Old people falling behavior identification method based on wireless sensing and related device

Abstract

The embodiment of the invention provides an old people falling behavior identification method based on wireless sensing and a related device. The method comprises the following steps: acquiring original CSI data of different human body behaviors in a simulated bathroom environment; the method comprises the following steps: firstly extracting an amplitude value from original CSI data, and then eliminating an abnormal value; carrying out dimensionality reduction on the normalized CSI data by adopting a singular value decomposition method to obtain dimensionality-reduced CSI data; for the CSI data after dimension reduction, calculating a Pearson's correlation coefficient as model input, and generating various behavior samples and labels; inputting the divided data into the trained embedded network for training to obtain prototype expressions of the support set and the query set of each category; and by calculating the Euclidean distance between the falling behavior prototype and other behavior prototypes, outputting an old people falling behavior identification result, thereby realizing cross-environment falling identification. According to the embodiment of the invention, the method can effectively improve the recognition accuracy and reliability of the falling behavior of the old people in different environments.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical fields of human body perception technology and deep learning technology based on WiFi channel state information, and in particular to a method and related device for identifying the falling behavior of the elderly based on wireless perception. Background Art

[0002] The aging of the population is intensifying day by day, resulting in the increasingly serious safety problems of the elderly living alone, especially the falling incidents that may occur when using the toilet and taking a bath. Therefore, the action monitoring of the elderly has become an urgent problem that needs attention.

[0003] However, the current methods for identifying human falling behavior have low recognition accuracy in new environments. Therefore, how to effectively improve the recognition accuracy and reliability of the elderly's falling behavior in different environments has become a technical problem to be solved urgently. Summary of the Invention

[0004] The embodiments of the present invention provide a method and related device for identifying the falling behavior of the elderly based on wireless perception, which can realize the cross-environment recognition of the elderly's falling behavior through a WiFi CSI sensor combined with a deep learning algorithm, and can effectively improve the recognition accuracy and reliability of the elderly's falling behavior in different environments.

[0005] In a first aspect, the embodiments of the present invention provide a method for identifying the falling behavior of the elderly based on wireless perception, including:

[0006] Step 1: Collect the original CSI data of different human behaviors in a simulated bathroom environment;

[0007] Step 2: Extract the amplitude information from the original CSI data, and then perform outlier elimination processing and normalization processing on the amplitude information to obtain the normalized CSI data;

[0008] Step 3: Use the singular value decomposition method to reduce the dimension of the normalized CSI data to obtain the dimension-reduced CSI data;

[0009] Step 4: Calculate the Pearson correlation coefficient for the dimension-reduced CSI data to obtain the input of the falling behavior recognition model;

[0010] Step 5: Divide the sample data of each environment for each action into a support set and a query set;

[0011] Step 6: Input the divided support set and query set into the trained embedding network for training to obtain the prototype expressions of the support set and query set for each category;

[0012] Step 7: Output the recognition result of the elderly's falling behavior by calculating the Euclidean distance between the falling behavior prototype and other behavior prototypes.

[0013] In some embodiments, two USRP devices are used for the acquisition of the original CSI data. One is used as a transmitter and the other as a receiver. The original CSI data collected using a 1x1 antenna transceiver system is represented as follows:

[0014]

[0015] where p represents a data packet and s represents a subcarrier.

[0016] In some embodiments, step 3 includes:

[0017] (1) Denote the original amplitude information obtained in step 2 as 500 is the number of data packets, and 57 is the number of subcarriers received by the antenna under a 20Mhz bandwidth. Then, the Francis algorithm is used to solve the eigenvalues and eigenvectors required for singular value decomposition:

[0018] ① For the matrix Calculate Denote it as the symmetric positive definite matrix A, and use the Francis algorithm to perform iterative calculations on A to obtain a series of similar matrices A k ;

[0019] ② Continuously iterate until the matrix A k converges to an upper triangular or diagonal matrix form. At this time, the elements on the main diagonal are the approximate eigenvalues λ i ;

