the target and the transceiver increases, the RCS (Radar cross-section) of the target decreases, resulting in energy attenuation in the corresponding area in the Doppler spectrum, such as Figure 1 In addition, in the process of gait feature construction, if the time sliding window segmentation method is used, the sample features will be different. The results of the time sliding window segmentation method are shown in Figure 1 As shown in Figure (b) in the figure. The sliding window segmentation method will destroy the feature alignment of the sample data, causing the local features of the sub-spectrogram to no longer correspond to the same gait phase, which will have a great impact on the sample distribution and thus reduce the classification effect of the model. Based on the above situation, it can be seen that the original gait Doppler spectrogram cannot provide consistent recognition performance in all scenarios and cannot meet the needs of gait-based identity recognition.

[0003] Existing gait recognition methods improve the robustness of network models by constructing large and diverse data in different directions and positions for training. However, these methods require a lot of manpower to collect data. In addition, some methods segment Doppler spectrograms based on speed in the hope of obtaining gait sub-spectrograms with unified features. However, if the extracted speed of this method cannot correctly reflect the changes in leg speed, using this speed for footstep segmentation will lead to erroneous results. Summary of the invention

[0004] In order to solve the above problems existing in the prior art, the present invention provides a gait feature construction method based on wireless signals, a training method for an identity recognition model, an identity recognition method, a training method for an abnormal gait recognition model, and a gait abnormality recognition method. The technical problem to be solved by the present invention is achieved through the following technical solutions:

[0005] In a first aspect, an embodiment of the present invention provides a method for constructing gait features based on wireless signals, the method for constructing gait features based on wireless signals comprising:

[0006] Acquire frequency modulated continuous wave signals of multiple users collected by radar transceiver equipment;

[0007] For each frequency modulated continuous wave signal, based on high-pass filtering, a filtered Doppler spectrum corresponding to the frequency modulated continuous wave signal is obtained;

[0008] Performing gait discretization processing on the filtered Doppler spectrum to obtain a gait signal scatter plot;

[0009] The gait signal scatter plot is subjected to regression processing and smoothing filtering processing in sequence to obtain a gait characteristic curve;

[0010] Determine the periodic peaks in the gait characteristic curve, obtain the segmentation starting point, and use the segmentation starting point and the fixed time window to perform gait segmentation on the gait characteristic curve to obtain multiple gait sub-spectrograms segmented from the frequency modulated continuous wave signal; wherein all the gait sub-spectrograms segmented from the frequency modulated continuous wave signal are used to train a preset model, and the preset model includes an identity recognition model and a gait abnormality recognition model.

[0011] In one embodiment of the present invention, the step of obtaining a filtered Doppler spectrum corresponding to each frequency modulated continuous wave signal based on high-pass filtering includes:

[0012] For each frequency modulated continuous wave signal, a distance dimension fast Fourier transform is performed on it to obtain a distance spectrum of the frequency modulated continuous wave signal;

[0013] Performing high-pass filtering on the distance spectrum;

[0014] The range spectrum after high-pass filtering is subjected to a velocity-dimensional fast Fourier transform to obtain a filtered Doppler spectrum corresponding to the frequency modulated continuous wave signal.

[0015] In one embodiment of the present invention, the filtered Doppler spectrogram is subjected to gait discretization processing to obtain a gait signal scatter plot, including:

[0016] Energy points below a preset energy threshold in the filtered Doppler spectrum are deleted to obtain a gait signal scatter plot.

[0017] In one embodiment of the present invention, determining the periodic peaks in the gait characteristic curve to obtain the segmentation starting point includes:

[0018] Derivative analysis is used to derive the gait characteristic curve to locate the peak position;

[0019] Based on the peak position analysis, a periodic peak is obtained as a segmentation starting point.

[0020] In one embodiment of the present invention, obtaining a periodic peak based on the peak position analysis as a segmentation starting point includes:

[0021] Based on the peak position, the peak period comparison method is used to screen out the periodic peak as the segmentation starting point.

[0022] In one embodiment of the present invention, obtaining a periodic peak based on the peak position analysis as a segmentation starting point includes:

[0023] An autocorrelation analysis is performed on the gait characteristic curve to determine a correct period, and the correct period is used to filter out periodic peaks from the peak positions as segmentation starting points.

[0024] In a second aspect, an embodiment of the present invention provides a method for training an identity recognition model, the method for training an identity recognition model comprising:

[0025] Obtain a training set of an identity recognition model; wherein the training set of the identity recognition model includes training data of multiple users, and the training data of each user is a plurality of gait sub-spectrograms segmented from the frequency modulated continuous wave signal of the user, which is obtained according to the gait feature construction method based on wireless signals described in the first aspect, and each gait sub-spectrogram is marked with user information as label data;

[0026] The training set of the identity recognition model is used to train a preset neural network model to obtain a trained identity recognition model.

[0027] In a third aspect, an embodiment of the present invention provides an identity recognition method, the identity recognition method comprising:

[0028] Obtaining a gait sub-spectrogram of a user to be identified;

[0029] The gait sub-spectrogram of the user identity to be identified is input into a pre-trained identity recognition model to obtain corresponding user identity information; wherein the identity recognition model is obtained according to the identity recognition model training method described in the second aspect.

