Behavior recognition method and device based on frequency domain channel response, electronic equipment and storage medium

Abstract

The application provides a behavior identification method and device based on a frequency domain channel response, electronic equipment and a storage medium, comprising: selecting target sampling points in a target power delay spectrum before a front window and after a rear window, and taking sampling points other than the target sampling points as noise sampling points; determining a noise threshold power based on the power of the plurality of noise sampling points; selecting target sampling points with power exceeding the noise threshold power in the front window and the rear window to obtain specified sampling points; determining whether the distance between the specified sampling point closest to the front end in the rear window and the center of gravity sampling point exceeds a sampling point quantity threshold; if yes, determining that the specified sampling point closest to the front end in the rear window is a LOS sampling point and the rest are NLOS sampling points; determining a cleaned frequency domain channel response corresponding to the frequency domain channel response based on the plurality of NLOS sampling points; and performing behavior identification on a passive target based on the cleaned frequency domain channel responses of the plurality of observation time points. The application can remove the influence of the LOS channel and accurately perform behavior identification.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of behavior recognition, and in particular to a behavior recognition method and device based on a frequency domain channel response, an electronic device, and a computer readable storage medium. BACKGROUND

[0002] Future Wi-Fi signals will have a sensing function for passive targets (people or other targets not carrying terminal devices) in addition to data transmission functions. When passive targets move in an indoor scene, they will affect indoor Wi-Fi signals, causing changes in multipath channels. In this case, channel state information changes, and passive targets can be recognized by channel state information. Channel state information includes the channel frequency response (CFR) of Wi-Fi signal multipath transmission, where the multiple channels of multipath transmission include LOS (Line of Sight) channels and NLOS (Not Line of Sight) channels. Since indoor stationary objects remain stationary, the channel state estimate of the LOS channel is not affected by passive targets, so the channel state estimate of the NLOS channel in the channel state information can be used to represent the impact of passive target activity on the wireless channel, thereby performing behavior recognition.

[0003] However, in related technologies, the process of cleaning the frequency domain channel response may not accurately remove the impact of the LOS channel, which can result in inaccurate behavior recognition results. SUMMARY

[0004] The embodiments of the present application aim to provide a behavior recognition method and device based on a frequency domain channel response, an electronic device, and a computer readable storage medium, which can accurately remove the impact of the LOS channel, thereby performing accurate behavior recognition based on the cleaned frequency domain channel response.

[0005] In one aspect, the present application provides a behavior recognition method based on a frequency domain channel response, comprising:

[0006] determining a corresponding power delay profile for the frequency domain channel response of the Wi-Fi signal;

[0007] searching for a sample point with the maximum power in the power delay profile as a gravity sample point, and cutting and splicing a plurality of sample points before the gravity sample point to the back end of the power delay profile to obtain a target power delay profile;

[0008] According to a preset first length of a front window and a preset second length of a rear window, a target sample point is screened out in the target power delay profile, and sample points other than the target sample point are taken as noise sample points; wherein the front window is placed at the front end of the target power delay profile, and the rear window is placed at the rear end of the target power delay profile;

[0009] Based on the power of the plurality of noise sample points, a noise threshold power is determined.

[0010] The target sample points whose power exceeds the noise threshold power are screened out in the front window and the rear window, and specified sample points are obtained.

[0011] The front and rear ends of the target power delay profile are cyclically connected, and it is judged whether the specified sample point closest to the front end in the rear window exceeds a sample point quantity threshold through a distance between the connection and the gravity sample point.

[0012] If yes, the specified sample point closest to the front end in the rear window is determined as a LOS sample point, and the remaining specified sample points are NLOS sample points.

[0013] Based on the plurality of NLOS sample points, a cleaned frequency domain channel response corresponding to the frequency domain channel response is determined.

[0014] Based on the cleaned frequency domain channel responses of the plurality of observation moments, behavior identification is performed on the passive target.

[0015] In an embodiment, the method further comprises:

[0016] If no, the gravity sample point is determined as a LOS sample point, and the remaining specified sample points are NLOS sample points.

[0017] Returning to the step of determining the cleaned frequency domain channel response from the frequency domain channel response based on the plurality of NLOS sample points.

[0018] In an embodiment, the noise threshold power is determined based on the power of the plurality of noise sample points, comprising:

[0019] The average value of the power of the plurality of noise sample points is calculated as the noise threshold power.