[0020] ③ For each eigenvalue λ i , by solving the linear equation system (A - λ i I)X = 0, the corresponding eigenvector X i can be obtained, where I is the identity matrix;

[0021] (2) Perform SVD decomposition:

[0022] ① For the obtained eigenvalues λ i and eigenvectors X i ;

[0023] ② Sort the eigenvalues, λ1 ≥ λ2 ≥ Λ ≥ λ n ≥ 0, and the corresponding eigenvectors X1, X2, Λ, X n constitute the column vectors of matrix V;

[0024] ③ Calculate the singular values and form the diagonal matrix Σ, whose main diagonal elements are σ i , and the remaining elements are 0;

[0025] ④ Calculate the column vectors \(u\) of \(U\). i , When \(\sigma\) i = 0, \(u\) i can be obtained by solving the fundamental solution system of \(A\) T \(u = 0\), so as to obtain the matrix \(U\), and thus the amplitude information

[0026] is obtained, and the singular value decomposition of the amplitude information is obtained;

[0027] ⑤ Take the product \(H\) of the left singular values and the diagonal matrix Ad \(= U\sum(:,1:57\) ) ;

[0028] (3) Indirectly calculate the matrix eigenvalues required for the SVD decomposition through the Francis algorithm. During the iterative process of calculating the matrix eigenvalues, the FPGA is used to perform matrix multiplication and QR decomposition operations in parallel.

[0029] In some embodiments, in step 3, for the \(H\) obtained by dimensionality reduction and feature extraction of the CSI data Ad After calculating the Pearson correlation coefficient, the dimension of each sample is reduced from 500x57 to 56x56.

[0030] In some embodiments, in the fall behavior recognition model, various types of samples are mapped to the feature space through CNN technology. According to the discrimination between the fall action and other actions in the feature space, the ProtoNet model is used for classification to obtain the recognition result of the elderly fall behavior.

[0031] In some embodiments, the verification method of the elderly fall behavior recognition method is as follows: collect the support set and the query set in an environment and train and optimize the model to obtain the optimized parameters; then, collect a small number of samples as the support set in the new environment to obtain the recognition effect, without the need to collect a large number of samples again and perform retraining or optimization.

[0032] In some embodiments, in the ProtoNet model, the Euclidean distance is used to calculate the distance between the prototype representations of the support set and the query set.

[0033] Second, the embodiments of the present invention also provide an elderly fall behavior recognition device based on wireless sensing. The device includes:

[0034] An acquisition module, configured to acquire the original CSI data of different human behaviors in a simulated bathroom environment;

[0035] A processing module, configured to extract amplitude information from the original CSI data, and then perform outlier elimination processing and normalization processing on the amplitude information to obtain the normalized CSI data;

[0036] A dimensionality reduction module, which is used to perform dimensionality reduction on the normalized CSI data by using the singular value decomposition method to obtain the CSI data after dimensionality reduction;

[0037] A calculation module, which is used to calculate the Pearson correlation coefficient for the CSI data after dimensionality reduction to obtain the input of the fall behavior recognition model;

[0038] A construction module, which is used to divide the sample data of each environment for each action into a support set and a query set;

[0039] An embedding module, which is used to input the divided support set and query set into a trained embedding network for training to obtain the prototype expressions of the support set and query set for each category;

[0040] An identification module, which is used to output the recognition result of the elderly fall behavior by calculating the Euclidean distance between the fall behavior prototype and other behavior prototypes.

[0041] In a third aspect, an embodiment of the present invention further provides an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein when the processor executes the computer program, the method for recognizing the elderly fall behavior based on wireless sensing described in the first aspect is implemented.

[0042] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, storing computer-executable instructions, and the computer-executable instructions are used to execute the method for recognizing the elderly fall behavior based on wireless sensing described in the first aspect.

