Behavior recognition method and device based on channel estimation, electronic device, and storage medium

Abstract

The present application provides a behavior recognition method and device based on channel estimation, an electronic device, and a storage medium, including: extracting the frequency domain channel estimation value of each symbol on multiple subcarriers in the channel estimation information to obtain the extracted channel estimation information; converting it into time domain channel estimation information through inverse Fourier transform, and determining the power corresponding to each channel; dividing multiple windows with the largest power by a preset window length, and calculating the average power corresponding to the channels outside the window as the noise power; determining the threshold power by the noise power, screening out several windows with power greater than the threshold power, filtering out the windows with the largest power, and determining the target time domain channel estimation information based on the windows with filtered out channels; Fourier transforming the target time domain channel estimation information to obtain the target channel estimation result; processing the target channel estimation result with a behavior recognition model to obtain the behavior recognition result. The present application realizes accurate behavior recognition of passive targets.

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 channel estimation, an electronic device, and a computer-readable storage medium. Background Art

[0002] In the future, Wi-Fi signals will not only have data transmission functions, but also have the ability to perceive passive targets. When passive targets (people or other targets without terminal devices) move around in indoor environments, they will affect indoor Wi-Fi signals, causing changes in multipath channels. In this case, the channel estimation information will change accordingly, so the behavior of passive targets can be identified with the help of the channel estimation information. Summary of the invention

[0003] The purpose of the embodiments of the present application is to provide a behavior recognition method and device based on channel estimation, an electronic device, and a computer-readable storage medium for accurately identifying the behavior of a passive target.

[0004] On the one hand, the present application provides a behavior recognition method based on channel estimation, comprising:

[0005] For the channel estimation information of the Wi-Fi signal, extract the frequency domain channel estimation value of each symbol on the first specified number of subcarriers to obtain extracted channel estimation information;

[0006] Converting the extracted channel estimation information into time domain channel estimation information by inverse Fourier transform, and determining the power corresponding to each channel based on the time domain channel estimation information;

[0007] Divide multiple channels in the window with the largest power according to the preset window length, and calculate the average value of the corresponding power of the channels outside the window as the noise power;

[0008] Screening out a number of in-window channels whose power is greater than a threshold power from a plurality of in-window channels, filtering out the in-window channel with the largest power among the screened in-window channels, and determining target time domain channel estimation information according to the filtered in-window channels; wherein the threshold power is determined by the noise power and a preset first adjustment coefficient;

[0009] Performing Fourier transform on the target time domain channel estimation information to obtain a target channel estimation result in the frequency domain;

[0010] Based on the trained behavior recognition model, the target channel estimation result is processed to obtain a behavior recognition result.

[0011] In one embodiment, the target channel estimation result is processed based on the trained behavior recognition model to obtain the behavior recognition result, including:

[0012] Based on the target channel estimation result, construct a target two-dimensional matrix;

[0013] Performing short-time Fourier transform on the target two-dimensional matrix to obtain a spectrum image;

[0014] Periodically cutting the spectrum image to obtain an image to be processed;

[0015] The image to be processed is input into the behavior recognition model to obtain the behavior recognition result.

[0016] In one embodiment, the behavior recognition model is trained in the following manner:

[0017] Input the sample images in the sample data set into the machine learning model to obtain the predicted behavior category output by the machine learning model; wherein each sample image in the sample data set carries a corresponding behavior category label;

[0018] Adjusting model parameters of the machine learning model according to the difference between the predicted behavior category and the behavior category label;

[0019] The above process is repeated until the machine learning model converges to obtain a behavior recognition model.

[0020] In one embodiment, before inputting the sample images in the sample data set into the machine learning model to obtain the predicted behavior category output by the machine learning model, the method further includes:

[0021] Acquire multiple sample channel estimation information of Wi-Fi signals, and convert each sample channel estimation information into a sample spectrum picture; wherein the multiple sample channel estimation information corresponds to multiple preset behavior categories;

[0022] According to each sample spectral image, a plurality of sample images are constructed, and behavior category labels corresponding to the sample spectral images are added to the sample images;

[0023] The sample data set is constructed based on the sample images to which the behavior category labels are added.

[0024] In one embodiment, before converting the extracted channel estimation information into time domain channel estimation information by inverse Fourier transform, the method further includes:

[0025] Expanding the number of subcarriers in the extracted channel estimation information to a second specified number by a boundary repetition method;

[0026] A circular shift operation is performed on the expanded decimated channel estimation information.