[0030] In a fourth aspect, an embodiment of the present invention provides a method for training a gait abnormality recognition model, the method for training a gait abnormality recognition model comprising:

[0031] Obtain a training set of a gait abnormality recognition model; wherein the training set of the gait abnormality recognition model includes training data of multiple users, and the training data of each user is a plurality of gait sub-spectrograms segmented from the frequency modulated continuous wave signal of the user, which is obtained according to the gait feature construction method based on wireless signals described in the first aspect, and each gait sub-spectrogram is marked with label data to indicate whether the gait is abnormal;

[0032] The training set of the gait abnormality recognition model is used to train a preset neural network model to obtain a trained gait abnormality recognition model.

[0033] In a fifth aspect, an embodiment of the present invention provides a method for identifying abnormal gait, the method comprising:

[0034] Obtaining a gait sub-spectrum to be recognized;

[0035] The gait sub-spectrum to be recognized is input into a pre-trained gait abnormality recognition model to obtain a gait recognition result, wherein the gait recognition result includes a normal gait or an abnormal gait; wherein the gait abnormality recognition model is obtained according to the training method of the gait abnormality recognition model described in the fourth aspect.

[0036] Beneficial effects of the present invention:

[0037] The gait feature construction method based on wireless signals provided in the embodiment of the present invention first extracts and restores gait information, and then uses the regularity of gait to effectively segment the Doppler spectrogram, thereby obtaining gait sub-spectrograms with consistent gait features and similar energy intensities. The gait time-varying problem that occurs during the construction of the Doppler spectrogram can be overcome, and these sub-spectrograms can be used to train identity recognition models and gait abnormality recognition models, which can improve the accuracy and robustness of recognition, so that the identity recognition model and gait abnormality recognition model implemented on this basis have higher recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a schematic diagram of random segmentation of gait Doppler spectrogram in the prior art, wherein (a) is a gait Doppler spectrogram, and (b) is a gait sub-spectrum;

[0039] Figure 2 It is a flow chart of a method for constructing gait features based on wireless signals provided by an embodiment of the present invention;

[0040] Figure 3 is an unfiltered Doppler spectrum of a frequency modulated continuous wave signal in an embodiment of the present invention;

[0041] Figure 4 Result diagrams of the distance spectrum before and after high-pass filtering in an embodiment of the present invention;

[0042] Figure 5 In the embodiment of the present invention Figure 3 The corresponding Doppler spectrum after filtering;

[0043] Figure 6 In the embodiment of the present invention Figure 5 Corresponding gait signal scatter plot;

[0044] Figure 7 is the result of performing regression processing on the gait signal scatter plot in the embodiment of the present invention;

[0045] Figure 8 is a schematic diagram of a gait characteristic curve in an embodiment of the present invention;

[0046] Fig. 9 It is a flowchart of a training method for an identity recognition model provided by an embodiment of the present invention;

[0047] Fig.10 It is a schematic diagram of a specific process of the training method of the identity recognition model provided by an embodiment of the present invention;

[0048] Fig.11 is a schematic diagram of a segmentation window and a segmented gait sub-spectrogram in an embodiment of the present invention;

[0049] Fig.12 is a structural diagram of an identity recognition model in an embodiment of the present invention;

[0050] Fig.13 It is a flowchart of an identity recognition method provided by an embodiment of the present invention;

[0051] Fig.14 It is a flowchart of a training method for an abnormal gait recognition model provided by an embodiment of the present invention;

[0052] Fig.15 It is a schematic diagram of a specific process of a training method for an abnormal gait recognition model provided in an embodiment of the present invention;

[0053] Fig.16 is a gait characteristic curve diagram of hemiplegic gait in an embodiment of the present invention;

[0054] Fig.17 1 is a schematic diagram of gait periodicity analysis in an embodiment of the present invention, wherein (a) is an autocorrelation analysis result of a normal gait, and (b) is an autocorrelation analysis result of a hemiplegic gait;

[0055] Fig.18It is a comparison diagram of gait segmentation results in an embodiment of the present invention. Among them, Figure (a) is a normal gait spectrogram, Figure (b) is a hemiplegic gait spectrogram, Figure (c) is a sub-spectrogram obtained after normal gait segmentation, and Figure (d) is a gait sub-spectrogram segmented after periodic analysis;

[0056] Fig.19 It is a schematic structural diagram of a gait abnormality recognition model in an embodiment of the present invention;

[0057] Fig. 20 It is a schematic flowchart of a gait abnormality recognition method provided by an embodiment of the present invention. Specific embodiments

[0058] The following further describes the present invention in detail with reference to specific embodiments, but the embodiments of the present invention are not limited thereto.

[0059] In order to overcome the problem of gait time-variation in the construction process of Doppler spectrograms, an embodiment of the present invention proposes a method for constructing gait features based on wireless signals, and on this basis provides a training method for an identity recognition model and an identity recognition method, which can improve the accuracy of user recognition; and on the basis of the proposed method for constructing gait features based on wireless signals, a training method for a gait abnormality recognition model and a gait abnormality recognition method are provided, which can achieve accurate diagnosis of abnormal gaits.