[0020] In an embodiment, the method further comprises:

[0021] After obtaining the noise threshold power corresponding to the plurality of continuous symbols, the plurality of noise threshold powers are smoothed to obtain a new noise threshold power.

[0022] In an embodiment, the cleaned frequency domain channel response corresponding to the frequency domain channel response is determined based on the plurality of NLOS sample points, comprising:

[0023] In the time domain channel impulse response corresponding to the frequency domain channel response, time domain channel estimation values corresponding to the plurality of NLOS sampling points are reserved, and time domain channel estimation values corresponding to the remaining sampling points are set to zero to obtain a cleaned time domain channel impulse response;

[0024] The cleaned time domain channel impulse response is subjected to zero padding processing.

[0025] The cleaned time domain channel impulse response subjected to zero padding processing is subjected to Fourier transform to obtain a cleaned frequency domain channel response.

[0026] In an embodiment, the cleaned frequency domain channel response based on a plurality of observation times is used to perform behavior recognition on a passive target, including:

[0027] The cleaned frequency domain channel response based on a plurality of observation times is used to construct a two-dimensional matrix.

[0028] The two-dimensional matrix is subjected to short-time Fourier transform to obtain a spectrum picture.

[0029] The spectrum picture is periodically cut to obtain a to-be-processed picture.

[0030] The to-be-processed picture is input into a trained behavior recognition model to obtain a behavior recognition result.

[0031] In an embodiment, the method further includes:

[0032] The first length is determined according to a first coefficient and a length of the cyclic prefix; wherein the first coefficient is greater than one.

[0033] The second length is determined according to a second coefficient and a length of the cyclic prefix; wherein the second coefficient is less than one.

[0034] On the other hand, the present application provides a behavior recognition device based on a frequency domain channel response, including:

[0035] A first determination module is configured to determine a corresponding power delay spectrum for a frequency domain channel response of a Wi-Fi signal.

[0036] A cutting module is configured to search for a sampling point with the maximum power in the power delay spectrum as a center sampling point, and cut and splice a plurality of sampling points before the center sampling point to the rear end of the power delay spectrum to obtain a target power delay spectrum.

[0037] A first screening module is configured to screen out target sampling points in the target power delay spectrum according to a preset first length front window and a preset second length rear window, and regard sampling points other than the target sampling points as noise sampling points; wherein the front window is placed at the front end of the target power delay spectrum, and the rear window is placed at the rear end of the target power delay spectrum.

[0038] a second determining module, configured to determine a noise threshold power based on the power of the plurality of noise sampling points;

[0039] a second screening module, configured to screen out target sampling points with power exceeding the noise threshold power in the front window and the rear window, to obtain specified sampling points;

[0040] a judging module, configured to judge whether a specified sampling point closest to the front end in the rear window exceeds a sampling point number threshold in a distance between the specified sampling point and the gravity sampling point after the front and rear ends of the target power delay profile are cyclically connected;

[0041] a third determining module, configured to determine the specified sampling point closest to the front end in the rear window as a LOS sampling point and the remaining specified sampling points as NLOS sampling points if the specified sampling point exceeds the sampling point number threshold;

[0042] a fourth determining module, configured to determine a cleaned frequency domain channel response corresponding to the frequency domain channel response based on the plurality of NLOS sampling points;

[0043] an identifying module, configured to perform behavior identification on a passive target based on the cleaned frequency domain channel responses at the plurality of observation moments.

[0044] Further, the present application provides an electronic device, which comprises:

[0045] a processor;

[0046] a memory for storing processor-executable instructions;

[0047] wherein the processor is configured to perform the above behavior identification method based on a frequency domain channel response.

[0048] In addition, the present application provides a computer-readable storage medium, which stores a computer program executable by a processor to complete the above behavior identification method based on a frequency domain channel response.

[0049] The present application scheme screens sampling points corresponding to an effective channel through a front window, a rear window and a noise threshold power, searches for a specified sampling point with a distance from a gravity sampling point exceeding a sampling point number threshold from the rear window, and thus firstly finds a sampling point corresponding to a LOS channel in time, and determines the gravity sampling point as a LOS sampling point in the case of being unable to find, which can more accurately determine data corresponding to a LOS channel, effectively clean a frequency domain channel response, and further achieve accurate behavior identification. BRIEF DESCRIPTION OF DRAWINGS

[0050] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly described below.

