Behavior recognition methods and devices based on exponential fitting, electronic devices, and storage media.

Abstract

This application provides a behavior recognition method, apparatus, electronic device, and storage medium based on exponential fitting, comprising: determining the corresponding power delay spectrum for the channel frequency response of a Wi-Fi signal at each observation time; fitting an exponential probability density function based on the power distribution of each sampling point in the power delay spectrum; performing a Fourier transform on the probability density function to obtain a frequency domain correlation function; determining filter coefficients based on the frequency domain correlation function and a preset frequency domain signal-to-noise ratio; filtering the channel frequency response based on the filter coefficients to obtain a denoised channel frequency response; and performing behavior recognition of passive targets based on the denoised channel frequency response at multiple observation times. This application's solution effectively denoises the channel frequency response of a Wi-Fi signal using exponential fitting, thereby enabling accurate behavior recognition based on the denoised channel frequency response.

Description

Technical Field

[0001] This application relates to the field of behavior recognition technology, and in particular to a behavior recognition method and apparatus, electronic device, and computer-readable storage medium based on exponential fitting. Background Technology

[0002] A key development direction for next-generation wireless communication technology is the integration of communication sensing and radar. Future Wi-Fi signals will not only transmit data but also sense passive targets (people or other targets without terminal devices). Utilizing wireless signals for behavior recognition of passive targets can play a crucial role in smart home scenarios. For example, it can free users from remote controls, improving the user experience of controlling smart devices, or it can effectively monitor a user's health status. In indoor scenarios, Wi-Fi transceivers (e.g., routers) can be placed in specific locations. During Wi-Fi signal transmission and reception, Channel State Information (CSI) can be determined for sensing the behavior of passive targets. However, CSI obtained through channel estimation often contains a large amount of noisy data, and directly applying CSI usually cannot achieve accurate behavior recognition. Summary of the Invention

[0003] The purpose of this application is to provide a behavior recognition method, device, electronic device, and storage medium based on exponential fitting, which is used to effectively reduce the noise of the channel frequency response of Wi-Fi signals by means of exponential fitting, so as to accurately recognize behavior based on the noise-reduced channel frequency response.

[0004] On the one hand, this application provides a behavior recognition method based on exponential fitting, including:

[0005] For each observation moment of the Wi-Fi signal, determine the corresponding power delay spectrum;

[0006] Based on the power distribution at each sampling point in the power delay spectrum, an exponential distribution probability density function is fitted.

[0007] Perform a Fourier transform on the probability density function to obtain the frequency domain correlation function;

[0008] The filter coefficients are determined based on the frequency domain correlation function and the preset frequency domain signal-to-noise ratio.

[0009] The channel frequency response is filtered based on the filter coefficients to obtain the noise-reduced channel frequency response.

[0010] Based on the denoised channel frequency response at multiple observation times, behavior recognition of passive targets is performed.

[0011] In one embodiment, fitting the probability density function of an exponential distribution based on the power distribution of each sampling point in the power delay spectrum includes:

[0012] The power delay spectrum is copied and stitched together to form a specified power delay spectrum; wherein the specified power delay spectrum includes two power delay spectra, with the rear end of the first power delay spectrum connected to the front end of the second power delay spectrum;

[0013] The sampling point corresponding to the maximum power in the second power delay spectrum of the specified power delay spectrum is taken as the specified sampling point;

[0014] Determine the zero-point sampling point by searching towards the front end from the specified sampling point;

[0015] The power distribution of multiple sampling points from the zero-point sampling point to the end of the specified power delay spectrum is fitted using an exponential distribution function to obtain the probability density function.

[0016] In one embodiment, determining the zero-point sampling point by searching towards the front end from the designated sampling point includes:

[0017] The zero-point sampling point is determined by taking the specified sampling point in the direction towards the front end and then setting a preset first number of sampling points as the zero-point sampling point.