[0027] In one embodiment, determining target time domain channel estimation information according to the filtered out in-window channel includes:

[0028] Constructing intermediate time-domain channel estimation information based on the time-domain channel estimation value of the channel within the window that has been filtered out;

[0029] Perform zero padding processing on the intermediate time domain channel estimation information to obtain target time domain channel estimation information; wherein the target time domain channel estimation information includes time domain channel estimation values ​​corresponding to a third specified number of channels.

[0030] In one embodiment, performing Fourier transform on the target time domain channel estimation information to obtain a target channel estimation result in the frequency domain includes:

[0031] Performing Fourier transform on the target time domain channel estimation information to obtain an intermediate channel estimation result in the frequency domain;

[0032] A circular shift operation is performed on the intermediate channel estimation result, and the intermediate channel estimation result after the circular shift operation is sampled to obtain the target channel estimation result.

[0033] On the other hand, the present application provides a behavior recognition device based on channel estimation, comprising:

[0034] An extraction module, configured to extract, for channel estimation information of the Wi-Fi signal, a frequency domain channel estimation value of each symbol on a first specified number of subcarriers to obtain extracted channel estimation information;

[0035] A first transformation module, configured to transform the extracted channel estimation information into time domain channel estimation information by inverse Fourier transformation, and determine the power corresponding to each channel based on the time domain channel estimation information;

[0036] A calculation module is used to divide multiple channels in the window with the largest power according to a preset window length, and calculate the average value of the power corresponding to the channels outside the window as the noise power;

[0037] A determination module, used to screen out several in-window channels whose power is greater than a threshold power from multiple in-window channels, and filter out the in-window channel with the largest power among the screened in-window channels, and determine the target time domain channel estimation information according to the filtered in-window channels; wherein the threshold power is determined by the noise power and a preset first adjustment coefficient;

[0038] A second transform module, used for performing Fourier transform on the target time domain channel estimation information to obtain a target channel estimation result in the frequency domain;

[0039] The recognition module is used to process the target channel estimation result based on the trained behavior recognition model to obtain a behavior recognition result.

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

[0041] processor;

[0042] a memory for storing processor-executable instructions;

[0043] Wherein, the processor is configured to execute the above-mentioned behavior recognition method based on channel estimation.

[0044] In addition, the present application provides a computer-readable storage medium, which stores a computer program. The computer program can be executed by a processor to complete the above-mentioned behavior recognition method based on channel estimation.

[0045] The solution of the present application can clean the channel estimation information of the Wi-Fi signal to remove noise and LOS background, obtain the target channel estimation result that characterizes the behavior of the passive target, and use the target channel estimation result to complete accurate behavior recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solution of the embodiments of the present application, the drawings required for use in the embodiments of the present application are briefly introduced below.

[0047] Figure 1 A schematic diagram of the structure of an electronic device provided in one embodiment of the present application;

[0048] Figure 2 A flow chart of a behavior recognition method based on channel estimation provided in one embodiment of the present application;

[0049] Figure 3 An embodiment of the present application provides Figure 2 Detailed flow chart of step 260;

[0050] Figure 4 A schematic diagram of a spectrum picture provided in an embodiment of the present application;

[0051] Figure 5 A flowchart of a method for training a behavior recognition model provided in one embodiment of the present application;

[0052] Figure 6 A schematic diagram of a flow chart of a method for constructing a sample data set provided in an embodiment of the present application;

[0053] Figure 7 A schematic diagram of cleaning channel estimation information provided by an embodiment of the present application;

[0054] Figure 8 A block diagram of a behavior recognition device based on channel estimation provided in one embodiment of the present application. DETAILED DESCRIPTION

[0055] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

[0056] Similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of this application, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood 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 A processor 11 is taken as an example. The processor 11 and the memory 12 are connected via a bus 10. The memory 12 stores instructions that can be executed by the processor 11. The instructions are executed by the processor 11 so that the electronic device 1 can execute all or part of the process of the method in the following embodiment. In one embodiment, the electronic device 1 can be a router, a communication terminal device (such as a mobile phone, a tablet computer), etc., which is used to execute a behavior recognition method based on channel estimation. The electronic device can be equipped with a transmitting antenna and a receiving antenna for transmitting and receiving Wi-Fi signals. The following describes the scheme with 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, disk or optical disk.