[0051] Figure 1 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application;

[0052] Figure 2 A flowchart illustrating a behavior recognition method based on frequency domain channel response provided in an embodiment of this application;

[0053] Figure 3 A schematic diagram of the power delay spectrum provided in an embodiment of this application;

[0054] Figure 4 Provided for an embodiment of this application Figure 3 A schematic diagram of the target power time delay spectrum corresponding to the medium power time delay spectrum;

[0055] Figure 5 A schematic diagram of windowing the target power delay spectrum provided in an embodiment of this application;

[0056] Figure 6 A schematic diagram of an OFDM symbol provided in an embodiment of this application;

[0057] Figure 7 A schematic diagram showing an OFDM symbol cutoff provided in an embodiment of this application;

[0058] Figure 8 Provided for an embodiment of this application Figure 2 A detailed flowchart of step 280 is provided.

[0059] Figure 9 Provided for an embodiment of this application Figure 2 A detailed flowchart of step 290 is shown below;

[0060] Figure 10 A block diagram of a behavior recognition device based on frequency domain channel response provided in an embodiment of this application. Detailed Implementation

[0061] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0062] Similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0063] like Figure 1As shown, this embodiment provides an electronic device 1, including: at least one processor 11 and a memory 12. Figure 1 Taking a processor 11 as an example, the processor 11 and memory 12 are connected via a bus 10. The memory 12 stores instructions that can be executed by the processor 11. These instructions are executed by the processor 11 to enable the electronic device 1 to perform all or part of the processes described in the embodiments below. In one embodiment, the electronic device 1 may be a router, a communication terminal device (e.g., a mobile phone, a tablet computer), etc., used to execute a behavior recognition method based on frequency domain channel response. The electronic device may be equipped with a transmitting antenna and a receiving antenna for transmitting and receiving Wi-Fi signals. The following description uses the electronic device as the execution subject.

[0064] The memory 12 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable red-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0065] This application also provides a computer-readable storage medium storing a computer program that can be executed by a processor 11 to perform the behavior recognition method based on frequency domain channel response provided in this application.

[0066] See Figure 2 The above is a flowchart illustrating a behavior recognition method based on frequency domain channel response provided in an embodiment of this application. Figure 2 As shown, the method may include steps 210-280.

[0067] Step 210: Determine the corresponding power delay spectrum for the frequency domain channel response of the Wi-Fi signal.

[0068] During Wi-Fi signal transmission and reception, electronic devices can determine the frequency domain channel response through channel estimation. The frequency domain channel response includes channel estimates at multiple sampling points in the frequency domain. In one embodiment, the electronic device can apply a window function to the frequency domain channel response, making the spectral leakage more concentrated and facilitating subsequent processing. Here, the window function can be, but is not limited to, a Hamming window.

[0069] Electronic devices can perform an inverse Fourier transform on the frequency domain channel response to obtain the time domain channel impulse response (CIR). The time domain channel impulse response includes channel estimates at multiple sampling points in the time domain. Based on the time domain channel impulse response, electronic devices can calculate the power delay profile (PDP), which includes the power at each sampling point, with each sampling point corresponding to a channel.

[0070] For each symbol in the Wi-Fi signal transmission and reception process, the electronic device can process the frequency domain channel response corresponding to the symbol into the time domain channel impulse response, and then calculate the power delay spectrum.

[0071] The transmission and reception of Wi-Fi signals may involve multiple transmitting antenna elements and multiple receiving antenna elements. In this case, a single symbol may have multiple frequency domain channel responses in the antenna domain, and correspondingly, multiple power delay spectra can be calculated. After obtaining multiple power delay spectra for the same symbol, these spectra can be averaged, and the results can be used for subsequent steps.

[0072] Step 220: Search for the sampling point with the highest power in the power delay spectrum and use it as the centroid sampling point. Then, cut and splice several sampling points before the centroid sampling point to the end of the power delay spectrum to obtain the target power delay spectrum.

[0073] The electronic device can search for the sampling point with the highest power in the power delay spectrum and use it as the centroid sampling point. The electronic device can then truncate all sampling points before the centroid sampling point and concatenate the truncated sampling points to the end of the power delay spectrum to obtain the target power delay spectrum.

[0074] See Figure 3 This is a schematic diagram of the power delay spectrum provided in an embodiment of this application, as shown below. Figure 3 As shown, the power delay spectrum contains 256 sampling points, and the sampling points within the small box are the sampling points before the centroid sampling point.

