Action recognition method, classification model training method and terminal equipment

Abstract

The invention discloses an action recognition method, a classification model training method and terminal equipment, and belongs to the technical field of action recognition. The action recognition method comprises the following steps: acquiring to-be-recognized channel state information of different beams by utilizing beam forming; determining a to-be-identified phase matrix and a to-be-identified amplitude matrix of to-be-identified channel state information matrixes of different beams with the same time sequence label according to the to-be-identified channel state information; fitting the to-be-identified phase matrix and the to-be-identified amplitude matrix to obtain a to-be-identified eigenvalue matrix; and inputting the to-be-identified characteristic value matrix into a preset classification model to identify an action type corresponding to the to-be-identified channel state information. According to the embodiment of the invention, the interference of other equipment can be reduced by using the beam forming technology, the recognition range is expanded, the phase matrix and the amplitude matrix are used for fitting to determine the characteristic value matrix, the noise interference can be reduced, and the recognition precision is improved.

Description

Technical Field

[0001] The present application belongs to the field of motion recognition technology, and specifically relates to a motion recognition method, a classification model training method and a terminal device. Background Art

[0002] With the continuous development of communication technology, the popularity of wireless network communication technology (WiFi) is increasing. People have proposed that the existing WiFi signals in the environment can be used to sense the environment while completing their main tasks. Compared with traditional cameras, infrared and other technologies, wireless sensing technology (WiFi Sensing) has the characteristics of easy deployment, low investment, passive, non-contact, non-intrusive monitoring, and non-line-of-sight monitoring.

[0003] However, the acquired channel state information (CSI) often contains a lot of noise, and in actual use, there may be multiple WiFi signals in the same space, which greatly affects the recognition range and accuracy. Summary of the invention

[0004] The present application aims to provide an action recognition method, a classification model training method and a terminal device, at least to solve the problem in the related art that the recognition accuracy is affected by the noise in the acquired CSI and the interference of multiple WiFi signals.

[0005] In a first aspect, an embodiment of the present application proposes an action recognition method, comprising:

[0006] Using beamforming to obtain channel state information to be identified for different beams;

[0007] Determine, according to the channel state information to be identified, a phase matrix to be identified and an amplitude matrix to be identified of the channel state information matrix to be identified of different beams of the same timing label;

[0008] Fitting the phase matrix to be identified and the amplitude matrix to be identified to obtain an eigenvalue matrix to be identified;

[0009] The eigenvalue matrix to be identified is input into a preset classification model to identify the action type corresponding to the channel state information to be identified.

[0010] In a second aspect, the present application embodiment proposes a classification model training method, including:

[0011] Acquire channel state information of multiple beams in different scenarios in a beam domain by using beamforming, and set corresponding scenario labels for the channel state information;

[0012] Determine, according to the channel state information, a phase matrix and an amplitude matrix of a channel state information matrix of different beams with the same timing label;

[0013] Fitting the phase matrix and the amplitude matrix to obtain an eigenvalue matrix;

[0014] The eigenvalue matrix and the scenario labels corresponding to the eigenvalue matrix are input into the action classification model for model training to obtain a target action classification model.

[0015] In a third aspect, an embodiment of the present application provides a terminal device, comprising one or more processors, one or more memories, and programs or instructions stored in the one or more memories and executable on the one or more processors, wherein when the program or instructions are executed by the processor, the steps of the method described in the first aspect are implemented, or the steps of the method described in the second aspect are implemented.

[0016] In a fourth aspect, a storage medium is provided, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the method described in the first aspect are implemented, or the steps of the method described in the second aspect are implemented.

[0017] In a fifth aspect, a chip is provided, comprising a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run a program or instructions to implement the steps of the method described in the first aspect, or to implement the steps of the method described in the second aspect.

[0018] In a sixth aspect, a computer program / program product is provided, wherein the computer program / program product is stored in a storage medium, and the computer program / program product is executed by at least one processor to implement the steps of the method described in the first aspect, or to implement the steps of the method described in the second aspect.

[0019] In an embodiment of the present application, beamforming is first used to obtain the channel state information to be identified of different beams, and then the phase matrix to be identified and the amplitude matrix to be identified of the channel state information matrix to be identified of different beams with the same timing label are determined according to the channel state information to be identified, and then the phase matrix to be identified and the amplitude matrix to be identified are fitted to obtain the eigenvalue matrix to be identified, and finally the eigenvalue matrix to be identified is input into a preset classification model to identify the action type corresponding to the channel state information to be identified. The embodiment of the present application uses beamforming technology to collect the channel state information to be identified, which can reduce interference from other devices and expand the recognition range. The eigenvalue matrix is ​​determined by fitting the phase matrix and the amplitude matrix, which can reduce noise interference and improve recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is a flowchart of a motion recognition method provided by an embodiment of the present application;

[0021] Figure 2 is a schematic diagram of a beamforming process provided by an embodiment of the present application;

[0022] Figure 3 It is a flowchart of a classification model training method provided by an embodiment of the present application;

[0023] Figure 4 is a schematic diagram of a motion recognition device provided by an embodiment of the present application;

[0024] Figure 5 is a schematic diagram of a training device for a classification model provided by an embodiment of the present application;

[0025] Figure 6 It is a schematic diagram of the structure of a terminal device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0026] The embodiments of the present application will be described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and should not be construed as limitations on the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application.