[0043] According to the method and related device for recognizing the elderly fall behavior based on wireless sensing provided by the embodiments of the present invention, wherein the method for recognizing the elderly fall behavior based on wireless sensing includes: collecting the original CSI data of different human behaviors in a simulated bathroom environment; extracting amplitude information from the original CSI data, and then performing outlier elimination processing and normalization processing on the amplitude information to obtain the normalized CSI data; performing dimensionality reduction on the normalized CSI data by using the singular value decomposition method to obtain the CSI data after dimensionality reduction; calculating the Pearson correlation coefficient for the CSI data after dimensionality reduction to obtain the input of the fall behavior recognition model; dividing the sample data of each environment for each action into a support set and a query set; inputting the divided support set and query set into a trained embedding network for training to obtain the prototype expressions of the support set and query set for each category; outputting the recognition result of the elderly fall behavior by calculating the Euclidean distance between the fall behavior prototype and other behavior prototypes, so as to realize cross-environment fall recognition. Based on this, the embodiments of the present invention can effectively improve the recognition accuracy and reliability of the elderly fall behavior in different environments. Description of the Drawings

[0044] Figure 1 It is a flowchart of a method for recognizing the falling behavior of the elderly based on wireless sensing provided by an embodiment of the present invention;

[0045] Figure 2 It is an original signal diagram of the collected falling actions provided by an embodiment of the present invention;

[0046] Figure 3 It is a signal diagram of the collected falling actions after simple processing in step 2 provided by an embodiment of the present invention;

[0047] Figure 4 It is a schematic diagram of a device for recognizing the falling behavior of the elderly based on wireless sensing provided by an embodiment of the present invention;

[0048] Figure 5 It is a schematic diagram of an electronic device provided by an embodiment of the present invention. Detailed implementation manners

[0049] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0050] It should be noted that although the functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order from the module division in the device or the flowchart in the flowchart. Terms such as "first" and "second" in the specification, claims and the following drawings are used to distinguish similar objects and do not have to be used to describe a specific order or sequence.

[0051] In the embodiments of the present invention, words such as "furthermore", "exemplarily" or "optionally" are used to represent examples, illustrations or explanations, and should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Using words such as "furthermore", "exemplarily" or "optionally" is intended to present related concepts in a specific manner.

[0052] In order to more conveniently describe the working principle of the embodiments of the present invention in the following, an introduction to the related technical scenarios will be given first.

[0053] However, the current methods for recognizing human falling behavior have low recognition accuracy in new environments. Therefore, how to effectively improve the recognition accuracy and reliability of the falling behavior of the elderly in different environments has become an urgent technical problem to be solved.

[0054] Based on this, the present invention provides a method and related device for recognizing the falling behavior of the elderly based on wireless sensing. Among them, the method for recognizing the falling behavior of the elderly based on wireless sensing includes: collecting the original CSI data of different human behaviors in a simulated bathroom environment; extracting amplitude information from the original CSI data, and then performing outlier elimination processing and normalization processing on the amplitude information to obtain the normalized CSI data; using the singular value decomposition method to perform dimensionality reduction on the normalized CSI data to obtain the dimensionality-reduced CSI data; calculating the Pearson correlation coefficient for the dimensionality-reduced CSI data to obtain the input of the falling behavior recognition model; dividing the sample data of each environment for each action into a support set and a query set; inputting the divided support set and query set into a trained embedding network for training to obtain the prototype expressions of the support sets and query sets of each category; and outputting the recognition result of the falling behavior of the elderly by calculating the Euclidean distance between the falling behavior prototype and other behavior prototypes, so as to achieve cross-environment fall recognition. Based on this, the embodiments of the present invention can effectively improve the recognition accuracy and reliability of the falling behavior of the elderly in different environments.

[0055] The following further elaborates on the embodiments of the present invention with reference to the accompanying drawings.