[0060] In the first aspect, an embodiment of the present invention provides a method for constructing gait features based on wireless signals, as Figure 2 shown, the method may include the following steps:

[0061] S1. Obtain the frequency-modulated continuous wave signals of multiple users collected by a radar transceiver device;

[0062] An embodiment of the present invention can use any radar transceiver device to receive the frequency-modulated continuous wave signals (i.e., FMCW signals) from users. For example, in an optional embodiment, the radar transceiver device can be a Texas Instrument AWR1443 FMCW radar transceiver device, which operates at 77 GHz and has a bandwidth set to 3.4 GHz, and can provide a distance resolution of 4.41 cm and a maximum detectable distance of 16 m.

[0063] S2. For each frequency-modulated continuous wave signal, obtain the filtered Doppler spectrogram corresponding to the frequency-modulated continuous wave signal based on high-pass filtering processing;

[0064] In addition to human motion information, the FMCW signal received by the radar also contains return signals from static objects in the environment, such as walls, ground, and other obstacles. These static information appear in the Doppler spectrum as low-frequency components with zero velocity and extremely high energy (see Figure 3 ), these low-frequency information will suppress the expression of dynamic components. In addition, when the target object is far away from the radar, the signal returned by the dynamic object becomes weaker and weaker due to the attenuation of signal strength, making it difficult to effectively extract useful gait information.

[0065] In order to solve the above problems, the present invention adopts a filter to suppress the low-frequency signal, and extracts the key information of gait so as to focus only on the dynamic component generated by the gait action.

[0066] Specifically, S2 may include the following steps:

[0067] S21, performing a distance-dimensional fast Fourier transform on each frequency modulated continuous wave signal to obtain a distance spectrum of the frequency modulated continuous wave signal;

[0068] The process of performing distance-dimensional fast Fourier transform can be found in the relevant technical understanding. We will not explain it in detail here. For the distance spectrum of FMCW signal, please refer to Figure 4 shown.

[0069] S22, performing high-pass filtering on the distance spectrum;

[0070] Specifically, this step adopts a signal processing method based on Butterworth high-pass filter. The design of the high-pass filter is based on the difference equation form, and the specific operation is expressed as follows:

[0071]

[0072] Among them, b i and a i are the Butterworth filter coefficients, x[n] and y[n] are the input signal and output signal respectively; m is the Butterworth filter order (i.e., m delayed signals); for example, if m=2, the Butterworth filter will use the current input signal x[n] and the input signals x[n-1] and x[n-2] of the previous two moments, as well as the output signals y[n-1] and y[n-2] of the previous two moments to calculate the output y[n] of the current moment. If the filtering result of the fourth frequency point is calculated, then:

[0073] y[4]=b 0 x[4]+b 1 x[3]+b 2 x[2]-a 1 y[3]-a 2 y[2].

[0074] Specifically, the input data of the Butterworth high-pass filter is the data of each row in the distance spectrum obtained after the distance dimension FFT (Fast Fourier Transform). The Butterworth filter is used to filter the N time points (i.e., each column) at each distance point (i.e., each row), and the value range of n is 1 to N-1 (this is to prevent cross-border access). Through the difference equation, the low-frequency signal will be filtered out due to its greater dependence on the previous data. High-frequency signals are less affected because their frequency changes are larger and the processing effect of the difference equation is weaker.

[0075] See the results before and after high-pass filtering. Figure 4 As shown in the left and right figures, in the range spectrum, each column contains the frequency information corresponding to different distance units, reflecting the relative position of the radar signal and the target object.

[0076] The present invention effectively removes low-frequency components in the signal by performing high-pass filtering operation on the distance unit in time series, thereby achieving suppression of static targets.

[0077] S23, performing a velocity-dimensional fast Fourier transform on the range spectrum after high-pass filtering to obtain a filtered Doppler spectrum corresponding to the frequency modulated continuous wave signal.

[0078] Please refer to the relevant technical understanding for the process of performing the velocity-dimensional fast Fourier transform. We will not explain it in detail here. Figure 3 , the Doppler spectrum of the FMCW signal after filtering can be found in Figure 5 shown.

[0079] S3, performing gait discretization processing on the filtered Doppler spectrum to obtain a gait signal scatter plot;

[0080] Specifically, the step may include:

[0081] Energy points below a preset energy threshold in the filtered Doppler spectrum are deleted to obtain a gait signal scatter plot.

[0082] In order to extract the key features of gait information, the present invention adopts a discretization processing method. Since the energy of the gait information after filtering is relatively high, the gait features are extracted from the Doppler spectrum by setting an energy threshold. The preset energy threshold can be set as needed. Figure 5 , generating Figure 6 The gait signal scatter plot shown in Figure 1 intuitively shows the changing pattern of gait information in the time dimension.

[0083] S4, performing regression processing and smoothing filtering processing on the gait signal scatter plot in sequence to obtain a gait characteristic curve;

[0084] Since the number of scattered points in the gait signal scatter plot is huge and contains abundant outlier noise, in order to solve this problem, the present invention performs regression processing on the gait signal scatter plot.