[0027] The term "first" or "second" in the specification and claims of this application may include one or more of the features explicitly or implicitly. In the description of this application, unless otherwise specified, "plurality" means two or more. In addition, "and / or" in the specification and claims means at least one of the connected objects, and the character " / " generally means that the objects connected before and after are in an "or" relationship.

[0028] In the description of the present application, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", "axial", "radial", "circumferential" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present application.

[0029] In the description of this application, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0030] Combine the following Figures 1 to 6 , an action recognition method, a classification model training method and a terminal device provided in an embodiment of the present application are explained through specific embodiments and their application scenarios.

[0031] like Figure 1 As shown, the present application embodiment provides a flow chart of an action recognition method. Figure 1 As shown, the action recognition method may include the contents shown in S101 to S104.

[0032] In S101, beamforming is used to obtain channel state information to be identified for different beams.

[0033] Among them, the channel state information (CSI) to be identified is extracted from the WiFi signal. Using beamforming to obtain CSI can improve signal strength and reduce interference from other devices.

[0034] In S102, according to the channel state information to be identified, a phase matrix to be identified and an amplitude matrix to be identified of the channel state information matrix to be identified of different beams of the same timing label are determined.

[0035] That is to say, the channel state matrix to be identified can be extracted from the channel state information to be identified, and then the phase matrix to be identified and the amplitude matrix to be identified can be determined. The channel state information to be identified can be denoised by determining the phase matrix to be identified and the amplitude matrix to be identified.

[0036] In S103, the phase matrix to be identified and the amplitude matrix to be identified are fitted to obtain the eigenvalue matrix to be identified.

[0037] In S104, the eigenvalue matrix to be identified is input into a preset classification model to identify the action type corresponding to the channel state information to be identified.

[0038] The action type may refer to walking, running, falling, gestures, and stillness, etc. In other embodiments, the action type may also be a type classified based on the magnitude of the action, or a type classified based on the action posture type.

[0039] In an embodiment of the present application, beamforming is first used to obtain the channel state information to be identified of different beams, and then the phase matrix to be identified and the amplitude matrix to be identified of the channel state information matrix to be identified of different beams with the same timing label are determined according to the channel state information to be identified, and then the phase matrix to be identified and the amplitude matrix to be identified are fitted to obtain the eigenvalue matrix to be identified, and finally the eigenvalue matrix to be identified is input into a preset classification model to identify the action type corresponding to the channel state information to be identified. The embodiment of the present application utilizes the characteristics of beamforming technology such as high signal strength and high signal-to-noise ratio to collect the channel state information to be identified, which can reduce interference from other devices and expand the recognition range. The eigenvalue matrix is ​​determined by fitting the phase matrix and the amplitude matrix, which can reduce noise interference and improve recognition accuracy.

[0040] In a possible implementation of the present application, using beamforming to obtain the channel state information to be identified may include the following steps 1 to 5. The beamforming process may be as follows: Figure 2 shown.

[0041] Step 1: Receive the empty data packet notification frame notification sent by the sender.

[0042] That is, the sender first sends a null data packet notification (NDP-Announcement) frame notification to the receiver, notifying the receiver that it will send a very high throughput (VHT) null data packet (NDP) frame, and then the receiver receives the null data packet notification frame notification sent by the sender.

[0043] Among them, NDP is a detection frame that does not contain a data field.

[0044] It is worth noting that the transmitting end can be a wireless access point (AP) device, for example, a router device that supports the Multiple-Input Multiple-Output (MIMO) function, and the receiving end can be a Spanning Tree Algorithm (STA) device, for example, a router, a mobile phone, a computer, and other devices that can connect to the router via WiFi.

[0045] Step 2: Receive the empty data packet frame sent by the sender.

[0046] That is, the sender will then send an NDP frame for detection, and then the receiver receives the NDP frame sent by the sender.

[0047] Step 3: Determine the feedback matrix according to the empty data packet frame.

[0048] The receiving end can analyze the leading sequence in the NDP frame to calculate a feedback matrix. The calculation method may include CSI, non-compressed beamforming, compressed beamforming, and compressed V matrix. In this embodiment, the compressed V matrix is ​​used as an example for calculation to obtain the feedback matrix V k .