[0018] In one embodiment, determining the zero-point sampling point by searching towards the front end from the designated sampling point includes:

[0019] Starting from the specified sampling point, find the sampling point corresponding to the minimum power among the second preset number of sampling points in the front-end direction, and use it as the zero-point sampling point.

[0020] In one embodiment, the step of performing behavior recognition of passive targets based on the denoised channel frequency response at multiple observation times includes:

[0021] From the denoised channel frequency responses at the multiple observation times, select one denoised channel frequency response as the target channel frequency response;

[0022] Each denoised channel frequency response is multiplied by the conjugate of the target channel frequency response to obtain a target vector. The mean of all values ​​in the target vector is subtracted from each value in the target vector to obtain a specified vector.

[0023] The specified vectors corresponding to multiple consecutive observation times within the observation window are used to construct an observation matrix, and the covariance matrix of the observation matrix is ​​calculated.

[0024] A spectral search is performed on the covariance matrix to obtain the channel length of the NLOS channel and the Doppler observation velocity of the passive target;

[0025] The behavior category of the passive target is determined based on the channel length and Doppler observation velocity corresponding to multiple observation windows.

[0026] In one embodiment, determining the behavior category of the passive target based on the channel length and Doppler observation velocity corresponding to multiple observation windows includes:

[0027] A behavior recognition parameter matrix is ​​constructed based on the channel length and Doppler observation velocity corresponding to multiple observation windows.

[0028] The behavior recognition parameter matrix is ​​converted into a spectral image, and sub-spectral images generated within a specified time period are cropped from the spectral image.

[0029] The sub-spectral image is input into the trained behavior recognition model to obtain the behavior recognition result corresponding to the sub-spectral image.

[0030] In one embodiment, after performing a spectral search on the covariance matrix to obtain the channel length of the NLOS channel and the Doppler observation velocity of the passive target, the method further includes:

[0031] After obtaining multiple channel lengths and multiple Doppler observation velocities from the covariance matrix through spectral search, the multiple channel lengths and multiple Doppler observation velocities are clustered to obtain several clusters;

[0032] The channel length and Doppler observation speed at the cluster center are selected for behavior recognition.

[0033] On the other hand, this application provides a behavior recognition device based on exponential fitting, comprising:

[0034] The first determining module is used to determine the corresponding power delay spectrum for the channel frequency response of the Wi-Fi signal at each observation time.

[0035] The fitting module is used to fit the probability density function of the exponential distribution based on the power distribution of each sampling point in the power delay spectrum.

[0036] The transformation module is used to perform a Fourier transform on the probability density function to obtain the frequency domain correlation function;

[0037] The second determining module is used to determine the filter coefficients based on the frequency domain correlation function and the preset frequency domain signal-to-noise ratio;

[0038] A filtering module is used to filter the channel frequency response based on the filter coefficients to obtain a noise-reduced channel frequency response.

[0039] The identification module is used to perform behavior identification of passive targets based on the noise-reduced channel frequency response at multiple observation times.

[0040] Furthermore, this application provides an electronic device, the electronic device comprising:

[0041] processor;

[0042] Memory used to store processor-executable instructions;

[0043] The processor is configured to execute the above-described behavior recognition method based on exponential fitting.

[0044] Furthermore, this application provides a computer-readable storage medium storing a computer program that can be executed by a processor to perform the above-described behavior recognition method based on exponential fitting.

[0045] The proposed solution determines the filter coefficients based on exponential fitting, and then uses these coefficients to filter the channel frequency response to obtain the denoised channel frequency response. Based on the denoised channel frequency response at multiple observation times, accurate behavior recognition can be achieved. Attached Figure Description

[0046] 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.