[0085] In an optional implementation, the regression process can be implemented using the KNN (K nearest neighbor) algorithm. For the gait signal scatter plot, regression prediction is performed by calculating the minimum Euclidean distance between the data point to be predicted and each point in the cluster to achieve data fitting. The corresponding calculation formula is as follows:

[0086]

[0087] Among them, x is the input point to be predicted, p j is the data point in the jth time dimension of the gait signal scatter plot, q j is with p j The corresponding Doppler dimension target output, K is the number of selected neighbor points, is the predicted value corresponding to the input point x. M is the number of all samples in the data set. The meaning of the first formula in formula (2) is to traverse all M data (p j ,q j ), calculate each p j The distance from the predicted point x is used to find the point closest to the predicted point. The second formula selects the K neighboring data points p closest to the predicted point x based on the distance calculation in the first step. j1 , p j2 ,…,p jK , their corresponding target values ​​are q j1 ,q j2 , …, q jK , that is (p jk ,q jk ). To perform regression prediction on the target values ​​of these K neighboring points, it is necessary to calculate the average value of the K targets It is the final predicted value of the predicted point x.

[0088] The results after regression processing are as follows Figure 7 As shown. It can be seen that the regression results have certain fluctuations and fail to reflect the smooth trend of gait well. In order to further optimize the smoothness of the KNN regression results, the present invention performs smoothing filtering on them.

[0089] In an optional implementation, the smoothing filter processing can be implemented by using a Savitzky-Golay filter to obtain a smoother and more accurate gait characteristic curve. The relevant formula for the smoothing filter processing is as follows:

[0090]

[0091] i=1,2,...,P

[0092] Among them, y i,smooth is the data after smoothing filtering, x i+j is the data point on the KNN regression result curve h j is the coefficient of the Butterworth filter, P is the total number of prediction points generated after KNN regression, and the present invention applies a window of size w to each point in P for smoothing filtering. For example, when m=3, the window size of each smoothing process is x. i Centered on x i-1 ,x i ,x i+1 The filter performs smoothing processing on the data within a window size of w, while effectively retaining the key change features in the signal. The gait characteristic curve is as follows: Figure 8 shown.

[0093] S5, determining the periodic peaks in the gait characteristic curve, obtaining a segmentation starting point, and performing gait segmentation on the gait characteristic curve using the segmentation starting point and a fixed time window to obtain a plurality of gait sub-spectragrams segmented from the frequency modulated continuous wave signal;

[0094] Specifically, in order to obtain a gait spectrum with stable characteristics, the present invention can adopt a preset method to first determine the periodic peaks in the gait characteristic curve, that is, the peaks in the gait characteristic curve that meet the periodic characteristics, and use them as the segmentation starting point, and then use each segmentation starting point as a starting point, and use a fixed time window to perform gait segmentation on the gait characteristic curve, thereby obtaining multiple gait sub-spectrograms segmented from the frequency modulated continuous wave signal.

[0095] The present invention does not limit the preset method for determining the periodic peak, and a specific example will be given below.

[0096] The above steps of the embodiment of the present invention show how to segment a gait sub-spectrogram from a frequency modulated continuous wave signal. According to the above method, all frequency modulated continuous wave signals can segment their own gait sub-spectrograms. The gait sub-spectrograms segmented from all frequency modulated continuous wave signals are used to train a preset model, which includes an identity recognition model and a gait abnormality recognition model. The details will be explained later.

[0097] In order to extract gait information from the gait Doppler spectrogram so that the segmentation task can focus on gait information, the present invention uses a high-pass filter to remove the static components generated by stationary objects in the Doppler spectrogram, and only retains the dynamic components generated by moving objects. The gait information with the highest energy is discretized, followed by KNN regression analysis, and finally SG smoothing filtering, so as to finally obtain a gait characteristic curve that can characterize the key characteristics of gait changes. In addition, the first significant peak point of the gait is obtained by periodic analysis to construct a gait sample data set for the recognition task. After the peak point is obtained on the gait characteristic curve, it is anchored and traced back to the scattered point position, and then the position of the segmentation point in the Doppler spectrogram is determined. The gait spectrogram is segmented using a fixed time window length.

[0098] It can be seen that the gait feature construction method based on wireless signals provided in the embodiment of the present invention first extracts and restores gait information, and then uses the regularity of gait to effectively segment the Doppler spectrogram, thereby obtaining gait sub-spectrograms with consistent gait features and similar energy intensity. The problem of gait time-varying in the process of Doppler spectrogram construction can be overcome, and these sub-spectrograms can be used to train identity recognition models and gait abnormality recognition models, which can improve the accuracy and robustness of recognition.

