A gesture recognition method and system based on mobile communication signals

Abstract

The application belongs to the field of wireless communication, signal processing and intelligent sensing technology, and relates to a gesture recognition method and system based on mobile communication signals, which comprises the following steps: judging the scene of a gesture to be recognized, and regarding the gesture as a close-range scene if the sensing range is within a set threshold, or regarding the gesture as a long-range scene; for the close-range scene, extracting CRS data in a single carrier of the mobile communication signal, and performing smoothing and denoising processing by using a sliding window filter; for the long-range scene, extracting and superimposing CRS data between different carriers of the mobile communication signal; calculating a static component by using a long-time window filter; calculating a dynamic component according to the static component; performing short-time Fourier transform on the dynamic component, constructing a gesture spectrum graph, and inputting the gesture spectrum graph into a trained convolutional neural network to realize gesture recognition. The application has a wider coverage range and higher network stability by using mobile communication signals, is suitable for more application scenes, and has high recognition accuracy.

Description

Technical Field

[0001] This invention belongs to the fields of wireless communication, signal processing and intelligent sensing technology, and particularly relates to a gesture recognition method and system based on mobile communication signals. Background Technology

[0002] Gesture recognition technology, as a natural human-computer interaction method, has brought new interactive experiences and convenient operation methods to users in fields such as smart homes, autonomous driving, entertainment games, medical rehabilitation, and public safety, and has broad application potential. The core principle of gesture recognition technology is to extract gesture-related feature information from data collected by sensors, cameras, or wireless devices, and convert it into instructions that computers can understand. Existing gesture recognition technologies are mainly divided into three types: vision-based, sensor-based, and wireless signal-based. Vision-based gesture recognition methods use cameras to capture hand images or video sequences, combining image processing and deep learning methods for analysis to identify and interpret the shape, position, and dynamic changes of the hand. Sensor-based gesture recognition technology captures the physical information of hand movements (such as acceleration, angular velocity, and orientation), converts it into digital signals, and processes and analyzes these signals to identify specific gestures. Wireless signal-based gesture recognition methods analyze the changes in wireless signals such as millimeter waves and WiFi under the influence of gesture movements to perceive and recognize different gestures. With the development of integrated communication and sensing technologies, how to achieve efficient communication while utilizing communication signals (such as WiFi signals and mobile communication signals) for gesture perception has become a hot research topic. WiFi-based solutions are typically limited to indoor environments, and achieving high sampling rates requires consuming communication resources, thus interfering with normal data transmission. In contrast, mobile communication signals offer wider coverage and higher network stability, making them suitable for more application scenarios. Therefore, this invention chooses mobile communication signals as the carrier for gesture perception. However, this also presents some challenges. For example, when gestures are performed in different locations, directions, or at long distances, their original spectrograms can vary significantly due to the environment, thus affecting the system's recognition accuracy.

[0003] To more clearly explain the underlying principles, three typical cases are listed, such as... Figure 1 As shown in (a), Target 1 and Target 2 perform the same gesture at different positions and directions within the sensing area, while Target 3 performs the same gesture at the boundary of the sensing area. This invention performs preliminary denoising on the raw CRS (cell reference signal) information of different carriers. The raw CRS information and preliminary denoising results for different carriers are shown below. Figure 1As shown in (b) above, the initial denoising result is subjected to a Short-Time Fourier Transform (STFT), and the resulting original spectrum is shown below. Figure 1 As shown in (c), when targets one and two perform the same gesture at different locations and in different directions, the original spectrograms corresponding to the CRS information in the mobile communication system are significantly different because the dynamic components generated by the gestures in different directions are inconsistent, and the static components of the signal after reflection from the environment are also different at different locations. For target three, since the location where the same gesture is performed is far from the transceiver device, the system cannot directly capture the changes in the original CRS information caused by the gesture, resulting in the dynamic component information of the gesture being annihilated by noise, and the obtained original spectrum cannot directly reflect the gesture information. Based on the above phenomena, it can be found that the original spectrogram of CRS information cannot represent the intrinsic characteristics of the gesture perception task. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and to propose a gesture recognition method and system based on mobile communication signals.

[0005] To achieve the above objectives, the present invention provides a gesture recognition method based on mobile communication signals, comprising:

[0006] Step 1) Determine the scene of the gesture to be recognized. If the perception range is within the set threshold, it is a near-distance scene; otherwise, it is a far-distance scene.