[0059] The present application also provides a computer-readable storage medium, which stores a computer program. The computer program can be executed by the processor 11 to complete the behavior recognition method based on channel estimation provided by the present application.

[0060] In an indoor environment, the electronic device that implements the solution of the present application can be placed in a specified position in the indoor scene (for example, the electronic device is a router, which is placed in the center of the TV wall in the living room). In the process of transmitting and receiving Wi-Fi signals by the electronic device, channel estimation information as channel state information (CSI) can be obtained through channel estimation. Since there are various objects in the indoor scene (such as people, furniture, electrical appliances, pets, etc.), various reflections will occur in the Wi-Fi signal. Therefore, there are multiple wireless channels between the transmitter and the receiver. When a passive target moves in the scene (such as running, making gestures, falling, etc.), since the indoor static objects are kept motionless, the channel state information can characterize the impact of the behavior of the passive target on the wireless channel. Therefore, the behavior of the passive target in the indoor scene can be identified with the help of the channel state information.

[0061] See also Figure 2 , is a flow chart of a behavior recognition method based on channel estimation provided by an embodiment of the present application, such as Figure 2 As shown, the method may include the following steps 210 - 260 .

[0062] Step 210: For the channel estimation information of the Wi-Fi signal, extract the frequency domain channel estimation value of each symbol on a first specified number of subcarriers to obtain extracted channel estimation information.

[0063] In the process of transmitting and receiving Wi-Fi signals, the electronic device may obtain channel estimation information of the Wi-Fi signal through channel estimation. Exemplarily, the electronic device may determine the channel estimation information in the frequency domain from the pilot signal of the Wi-Fi signal through the LS (Least Square) algorithm. In one embodiment, in order to make the frequency leakage more concentrated, the channel estimation information may be windowed, and the window function used for the windowing process may be but is not limited to the Hamming window.

[0064] After obtaining the channel estimation information, for each symbol sent by the Wi-Fi signal, the electronic device can extract the frequency domain channel estimation value on the first specified number of subcarriers, thereby obtaining the extracted channel estimation information. Here, the first specified number can be pre-configured as needed, or determined by the hardware conditions of the electronic device.

[0065] Step 220: Convert the extracted channel estimation information into time domain channel estimation information through inverse Fourier transform, and determine the power corresponding to each channel based on the time domain channel estimation information.

[0066] The electronic device can transform the extracted channel estimation information into the time domain through Inverse Fast Fourier Transform (IFFT) to obtain the time domain channel estimation information. Exemplarily, the transformation process can refer to the following formula (1):

[0067]

[0068] in, is the channel estimation information in the frequency domain; is the time domain channel estimation information; q is the receiving antenna index; k is the subcarrier index; l is the symbol index.

[0069] After obtaining the time domain channel estimation information, the power corresponding to each channel indicated by the time domain channel estimation information can be determined by the following formula (2):

[0070]

[0071] in, Indicates power; is the time domain channel estimation information; q is the receiving antenna index; k is the subcarrier index; l is the symbol index.

[0072] Step 230: divide a plurality of channels in the window with the largest power according to the preset window length, and calculate the average value of the power corresponding to the channels outside the window as the noise power.

[0073] The window length may be a preset empirical value, or the window length may be determined according to the length of a cyclic prefix (CP). For example, the relationship between the window length L and the cyclic prefix length CP may be expressed as L≤2*CP-1.

[0074] The electronic device can select multiple time domain channels with the largest power according to the window length as channels within the window, and use time domain channels other than the channels within the window as channels outside the window. Exemplarily, the time domain channel estimation information includes channel estimation values ​​of 60 time domain channels, 60 powers can be calculated, the window length is 10, and the time domain channels corresponding to the 10 largest powers are selected as channels within the window, and the remaining time domain channels are channels outside the window.

[0075] The channel outside the window can be regarded as the channel where the noise is located, and the electronic device can calculate the average value of the corresponding powers of multiple channels outside the window as the noise power.

[0076] Step 240: Filter out several in-window channels whose power is greater than a threshold power from multiple in-window channels, and filter out the in-window channel with the largest power among the filtered in-window channels, and determine the target time domain channel estimation information based on the filtered in-window channels; wherein the threshold power is determined by the noise power and a preset first adjustment coefficient.

[0077] Here, the first adjustment coefficient is used to adjust the noise power to obtain the threshold power. The size of the first adjustment coefficient can be configured as needed. Exemplarily, the first adjustment coefficient can be 1.5.