[0099] In a second aspect, an embodiment of the present invention further provides a method for training an identity recognition model, such as Fig. 9 As shown, the training method of the identity recognition model may include the following steps:

[0100] Step A1, obtaining a training set of an identity recognition model;

[0101] The training set of the identity recognition model includes training data of multiple users, and the training data of each user is a plurality of gait sub-spectrograms segmented from the frequency modulated continuous wave signal of the user, which is obtained according to the gait feature construction method based on wireless signals described in the first aspect, and each gait sub-spectrogram is marked with user information as label data;

[0102] Step A2: train a preset neural network model using the training set of the identity recognition model to obtain a trained identity recognition model.

[0103] For the specific process of the training method of this identity recognition model, please refer to Fig.10Understand. The overall workflow is divided into a data preprocessing stage and a gait segmentation stage. In the data preprocessing stage, since static objects are usually manifested as static components with zero velocity and extremely high energy in the Doppler spectrum, these static components may inhibit the expression of gait signals. According to the foregoing, the present invention proposes a method to effectively remove static information based on the difference between dynamic and static information. In order to solve the problem of restoring dynamic information, the present invention first discretizes the gait features and extracts the gait information from the spectrum. Then, the KNN algorithm and SG filter are used to smooth the discrete information, thereby realizing the restoration of gait information. In the gait segmentation stage, firstly, based on the extracted gait feature curve, the potential gait cycle is found through gait periodicity analysis. Finally, with the help of sliding window technology and periodic stability rules, the gait Doppler spectrum is segmented to obtain a gait Doppler sub-spectrum with stable features for identity recognition model training.

[0104] Specifically, in step A1, multiple gait sub-spectrograms segmented from all frequency modulated continuous wave signals are obtained according to steps S1 to S5 of the first aspect, and multiple gait sub-spectrograms segmented from each frequency modulated continuous wave signal are used as training data for the corresponding user, and each gait sub-spectrogram is marked with user information as label data, and the user information can be the user's name, number, and other information that can indicate the user's identity. The training data of all users constitute the training set of the identity recognition model.

[0105] Wherein, for the identity recognition model, determining the periodic peaks in the gait characteristic curve in S5 to obtain the segmentation starting point may include:

[0106] Step 1), using a derivative analysis method to derive the gait characteristic curve to locate the peak position;

[0107] Step 2), based on the peak position analysis, a periodic peak is obtained as a segmentation starting point.

[0108] Wherein, step 2) may include:

[0109] Based on the peak position, the peak period comparison method is used to screen out the periodic peak as the segmentation starting point.

[0110] Specifically, in order to obtain a gait sub-spectrogram with stable characteristics, the present invention combines the derivative analysis method with the peak period comparison method to jointly determine the segmentation starting point, and segments the gait characteristic curve through a fixed time window.

[0111] Specifically, for step 1), the derivative of the gait characteristic curve is obtained by using the derivative analysis method to preliminarily screen out the positions of the peaks and troughs. Among them, the derivative analysis method is used to obtain the derivative of the curve and distinguish the positions of the peaks and troughs. Please refer to the relevant technical understanding and will not be explained here.

[0112] Regarding step 2), the present invention takes into account that the periodicity of normal gait is relatively stable, and adopts the peak-cycle comparison method to exclude abnormal peaks, thereby determining the segmentation starting point of normal gait. The specific method is to take each peak as the starting point of the cycle, and compare the length consistency of adjacent cycles backwards. Only when the current peak is similar to the length of several subsequent cycles, the peak is determined to be the correct segmentation starting point, that is, it belongs to a periodic peak. In addition, by setting a protection time band and avoiding cross-segmentation, the consistency of gait segmentation and the purity of the data are ensured. The segmentation window is as follows: Fig.11 As shown in the upper middle part of the figure, the starting column index and the ending column index of the sub-spectrogram are determined to complete the gait division. The gait sub-spectrogram obtained after the segmentation operation is as follows Fig.11 As shown in the lower middle part of the figure.

[0113] For step A2, the structure of the preset neural network model used can be any existing classification network.

[0114] In order to implement the gait-based user recognition method, the preset neural network model of the present invention can be implemented based on a convolutional neural network layer (CNN), such as Fig.12 As shown in the figure. The segmented gait sub-spectrogram is first extracted through the convolution layer, and then further compressed through the pooling layer. After being processed by the ReLU activation function, it is input into the fully connected layer to obtain a higher-level feature representation. Finally, the data enters the fully connected layer for feature extraction, and the Softmax function is used to output the identity recognition result. In order to simplify, Fig.12 The ReLU activation function and Softmax function are not shown.

[0115] in, Fig.12 In the figure, the convolutional layer (64) indicates that the convolutional layer has 64 output channels, the fully connected layer (500) indicates that there are 500 neurons, and the output of this layer is a one-dimensional vector with a shape of 500. The following are similar; the pooling layer (2×2) indicates that a window of size (2×2) is used, and each pooling will downsample the (2×2) area; the number 10 in the last fully connected layer indicates that the number of categories is 10, that is, the number of categories of user identity is 10.

[0116] The preset neural network model is trained based on RTX 4090 GPU and Intel i9-14900K device. For the process of training the preset neural network model using the training set of the identity recognition model, please refer to the existing neural network model training process, which will not be described in detail here.