[0047] Figure 1 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application;

[0048] Figure 2 A flowchart illustrating an embodiment of the behavior recognition method based on exponential fitting provided in this application;

[0049] Figure 3 A schematic diagram of the probability density function of the exponential distribution provided in an embodiment of this application;

[0050] Figure 4 Provided for an embodiment of this application Figure 2 A detailed flowchart of step 220 is shown below;

[0051] Figure 5 A schematic diagram of a specified power delay spectrum provided in an embodiment of this application;

[0052] Figure 6 Provided for an embodiment of this application Figure 2A detailed flowchart of step 260;

[0053] Figure 7 A flowchart illustrating a behavior recognition method provided in an embodiment of this application;

[0054] Figure 8 A block diagram of an exponential fitting-based behavior recognition device provided in an embodiment of this application. Detailed Implementation

[0055] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0056] 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.

[0057] like Figure 1 As 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 of the methods described in the following embodiments. 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 exponential fitting. The electronic device may receive Wi-Fi signals emitted by other devices. The following description uses the electronic device as the execution subject.

[0058] 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.

[0059] 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 exponential fitting provided in this application.

[0060] See Figure 2 The above is a flowchart illustrating a behavior recognition method based on exponential fitting provided in an embodiment of this application. Figure 2 As shown, the method may include steps 210-260.

[0061] Step 210: For the channel frequency response of the Wi-Fi signal at each observation time, determine the corresponding power delay spectrum.

[0062] During the implementation of this application's solution, the Channel Frequency Response (CFR), which serves as the channel state, needs to be determined from the Wi-Fi signal through channel estimation at multiple consecutive observation times. Here, the time interval between adjacent observation times can be configured as needed. For example, each data packet transmitted by the Wi-Fi signal corresponds to a time interval, or the time interval can be any of 0.01 seconds, 0.02 seconds, etc.

[0063] After obtaining the channel frequency response at each observation time, the channel frequency response in the frequency domain can be transformed to the time domain using an inverse Fourier transform to obtain the channel impulse response (CIR). Here, the inverse Fourier transform can be an inverse fast Fourier transform (IFFT). The electronic device can square the modulus of each tap of each channel impulse response to obtain the power delay profile (PDP).

[0064] Step 220: Based on the power distribution of each sampling point in the power delay spectrum, fit the probability density function of the exponential distribution.

[0065] The power delay spectrum can include the power corresponding to multiple sampling points, with each sampling point corresponding to a channel. The power distribution of each sampling point in the power delay spectrum conforms to an exponential distribution. Electronic devices can fit the power distribution to obtain the probability density function of the exponential distribution.

[0066] See Figure 3 This is a schematic diagram of the probability density function of the exponential distribution provided in an embodiment of this application, as shown below. Figure 3 As shown, the power distribution of multiple sampling points in the power delay spectrum follows an exponential distribution, therefore it can be fitted with... Figure 3The probability density function corresponding to the curve in the figure.

[0067] For example, the probability density function of the exponential distribution can be expressed by the following formula (1):

[0068] - / d

[0069] p(k)=σ*k*e (1)

[0070] Where k is the sampling point number, k is a positive integer; d is the x-coordinate corresponding to the maximum function value, d is greater than 0; e is the natural base; σ is the constant obtained by fitting.

[0071] The power distribution at each sampling point in the power delay spectrum reflects the intensity distribution of the channel corresponding to each sampling point. Therefore, the envelope of the curve indicated by the probability density function of the exponential distribution fitted with the power distribution is essentially determined by the maximum delay spread. Thus, the frequency domain correlation can be determined based on the probability density function.

[0072] Step 230: Perform a Fourier transform on the probability density function to obtain the frequency domain correlation function.

[0073] Electronic devices can perform a Fourier transform on the probability density function. Here, the Fourier transform can be a Fast Fourier Transform (FFT). After the Fourier transform, the frequency domain correlation function can be obtained. This frequency domain correlation function can be expressed as R. HH (Δf) represents the mapping relationship between the carrier spacing Δf (in Hertz) of adjacent sub-subs and the frequency domain correlation. Here, the carrier spacing Δf is a known quantity. In the frequency domain correlation function, 1 Δf, 2 Δf, 3 Δf...n Δf correspond to 1 frequency domain correlation.