[0075] See Figure 4 This is provided as an embodiment of the present application. Figure 3 A schematic diagram of the target power time delay spectrum corresponding to the medium power time delay spectrum, as shown below.Figure 4 As shown, Figure 3 By extracting and splicing multiple sampling points before the centroid sampling point in the medium power delay spectrum and then adding them to the 256th sampling point in the original power delay spectrum, the target power delay spectrum can be obtained.

[0076] Step 230: Based on the preset first length front window and the preset second length rear window, select target sampling points in the target power delay spectrum, and use the sampling points other than the target sampling points as noise sampling points; wherein, the front window is placed at the front end of the target power delay spectrum, and the rear window is placed at the back end of the target power delay spectrum.

[0077] The electronic device can place a first-length front window at the very beginning of the target power delay spectrum and a second-length rear window at the very end, thus using the sampling points within the front and rear windows as target sampling points. Sampling points on the target power delay spectrum other than the target sampling points are considered noise sampling points. Noise sampling points refer to sampling points corresponding to channels other than the effective channels (including LOS and NLOS channels) in multipath transmission.

[0078] See Figure 5 This is a schematic diagram of windowing the target power delay spectrum according to an embodiment of this application, as shown below. Figure 5 As shown, the front window is positioned at the very beginning of the target power delay spectrum, and the rear window is positioned at the very end. The area between the front and rear windows is a noise region containing multiple noise sampling points.

[0079] For example, the target power delay spectrum includes 256 sampling points. The first length of the front window is 90 and the second length of the back window is 30. The first 90 sampling points of the target power delay spectrum are selected as target sampling points using the front window and the last 30 sampling points of the target power delay spectrum are selected as target sampling points using the back window. Then, the 91st to 226th sampling points are noise sampling points.

[0080] See Figure 6 This is a schematic diagram of OFDM (Orthogonal Frequency Division Multiplexing) symbols provided in an embodiment of this application, as shown below. Figure 6 As shown, the OFDM mechanism will process the front part of a symbol ( Figure 6 The gray portion on the right is copied to the next symbol as a cyclic prefix (CP) to eliminate interference to the next symbol. Since the time length of the cyclic prefix is ​​greater than the maximum time domain spread of the wireless channel, the multipath component of one symbol will not interfere with the next symbol.

[0081] See Figure 7This is a schematic diagram showing an OFDM symbol cutoff from an embodiment of this application, as shown below. Figure 7 As shown, the positions of each symbol are determined in time. When truncating data according to the FFT (fast Fourier transform) window, due to the existence of the cyclic prefix, the FFT window can still ensure that the complete data is truncated even if it is offset by at most one cyclic prefix length.

[0082] Depending on the application requirements, the timing position for data truncation in the FFT window can be preset. For example, the timing position can be set to the middle of the cyclic prefix. For instance, if the number of sampling points for a symbol is 256 without the cyclic prefix, the cyclic prefix length is 16, and the FFT window length is 256, then when truncating data from the 272 sampling points including the cyclic prefix, as long as the offset is within the range of the cyclic prefix, the complete data from all 256 sampling points can be truncated.

[0083] By using the aforementioned truncation and splicing, and the filtering of the front and back windows in this application, it can be ensured that the selected target sampling points include the sampling points corresponding to all valid channels.

[0084] In one embodiment, before performing step 230, the electronic device can determine the first length based on the first coefficient and the length of the cyclic prefix. The first coefficient is greater than one. For example, the first length can be determined using the following formula (1):

[0085] (1)

[0086] Where w1 represents the first length; β is the first coefficient, which is greater than one, for example, the first coefficient is 2; CP is the length of the cyclic prefix.

[0087] The electronic device can determine the second length based on the second coefficient and the length of the cyclic prefix. The second length is less than one. For example, the second length can be determined using the following formula (2):

[0088] (2)

[0089] Where w2 represents the second length; α is the second coefficient, which is less than one, for example, the second coefficient is 0.5; CP is the length of the cyclic prefix.

[0090] Step 240: Determine the noise threshold power based on the power of multiple noise sampling points.

[0091] Electronic devices can calculate the average power of multiple noise sampling points and use this average noise power as the noise threshold power.

[0092] Step 250: Select target sampling points whose power exceeds the noise threshold in the front and rear windows to obtain the specified sampling points.