[0049] Step 4: Send the feedback matrix to the transmitting end, so that the transmitting end uses the feedback matrix to perform beamforming.

[0050] Finally, the receiving end sends the feedback matrix to the transmitting end, and the transmitting end uses the feedback matrix to perform beamforming.

[0051] The above beamforming process can be described as follows: Figure 2 shown.

[0052] Step 5: Use beamforming to obtain the channel state information to be identified for different beams.

[0053] This embodiment may adopt the above-determined beamforming to obtain the to-be-identified channel state information of different beams, which may improve signal strength and reduce interference from other devices.

[0054] In a possible implementation manner of the present application, determining a feedback matrix according to an empty data packet frame may include the contents shown in the following steps 1 to 3.

[0055] Step 1: Analyze the leading sequence in the empty data packet frame to obtain a receiving matrix, which contains a search space.

[0056] In this embodiment, the receiving end can analyze the leading sequence in the NDP frame to obtain a receiving matrix A(t), which contains the search space E N .

[0057] Step 2: Determine the search vector and the signal source direction estimation value according to the receiving matrix and the search space.

[0058] In this embodiment, according to the E obtained above, N Construct a search vector a(θ i ), and then calculate the estimated value of the signal source direction F(θ i ), F(θ i )=1 / |a H (θ i )E N E N H a(θ i )| 2 , where a H (θi ) is a(θ i ), E N H For E N The transposed vector of .

[0059] Step 3: Determine the feedback matrix based on the estimated value of the signal source direction using the linear constrained minimum variance criterion.

[0060] That is to say, the feedback matrix V can be calculated based on the estimated value of the signal source direction using the linear constrained minimum variance criterion. k .

[0061] Specifically, the feedback matrix can be determined according to the following formula:

[0062]

[0063] R N =A(t)A H (t)

[0064]

[0065] Among them, V k is the feedback matrix; A(t) is the receiving matrix; A H (t) is the transposed matrix of A(t); F(θ i ) is the estimated value of the signal source direction; F H (θ i ) is F(θ i ) is the transposed matrix; a(θ i ) is the search vector; a H (θ i ) is a(θ i ) is the transposed matrix; E N is the search space; For E N The transposed matrix of For R N The inverse matrix of .

[0066] In this embodiment, the feedback matrix is ​​determined by the above method and using Lagrange multipliers, so that the position of the STA can be located more accurately.

[0067] In a possible implementation of the present application, determining the phase matrix to be identified and the amplitude matrix to be identified of the channel state information matrix to be identified for different beams of the same timing label according to the channel state information to be identified may include the contents shown in the following steps 1 to 4.

[0068] Step 1: Determine a channel state information matrix according to the channel state information to be identified, where the channel state information to be identified carries a timing label.

[0069] Step 2: According to the channel state information matrix to be identified, determine the target channel state information matrix to be identified, where the target channel state information matrix to be identified is a matrix composed of the channel state information to be identified whose subcarrier fluctuation values ​​are greater than the fluctuation threshold.

[0070] Among them, the sub-wave fluctuation value is in, is the channel state information matrix H to be identified i The j-th row and k-th column element of is the matrix H i-1 The j-th row and k-th column element of H i-1 is the matrix H i The CSI matrix of the previous time sequence of the same beam.

[0071] Among them, the fluctuation threshold can be 0.1 or other values, which is determined according to actual applications and is not limited in the embodiments of the present application.

[0072] In this embodiment, the CSI information of n subcarriers greater than the threshold is recorded as the target channel state information matrix β to be identified i , that is, β i is the channel state information matrix to be identified after screening.

[0073] By screening, it is possible to obtain relatively easy-to-identify subcarriers in various situations, filter out invalid data, facilitate subsequent data processing, and reduce the complexity of subsequent data processing.

[0074] Step 3: Divide the target channel state information matrices to be identified of different beams with the same timing label by two by two to obtain a ratio matrix to be identified. The phase information of the ratio matrix to be identified is the phase matrix to be identified.

[0075] In this embodiment, the target CSI matrices to be identified of different beams of the same time sequence are divided two by two to obtain a ratio matrix to be identified, and the phase information of the ratio matrix to be identified is saved and recorded as the phase matrix to be identified θ i .

[0076] Step 4: Use wavelet transform to perform noise reduction on the target channel state information matrix to be identified in different beams with the same timing label to obtain the amplitude matrix to be identified.

[0077] In this embodiment, the target CSI matrix to be identified of different beams of the same time sequence is subjected to wavelet transform noise reduction processing, and the amplitude is recorded as the amplitude matrix to be identified D i .

[0078] In the embodiment of the present application, noise reduction processing is performed by means of CSI ratio and wavelet transform, which can reduce noise interference and improve recognition accuracy.