[0074] Step 240: Determine the filter coefficients based on the frequency domain correlation function and the preset frequency domain signal-to-noise ratio.

[0075] Here, the frequency domain signal-to-noise ratio can be obtained from the channel estimation hardware and software system, or the frequency domain signal-to-noise ratio can be configured empirically.

[0076] Electronic devices can calculate the corresponding frequency domain correlation for multiple carrier number intervals using a frequency domain correlation function. The carrier number interval is the interval between other subcarriers selected for filtering and the subcarrier itself when filtering any given subcarrier. The carrier number interval is determined by the filtering order, which can be configured as needed. The filtering order tapLen ≤ 2d, where d is the abscissa corresponding to the maximum function value in formula (1). For example, if the filtering order is 3, the carrier number intervals are 1, 2, and 3. If filtering is applied to the 124th subcarrier, the 123rd and 125th subcarriers with a carrier number interval of 1, the 122nd and 126th subcarriers with a carrier number interval of 2, and the 121st and 127th subcarriers with a carrier number interval of 3 will be selected.

[0077] After obtaining multiple frequency domain correlations, the corresponding MMSE (Minimum Mean Squared Error) filter coefficients can be calculated for each frequency domain correlation. For example, the filter coefficients can be calculated using the following formula (2):

[0078]

[0079] Where coeff represents the filter coefficients; γ FD R is the frequency domain signal-to-noise ratio; HH For frequency domain correlation; I is the identity matrix.

[0080] Step 250: Filter the channel frequency response based on the filter coefficients to obtain the noise-reduced channel frequency response.

[0081] Electronic devices can filter the channel frequency response based on filter coefficients corresponding to multiple carrier number intervals to obtain a denoised channel frequency response. The denoised channel frequency response removes data from the noisy channel.

[0082] Step 260: Based on the noise-reduced channel frequency response at multiple observation times, perform behavior recognition for passive targets.

[0083] After performing the above process on the channel frequency response at each observation time, the denoised channel frequency response at each observation time can be obtained. Based on the denoised channel frequency response at multiple observation times, the electronic device can determine the feature information generated by the behavior of passive targets within the corresponding time period at multiple observation times, and then use this to perform behavior recognition.

[0084] Through the above measures, without calculating the noise threshold, without screening effective channels, and without specifically calculating the delay spread, the delay spread, the most core parameter of the fading channel, is determined by the change trend of the multipath indicated by the envelope. Then, the delay spread is mapped to the frequency domain to obtain the frequency domain correlation (the frequency domain correlation function is used to characterize the frequency domain correlation). Furthermore, noise reduction can be performed through the Wiener criterion, thereby enabling accurate behavior recognition based on the frequency response of the denoised channel.

[0085] In one embodiment, see Figure 4 This is provided as an embodiment of the present application. Figure 2 A detailed flowchart of step 220 is shown below. Figure 4 As shown, when executing step 220, steps 221 to 224 can be executed.

[0086] Step 221: Copy the power delay spectrum and stitch it together to form the specified power delay spectrum; wherein, the specified power delay spectrum includes two power delay spectra, with the back end of the first power delay spectrum connected to the front end of the second power delay spectrum.

[0087] After the electronic device copies the power delay spectrum, it obtains two power delay spectra. Then, it stitches the two power delay spectra together to obtain the specified power delay spectrum.

[0088] See Figure 5 This is a schematic diagram of a specified power delay spectrum provided in an embodiment of this application, as shown below. Figure 5 As shown, the specified power delay spectrum includes two power delay spectra. The sampling point numbers of the second power delay spectrum remain unchanged, while the sampling point numbers of the first power delay spectrum are all reduced by 256. Therefore, the sampling point numbers in the specified power delay spectrum range from -256 to 255.