[0093] Electronic devices can check whether the power corresponding to each target sampling point exceeds the noise threshold power, thereby filtering out target sampling points whose power exceeds the noise threshold power. These filtered target sampling points are called designated sampling points.

[0094] Step 260: Connect the front and back ends of the target power delay spectrum in a loop, and determine whether the distance between the specified sampling point closest to the front end in the back window and the centroid sampling point exceeds the sampling point number threshold.

[0095] Step 270: If yes, determine the specified sampling point closest to the front end in the back window as the LOS sampling point, and the remaining specified sampling points as NLOS sampling points.

[0096] The electronic device can cyclically connect the front and back ends of the target power delay spectrum, in which case the last sampling point of the target power delay spectrum is connected to the first sampling point. The electronic device can determine whether the distance between the specified sampling point closest to the front end in the back window and the centroid sampling point exceeds the sampling point number threshold. Here, the sampling point number threshold can be configured as needed; an exemplary sampling point number threshold is 2.

[0097] by Figure 5 For example, the designated sampling point closest to the front end in the back window is the first designated sampling point from left to right. The distance between this designated sampling point and the centroid sampling point after passing the 256th sampling point from left to right is checked to see if it exceeds the sampling point number threshold.

[0098] In one scenario, if so, it means that the distance between the designated sampling point closest to the front end in the back window and the centroid sampling point is large enough that the power is not affected by the power of the centroid sampling point. In this case, the designated sampling point closest to the front end in the back window can be determined as the LOS sampling point, while the other designated sampling points are NLOS sampling points.

[0099] In another scenario, if not, it indicates that the distance between the designated sampling point closest to the front end within the rear window and the centroid sampling point is relatively small, and the power may be affected by the power of the centroid sampling point. In this case, if there are other designated sampling points within the rear window that are even closer to the centroid sampling point, they will also be affected by the centroid sampling point. Therefore, the centroid sampling point can be directly identified as the LOS sampling point, and the remaining designated sampling points can be identified as NLOS sampling points.

[0100] Furthermore, if there are no designated sampling points in the back window, in other words, if the power of all target sampling points in the back window does not exceed the noise power threshold, the centroid sampling point can be directly determined as the LOS sampling point, and the remaining designated sampling points can be determined as NLOS sampling points.

[0101] Step 280: Based on multiple NLOS sampling points, determine the cleaned frequency domain channel response corresponding to the frequency domain channel response.

[0102] After determining multiple NLOS sampling points, the electronic device can determine the cleaned frequency domain channel response corresponding to the frequency domain channel response.

[0103] In one embodiment, see Figure 8 This is provided as an embodiment of the present application. Figure 2 A detailed flowchart of step 280 is shown below. Figure 8 As shown, the specific process for determining the frequency domain channel response after cleaning may include the following steps 281 to 283.

[0104] Step 281: In the time-domain channel impulse response corresponding to the frequency-domain channel response, retain the time-domain channel estimates corresponding to multiple NLOS sampling points, and set the time-domain channel estimates corresponding to the remaining sampling points to zero to obtain the cleaned time-domain channel impulse response.

[0105] The electronic device may include time-domain channel estimates of multiple NLOS sampling points in the time-domain channel impulse response, and set the time-domain channel estimates of the remaining sampling points (including LOS sampling points) to zero, thereby obtaining the cleaned time-domain channel impulse response.

[0106] Step 282: Perform zero-padding on the time-domain channel impulse response after cleaning.

[0107] Electronic devices can perform zero-padding on the cleaned time-domain channel impulse response, thereby increasing the number of sampling points in the cleaned time-domain channel impulse response. The added sampling points correspond to zero power. For example, the cleaned time-domain channel impulse response includes M sampling points, and adding M sampling points with zero power makes the cleaned time-domain channel impulse response include 2M sampling points.

[0108] Step 283: Perform a Fourier transform on the cleaned time-domain channel impulse response after zero-padding to obtain the cleaned frequency-domain channel response.

[0109] Electronic devices can perform a Fourier transform on the zero-padded time-domain channel impulse response to obtain the cleaned frequency-domain channel response. Since the time-domain channel estimates corresponding to sampling points other than NLOS sampling points in the cleaned time-domain channel impulse response are set to zero, noise removal and LOS background removal are achieved. The zero-padded processing further reduces noise. Therefore, the cleaned frequency-domain channel response can accurately characterize the impact of passive target behavior on the wireless channel.