[0078] The electronic device can multiply the noise power by the first adjustment coefficient to obtain the threshold power. By comparing the corresponding power of the channel in the window with the threshold power, several channels in the window with power greater than the threshold power can be screened out, and the channel in the window with the largest power can be filtered out from the screened several channels in the window. Here, the filtered channels in the window with power not greater than the threshold power and the channels in the window with the largest power can both set the time domain channel estimation value to zero.

[0079] The electronic device can construct the target time domain channel estimation information based on the time domain channel estimation value of the channel in the window that has been filtered out and other time domain channel estimation values ​​that are set to zero. The target time domain channel estimation information removes the noise and LOS (Line of Sight) background in the time domain, and can more accurately characterize the impact of the behavior of the passive target on the NLOS (Not Line of Sight) channel.

[0080] Step 250: Perform Fourier transform on the target time domain channel estimation information to obtain a target channel estimation result in the frequency domain.

[0081] After obtaining the target time domain channel estimation information, the electronic device may convert the target time domain channel estimation information into the frequency domain through a Fourier transform (Fast Fourier Transform), thereby obtaining a target channel estimation result in the frequency domain.

[0082] Step 260: Based on the trained behavior recognition model, the target channel estimation result is processed to obtain a behavior recognition result.

[0083] Among them, the behavior recognition model can be a machine learning model used to classify behavior categories.

[0084] After obtaining the cleaned target channel estimation result, the target channel estimation result can be processed with the help of the behavior recognition model to obtain the behavior recognition result. The electronic device can construct input parameters of the behavior recognition model based on the target channel estimation result, and input the input parameters into the behavior recognition model to obtain the behavior recognition result corresponding to the input parameters, and the behavior recognition result characterizes the behavior category of the behavior performed by the passive target.

[0085] Through the above measures, the channel estimation information of the Wi-Fi signal can be cleaned to remove noise and LOS background, obtain the target channel estimation result that characterizes the behavior of the passive target, and complete accurate behavior recognition with the target channel estimation result.

[0086] In one embodiment, see Figure 3 , provided in one embodiment of the present application Figure 2 The detailed flow chart of step 260 is as follows: Figure 3 As shown, when executing step 260, the following steps 261 to 264 may be specifically executed.

[0087] Step 261: construct a target two-dimensional matrix based on the target channel estimation result.

[0088] After obtaining the target channel estimation result, the electronic device can arrange multiple frequency domain channel estimation values ​​in the target channel estimation result by subcarriers on the vertical axis and by time on the horizontal axis, so as to construct a target two-dimensional matrix. Since there are multiple symbols in each data packet and different symbols are sent successively, the frequency domain channel estimation value of the first or last symbol in multiple subcarriers can be selected from each data packet to construct the target two-dimensional matrix; or, the frequency domain channel estimation values ​​of multiple symbols in each data packet on multiple subcarriers can be averaged respectively to obtain the average of the frequency domain channel estimation values ​​of multiple symbols on each subcarrier, which is used to construct the target two-dimensional matrix.

[0089] Exemplarily, the target channel estimation result includes frequency domain channel estimation values ​​corresponding to 50 data packets, and the frequency domain channel estimation values ​​are distributed on 100 subcarriers, then a 50*100 target two-dimensional matrix can be constructed.

[0090] Step 262: Perform short-time Fourier transform on the target two-dimensional matrix to obtain a spectrum image.

[0091] After obtaining the target two-dimensional matrix, the electronic device can perform a short-time Fourier transform (STFT) on the target two-dimensional matrix to obtain a spectrum image. Here, the step size and window length of the short-time Fourier transform can be pre-configured as needed.

[0092] See also Figure 4 , is a schematic diagram of a spectrum picture provided in an embodiment of the present application, such as Figure 4 As shown, the vertical axis of the spectrum picture is frequency and the horizontal axis is time. During the execution of the present application solution, the channel estimation information of the Wi-Fi signal can be continuously processed to continuously update the spectrum picture. In other words, the horizontal axis of the spectrum picture will continue to increase during this process.

[0093] Step 263: Periodically cut the spectrum image to obtain the image to be processed.

[0094] After obtaining the spectral image, the electronic device can periodically segment the spectral image to obtain the image to be processed. Here, the cycle duration can be pre-configured based on experience, generally the time required for the user to complete a behavior, and an exemplary cycle duration is 2 seconds. During the application process, the electronic device can cut the spectral image within the latest cycle duration to identify the current behavior of the passive target.