[0079] In a possible implementation manner of the present application, fitting the phase matrix to be identified and the amplitude matrix to be identified to obtain the eigenvalue matrix to be identified may include the contents shown in the following steps 1 to 6.

[0080] Step 1: Determine the initial sample weights according to the phase matrix to be identified and the amplitude matrix to be identified, wherein the initial sample weights include the constants to be decided.

[0081] The sample weight is Among them, x i is an element in the phase matrix to be identified and the amplitude matrix to be identified, j represents that the matrix is ​​the phase matrix to be identified or the amplitude matrix to be identified, and k is a constant to be decided.

[0082] Step 2: Determine the goodness of fit based on the fitted data of the phase matrix to be identified and the amplitude matrix to be identified, as well as the original data of the phase matrix to be identified and the amplitude matrix to be identified.

[0083] Among them, the goodness of fit is a value that intuitively identifies the fitting effect, which is expressed as in, is the data after fitting the phase matrix to be identified and the amplitude matrix to be identified, y i is the original data of the phase matrix and amplitude matrix to be identified.

[0084] Step 3: Iterate the decision constant according to the preset step value until the goodness of fit meets the goodness threshold and the target decision constant is obtained.

[0085] The step value may be 0.05 or other values, and the range of the step value is (0, 1). The goodness threshold may be (0.85, 0.95) or other values, which are not limited in this embodiment.

[0086] That is to say, when the goodness of fit meets the goodness threshold for the first time, the iteration ends and the k value at this time is used as the value of the target decision constant in the weight calculation.

[0087] Step 4: Determine the target sample weight based on the target decision constant.

[0088] According to the target decision constant determined above, the target sample weight can be determined.

[0089] Step 5: Based on the target sample weights, obtain the phase weight matrix to be identified and the amplitude weight matrix to be identified.

[0090] Step 6: Obtain the eigenvalue matrix to be identified based on the phase weight matrix to be identified, the amplitude weight matrix to be identified, the phase matrix to be identified and the amplitude matrix to be identified.

[0091] That is to say, all matrices are iteratively fitted with local weighted linear regression to obtain the eigenvalue matrix C to be identified i =W1θ i +W2D i , where W1 and W2 are matrices composed of w(i,j) calculated by the above method.

[0092] In the embodiment of the present application, an optimal parameter combination is obtained by fitting the existing data, reducing the data dimension, removing outliers or filling missing values, thereby improving the quality and reliability of the data, and intuitively displaying the relationship between each parameter and the result. In the above calculation, by limiting the k value, it is also ensured that overfitting is avoided under the condition that the fitting result is credible, thereby improving universality.

[0093] It should be noted that the action recognition method provided in the embodiment of the present application can be performed by a motion recognition device or a control module in the motion recognition device for performing the motion recognition method. In the embodiment of the present application, the motion recognition device provided in the embodiment of the present application is described by taking the motion recognition device performing the motion recognition method as an example.

[0094] like Figure 3 As shown, it is a flow chart of a training method of a classification model provided in an embodiment of the present application. Figure 3 As shown, the training method of the classification model may include the contents shown in S301 to S304.

[0095] In S301, beamforming is used to obtain channel state information of multiple beams in different scenarios in a beam domain, and corresponding scenario labels are set for the channel state information.

[0096] Among them, the beamforming is also formed by using the beamforming method in the above embodiment, which will not be described in detail in this embodiment.

[0097] That is to say, the training phase uses the channel state information collected in a certain scenario to train the model, and sets a corresponding scenario label for the channel state information according to the scenario. The scenario label corresponds to the action type, for example, the action type corresponding to the motion scenario label is motion, and the action type corresponding to the static scenario label is static.

[0098] In S302, according to the channel state information, the phase matrix and the amplitude matrix of the channel state information matrix of different beams with the same timing label are determined.

[0099] In S303, the phase matrix and the amplitude matrix are fitted to obtain an eigenvalue matrix.

[0100] Among them, the contents shown in S302 to S303 are the same as the contents shown in S102 to S103, and are not described in detail in this embodiment. For details, please refer to the above corresponding embodiments.

[0101] In S304, the eigenvalue matrix and the scenario labels corresponding to the eigenvalue matrix are input into the action classification model for model training to obtain a target action classification model.

[0102] In this embodiment, the above-determined eigenvalue matrix and its corresponding scenario label are input into the motion classification model for model training to obtain the target action classification model, that is, the preset classification model in the above-mentioned motion recognition method embodiment is obtained.

[0103] Among them, the action classification model can be any machine learning algorithm model, such as the LightGBM algorithm, or other machine learning models, which shall be subject to the actual application and is not limited in this embodiment.