[0110] Step 290: Based on the cleaned frequency domain channel response at multiple observation times, perform behavior recognition on passive targets.

[0111] Here, the observation time can be the time when the frequency domain channel response is obtained after cleaning. The observation time can be configured as needed. For example, each data packet during Wi-Fi signal transmission can be set to correspond to one observation time, or every two data packets can be set to correspond to one observation time.

[0112] Electronic devices accurately identify the behavior of passive targets during a given period based on the cleaned frequency domain channel response at multiple consecutive observation times.

[0113] The above measures can resolve the errors that occur when directly selecting the LOS channel based on the power corresponding to the effective channel. In the actual transmission of Wi-Fi signals, due to factors such as the transmission medium, the sampling point corresponding to the maximum power in the power delay spectrum may not be the sampling point corresponding to the LOS channel. This application's solution first filters the sampling points corresponding to the effective channel using a front window, a back window, and a noise threshold power. Starting from the back window, it searches for specified sampling points whose distance from the centroid sampling point exceeds a threshold for the number of sampling points. This first searches for the sampling point corresponding to the LOS channel in time. Only when it cannot be found is the centroid sampling point determined as the LOS sampling point. In this case, the data corresponding to the LOS channel can be determined more accurately, effectively cleaning the frequency domain channel response, thereby achieving accurate behavior recognition.

[0114] In one embodiment, during the determination of the noise threshold power, noise threshold powers corresponding to multiple consecutive symbols can be obtained, and these multiple noise threshold powers can be smoothed to obtain a new noise threshold power. Here, the number of symbols used for smoothing can be configured as needed.

[0115] For example, smoothing can be performed using the following formula (3):

[0116] (3)

[0117] Where, n lThe noise threshold power corresponding to the l-th symbol in the data packet is represented by γ; γ represents the noise smoothing factor, which is greater than zero and less than one; n l-1 This represents the noise threshold power corresponding to the (l-1)th symbol in the data packet; m l This represents the noise threshold power after smoothing.

[0118] When smoothing is performed using the noise threshold power corresponding to two consecutive symbols, it can be done directly using formula (3). If smoothing is performed using the noise threshold power corresponding to at least three consecutive symbols, it can be done first based on the noise threshold power corresponding to the first two symbols, and then the smoothing result can be smoothed with the noise threshold power corresponding to the next symbol, and so on, until the noise threshold power of all symbols is smoothed.

[0119] A new noise threshold power can be obtained by smoothing the noise threshold power corresponding to multiple symbols, which can yield a more accurate noise threshold power.

[0120] In one embodiment, see Figure 9 This is provided as an embodiment of the present application. Figure 2 A detailed flowchart of step 290 is shown below. Figure 9 As shown, the specific process for executing step 290 may include steps 291 to 294.

[0121] Step 291: Construct a two-dimensional matrix based on the cleaned frequency domain channel response at multiple observation times;

[0122] The frequency domain channel estimate corresponding to multiple sampling points in the cleaned frequency domain channel response at each observation time can be used as a column in a two-dimensional matrix. Based on the cleaned frequency domain channel response corresponding to multiple consecutive observation times, a two-dimensional matrix can be constructed.

[0123] Step 292: Perform a short-time Fourier transform on the two-dimensional matrix to obtain a spectral image;

[0124] After obtaining the two-dimensional matrix, a short-time Fourier transform (STFT) can be performed on the matrix to obtain a spectral image.

[0125] Step 293: Periodically cut the spectral image to obtain the image to be processed;

[0126] Here, the cycle duration can be configured based on experience. Generally, the cycle duration is the time required to complete one action in the application scenario, such as two seconds.

[0127] Electronic devices can cut out spectral images generated in the most recent period from a spectral image and use them as images to be processed.

[0128] Step 294: Input the image to be processed into the trained behavior recognition model to obtain the behavior recognition result.

[0129] The behavior recognition model can be trained from any network model used for classification, such as CNN (Convolutional Neural Networks), TCN (Temporal Convolutional Network), or a hybrid model of CNN and RNN (Recurrent Neural Networks).

[0130] After obtaining the image to be processed, it can be input into the behavior recognition model, which will then process the image to obtain the behavior recognition result.

[0131] By employing the above measures, neural network models can be used to process the cleaned frequency domain channel response data, thereby obtaining behavior recognition results.