[0007] Step 2) For near-field scenarios, extract the CRS data from a single carrier of the mobile communication signal and use a sliding window filter for smoothing and noise reduction; for long-field scenarios, extract and overlay the CRS data between different carriers of the mobile communication signal.

[0008] Step 3) Calculate the static components using a long-window filter;

[0009] Step 4) Calculate the dynamic components based on the static components;

[0010] Step 5) Perform short-time Fourier transform on the dynamic components to construct a gesture spectrogram, and input it into the trained convolutional neural network to achieve gesture recognition.

[0011] Preferably, the mobile communication system is an LTE system, including two omnidirectional antennas, one for transmitting and one for receiving.

[0012] Preferably, the threshold value in step 1) is set to 30 square meters.

[0013] Preferably, in step 2), for close-range scenes, the signal after smoothing and denoising is... Satisfy the following formula:

[0014]

[0015] wherein n∈[1,N] represents the index of the sampling point, N represents the number of sampling points in the sample, is the first point of the sliding window, is the last point of the sliding window, L(n)=k2-k1+1 represents the number of sampling points in the sliding window, and W is the window length parameter.

[0016] Preferably, the step 2) extracts and superimposes the CRS data between different carriers of the mobile communication signal for a long-distance scene.

[0017] For a long-distance scene, the CRS data between different carriers is extracted and superimposed based on the correlation of the information between the multiple carriers of the mobile communication signal, so as to realize the effect of signal quality enhancement.

[0018] Preferably, the static component obtained in the step 3) is satisfies the following formula:

[0019]

[0020] wherein, is the first point of the long-time window, is the last point of the long-time window, L long (n)=k long2 -k long1 +1 represents the number of samples in the long-time window, and W long is the window length parameter.

[0021] Preferably, the dynamic component obtained in the step 4) is satisfies the following formula:

[0022]

[0023] Preferably, the convolutional neural network in the step 5) includes four layers, the first three layers are composed of a convolutional layer, a batch normalization layer, a ReLU activation layer and a maximum pooling layer; the fourth layer is composed of a convolutional layer, a batch normalization layer and a ReLU activation layer; the number of convolutional kernels in each layer is 16, 32, 64 and 256 respectively, the size of the convolutional kernel is 3*3, the step and the padding size are both 1, the classifier includes a first fully connected layer with 100 neurons, a second fully connected layer with 5 neurons, and a Softmax layer with 5 class labels.

[0024] Preferably, the input of the convolutional neural network is a gesture spectrum graph, and the output is a gesture type, including:

[0025] a "push" gesture, a "rectangle" gesture, a "circle" gesture, a "push-pull" gesture, and an "X" gesture.

[0026] In another aspect, the present application provides a gesture recognition system based on mobile communication signals, comprising:

[0027] a near-far scene judgment module for judging the scene of the gesture to be recognized, and the range within the set threshold is a near distance scene, and otherwise is a far distance scene;

[0028] a CRS data processing module for extracting CRS data in a single carrier of the mobile communication signal for the near distance scene, and using a sliding window filter for smoothing and denoising processing; and for extracting and superimposing the CRS data between different carriers of the mobile communication signal for the far distance scene;

[0029] a static component calculation module for calculating a static component using a long time window filter;

[0030] a dynamic component calculation module for calculating a dynamic component according to the static component;

[0031] a gesture recognition module for performing short-time Fourier transform on the dynamic component, constructing a gesture spectrum graph, and inputting a trained convolutional neural network to realize gesture recognition.

[0032] Compared with the prior art, the present application has the following advantages:

[0033] 1. The present application proposes a signal quality enhancement method based on LTE (Long Term Evolution) signals, which enhances the signal quality by superimposing the CRS information in multiple carriers, makes the noise distribution more concentrated to improve the signal quality, and enhances the perception range of the system to realize far distance gesture recognition;

[0034] 2. The present application adopts a dynamic component capture method, which is based on CRS signals, uses a sliding window filter to denoise the original CRS signal and extract a static component, and then subtracts the static component from the denoised signal to obtain dynamic component information reflecting the intrinsic gesture action;