[0104] In an embodiment of the present application, beamforming is first used to obtain channel state information of multiple beams in different scenarios in the beam domain, and corresponding scenario labels are set for the channel state information. Then, based on the channel state information, the phase matrix and amplitude matrix of the channel state information matrix of different beams with the same timing label are determined, and then the phase matrix and the amplitude matrix are fitted to obtain the eigenvalue matrix. Finally, the eigenvalue matrix and the scenario labels corresponding to the eigenvalue matrix are input into the action classification model for model training to obtain the target action classification model. The embodiment of the present application utilizes the characteristics of beamforming technology such as high signal strength and high signal-to-noise ratio to collect the channel state information to be identified, which can reduce interference from other devices and expand the recognition range. The eigenvalue matrix is ​​determined by fitting the phase matrix and the amplitude matrix, which can reduce noise interference and improve recognition accuracy.

[0105] In a possible implementation of the present application, determining the phase matrix and amplitude matrix of the channel state information matrix of different beams with the same timing label according to the channel state information may include: determining the channel state information matrix according to the channel state information, the channel state information carrying the timing label; determining the target channel state information matrix according to the channel state information matrix, the target channel state information matrix being a matrix composed of channel state information with subcarrier fluctuation values ​​greater than a fluctuation threshold; obtaining the phase matrix and amplitude matrix based on the target channel state information matrix of different beams with the same timing label under the same scenario label.

[0106] That is to say, in this embodiment, the phase matrix and the amplitude matrix are obtained according to the target channel state information matrices of different beams with the same timing label under the same scenario label. Since the corresponding scenario label is known, it can be used for subsequent model training.

[0107] In a possible implementation of the present application, a phase matrix and an amplitude matrix are obtained based on target channel state information matrices of different beams with the same timing label under the same scenario label, which may include: dividing the target channel state information matrices of different beams with the same timing label under the same scenario label by two or three to obtain a channel state information ratio matrix, wherein the phase information of the channel state information ratio matrix is ​​the phase matrix; and performing denoising processing on the target channel state information matrices of different beams with the same timing label under the same scenario label using wavelet transform to obtain an amplitude matrix.

[0108] In this embodiment, the CSI matrices of different beams with the same timing label under the same scenario label are divided by two by two to obtain a CSI ratio matrix, and the phase information of the CSI ratio matrix is ​​saved and recorded as a phase matrix. The CSI matrices of different beams with the same timing label under the same scenario label are subjected to denoising processing using wavelet transform, and the amplitude is recorded as an amplitude matrix. Noise reduction processing is performed by the above CSI ratio and wavelet transform method, which can reduce noise interference and improve recognition accuracy.

[0109] It should be noted that the training method for the classification model provided in the embodiment of the present application can be executed by a training device for the classification model, or a control module in the training device for the classification model for executing the training method for the classification model. In the embodiment of the present application, the training device for the classification model provided in the embodiment of the present application is described by taking the training method for the classification model executed by the training device for the classification model as an example.

[0110] like Figure 4 As shown in FIG. 1 , a schematic diagram of a motion recognition device provided in an embodiment of the present application is shown. Figure 4 As shown, the action recognition device may include: a first acquisition module 401 , a first determination module 402 , a first fitting module 403 and a recognition module 404 .

[0111] Among them, the first acquisition module 401 is used to obtain the channel state information to be identified of different beams by using beamforming; the first determination module 402 is used to determine the phase matrix to be identified and the amplitude matrix to be identified of the channel state information matrix to be identified of different beams with the same timing label according to the channel state information to be identified; the first fitting module 403 is used to fit the phase matrix to be identified and the amplitude matrix to be identified to obtain the eigenvalue matrix to be identified; the identification module 404 is used to input the eigenvalue matrix to be identified into a preset classification model to identify the action type corresponding to the channel state information to be identified.

[0112] In an embodiment of the present application, first, the first acquisition module 401 uses beamforming to obtain the channel state information to be identified of different beams, and then the first determination module 402 determines the phase matrix to be identified and the amplitude matrix to be identified of the channel state information matrix to be identified of different beams with the same timing label according to the channel state information to be identified, and the first fitting module 403 fits the phase matrix to be identified and the amplitude matrix to be identified to obtain the eigenvalue matrix to be identified, and finally the identification module 404 inputs the eigenvalue matrix to be identified into a preset classification model to identify the action type corresponding to the channel state information to be identified. The embodiment of the present application uses beamforming technology with the characteristics of high signal strength and high signal-to-noise ratio to collect the channel state information to be identified, which can reduce interference from other devices and expand the identification range. The eigenvalue matrix is ​​determined by fitting the phase matrix and the amplitude matrix, which can reduce noise interference and improve recognition accuracy.