[0089] Step 222: The sampling point corresponding to the maximum power in the second power delay spectrum of the specified power delay spectrum is taken as the specified sampling point.

[0090] The sampling point corresponding to the maximum power in the second power delay spectrum is taken as the designated sampling point. Generally, the designated sampling point is in the middle part of the designated power delay spectrum. Ideally, the designated sampling point is the sampling point with the number 0, or the designated sampling point is around the sampling point with the number 0.

[0091] Step 223: Determine the zero sampling point by searching towards the front end from the specified sampling point.

[0092] After determining the specified sampling points, a sampling point can be searched in the forward direction (from the sampling point index in descending order) as the zero-point sampling point. This zero-point sampling point is used as the first sampling point when subsequently fitting the probability density function.

[0093] In one embodiment, the electronic device can determine zero-point sampling points at preset first intervals in the direction from a specified sampling point towards the front end. Here, the first number can be configured as needed; for example, it can be any number such as 6, 7, or 8. For example, if the specified sampling point number is 3, the zero-point sampling point number at 7 sampling points towards the front end is -5.

[0094] In one embodiment, the electronic device can start from a specified sampling point and search for the sampling point corresponding to the minimum power among a preset second number of sampling points in the front-end direction, and use this as the zero-point sampling point. Here, the second number can be configured as needed; for example, the second number can be any number such as 10 or 15. For example, if the specified sampling point number is 3 and the second number is 10, the sampling point corresponding to the minimum power is found from the sampling points with numbers from -7 to 2, and used as the zero-point sampling point.

[0095] Step 224: Fit the power distribution of multiple sampling points from the zero-point sampling point to the end of the specified power delay spectrum using the exponential distribution function to obtain the probability density function.

[0096] Using the zero-point sampling point as the first sampling point, the probability density function of the specified distribution is fitted to the power distribution of multiple sampling points from the zero-point sampling point to the end of the specified power delay spectrum. Since the zero-point sampling point may not be the sampling point with index 1 in the specified power delay spectrum, the index of each sampling point is changed when selecting the zero-point sampling point as the first sampling point to fit the probability density function.

[0097] In this case, in the fitted formula (1) above, d is the difference between the index of the specified sampling point and the index of the zero sampling point plus one. For example, if the index of the specified sampling point in the specified power delay spectrum is 3 and the index of the zero sampling point in the specified power delay spectrum is -5, then d is 9.

[0098] When k = d, p(k) reaches its maximum value. Therefore, the constant in formula (1) can be expressed as: In this case, formula (1) can be converted into formula (3):

[0099]

[0100] Where k is the sampling point number, k is a positive integer; d is the x-coordinate corresponding to the maximum function value, d is greater than 0; e is the natural base; p(d) is the maximum function value.

[0101] By taking the above measures, when the specified sampling point is too far ahead in the power delay spectrum, multiple sampling points can be set before the specified sampling point by splicing and reselecting the zero sampling point, so that the power distribution of multiple sampling points from the specified sampling point onwards satisfies the exponential distribution.

[0102] The behavior of passive targets can affect the multipath channels of Wi-Fi signals, causing changes in the channel frequency response, which serves as channel state information. These multipath transmission channels include LOS (Line of Sight) and NLOS (Not Line of Sight) channels. Indoor still objects remain stationary, and the LOS channel data is unaffected by passive targets. Therefore, the NLOS channel data in the channel state information represents the impact of passive target activity on the wireless channel. Thus, after denoising the channel frequency response, LOS channel data can be further eliminated.

[0103] In one embodiment, see Figure 6 This is provided as an embodiment of the present application. Figure 2 A detailed flowchart of step 260 is shown below. Figure 6 As shown, when executing step 260, steps 261 to 265 can be executed in detail.

[0104] Step 261: Select one of the denoised channel frequency responses from multiple observation times as the target channel frequency response.

[0105] Electronic devices can select the denoised channel frequency response at the first observation time as the target channel frequency response.