[0035] 3. Since the amplitude and trend of the dynamic component are related to the position and direction of the human gesture, the present application performs short-time Fourier transform on the obtained dynamic component to generate a gesture intrinsic spectrum graph, and then recognizes it through a convolutional neural network to realize position and direction independent gesture recognition. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 is the spectrum graph difference of the same gesture original CRS information in different positions, directions or far distances;

[0037] Figure 2 is a method flowchart of the present application;

[0038] Figure 3 is a comparison of the effect of the dynamic component capture method, wherein Figure 3(a) is the result of the preliminary denoising of the original CRS information of the target one and target two performing the "push" gesture action, and Figure 3(b) is the result after processing by the dynamic component capture method; Figure 1

[0039] Figure 4 is a visual comparison of the original spectrum and the gesture spectrum;

[0040] Figure 5 is the noise statistical characteristics and CRS amplitude information before and after the perceptual distance enhancement method, wherein Figure 5(a) is the noise statistical characteristics, and Figure 5(b) is the CRS amplitude information;

[0041] Figure 6 is a comparison of the effect of the original spectrum and the gesture spectrum at a long distance, wherein Figure 6(a) is the original spectrum at a long distance, and Figure 6(b) is the gesture spectrum. DETAILED DESCRIPTION

[0042] The present application proposes a signal quality enhancement method based on mobile communication signals and a construction method of a gesture spectrum, solves the technical problem of inconsistency of the original spectrum of a gesture recognition system based on mobile communication signals at different positions, directions or long distance scenes of gestures, and realizes the consistency of the features of the same gesture at any position, direction or long distance and the difference of the features of different gestures. The core idea of the present application is to enhance the signal quality of the dynamic component of the gesture by superimposing the CRS information of multiple carriers in the mobile communication system, so as to increase the gesture perception range to realize the perception task at a long distance. Then, the dynamic component related to the gesture is extracted from the CRS information, and a gesture spectrum is constructed to represent the intrinsic motion mode of the gesture independent of the scene.

[0043] In daily gesture recognition tasks, the perception scene of the system can be divided into two cases: a near distance and a long distance. The near distance scene refers to the case that the target is located within the initial perception range of the system (after test evaluation, the initial perception range of the present system covers an area of 30 square meters); the long distance scene refers to the case that the target position is beyond the initial perception range (after test evaluation, the perception range of the system after signal quality enhancement covers an area of 228 square meters). For the two different scenes, the present application proposes a robust gesture recognition method based on a gesture spectrum, which is as follows.

[0044] The workflow of the present application is as follows Figure 2 ​As shown, the whole is divided into signal quality enhancement stage and gesture spectrum construction stage. In the signal quality enhancement stage, for the gesture recognition task of any position and direction in the near distance scene, the application can complete the sensing task directly using the CRS information in the single carrier. For the gesture recognition task of any position and direction in the long distance scene, the application selects the CRS amplitude information of 30 different subcarriers in the superposition of the mobile communication system. According to the Wiener-Sinai large number theorem, based on the superposition of multiple independent and identically distributed samples, the distribution of noise can be significantly reduced, and the CRS information related to the gesture is highlighted. In the gesture spectrum construction stage, based on the CRS information processed in the signal quality enhancement stage, first, a smoothing filter with a window length of 100 is used to further eliminate the interference of noise, second, a long-time sliding window filter with a window length of 400 is used to calculate the static component related to the environment, third, the dynamic component of the gesture action is obtained by subtracting the static component from the de-noised data, and then the CRS information containing the dynamic component related to the gesture motion process is obtained, and finally, the short-time Fourier transform is performed based on the dynamic component of the CRS to obtain the gesture spectrum graph representing the intrinsic characteristics of the gesture, which is input into the pre-established and trained convolutional neural network to obtain the gesture type and complete the gesture recognition task.

[0045] The technical solutions of the application will be described in detail below in combination with the drawings and embodiments.

[0046] Embodiment 1

[0047] Embodiment 1: Gesture recognition of any position and direction in the near distance scene

[0048] The system configuration is as follows:

[0049] 1. The system works on the YunSDR Y750 software radio platform.

[0050] 2. The LTE system bandwidth built is 20MHz, and the system sampling rate is 30.72MHz. Only the CRS data in the single carrier is collected, and the number of CRS information sample points per second is 2000. LTE (Long Term Evolution) is the long-term evolution of the Universal Mobile Telecommunications System (UMTS) technical standard formulated by the 3rd Generation Partnership Project (3GPP) organization.