[0113] In a possible implementation of the present application, the first acquisition module 401 can be used to: receive an empty data packet notification frame notification sent by the transmitter; receive an empty data packet frame sent by the transmitter; determine a feedback matrix based on the empty data packet frame; send the feedback matrix to the transmitter so that the transmitter uses the feedback matrix to perform beamforming; and use beamforming to obtain channel state information to be identified for different beams.

[0114] In a possible implementation of the present application, the first acquisition module 401 can be used to: analyze the leading sequence in the empty data packet frame to obtain a receiving matrix, which includes a search space; determine a search vector and a signal source direction estimate based on the receiving matrix and the search space; and determine a feedback matrix based on the signal source direction estimate using a linear constrained minimum variance criterion.

[0115] In a possible implementation manner of the present application, the first acquisition module 401 may be used to determine a feedback matrix according to the following formula:

[0116]

[0117] R N =A(t)A H (t)

[0118]

[0119] Among them, V k is the feedback matrix; A(t) is the receiving matrix; A H (t) is the transposed matrix of A(t); F(θ i ) is the estimated value of the signal source direction; F H (θ i ) is F(θ i ) is the transposed matrix; a(θ i) is the search vector; a H (θ i ) is a(θ i ) is the transposed matrix; E N is the search space; For E N The transposed matrix of For R N The inverse matrix of .

[0120] In a possible implementation of the present application, the first determination module 402 can be used to: determine a channel state information matrix to be identified based on the channel state information to be identified, where the channel state information to be identified carries a timing label; determine a target channel state information matrix to be identified based on the channel state information matrix to be identified, where the target channel state information matrix to be identified is a matrix composed of channel state information to be identified whose subcarrier fluctuation values ​​are greater than a fluctuation threshold; divide the target channel state information matrices to be identified of different beams with the same timing label by two by two to obtain a ratio matrix to be identified, where the phase information of the ratio matrix to be identified is a phase matrix to be identified; perform denoising on the target channel state information matrices to be identified of different beams with the same timing label using wavelet transform to obtain an amplitude matrix to be identified.

[0121] In a possible implementation of the present application, the first fitting module 403 can be used to: determine the initial sample weights according to the phase matrix to be identified and the amplitude matrix to be identified, wherein the initial sample weights include the constants to be decided; determine the goodness of fit based on the data after fitting the phase matrix to be identified and the amplitude matrix to be identified, as well as the original data of the phase matrix to be identified and the amplitude matrix to be identified; iteratively process the constants to be decided according to a preset step value until the goodness of fit meets the goodness threshold to obtain the target decision constant; determine the target sample weights according to the target decision constant; obtain the phase weight matrix to be identified and the amplitude weight matrix to be identified based on the target sample weights; obtain the eigenvalue matrix to be identified according to the phase weight matrix to be identified, the amplitude weight matrix to be identified, the phase matrix to be identified and the amplitude matrix to be identified.

[0122] The motion recognition device in the embodiment of the present application can be a device, or a component, integrated circuit, or chip in a terminal. The device can be a mobile electronic device or a non-mobile electronic device, which is not specifically limited in the embodiment of the present application.

[0123] The motion recognition device in the embodiment of the present application may be a device having an operating system. The operating system may be an Android operating system, an iOS operating system, or other possible operating systems, which are not specifically limited in the embodiment of the present application.

[0124] The motion recognition device provided in the embodiment of the present application can achieve Figure 1 to Figure 2The various processes implemented in the method embodiment achieve the same technical effect and will not be described again here to avoid repetition.

[0125] like Figure 5 As shown, it is a schematic diagram of a training device for a classification model provided in an embodiment of the present application. Figure 5 As shown, the training device of the classification model may include: a second acquisition module 501, a second determination module 502, a second fitting module 503 and a training module 504.

[0126] Among them, the second acquisition module 501 is used to use beamforming to obtain the channel state information of multiple beams in different scenarios in the beam domain, and set corresponding scenario labels for the channel state information; the second determination module 502 is used to determine the phase matrix and amplitude matrix of the channel state information matrix of different beams with the same timing labels according to the channel state information; the second fitting module 503 is used to fit the phase matrix and the amplitude matrix to obtain the eigenvalue matrix; the training module 504 is used to input the eigenvalue matrix and the scenario labels corresponding to the eigenvalue matrix into the action classification model for model training to obtain the target action classification model.