[0132] In one embodiment, for each behavior category to be identified, people of different body types can continuously perform the action of that behavior category in an indoor environment to obtain the frequency domain channel response corresponding to that behavior category. For example, if the behavior categories to be identified include waving, spreading hands, clapping, running, and falling, then the frequency domain channel response corresponding to waving can be obtained when people of different body types continuously perform the waving action in an indoor environment; similarly, the frequency domain channel responses corresponding to spreading hands, clapping, clapping, and falling can be obtained.

[0133] After obtaining the frequency domain channel response corresponding to each behavior category, the electronic device can clean the frequency domain channel response and convert it into a sample spectral image. The cleaning process and conversion method can be referred to the relevant description above, and will not be repeated here.

[0134] The electronic device can periodically crop multiple sample images from each sample spectral image. Here, the period duration can be the same as the period duration for cropping spectral images in subsequent recognition processes. After cropping the sample images, behavior category labels can be added to the sample images based on the behavior category corresponding to the sample spectral images.

[0135] After obtaining multiple sample images carrying behavior category labels, supervised learning can be performed on the network model based on the sample images to train a behavior recognition model.

[0136] Figure 10 This is a block diagram of a behavior recognition device based on frequency domain channel response according to an embodiment of the present invention, as shown below. Figure 10 As shown, the device may include:

[0137] The first determining module 1010 is used to determine the corresponding power delay spectrum for the frequency domain channel response of the Wi-Fi signal;

[0138] The interception module 1020 is used to search for the sampling point with the largest power in the power delay spectrum, take it as the centroid sampling point, and intercept and splice several sampling points before the centroid sampling point to the end of the power delay spectrum to obtain the target power delay spectrum.

[0139] The first screening module 1030 is used to screen target sampling points in the target power delay spectrum according to a preset first length front window and a preset second length rear window, and to treat sampling points other than the target sampling points as noise sampling points; wherein, the front window is placed at the beginning of the target power delay spectrum and the rear window is placed at the end of the target power delay spectrum;

[0140] The second determining module 1040 is used to determine the noise threshold power based on the power of multiple noise sampling points;

[0141] The second filtering module 1050 is used to filter out target sampling points whose power exceeds the noise threshold power in the front window and the rear window, and obtain the specified sampling points.

[0142] The judgment module 1060 is used to connect the front and back ends of the target power delay spectrum in a loop and determine whether the distance between the specified sampling point closest to the front end in the back window and the centroid sampling point at the connection point exceeds the sampling point number threshold.

[0143] The third determining module 1070 is used to determine, if so, the specified sampling point closest to the front end in the rear window as the LOS sampling point, and the other specified sampling points as NLOS sampling points;

[0144] The fourth determining module 1080 is used to determine the cleaned frequency domain channel response corresponding to the frequency domain channel response based on multiple NLOS sampling points;

[0145] The identification module 1090 is used to identify the behavior of passive targets based on the cleaned frequency domain channel response at multiple observation times.

[0146] The specific implementation process of the functions and roles of each module in the above-mentioned device can be found in the implementation process of the corresponding steps in the behavior recognition method based on frequency domain channel response, and will not be repeated here.

[0147] The apparatuses and methods disclosed in the several embodiments provided in this application can also be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatuses, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0148] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0149] If a function is implemented as a software module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

Claims

1. A behavior recognition method based on frequency domain channel response, characterized in that, include: Determine the corresponding power delay spectrum for the frequency domain channel response of the Wi-Fi signal; Search for the sampling point with the highest power in the power delay spectrum and use it as the centroid sampling point. Then, cut and splice several sampling points before the centroid sampling point to the end of the power delay spectrum to obtain the target power delay spectrum. Based on a preset first-length front window and a preset second-length rear window, target sampling points are selected from the target power delay spectrum, and sampling points other than the target sampling points are used as noise sampling points; wherein, the front window is placed at the very beginning of the target power delay spectrum, and the rear window is placed at the very end of the target power delay spectrum, the first length is determined based on a first coefficient and the length of the cyclic prefix, the first coefficient being greater than one, and the second length is determined based on a second coefficient and the length of the cyclic prefix, the second coefficient being less than one; The noise threshold power is determined based on the power of multiple noise sampling points; Target sampling points whose power exceeds the noise threshold are selected from the front window and the rear window to obtain designated sampling points; The front and back ends of the target power delay spectrum are cyclically connected. It is determined whether the distance between the designated sampling point closest to the front end in the back window and the centroid sampling point at the connection point exceeds the sampling point number threshold. If so, the designated sampling point closest to the front end within the rear window is determined as the LOS sampling point, and the remaining designated sampling points are determined as NLOS sampling points; Based on multiple NLOS sampling points, the cleaned frequency domain channel response corresponding to the frequency domain channel response is determined; Based on the cleaned frequency domain channel response at multiple observation times, behavior recognition is performed on passive targets.