[0051] 3. The system uses one transmitting and one receiving omnidirectional antenna, and the initial sensing capability coverage range of the device is 30m 2 .

[0052] Task: Construct gesture spectrogram independent of position and direction of gesture action in close-range scenario.

[0053] For gesture recognition, the present application takes the dynamic component only affected by target gesture action as the intrinsic information of gesture spectrogram. Specifically, in wireless communication channel, in addition to LOS propagation, there also exists multipath effect caused by environmental diffraction, reflection, refraction and scattering. When there is human gesture action in the propagation environment, the signals reflected or scattered by arms and palms will form dynamic component, thus the signal at the receiver can be further written as:

[0054] h(n) = h s (n) + h d (n) + h n (n)

[0055] where h s (n) represents static component due to environmental reflection, h d (n) represents dynamic component due to human gesture action reflection, and h n (n) represents noise. Since the signal environmental reflection path length is inconsistent at different positions and directions of the same gesture, the waveform of CRS information and the original spectrogram of the waveform will be different after superposition of noise and dynamic component generated by the gesture itself.

[0056] To solve this problem, the present application uses dynamic component capture method to filter out the interference of noise component h n (n) and static component h s (n), and only focuses on the dynamic component h d (n) generated by gesture action. Specifically, the present application first performs smoothing denoising on the original CRS data h(n) to obtain the signal after removing noise The formula is as follows:

[0057]

[0058] where n ∈ [1, N] represents the index of sampling points, N represents the number of sampling points in the sample,

[0059] is the first point of the sliding window, is the last point of the sliding window, L(n) = k2-k1+1 represents the number of sampling points in the sliding window, and W is the window length parameter, which is set to 100 in the present application. For Figure 1 The results of the original CRS information collected by target one and target two after preliminary denoising are shown in FIG. 3(a), and it can be seen that for the same gesture performed at different positions and directions, their differences are large.

[0060] For static component This invention uses a sliding window filter. The filtering calculation formula is as follows:

[0061]

[0062] in It is the first point of the long window. L represents the last point of the long window. long (n)=k long2 -k long1 +1 indicates the number of samples within the long window.

[0063] W long This is the window length parameter, which is set to 400 sampling points in this invention. This invention utilizes denoised data... Subtract static components To obtain the dynamic components of gestures

[0064]

[0065] The result after processing by the dynamic component capture method is shown in Figure 3(b). Since the "push" gesture has a motion phase, the CRS information after processing also has a dynamic component fluctuation phase. Therefore, this method can effectively extract the intrinsic motion information of the gesture, so that the same gesture has similar characteristic patterns in different positions and directions.

[0066] Because the amplitude and trend of dynamic components are related to the size, position, and direction of human gestures, the dynamic components of the same gesture will differ in different scenarios, thus affecting the accuracy of gesture recognition. To solve this problem, this invention [addresses the issue of dynamic components]. Perform a short-time Fourier transform and set the time window length to N. s =1500, yielding a frequency resolution of approximately 1.34 Hz, ultimately resulting in a gesture spectrogram related to the gesture movement. For example... Figure 4As shown, the upper row of figures is the original spectrum diagram of the gesture, and the lower row of figures is the gesture spectrum diagram obtained by the method proposed in the application. The gesture recognition method adopted is to take the obtained gesture spectrum diagram as the input of the network through a convolutional neural network. Specifically, we designed a convolutional neural network based on deep learning. The network is composed of four layers, and the number of convolution kernels in each layer is 16, 32, 64 and 256 respectively, the size of the convolution kernel is 3*3, and the step and padding size are 1. The first three layers are composed of convolutional layers, batch normalization layers, ReLU activation layers and maximum pooling layers. The fourth layer is composed of convolutional layers, batch normalization layers and ReLU activation layers. The classifier is composed of a fully connected layer 1 with 100 neurons, a fully connected layer 2 with 5 neurons and a Softmax layer with 5 class labels.

[0067] The currently selected gesture types include but are not limited to the following five types: "push" gesture, "rectangle" gesture, "circle" gesture, "push-pull" gesture and "X" gesture.