[0117] The training method of the identity recognition model provided by the embodiments of the present invention first constructs a training set of the identity recognition model by using the proposed gait feature construction method based on wireless signals, and then trains a preset neural network model by using the training set of the identity recognition model to obtain a trained identity recognition model. In the above process, the present invention analyzes the extracted gait feature curve, identifies its gait pattern, and further determines the potential gait cycle; then, the Doppler spectrogram is segmented by the sliding window technique to construct sample data with stable features and strong consistency, which is finally used for model training to realize user identity recognition; so that the trained identity recognition model only uses the dynamic component caused by the target gait action as the eigeninformation of the sample data during the identity recognition process, which can improve the recognition accuracy.

[0118] In a third aspect, an embodiment of the present invention further provides an identity recognition method, as Fig.13 shown, the identity recognition method includes:

[0119] Step B1, obtaining a gait sub-spectrogram of the user identity to be recognized;

[0120] Among them, the gait sub-spectrogram of the user identity to be recognized can also be obtained in the manner described above, which will not be elaborated here.

[0121] Step B2, inputting the gait sub-spectrogram of the user identity to be recognized into a pre-trained identity recognition model to obtain corresponding user identity information;

[0122] Among them, the identity recognition model is obtained according to the training method of the identity recognition model described in the second aspect. Specifically, reference can be made to the relevant content above, which will not be elaborated here. The user identity information can be the user name and number, etc., corresponding to the labeled data used in training, and obtaining the user identity information can realize the recognition of user categories.

[0123] The identity recognition method provided by the embodiments of the present invention can be implemented based on the model obtained by the provided training method of the identity recognition model, and can recognize the user identity and improve the recognition accuracy.

[0124] In a fourth aspect, an embodiment of the present invention provides a training method of a gait abnormality recognition model, as Fig.14 shown, the training method of the gait abnormality recognition model may include the following steps:

[0125] Step C1, obtaining a training set of the gait abnormality recognition model;

[0126] The training set of the abnormal gait recognition model includes training data of multiple users, and the training data of each user is a plurality of gait sub-spectrograms segmented from the frequency modulated continuous wave signal of the user, which is obtained according to the gait feature construction method based on wireless signals described in the first aspect, and each gait sub-spectrogram is marked with label data to indicate whether the gait is abnormal;

[0127] Step C2: using the training set of the gait abnormality recognition model to train a preset neural network model to obtain a trained gait abnormality recognition model.

[0128] For the specific process of the training method of the gait abnormality recognition model, please refer to Fig.15 Understand. The overall workflow is divided into a data preprocessing stage and a gait segmentation stage. In the data preprocessing stage, since static objects are usually manifested as static components with zero velocity and extremely high energy in the Doppler spectrum, these static components may inhibit the expression of gait signals. According to the foregoing, the present invention proposes a method to effectively remove static information based on the difference between dynamic and static information. In order to solve the problem of restoring dynamic information, the present invention first discretizes the gait features and extracts the gait information from the spectrum. Then, the KNN algorithm and SG filter are used to smooth the discrete information, thereby realizing the restoration of gait information. In the gait segmentation stage, periodic peaks are first found based on the extracted gait feature curve. Finally, with the help of sliding window technology and periodic stability rules, the gait Doppler spectrum is segmented to obtain a gait Doppler sub-spectrum with stable features for training the gait abnormality recognition model.

[0129] Specifically, for step C1, multiple gait sub-spectrograms segmented from all frequency modulated continuous wave signals are obtained according to steps S1 to S5 of the first aspect, and multiple gait sub-spectrograms segmented from each frequency modulated continuous wave signal are used as training data for the corresponding user, wherein each gait sub-spectrogram is marked with label data indicating whether it is a normal gait or an abnormal gait. The training data of all users constitute the training set of the gait abnormality recognition model.

[0130] Specifically, abnormal or irregular walking patterns in humans caused by various factors such as neurological diseases, musculoskeletal diseases, post-traumatic recovery or postoperative rehabilitation are usually referred to as abnormal gaits. Types of abnormal gaits include spastic gait, scissor gait, straddling gait, staggering gait and propulsion gait. These abnormal gaits usually show a walking speed that is slightly slower than normal gait, and the gait cycle is also different from normal gait.

[0131] Considering that the walking speed of abnormal gait is slow, the present invention adopts a lower high-pass filter cutoff frequency to retain more gait detail information. Subsequently, the key features of gait are further retained through gait discretization processing, regression processing and smoothing filtering processing to obtain a gait characteristic curve.

[0132] The present invention uses hemiplegic gait as an abnormal gait case. Due to insufficient flexor strength, the lower limbs are thrown forward by leaning forward and driven by the pelvis, as if drawing circles on the ground. At the same time, due to reasons such as extensor spasm, the affected lower limb cannot support the body weight well, resulting in leg dragging. The gait characteristic curve of hemiplegic gait is shown in Figure 1. Fig.16 As shown in the figure, since there are many noise peak points in abnormal gait, segmentation errors may occur when using the peak period comparison method to find the segmentation starting point.