2. The method according to claim 1, characterized in that, The method further includes: If not, the centroid sampling point is determined to be a LOS sampling point, and the remaining designated sampling points are NLOS sampling points; Return to the step of determining the cleaned frequency domain channel response from the frequency domain channel response based on multiple NLOS sampling points.

3. The method according to claim 1, characterized in that, The determination of the noise threshold power based on the power of multiple noise sampling points includes: Calculate the average power of multiple noise sampling points as the noise threshold power.

4. The method according to claim 3, characterized in that, The method further includes: After obtaining the noise threshold power corresponding to multiple consecutive symbols, the multiple noise threshold powers are smoothed to obtain a new noise threshold power.

5. The method according to claim 1, characterized in that, The step of determining the cleaned frequency domain channel response corresponding to the frequency domain channel response based on multiple NLOS sampling points includes: In the time-domain channel impulse response corresponding to the frequency-domain channel response, the time-domain channel estimate corresponding to the plurality of NLOS sampling points is retained, and the time-domain channel estimate corresponding to the remaining sampling points is set to zero to obtain the cleaned time-domain channel impulse response. The time-domain channel impulse response after cleaning is zero-padding. The time-domain channel impulse response after zero-padding is subjected to Fourier transform to obtain the cleaned frequency-domain channel response.

6. The method according to claim 1, characterized in that, The method for identifying the behavior of passive targets based on the cleaned frequency domain channel response at multiple observation times includes: A two-dimensional matrix is ​​constructed based on the cleaned frequency domain channel response at multiple observation times; Perform a short-time Fourier transform on the two-dimensional matrix to obtain a spectral image; The spectral image is periodically cut to obtain the image to be processed; The image to be processed is input into the trained behavior recognition model to obtain the behavior recognition result.

7. A behavior recognition device based on frequency domain channel response, characterized in that, include: The first determining module is used to determine the corresponding power delay spectrum for the frequency domain channel response of the Wi-Fi signal; The truncation module is used to search for the sampling point with the highest power in the power delay spectrum, take it as the centroid sampling point, and truncate and splice several sampling points before the centroid sampling point to the end of the power delay spectrum to obtain the target power delay spectrum. The first filtering module is used to filter target sampling points in the target power delay spectrum according to a preset first length front window and a preset second length back window, and to treat sampling points other than the target sampling points as noise sampling points; wherein, the front window is placed at the beginning of the target power delay spectrum, the back window is placed at the end of the target power delay spectrum, the first length is determined based on a first coefficient and the length of the cyclic prefix, the first coefficient is greater than one, and the second length is determined based on a second coefficient and the length of the cyclic prefix, the second coefficient is less than one; The second determining module is used to determine the noise threshold power based on the power of multiple noise sampling points; The second filtering module is used to filter out target sampling points whose power exceeds the noise threshold power in the front window and the rear window, and obtain the specified sampling points. The judgment module is used to connect the front and back ends of the target power delay spectrum in a loop and determine whether the distance between the specified sampling point closest to the front end in the back window and the centroid sampling point at the connection point exceeds the sampling point number threshold. The third determining module is used to determine, if so, the specified sampling point closest to the front end in the rear window as the LOS sampling point, and the other specified sampling points as NLOS sampling points; The fourth determining module is used to determine the cleaned frequency domain channel response corresponding to the frequency domain channel response based on multiple NLOS sampling points; The identification module is used to identify the behavior of passive targets based on the cleaned frequency domain channel response at multiple observation times.

8. An electronic device, characterized in that, The electronic device includes: processor; Memory used to store processor-executable instructions; The processor is configured to execute the behavior recognition method based on frequency domain channel response as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The storage medium stores a computer program that can be executed by a processor to perform the behavior recognition method based on frequency domain channel response as described in any one of claims 1-6.

Images