[0068] Among them, Figure 4 (a) and (b) in are the original spectrum diagrams and gesture spectrum diagrams of target one and target two performing "push" gesture actions, respectively. It can be seen that the method proposed in the application can significantly reduce the influence of gesture position and direction, so that the same gesture features have consistency. Figure 4 (c)-(f) in are the original spectrum diagrams and gesture spectrum diagrams of performing "rectangle", "circle", "push-pull" and "X" gesture actions, respectively. It can be seen that the features of the method proposed in the application have significant differences for different gesture actions.

[0069] Example 2: Gesture recognition of arbitrary position and direction in a long-distance scene

[0070] The system configuration is as follows:

[0071] 1. The system works on the YunSDR Y750 software radio platform.

[0072] 2. The LTE system bandwidth built is 20MHz, and the system sampling rate is 30.72MHz. The subcarrier frequency interval is 15kHz, and there are 1200 subcarriers. The CRS data in 30 carriers is collected, and the number of CRS information sample points per second is 2000.

[0073] 3. The system uses one transmitting and one receiving omnidirectional antenna, and the sensing coverage range of the system after signal quality enhancement is 228m 2 .

[0074] Task: Within the sensing coverage range of the system after enhancement, construct a gesture spectrum diagram independent of the position and direction of the gesture action.

[0075] It is known that the measurement noise of CRS information follows a zero-mean, σ 2 (f) additive white Gaussian distribution. When the position of the gesture action is far away from the transceiver, the noise power will be greater than the dynamic power, at this time the change of the dynamic component caused by the human gesture action will be drowned in the noise, so it is difficult to extract the change of the CRS caused by the human gesture action. Due to the influence of noise, the directly obtained CRS change cannot accurately reflect the gesture action.

[0076] In order to solve the above problems, the present application proposes a signal quality enhancement method. The idea of this method is to superimpose multiple samplings of independent and identically distributed noise, which can significantly reduce the noise level. And this idea conforms to the Wiener-Sinai law of large numbers: let X n ,n=1,2,... are independent, identically distributed random variables, and as the sample number n increases, the average of all samples will converge to the expectation, as shown in the following formula:

[0077]

[0078] Since the LTE system can divide the signal into multiple subcarriers in the frequency domain, the present application uses the correlation of information between multiple carriers to superimpose the CRS information between different carriers, increases the number of data samples, and achieves the purpose of enhancing the quality of the CRS signal.

[0079] Specifically, in the signal quality enhancement stage, the present application selects to use 30 subcarriers. First, the CRS information of each carrier is collected. Second, the CRS information of different subcarriers is superimposed, so that the noise distribution is more narrow. Figure 5(a) shows the noise statistical characteristics of the CRS information obtained before and after processing under the condition of the target three far distances, it can be seen that the noise distribution after processing is more narrow. Figure 5(b) shows the CRS amplitude information before and after processing by the method, it can be seen that the method can effectively enhance the dynamic component, and can effectively obtain gesture information under the condition of far distance. Figure 1

[0080] In the gesture spectrum construction stage, first, based on the CRS information obtained by the signal quality enhancement method, the dynamic component related to the gesture action is extracted by using the dynamic component extraction method, and the static component related to the gesture position, direction and environment is filtered out. Second, the gesture spectrum graph capable of representing the intrinsic characteristics of the gesture is constructed, as shown in Figure 6(b).

[0081] The gesture recognition method adopted is to take the obtained gesture spectrum graph as the input of the network through the convolutional neural network, and the specific network parameters are the same as those of embodiment 1.

[0082] ​Compared with the original spectrum graph 6(a) of the target three, the method of the present application can extract the essential features of the gesture action in a long distance, and the extracted gesture spectrum graph is similar to the "push" gesture feature of (a) and (b) in Figure 4 Compared with (a) and (b) in the above, the "push" gesture feature is similar, which reflects the consistency of the same gesture feature.

[0083] Embodiment 3

[0084] Embodiment 3 of the present application provides a gesture recognition system based on mobile communication signals, which is realized based on the method of embodiment 1 and embodiment 2, comprising:

[0085] A near-far scene judgment module is used for scene judgment of the gesture to be recognized, and the range within the set threshold is a near distance scene, otherwise it is a long distance scene;

[0086] A CRS data processing module is used for extracting the CRS data in the single carrier of the mobile communication signal for the near distance scene, and using a sliding window filter for smoothing and denoising processing; for the long distance scene, the CRS data between different carriers of the mobile communication signal is extracted and superimposed;

[0087] A static component calculation module is used for calculating the static component using a long time window filter;

[0088] A dynamic component calculation module is used for calculating the dynamic component according to the static component;

[0089] A gesture recognition module is used for short-time Fourier transform of the dynamic component, construction of the gesture spectrum graph, and input of the trained convolutional neural network to realize gesture recognition.