[0127] In an embodiment of the present application, first, the second acquisition module 501 uses beamforming to acquire the channel state information of multiple beams in different scenarios in the beam domain, and sets corresponding scenario labels for the channel state information. Then, the second determination module 502 determines the phase matrix and amplitude matrix of the channel state information matrix of different beams with the same timing label according to the channel state information. The second fitting module 503 fits the phase matrix and the amplitude matrix to obtain the eigenvalue matrix. Finally, the training module 504 inputs the eigenvalue matrix and the scenario labels corresponding to the eigenvalue matrix into the action classification model for model training to obtain the target action classification model. The embodiment of the present application uses beamforming technology with the characteristics of high signal strength and high signal-to-noise ratio to collect the channel state information to be identified, which can reduce interference from other devices and expand the recognition range. The eigenvalue matrix is ​​determined by fitting the phase matrix and the amplitude matrix, which can reduce noise interference and improve recognition accuracy.

[0128] In a possible implementation of the present application, the second determination module 502 can be used to: determine a channel state information matrix based on channel state information, where the channel state information carries a timing label; determine a target channel state information matrix based on the channel state information matrix, where the target channel state information matrix is ​​a matrix composed of channel state information with subcarrier fluctuation values ​​greater than a fluctuation threshold; obtain a phase matrix and an amplitude matrix based on the target channel state information matrix of different beams with the same timing label under the same scenario label.

[0129] In a possible implementation of the present application, the second determination module 502 can be used to: divide the target channel state information matrices of different beams with the same timing label under the same scenario label by two by two to obtain a channel state information ratio matrix, wherein the phase information of the channel state information ratio matrix is ​​a phase matrix; and perform denoising processing on the target channel state information matrices of different beams with the same timing label under the same scenario label using wavelet transform to obtain an amplitude matrix.

[0130] The training device of the classification model in the embodiment of the present application can be a device, or a component, integrated circuit, or chip in a terminal. The device can be a mobile electronic device or a non-mobile electronic device, which is not specifically limited in the embodiment of the present application.

[0131] The training device of the classification model in the embodiment of the present application may be a device having an operating system. The operating system may be an Android operating system, an iOS operating system, or other possible operating systems, which are not specifically limited in the embodiment of the present application.

[0132] The training device of the classification model provided in the embodiment of the present application can achieve Figures 2 to 3 The various processes implemented in the method embodiment achieve the same technical effect and will not be described again here to avoid repetition.

[0133] Optional, such as Figure 6 As shown, the embodiment of the present application also provides a terminal device 600, including one or more processors 601, one or more memories 602, and programs or instructions stored in one or more memories 602 and executable on one or more processors 601. When the program or instruction is executed by the processor 601, the various processes of the above-mentioned action recognition method embodiment are implemented, or the various processes of the above-mentioned classification model training method embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0134] The embodiment of the present application also provides a storage medium, on which a program or instruction is stored, and when the program or instruction is executed by a processor, each process of the action recognition method embodiment provided in any of the above embodiments is implemented, or each process of the training method embodiment of the above classification model is implemented. And the same technical effect can be achieved, so it will not be repeated here to avoid repetition.

[0135] The processor is the processor in the terminal device described in the above embodiment. The storage medium includes computer storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, etc.

[0136] An embodiment of the present application further provides a chip, which includes a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run programs or instructions to implement the various processes of the above-mentioned action recognition method embodiment, or to implement the various processes of the above-mentioned classification model training method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0137] It should be understood that the chip mentioned in the embodiments of the present application can also be called a system-level chip, a system chip, a chip system or a system-on-chip chip, etc.

[0138] The embodiments of the present application further provide a computer program / program product, which is stored in a storage medium, and is executed by at least one processor to implement the various processes of the above-mentioned action recognition method embodiment, or to implement the various processes of the above-mentioned classification model training method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0139] An embodiment of the present application further provides a processing device, which is configured to execute the various processes of the above-mentioned action recognition method embodiment, or to implement the various processes of the above-mentioned classification model training method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0140] It should be noted that, in this article, the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise one..." do not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0141] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, a disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0142] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

Claims

1. A method for motion recognition, characterized in that: include Using beamforming to obtain channel state information to be identified for different beams; Determine, according to the channel state information to be identified, a phase matrix to be identified and an amplitude matrix to be identified of the channel state information matrix to be identified of different beams of the same timing label; Fitting the phase matrix to be identified and the amplitude matrix to be identified to obtain an eigenvalue matrix to be identified; The eigenvalue matrix to be identified is input into a preset classification model to identify the action type corresponding to the channel state information to be identified.

2. The action recognition method according to claim 1, characterized in that: The method of obtaining the channel state information to be identified by using beamforming includes: Receive the empty data packet notification frame notification sent by the sender; Receiving an empty data packet frame sent by the transmitting end; Determining a feedback matrix according to the empty data packet frame; Sending the feedback matrix to the transmitting end so that the transmitting end performs beamforming using the feedback matrix; The beamforming is used to obtain the to-be-identified channel state information of different beams.