[0095] Step 264: Input the image to be processed into the behavior recognition model to obtain a behavior recognition result.

[0096] After obtaining the image to be processed, the electronic device can input the image to be processed into a behavior recognition model, classify the image to be processed through the behavior recognition model, and obtain a behavior recognition result. The behavior recognition result is a predicted behavior category output by the behavior recognition model.

[0097] Through the above measures, after constructing the target two-dimensional matrix with the target channel estimation result, it is converted into a spectral image through short-time Fourier transform, so that the frequency change in time is more accurate, and then the image to be processed cut out from the spectral image is processed by the behavior recognition model to obtain accurate behavior recognition results.

[0098] In one embodiment, before executing the solution of the present application, a behavior recognition model can be trained. Figure 5 , is a flow chart of a method for training a behavior recognition model provided in an embodiment of the present application, such as Figure 5 As shown, the behavior recognition model can be trained through the following steps 510 to 530.

[0099] Step 510: Input the sample images in the sample data set into the machine learning model to obtain the predicted behavior category output by the machine learning model; wherein each sample image in the sample data set carries a corresponding behavior category label.

[0100] The sample data set includes a large number of sample images, which are images cut out from the spectral images. Each sample image carries a corresponding behavior category label, which indicates the behavior category represented by the sample image.

[0101] The machine learning model used in training may include but is not limited to CNN (Convolutional Neural Networks), TCN (Temporal Convolutional Network), a hybrid model of CNN plus RNN (Recurrent Neural Networks), and the like.

[0102] The electronic device can use the sample image as input to the machine learning model to obtain the predicted behavior category output by the machine learning model.

[0103] Step 520: Adjust the model parameters of the machine learning model based on the difference between the predicted behavior category and the behavior category label.

[0104] Step 530: Repeat the above process until the machine learning model converges to obtain a behavior recognition model.

[0105] The electronic device can evaluate the difference between the predicted behavior category and the behavior category label corresponding to the same sample image according to a preset loss function, obtain a function value, and adjust the model parameters of the machine learning model with the function value.

[0106] After adjusting the machine learning model, the process can return to step 510 and re-input the sample image into the adjusted machine learning model to obtain a new predicted behavior category, thereby re-evaluating the difference between the behavior category label and the predicted behavior category, and re-adjusting the model parameters based on the evaluation results. After repeated iterations, when the function value of the loss function tends to be stable, the machine learning model can be considered to have converged. At this point, the converged machine learning model can be used as a behavior recognition model.

[0107] Through the above measures, a behavior recognition model for identifying user behaviors can be trained.

[0108] In one embodiment, before training the behavior recognition model, a sample data set may be constructed. Figure 6 , is a flow chart of a method for constructing a sample data set provided in an embodiment of the present application, such as Figure 6 As shown, a sample data set may be constructed through the following steps 610 to 630.

[0109] Step 610: Acquire multiple sample channel estimation information of the Wi-Fi signal, and convert each sample channel estimation information into a sample spectrum picture; wherein the multiple sample channel estimation information corresponds to a plurality of preset behavior categories.

[0110] For each behavior category that needs to be identified later, people of different body sizes can continue to perform actions of the behavior category in an indoor environment, thereby obtaining sample channel estimation information corresponding to the behavior category. Exemplarily, the behavior categories that need to be identified later include: waving, spreading hands, clapping, running, and falling. In the case where people of different body sizes continue to wave their hands in an indoor environment, sample channel estimation information corresponding to waving can be obtained; similarly, sample channel estimation information corresponding to spreading hands, clapping, flapping, and falling can be obtained.

[0111] After obtaining the sample channel estimation information corresponding to each behavior category, the electronic device can clean the sample channel estimation information and convert it into a sample spectrum image. The cleaning process and conversion method can refer to the relevant description above and will not be repeated here.

[0112] Step 620: construct multiple sample images based on each sample spectrum image, and add behavior category labels corresponding to the sample spectrum images to the sample images.

[0113] The electronic device can periodically cut out multiple sample images from each sample spectrum image. Here, the period duration can be consistent with the period duration of the spectrum image cut out in the subsequent recognition process. After cutting out the sample image, a behavior category label can be added to the sample image according to the behavior category corresponding to the sample spectrum image. Exemplarily, for the sample channel estimation information corresponding to the behavior category "fall", and the multiple sample images cut out from the converted sample spectrum image, a behavior category label indicating "fall" is added.