[0133] In order to analyze the potential gait cycle of abnormal gait to achieve correct gait segmentation, the present invention uses autocorrelation function to analyze the gait cycle. Specifically, for the gait abnormality recognition model, determining the periodic peaks in the gait characteristic curve in S5 to obtain the segmentation starting point may include:

[0134] Step 1), using a derivative analysis method to derive the gait characteristic curve to locate the peak position;

[0135] Step 2), based on the peak position analysis, a periodic peak is obtained as a segmentation starting point.

[0136] Wherein, step 2) may include:

[0137] An autocorrelation analysis is performed on the gait characteristic curve to determine a correct period, and the correct period is used to filter out periodic peaks from the peak positions as segmentation starting points.

[0138] Specifically, the unbiased estimation autocorrelation method is used to perform periodic analysis on the gait characteristic curve, using the following formula:

[0139]

[0140] Where R(τ) represents the autocorrelation value of the signal x(t) at the lag time τ, T is the total number of sample points on the gait characteristic curve, and x(t) is the sample point on the gait characteristic curve (the filtered y i,smooth ) as input for analysis. This ensures that the result is not affected by the signal length, and then extracts the positive lag part from the autocorrelation result, which best reflects the regularity of the signal over time.

[0141] Autocorrelation analysis of gait signals Fig.17 As shown, Figure a is the autocorrelation analysis of normal gait, and it can be observed that the autocorrelation of the gait cycle is strong and the average cycle length is small; Figure b is the autocorrelation analysis of hemiplegic gait, the gait autocorrelation is low, and the cycle length is long.

[0142] The peak of the autocorrelation function is an important sign of periodicity, and the lag time of the peak corresponds to the potential period of the signal. In order to locate the segmentation point of the gait characteristic curve of the abnormal gait, the present invention first directly uses autocorrelation analysis to analyze the gait characteristic curve to find the correct period in the abnormal gait. The correct period is used to identify the periodic peak as the segmentation starting point.

[0143] Specifically, the lag time of the autocorrelation analysis can directly indicate the periodicity of the signal, and can help determine the length of the gait cycle and the location of the peak point. The time difference is calculated by the lag time and the nearby peak points, and the peak point with the smallest time difference is set as the peak point of the correct cycle. After that, the present invention marks the peak point of the correct cycle on the gait characteristic curve as the starting point of gait segmentation. The gait segmentation algorithm is then used to segment the Doppler spectrogram of the abnormal gait, thereby obtaining a gait sub-spectrum for model training. The comparison diagram of gait segmentation results is shown in the figure below. Fig.18 As shown in the figure, (a) is a normal gait spectrogram, (b) is a hemiplegic gait spectrogram, (c) is a sub-spectrum obtained after normal gait segmentation, and (d) is a gait sub-spectrum segmented after period analysis. The segmented gait sub-spectrum is constructed as a training set for the gait abnormality recognition model.

[0144] Because there is a difference in the potential gait cycle length between normal gait and abnormal gait. In normal gait, the cycle length of each step is usually similar; while the cycle of abnormal gait is unstable, or its cycle length is different from that of normal gait. The present invention uses an autocorrelation function to perform autocorrelation analysis on the gait characteristic curve, calculates the cycle length between adjacent peaks, and thus evaluates the stability of the signal cycle. By using the autocorrelation analysis function to perform gait cycle analysis, and using the gait segmentation algorithm to segment the abnormal gait spectrum, it is possible to construct abnormal gait sample data for model recognition.

[0145] For step B2, the structure of the preset neural network model used can be any existing classification network. The present invention can achieve the classification task of abnormal gait through the CNN network. The structure of the CNN network is as follows: Fig.19 shown. Fig.19 and Fig.12 The difference is that the number of classifications in the last fully connected layer is 2, which means two classifications, normal gait or abnormal gait.

[0146] The network model is trained based on RTX 4090 GPU and Intel i9-14900K device. For the process of training the preset neural network model using the training set of the gait abnormality recognition model, please refer to the existing neural network model training process for understanding, which will not be described in detail here.

[0147] The training method of the gait abnormality recognition model provided in the embodiment of the present invention first uses the proposed gait feature construction method based on wireless signals to construct a training set of the gait abnormality recognition model, and then uses the training set of the gait abnormality recognition model to train a preset neural network model to obtain a trained gait abnormality recognition model. In the above process, the present invention analyzes the extracted gait characteristic curve, identifies its gait law, and further determines the potential gait cycle; then, the Doppler spectrum is segmented by the sliding window technology to construct sample data with stable features and strong consistency, which is finally used for model training to realize gait abnormality recognition; so that the trained gait abnormality recognition model only uses the dynamic component caused by the target gait action as the intrinsic information of the sample data during the gait recognition process, which can improve the recognition accuracy.

[0148] In a fifth aspect, an embodiment of the present invention further provides a method for identifying abnormal gait, such as Fig. 20 As shown, the gait abnormality recognition method includes:

[0149] Step D1, obtaining a gait sub-spectrum to be recognized;

[0150] The gait sub-spectrogram to be recognized can also be obtained in the manner described above, which will not be described in detail here.