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not limited. Although the present application has been described in detail with reference to the embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, and they should be covered in the scope of the claims of the present application.

Claims

1. A gesture recognition method based on mobile communication signals, comprising: Step 1) Determine the scene of the gesture to be recognized. If the perception range is within the set threshold, it is a near-distance scene; otherwise, it is a far-distance scene. Step 2) For near-field scenarios, extract the CRS data from a single carrier of the mobile communication signal and perform smoothing and noise reduction using a sliding window filter; for long-field scenarios, extract and overlay the CRS data between different carriers of the mobile communication signal. Step 3) Calculate the static components using a long-window filter; Step 4) Calculate the dynamic components based on the static components; Step 5) Perform short-time Fourier transform on the dynamic components to construct a gesture spectrogram, and input it into the trained convolutional neural network to realize gesture recognition.

2. The gesture recognition method based on mobile communication signals according to claim 1, characterized in that, The mobile communication system is an LTE system, which includes two omnidirectional antennas, one for transmitting and one for receiving.

3. The gesture recognition method based on mobile communication signals according to claim 1, characterized in that, The threshold for step 1) is set to 30 square meters.

4. The gesture recognition method based on mobile communication signals according to claim 1, characterized in that, Step 2) For close-range scenes, the signal after smoothing and denoising is: The following equation is satisfied: ； in, This represents the original CRS data. Indicates the index of the sampling point. N Indicates the number of sampling points in the sample. It is the first point of the sliding window. This represents the last point of the sliding window. This indicates the number of sampling points within the sliding window. W It is the window length parameter.

5. The gesture recognition method based on mobile communication signals according to claim 1, characterized in that, Step 2) for long-distance scenarios involves extracting and overlaying CRS data between different carriers of the mobile communication signal; including: For long-distance scenarios, based on the correlation between information of multiple carriers in mobile communication signals, CRS data between different carriers is extracted and superimposed to achieve the effect of signal quality enhancement.

6. The gesture recognition method based on mobile communication signals according to claim 4, characterized in that, The static component obtained in step 3) is The following equation is satisfied: ； in, It is the first point of the long window. This indicates the last point of the long window. This represents the number of samples within a long time window. It is the window length parameter.

7. The gesture recognition method based on mobile communication signals according to claim 6, characterized in that, The dynamic component obtained in step 4) is The following equation is satisfied: 。 8. The gesture recognition method based on mobile communication signals according to claim 1, characterized in that, The convolutional neural network in step 5) consists of four layers. The first three layers consist of a convolutional layer, a batch normalization layer, a ReLU activation layer, and a max pooling layer. The fourth layer consists of a convolutional layer, a batch normalization layer, and a ReLU activation layer. The number of convolutional kernels in each layer are 16, 32, 64, and 256, respectively. The kernel size is 3*3, and the stride and padding size are both 1. The classifier includes a first fully connected layer with 100 neurons, a second fully connected layer with 5 neurons, and a Softmax layer with 5 class labels.

9. The gesture recognition method based on mobile communication signals according to claim 1, characterized in that, The input to the convolutional neural network is a gesture spectrogram, and the output is the gesture type, including: "push" gesture, "rectangle" gesture, "circle" gesture, "push-pull" gesture and "X" shaped gesture.

10. A gesture recognition system based on mobile communication signals, characterized in that, include: The near-far scene judgment module is used to judge the scene of the gesture to be recognized. If the perception range is within a set threshold, it is a near-far scene; otherwise, it is a far-far scene. The CRS data processing module is used to extract CRS data from a single carrier of a mobile communication signal for near-field scenarios and perform smoothing and noise reduction using a sliding window filter; for long-field scenarios, it is used to extract and superimpose CRS data between different carriers of the mobile communication signal. The static component calculation module is used to calculate static components using a long-time window filter. The dynamic component calculation module is used to calculate the dynamic components based on the static components. and The gesture recognition module is used to perform short-time Fourier transform on the dynamic components, construct a gesture spectrogram, and input it into a trained convolutional neural network to achieve gesture recognition.

Images