[0114] Step 630: Construct a sample data set based on the sample images with the behavior category labels added.

[0115] The electronic device can construct a sample data set based on multiple sample images with added behavior category labels for use in the subsequent behavior recognition model training process.

[0116] In one embodiment, before executing step 220, the electronic device may expand the number of subcarriers in the extracted channel estimation information to a second specified number by a boundary repetition method. Here, the second specified number is the number of sampling points that can be processed by a subsequent inverse Fourier transform. In one embodiment, the second specified number may be determined by the first specified number K. For example, the logarithm of K with a base of 2 may be obtained, the logarithm may be rounded down and added to obtain a new integer, and the new integer may be used as a power of 2 to obtain the second specified number.

[0117] The electronic device may repeat the frequency domain channel estimation values ​​of the first subcarrier and the last subcarrier in the extracted channel estimation information for multiple times, thereby expanding the number of subcarriers to the second specified number. The number of repetitions of the first subcarrier and the last subcarrier of the extracted channel estimation information may be the same. Exemplarily, taking the first specified number as 60 and the second specified number as 128 as an example, the expansion process may be represented by the following formula (3):

[0118]

[0119] Wherein, m is 0 to 59, indicating the decimated channel estimation information in the middle, including the frequency domain channel estimation values ​​of 60 subcarriers; q is the index of the receiving antenna; k is the subcarrier index; l is the symbol index. As shown in formula (3), the frequency domain channel estimation value of the first subcarrier is repeated 34 times, and the frequency domain channel estimation value of the last subcarrier is repeated 34 times, thereby obtaining the expanded decimated channel estimation information.

[0120] The electronic device may perform a circular shift operation on the expanded decimated channel estimation information. Exemplarily, the circular shift operation may be expressed by the following formula (4):

[0121]

[0122] Among them, q is the index of the receiving antenna; k is the subcarrier index; l is the symbol index.

[0123] In this embodiment, after the circular shift operation is performed on the expanded decimated channel estimation information, the decimated channel estimation information after the circular shift operation can be processed by inverse Fourier transform, so as to perform subsequent steps.

[0124] In one embodiment, if the number of subcarriers in the extracted channel estimation information is expanded and a circular shift operation is performed on the expanded extracted channel estimation information, when the target time domain channel estimation information is determined according to the filtered out window channel in step 240, the electronic device can construct intermediate time domain channel estimation information based on the time domain channel estimation value of the filtered out window channel and other time domain channel estimation values ​​that are set to zero.

[0125] The electronic device may perform zero padding processing for the intermediate time domain channel estimation information, and the zero padding processing is to add a time domain channel whose time domain channel estimation value is zero to the intermediate time domain channel estimation information. The electronic device may expand the number of channels in the intermediate time domain channel estimation information to a third specified number by zero padding, thereby obtaining the target time domain channel estimation information. Among them, the target time domain channel estimation information includes the time domain channel estimation information corresponding to the third specified number of channels. In one embodiment, when the electronic device performs zero padding processing, multiple zeros may be added between multiple time domain channel estimation values ​​of the intermediate time domain channel estimation information.

[0126] Exemplarily, the zero padding process can be expressed by the following formula (5):

[0127]

[0128] Wherein, n is the third specified number; q is the index of the receiving antenna; l is the symbol index. The example in formula (5) is to fill the intermediate time domain information estimation information containing 128 sampling points with zeros to 512 sampling points. The actual number of sampling points can be configured as needed.

[0129] In one embodiment, when the target time domain channel estimation information is obtained by padding the intermediate time domain channel estimation information with zeros, when executing step 250, the electronic device may perform Fourier transform on the target time domain channel information to obtain the intermediate channel estimation result in the frequency domain.

[0130] The electronic device may perform a circular shift operation on the intermediate channel estimation result, and sample the intermediate channel estimation result after the circular shift operation to obtain a target channel estimation result. Exemplarily, a fourth specified number of frequency domain channel estimation values ​​in the middle of the intermediate channel estimation result after the circular shift operation may be selected during sampling. Here, the fourth specified number may be configured as needed. Exemplarily, if the third specified number is 512, the fourth specified number may be 240.