[0151] Step D2, inputting the gait sub-spectrum to be recognized into a pre-trained gait abnormality recognition model to obtain a gait recognition result, wherein the gait recognition result includes a normal gait or an abnormal gait;

[0152] The abnormal gait recognition model is obtained according to the training method of the abnormal gait recognition model described in the fourth aspect. For details, please refer to the relevant content in the above text, which will not be repeated here.

[0153] The abnormal gait recognition method provided in the embodiment of the present invention is implemented based on a model obtained by the training method of the provided abnormal gait recognition model, and can identify whether the user's gait is abnormal, thereby improving the accuracy of recognition.

[0154] It should be noted that, in the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification.

[0155] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A method for constructing gait features based on wireless signals, characterized in that: include: Acquire frequency modulated continuous wave signals of multiple users collected by radar transceiver equipment; For each frequency modulated continuous wave signal, based on high-pass filtering, a filtered Doppler spectrum corresponding to the frequency modulated continuous wave signal is obtained; Performing gait discretization processing on the filtered Doppler spectrum to obtain a gait signal scatter plot; The gait signal scatter plot is subjected to regression processing and smoothing filtering processing in sequence to obtain a gait characteristic curve; Determine the periodic peaks in the gait characteristic curve, obtain the segmentation starting point, and use the segmentation starting point and the fixed time window to perform gait segmentation on the gait characteristic curve to obtain multiple gait sub-spectrograms segmented from the frequency modulated continuous wave signal; wherein all the gait sub-spectrograms segmented from the frequency modulated continuous wave signal are used to train a preset model, and the preset model includes an identity recognition model and a gait abnormality recognition model.

2. The method for constructing gait features based on wireless signals according to claim 1, characterized in that: The method of obtaining a filtered Doppler spectrum corresponding to each frequency modulated continuous wave signal based on high-pass filtering includes: For each frequency modulated continuous wave signal, a distance dimension fast Fourier transform is performed on it to obtain a distance spectrum of the frequency modulated continuous wave signal; Performing high-pass filtering on the distance spectrum; The range spectrum after high-pass filtering is subjected to a velocity-dimensional fast Fourier transform to obtain a filtered Doppler spectrum corresponding to the frequency modulated continuous wave signal.

3. The method for constructing gait features based on wireless signals according to claim 1, characterized in that: The filtered Doppler spectrogram is subjected to gait discretization processing to obtain a gait signal scatter plot, including: Energy points below a preset energy threshold in the filtered Doppler spectrum are deleted to obtain a gait signal scatter plot.

4. The method for constructing gait features based on wireless signals according to claim 1, characterized in that: Determining the periodic peaks in the gait characteristic curve to obtain the segmentation starting point includes: Derivative analysis is used to derive the gait characteristic curve to locate the peak position; Based on the peak position analysis, a periodic peak is obtained as a segmentation starting point.

5. The method for constructing gait features based on wireless signals according to claim 4, characterized in that: Based on the peak position analysis, a periodic peak is obtained as a segmentation starting point, including: Based on the peak position, the peak period comparison method is used to screen out the periodic peak as the segmentation starting point.

6. The method for constructing gait features based on wireless signals according to claim 4, characterized in that: Based on the peak position analysis, a periodic peak is obtained as a segmentation starting point, including: An autocorrelation analysis is performed on the gait characteristic curve to determine a correct period, and the correct period is used to filter out periodic peaks from the peak positions as segmentation starting points.

7. A method for training an identity recognition model, characterized in that: include: Obtain a training set of an identity recognition model; wherein the training set of the identity recognition model includes training data of multiple users, and the training data of each user is a plurality of gait sub-spectrograms segmented from the frequency modulated continuous wave signal of the user, which is obtained according to the gait feature construction method based on wireless signals according to any one of claims 1 to 5, and each gait sub-spectrogram is marked with user information as label data; The training set of the identity recognition model is used to train a preset neural network model to obtain a trained identity recognition model.

8. An identity recognition method, characterized in that: include: Obtaining a gait sub-spectrogram of a user to be identified; The gait sub-spectrogram of the user identity to be identified is input into a pre-trained identity recognition model to obtain corresponding user identity information; wherein, the identity recognition model is obtained according to the identity recognition model training method according to claim 7.

9. A training method for an abnormal gait recognition model, characterized in that: include: Obtain a training set of a gait abnormality recognition model; wherein the training set of the gait abnormality recognition model includes training data of multiple users, and the training data of each user is a plurality of gait sub-spectrograms segmented from the frequency modulated continuous wave signal of the user, which is obtained according to the gait feature construction method based on wireless signals according to any one of claims 1, 2, 3, 4, and 6, and each gait sub-spectrogram is marked with label data to indicate whether the gait is abnormal; The training set of the gait abnormality recognition model is used to train a preset neural network model to obtain a trained gait abnormality recognition model.

10. A method for identifying abnormal gait, characterized in that: include: Obtaining a gait sub-spectrum to be recognized; The gait sub-spectrum to be recognized is input into a pre-trained gait abnormality recognition model to obtain a gait recognition result, wherein the gait recognition result includes a normal gait or an abnormal gait; wherein the gait abnormality recognition model is obtained according to the training method of the gait abnormality recognition model according to claim 9.

Images