3. The action recognition method according to claim 2, characterized in that: The step of determining a feedback matrix according to the empty data packet frame comprises: Analyze the leading sequence in the empty data packet frame to obtain a receiving matrix, wherein the receiving matrix includes a search space; Determine a search vector and a signal source direction estimation value according to the receiving matrix and the search space; A feedback matrix is ​​determined according to the signal source direction estimation value using a linear constrained minimum variance criterion.

4. The action recognition method according to claim 3, characterized in that: The feedback matrix is ​​determined as shown below: V k ={R N -1 F(θ i )} / {F H (i i )R N -1 F(θ i )} R N =A(t)A H (t) Among them, V k is the feedback matrix; A(t) is the receiving matrix; A H (t) is the transposed matrix of A(t); F(θ i ) is the estimated value of the signal source direction; F H (θ i ) is F(θ i ) is the transposed matrix; a(θ i ) is the search vector; a H (θ i ) is a(θ i ) is the transposed matrix; E N is the search space; For E N The transposed matrix of For R N The inverse matrix of .

5. The action recognition method according to claim 1, characterized in that: The step of determining, according to the channel state information to be identified, a phase matrix to be identified and an amplitude matrix to be identified of channel state information matrices to be identified for different beams of the same timing label, comprises: Determine a channel state information matrix to be identified according to the channel state information to be identified, wherein the channel state information to be identified carries a timing label; Determine a target channel state information matrix to be identified according to the channel state information matrix to be identified, wherein the target channel state information matrix to be identified is a matrix composed of channel state information to be identified whose subcarrier fluctuation value is greater than a fluctuation threshold; Divide the target channel state information matrices to be identified of different beams with the same timing label by two by two to obtain a ratio matrix to be identified, wherein the phase information of the ratio matrix to be identified is a phase matrix to be identified; Wavelet transform is used to perform denoising on the target channel state information matrix to be identified in different beams with the same timing label to obtain the amplitude matrix to be identified.

6. The action recognition method according to claim 5, characterized in that: The step of fitting the phase matrix to be identified and the amplitude matrix to be identified to obtain the eigenvalue matrix to be identified includes: Determining initial sample weights according to the phase matrix to be identified and the amplitude matrix to be identified, wherein the initial sample weights include a constant to be decided; Determine the goodness of fit based on the fitted data of the phase matrix to be identified and the amplitude matrix to be identified, and the original data of the phase matrix to be identified and the amplitude matrix to be identified; Iteratively process the constant to be decided according to a preset step value until the goodness of fit meets the goodness threshold, thereby obtaining a target decision constant; Determining a target sample weight according to the target decision constant; Based on the target sample weights, a phase weight matrix to be identified and an amplitude weight matrix to be identified are obtained; A to-be-identified eigenvalue matrix is ​​obtained according to the to-be-identified phase weight matrix, the to-be-identified amplitude weight matrix, the to-be-identified phase matrix and the to-be-identified amplitude matrix.

7. A classification model training method, characterized in that: include: Acquire channel state information of multiple beams in different scenarios in a beam domain by using beamforming, and set corresponding scenario labels for the channel state information; Determine, according to the channel state information, a phase matrix and an amplitude matrix of a channel state information matrix of different beams with the same timing label; Fitting the phase matrix and the amplitude matrix to obtain an eigenvalue matrix; The eigenvalue matrix and the scenario labels corresponding to the eigenvalue matrix are input into the action classification model for model training to obtain a target action classification model.

8. The training method of the classification model according to claim 7, characterized in that: The determining, according to the channel state information, a phase matrix and an amplitude matrix of a channel state information matrix of different beams with the same timing label, comprises: Determine a channel state information matrix according to the channel state information, wherein the channel state information carries a timing label; Determine a target channel state information matrix according to the channel state information matrix, wherein the target channel state information matrix is ​​a matrix composed of channel state information whose subcarrier fluctuation values ​​are greater than a fluctuation threshold; Based on the target channel state information matrix of different beams with the same timing label under the same scenario label, a phase matrix and an amplitude matrix are obtained.

9. The training method of the classification model according to claim 8, characterized in that: The target channel state information matrix of different beams with the same timing label under the same scenario label is used to obtain a phase matrix and an amplitude matrix, including: Divide the target channel state information matrices of different beams with the same timing label under the same scenario label by two by two to obtain a channel state information ratio matrix, wherein the phase information of the channel state information ratio matrix is ​​a phase matrix; Wavelet transform is used to perform denoising on the target channel state information matrix of different beams with the same time sequence label under the same scenario label to obtain an amplitude matrix.

10. A terminal device, characterized in that: The method comprises one or more processors, one or more memories, and programs or instructions stored in the one or more memories and executable on the one or more processors, wherein the programs or instructions, when executed by the processors, implement the steps of the method according to any one of claims 1 to 6, or implement the steps of the method according to any one of claims 7 to 9.

Images