[0131] See also Figure 7 , is a schematic diagram of a cleaning process of channel estimation information provided in an embodiment of the present application, such as Figure 7 As shown, for the channel estimation information of any symbol, the frequency domain channel estimation information of the symbol on the first specified number of subcarriers is extracted to obtain the extracted channel estimation information. Figure 7For example, the first specified number is 60. At this time, for any symbol, the frequency domain channel estimation information of the symbol on 60 subcarriers can be obtained. The number of subcarriers in the extracted channel estimation information is expanded to the second specified number by the boundary repetition method. Here, the second specified number is 128. For any symbol, the frequency domain channel estimation value of the symbol on the first subcarrier can be repeated 34 times, and the frequency domain channel estimation value of the symbol on the 60th subcarrier is added to obtain the expanded extracted channel estimation information corresponding to the symbol.

[0132] The expanded extracted channel estimation information corresponding to each symbol is subjected to noise reduction processing: a circular shift operation is performed on the expanded extracted channel estimation information, and then the extracted channel estimation information is converted into time domain channel estimation information through inverse Fourier transform, and the power corresponding to each channel is determined based on the time domain channel estimation information. According to the preset window length (for example: 16), multiple in-window channels with the largest power are divided, and the average value of the power corresponding to 112 out-of-window channels is calculated as the noise power. After the threshold power is determined by the third adjustment coefficient and the noise power, several in-window channels with power greater than the threshold power can be screened out from the multiple in-window channels, and the in-window channels with the largest power are filtered out from the screened in-window channels, and the time domain channel estimation value of the in-window channel that has been filtered out is retained, and the time domain channel estimation values ​​of other in-window channels are set to zero, so as to construct the intermediate time domain channel estimation information.

[0133] Perform zero padding processing on the intermediate time domain channel estimation information, so that the time domain channel estimation information after zero padding processing includes a third specified number ( Figure 7 12) sampling points to obtain the target time domain channel estimation information. Each sampling point represents a time domain channel estimation value corresponding to a channel.

[0134] The target time domain channel estimation information of each symbol is Fourier transformed to obtain the intermediate channel estimation result of each symbol in the frequency domain. The intermediate channel estimation result includes 512 sampling points, each sampling point is a frequency domain channel estimation value in the frequency domain. A circular shift operation is performed on each intermediate channel estimation result, and 240 sampling points in the middle are sampled from the intermediate channel estimation result after the circular shift operation as the target channel estimation result of each symbol.

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

[0136] The extraction module 810 is used to extract the frequency domain channel estimation value of each symbol on the first specified number of subcarriers for the channel estimation information of the Wi-Fi signal to obtain the extracted channel estimation information;

[0137] A first transform module 820, configured to transform the extracted channel estimation information into time domain channel estimation information by inverse Fourier transform, and determine the power corresponding to each channel based on the time domain channel estimation information;

[0138] A calculation module 830 is used to divide multiple channels in the window with the largest power according to a preset window length, and calculate the average power corresponding to the channels outside the window as the noise power;

[0139] The determination module 840 is used to screen out a plurality of in-window channels whose power is greater than a threshold power from a plurality of in-window channels, and filter out the in-window channel with the largest power among the screened in-window channels, and determine the target time domain channel estimation information according to the filtered in-window channels; wherein the threshold power is determined by the noise power and a preset first adjustment coefficient;

[0140] A second transform module 850 is used to perform Fourier transform on the target time domain channel estimation information to obtain a target channel estimation result in the frequency domain;

[0141] The identification module 860 is used to process the target channel estimation result based on the trained behavior recognition model to obtain a behavior recognition result.

[0142] The implementation process of the functions and effects of each module in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned behavior recognition method based on channel estimation, and will not be repeated here.

[0143] In several embodiments provided in the present application, the disclosed devices and methods may also be implemented in other ways. The device embodiments described above are merely schematic, for example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to the multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram may represent a module, a program segment or a part of a code, and a module, a program segment or a part of a code contains one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the box may also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they may sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or the flowchart, and the combination of boxes in the block diagram and / or the flowchart, can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0144] In addition, the functional modules in the various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0145] If the function is implemented in the form of a software function 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 the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program codes.

Claims

1. A behavior recognition method based on channel estimation, characterized in that: include: For the channel estimation information of the Wi-Fi signal, extract the frequency domain channel estimation value of each symbol on the first specified number of subcarriers to obtain extracted channel estimation information; Converting the extracted channel estimation information into time domain channel estimation information by inverse Fourier transform, and determining the power corresponding to each channel based on the time domain channel estimation information; Divide multiple channels in the window with the largest power according to the preset window length, and calculate the average value of the corresponding power of the channels outside the window as the noise power; Screening out a number of in-window channels whose power is greater than a threshold power from a plurality of in-window channels, filtering out the in-window channel with the largest power among the screened in-window channels, and determining target time domain channel estimation information according to the filtered in-window channels; wherein the threshold power is determined by the noise power and a preset first adjustment coefficient; Performing Fourier transform on the target time domain channel estimation information to obtain a target channel estimation result in the frequency domain; Based on the trained behavior recognition model, the target channel estimation result is processed to obtain a behavior recognition result.