[0106] Step 262: Multiply each denoised channel frequency response by the target channel frequency response using the conjugate multiplication to obtain the target vector, and subtract the mean of all values ​​in the target vector from each value in the target vector to obtain the specified vector.

[0107] After determining the target channel frequency response, the other denoised channel frequency response distributions are multiplied by the target channel frequency response using the conjugate multiplication to obtain the target vector. Then, the mean of all values ​​in the target vector is subtracted from each value in the target vector to obtain the specified vector.

[0108] For example, the target vector can be calculated using the following formula (4):

[0109]

[0110] in, This represents the denoised channel frequency response from observation time 1 to observation time T; This indicates the frequency response of the selected target channel.

[0111] After obtaining the target vector, the mean of each value in the target vector is calculated, and then the mean is subtracted from each value to obtain the specified vector.

[0112] This measure can eliminate the observations from the LOS channel.

[0113] Step 263: Construct an observation matrix from the specified vectors corresponding to multiple consecutive observation times within the observation window, and calculate the covariance matrix of the observation matrix.

[0114] The observation window can be configured as needed. For example, if the time interval between adjacent observation times is 0.01 seconds, then the observation window can be 0.1 seconds.

[0115] The electronic device can divide multiple observation times based on the observation window, thereby assigning the multiple observation times to different observation windows. For a specified vector corresponding to multiple observation times within each observation window, a two-dimensional matrix is ​​constructed with the vertical axis representing the observation time and the horizontal axis representing the subcarrier index, serving as the observation matrix corresponding to that observation window. The electronic device can calculate the corresponding covariance matrix for this observation matrix. For example, the covariance matrix can be calculated using the following formula (5):

[0116] R XX =E[XX * (5)

[0117] Where X is the observation matrix; E is the identity matrix; R XX Let be the covariance matrix.

[0118] Step 264: Perform a spectral search on the covariance matrix to obtain the channel length of the NLOS channel and the Doppler observation velocity of the passive target.

[0119] Electronic devices can use the vertical axis as the guide vector and search for the channel length L of the NLOS channel in the covariance matrix using any of the spectrum peak search methods such as MUSIC (multiple signal classification algorithm), ESPRIT (Estimation of Signal Parameters using Rotational Invariance Techniques), or Pisarenko algorithm.

[0120] Electronic devices can use the horizontal axis as the guide vector and any spectral peak search method to search for the Doppler observation velocity V of the passive target in the covariance matrix.

[0121] In one embodiment, after the spectral search, the electronic device can obtain multiple channel lengths and multiple Doppler observation velocities from the covariance matrix, the multiple channel lengths and multiple Doppler observation velocities constituting multiple (V, L) combinations.

[0122] Electronic devices can cluster multiple channel lengths and multiple Doppler observation velocities; in other words, they can cluster multiple combinations of (V, L) to obtain several clusters. The electronic devices can then select the channel length and Doppler observation velocity at the cluster center for behavior recognition. In this case, the channel length and Doppler observation velocity obtained after clustering can more accurately characterize the actual behavior of passive targets.

[0123] Step 265: Determine the behavior category of the passive target based on the channel length and Doppler observation speed corresponding to multiple observation windows.

[0124] Electronic devices can construct technical features for behavior recognition based on the channel length and Doppler observation speed corresponding to multiple observation windows, thereby determining the behavior category of passive targets.

[0125] By taking the above measures, the LOS channel data in the channel frequency response after noise reduction can be eliminated, thereby achieving more accurate behavior recognition.

[0126] In one embodiment, see Figure 7 The above is a flowchart illustrating a behavior recognition method provided in an embodiment of this application. Figure 7 As shown, steps 710 to 730 are performed using a machine learning model to perform behavior recognition.

[0127] Step 710: Construct the behavior recognition parameter matrix based on the channel length and Doppler observation speed corresponding to multiple observation windows.