2. The method according to claim 1, characterized in that: The step of processing the target channel estimation result based on the trained behavior recognition model to obtain the behavior recognition result includes: Based on the target channel estimation result, construct a target two-dimensional matrix; Performing short-time Fourier transform on the target two-dimensional matrix to obtain a spectrum image; Periodically cutting the spectrum image to obtain an image to be processed; The image to be processed is input into the behavior recognition model to obtain the behavior recognition result.

3. The method according to claim 2, characterized in that The behavior recognition model is trained in the following way: Input the sample images in the sample data set into the machine learning model to obtain the predicted behavior category output by the machine learning model; wherein each sample image in the sample data set carries a corresponding behavior category label; Adjusting model parameters of the machine learning model according to the difference between the predicted behavior category and the behavior category label; The above process is repeated until the machine learning model converges to obtain a behavior recognition model.

4. The method according to claim 3, characterized in that Before inputting the sample images in the sample data set into the machine learning model to obtain the predicted behavior category output by the machine learning model, the method further includes: Acquire multiple sample channel estimation information of Wi-Fi signals, and convert each sample channel estimation information into a sample spectrum picture; wherein the multiple sample channel estimation information corresponds to multiple preset behavior categories; According to each sample spectral image, a plurality of sample images are constructed, and behavior category labels corresponding to the sample spectral images are added to the sample images; The sample data set is constructed based on the sample images to which the behavior category labels are added.

5. The method according to claim 1, characterized in that Before converting the extracted channel estimation information into time domain channel estimation information by inverse Fourier transform, the method further includes: Expanding the number of subcarriers in the extracted channel estimation information to a second specified number by a boundary repetition method; A circular shift operation is performed on the expanded decimated channel estimation information.

6. The method according to claim 5, characterized in that The step of determining target time domain channel estimation information according to the filtered out window channel includes: Constructing intermediate time-domain channel estimation information based on the time-domain channel estimation value of the channel within the window that has been filtered out; Perform zero padding processing on the intermediate time domain channel estimation information to obtain target time domain channel estimation information; wherein the target time domain channel estimation information includes time domain channel estimation values ​​corresponding to a third specified number of channels.

7. The method according to claim 6, characterized in that The step of performing Fourier transform on the target time domain channel estimation information to obtain a target channel estimation result in the frequency domain includes: Performing Fourier transform on the target time domain channel estimation information to obtain an intermediate channel estimation result in the frequency domain; A circular shift operation is performed on the intermediate channel estimation result, and the intermediate channel estimation result after the circular shift operation is sampled to obtain the target channel estimation result.

8. A behavior recognition device based on channel estimation, characterized in that: include: An extraction module, configured to extract, for channel estimation information of the Wi-Fi signal, a frequency domain channel estimation value of each symbol on a first specified number of subcarriers to obtain extracted channel estimation information; A first transformation module, configured to transform the extracted channel estimation information into time domain channel estimation information by inverse Fourier transformation, and determine the power corresponding to each channel based on the time domain channel estimation information; A calculation module is used to divide multiple channels in the window with the largest power according to a preset window length, and calculate the average value of the power corresponding to the channels outside the window as the noise power; A determination module, used to screen out several in-window channels whose power is greater than a threshold power from multiple in-window channels, and filter out the in-window channel with the largest power among the screened in-window channels, and determine the target time domain channel estimation information according to the filtered in-window channels; wherein the threshold power is determined by the noise power and a preset first adjustment coefficient; A second transform module, used for performing Fourier transform on the target time domain channel estimation information to obtain a target channel estimation result in the frequency domain; The recognition module is used to process the target channel estimation result based on the trained behavior recognition model to obtain a behavior recognition result.

9. An electronic device, characterized in that: The electronic device comprises: processor; a memory for storing processor-executable instructions; The processor is configured to execute the behavior recognition method based on channel estimation as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and the computer program can be executed by a processor to complete the behavior recognition method based on channel estimation as described in any one of claims 1-7.

Images