[0128] The electronic device can construct a column from the channel length and Doppler observation velocity corresponding to each observation window, thereby constructing an identification parameter matrix from columns corresponding to multiple consecutive observation windows. The horizontal axis of this identification parameter matrix corresponds to time.

[0129] Step 720: Convert the behavior recognition parameter matrix into a spectral image, and crop out the sub-spectral images generated within the specified time period from the spectral image.

[0130] The specified time period can be set as needed, and its duration can be the duration of a single completion of the behavior to be identified. For example, if the behavior to be identified includes hand gestures such as raising, waving, and clapping, and the duration of a single completion is 2 seconds, and the application scenario requires real-time identification of the behavior category, then the specified time period can be the most recent 2 seconds.

[0131] Step 730: Input the subspectral image into the trained behavior recognition model to obtain the behavior recognition result corresponding to the subspectral image.

[0132] The behavior recognition model is a network model used to classify behavior categories. This network model can be any of the following: CNN (Convolutional Neural Networks), TCN (Temporal Convolutional Network), or a combination of CNN and RNN (Recurrent Neural Networks). The behavior recognition model can be trained using supervised learning methods.

[0133] The electronic device inputs the cropped sub-spectral image into the behavior recognition model, and then processes the sub-spectral image through the behavior recognition model to obtain the behavior recognition result.

[0134] Through the above measures, behavior recognition can be achieved quickly and accurately with the help of machine learning methods.

[0135] Figure 8 This is a block diagram of a behavior recognition device based on exponential fitting according to an embodiment of the present invention, as shown below. Figure 8 As shown, the device may include:

[0136] The first determining module 810 is used to determine the corresponding power delay spectrum for the channel frequency response of the Wi-Fi signal at each observation time.

[0137] The fitting module 820 is used to fit an exponential distribution probability density function based on the power distribution of each sampling point in the power delay spectrum.

[0138] Transformation module 830 is used to perform a Fourier transform on the probability density function to obtain a frequency domain correlation function;

[0139] The second determining module 840 is used to determine the filter coefficients based on the frequency domain correlation function and the preset frequency domain signal-to-noise ratio;

[0140] The filtering module 850 is used to filter the channel frequency response based on the filter coefficients to obtain the noise-reduced channel frequency response.

[0141] The identification module 860 is used to perform behavior identification of passive targets based on the noise-reduced channel frequency response at multiple observation times.

[0142] 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 exponential fitting, and will not be repeated here.

[0143] 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.

[0144] 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.

[0145] 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 exponential fitting, characterized in that, include: For each observation moment of the Wi-Fi signal, determine the corresponding power delay spectrum; Based on the power distribution of each sampling point in the power delay spectrum, an exponential distribution probability density function is fitted, including: copying the power delay spectrum and stitching it together to form a specified power delay spectrum; wherein, the specified power delay spectrum includes two power delay spectra, the rear end of the first power delay spectrum is connected to the front end of the second power delay spectrum; the sampling point corresponding to the maximum power in the second power delay spectrum of the specified power delay spectrum is used as the specified sampling point; a zero-point sampling point is searched from the specified sampling point towards the front end; the power distribution of multiple sampling points from the zero-point sampling point to the rear end of the specified power delay spectrum is fitted using an exponential distribution function to obtain the probability density function; Perform a Fourier transform on the probability density function to obtain the frequency domain correlation function; The filter coefficients are determined based on the frequency domain correlation function and the preset frequency domain signal-to-noise ratio. The channel frequency response is filtered based on the filter coefficients to obtain the noise-reduced channel frequency response. Based on the denoised channel frequency responses at multiple observation times, behavior recognition of passive targets is performed, including: selecting a denoised channel frequency response as the target channel frequency response from the multiple observation times; performing conjugate multiplication of each denoised channel frequency response with the target channel frequency response to obtain a target vector, and subtracting the mean of all values ​​in the target vector from each value in the target vector to obtain a specified vector; constructing an observation matrix from the specified vectors corresponding to multiple consecutive observation times within the observation window, and calculating the covariance matrix of the observation matrix; performing a spectral search on the covariance matrix to obtain the channel length of the NLOS channel and the Doppler observation velocity of the passive target; and determining the behavior category of the passive target based on the channel length and Doppler observation velocity corresponding to multiple observation windows.

2. The method according to claim 1, characterized in that, The step of determining the zero-point sampling point by searching towards the front end from the specified sampling point includes: The zero-point sampling point is determined by taking the specified sampling point in the direction towards the front end and then setting a preset first number of sampling points as the zero-point sampling point.

3. The method according to claim 1, characterized in that, The step of determining the zero-point sampling point by searching towards the front end from the specified sampling point includes: Starting from the specified sampling point, find the sampling point corresponding to the minimum power among the second preset number of sampling points in the front-end direction, and use it as the zero-point sampling point.

4. The method according to claim 1, characterized in that, The determination of the passive target's behavior category based on the channel length and Doppler observation velocity corresponding to multiple observation windows includes: A behavior recognition parameter matrix is ​​constructed based on the channel length and Doppler observation velocity corresponding to multiple observation windows. The behavior recognition parameter matrix is ​​converted into a spectral image, and sub-spectral images generated within a specified time period are cropped from the spectral image. The sub-spectral image is input into the trained behavior recognition model to obtain the behavior recognition result corresponding to the sub-spectral image.

5. The method according to claim 1, characterized in that, After performing a spectral search on the covariance matrix to obtain the channel length of the NLOS channel and the Doppler observation velocity of the passive target, the method further includes: After obtaining multiple channel lengths and multiple Doppler observation velocities from the covariance matrix through spectral search, the multiple channel lengths and multiple Doppler observation velocities are clustered to obtain several clusters; The channel length and Doppler observation speed at the cluster center are selected for behavior recognition.

6. A behavior recognition device based on exponential fitting, characterized in that, include: The first determining module is used to determine the corresponding power delay spectrum for the channel frequency response of the Wi-Fi signal at each observation time. The fitting module is used to fit an exponential distribution probability density function based on the power distribution of each sampling point in the power delay spectrum. The process includes: copying the power delay spectrum and stitching it together to form a specified power delay spectrum; wherein the specified power delay spectrum includes two power delay spectra, with the rear end of the first power delay spectrum connected to the front end of the second power delay spectrum; using the sampling point corresponding to the maximum power in the second power delay spectrum as a specified sampling point; determining a zero-point sampling point and searching for it towards the front end; and fitting the power distribution of multiple sampling points from the zero-point sampling point to the rear end of the specified power delay spectrum using an exponential distribution function to obtain the probability density function. The transformation module is used to perform a Fourier transform on the probability density function to obtain the frequency domain correlation function; The second determining module is used to determine the filter coefficients based on the frequency domain correlation function and the preset frequency domain signal-to-noise ratio; The filtering module is used to filter the channel frequency response based on the filter coefficients to obtain the noise-reduced channel frequency response. The identification module is used to perform behavior identification of passive targets based on the denoised channel frequency responses at multiple observation times. This includes: selecting a denoised channel frequency response as the target channel frequency response from the multiple observation times; performing conjugate multiplication of each denoised channel frequency response with the target channel frequency response to obtain a target vector; subtracting the mean of all values ​​in the target vector from each value in the target vector to obtain a specified vector; constructing an observation matrix from the specified vectors corresponding to multiple consecutive observation times within the observation window; calculating the covariance matrix of the observation matrix; performing a spectral search on the covariance matrix to obtain the channel length of the NLOS channel and the Doppler observation velocity of the passive target; and determining the behavior category of the passive target based on the channel length and Doppler observation velocity corresponding to the multiple observation windows.

7. 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 exponential fitting as described in any one of claims 1-5.

8. 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 exponential fitting as described in any one of claims 1-5.

Images