[0056] As Figure 1 shown, Figure 1 is a flowchart of a method for recognizing the falling behavior of the elderly based on wireless sensing provided by an embodiment of the present invention. By collecting the original CSI data of different human behaviors in a simulated bathroom environment; extracting amplitude information from the original CSI data, and then performing outlier elimination processing and normalization processing on the amplitude information to obtain the normalized CSI data; using the singular value decomposition method to perform dimensionality reduction on the normalized CSI data to obtain the dimensionality-reduced CSI data; calculating the Pearson correlation coefficient for the dimensionality-reduced CSI data to obtain the input of the falling behavior recognition model; dividing the sample data of each environment for each action into a support set and a query set; inputting the divided support set and query set into a trained embedding network for training to obtain the prototype expressions of the support sets and query sets of each category; and outputting the recognition result of the falling behavior of the elderly by calculating the Euclidean distance between the falling behavior prototype and other behavior prototypes, so as to achieve cross-environment fall recognition.

[0057] It should be noted that the current human fall detection methods mainly include four technologies based on radar, wearable devices, video analysis, and WiFi. Video analysis has privacy issues, and wearable devices have relatively high deployment costs. In contrast, the fall detection method based on WiFi signals has obvious advantages in terms of privacy protection, cost reduction, and coverage. Therefore, the present invention adopts the fall detection technology based on WiFi signals. Considering compatibility, a 20MHz bandwidth is the basic standard for WiFi, which can ensure wide device compatibility and provide a longer transmission distance. A wider bandwidth is usually suitable for short-distance communication, while a 20MHz bandwidth provides a more stable connection in an environment where the signal needs to penetrate obstacles or is more interfered.

[0058] In the embodiment of the present invention, a software-defined radio device is used as a WiFi CSI sensor to collect channel state information (CSI) of six human behaviors, such as falling, walking, and standing still, in two different simulated bathroom environments. The duration of each behavior is about 2 seconds. For the original CSI data, the amplitude is first extracted and then the outliers are removed. Then, the singular value decomposition (SVD) method is used to reduce the dimension of the normalized CSI data and reduce the number of data packets (time-domain features). The Pearson correlation coefficient is calculated for the dimension-reduced CSI data to obtain the input of the model, and samples and labels of various behaviors are generated. After the divided data is sent to the trained embedding network for training, and the prototype expressions of the support set and query set of each category are obtained, the Euclidean distance between the prototype of the fall behavior and the prototypes of other behaviors is calculated for cross-environment fall recognition.

[0059] Specifically, the recognition method may include the following steps:

[0060] Step 1, CSI data collection: As Figure 2 shown, a software-defined radio device (USRP) is used as a WiFi CSI sensor to collect channel state information (CSI) of six human behaviors, such as falling, walking, and standing still, in two different simulated bathroom environments. The duration of each behavior is about 2 seconds.

[0061] Step 2, Perform a series of preprocessing operations on the collected CSI signal. First, extract the amplitude information and remove the outliers, and then normalize the signal to ensure the consistency and comparability of the data. The result is as Figure 3 shown.

[0062] Step 3, Feature dimension reduction and extraction: The singular value decomposition (SVD) method is used to reduce the dimension of the normalized CSI signal to reduce the number of data packets (time-domain features). Through SVD decomposition, the subcarrier resolution of the signal is retained, and the reduced feature data is provided for subsequent processing.

[0063] Step 4, Calculate the Pearson correlation coefficient: Calculate the Pearson correlation coefficient for the dimension-reduced signal to obtain the similarity matrix for each action category, further compressing the feature dimension and improving the model performance.

[0064] Step 5, Behavior sample construction: Divide the sample data of each action into a support set and a query set. The support set is used for training, and the query set is used for testing. Through the prototype representation of these samples in the embedding space, it helps the model identify different action categories, especially the fall behavior.

[0065] Step 6, Input the embedding model: Use the deep learning embedding model (ProtoNet) to train various actions. The ProtoNet model optimizes the embedding representation of various behaviors by calculating the prototype representation of the support set and the query set in the feature space. In particular, the prototype representation of the fall behavior will be specially optimized to improve the cross-environment recognition rate of the fall action.

[0066] Step 7, Fall behavior recognition: Perform cross-environment fall recognition by calculating the Euclidean distance between the fall behavior prototype and other behavior prototypes. The ProtoNet model optimizes the support set and the query set, and finally achieves a high recognition accuracy for the fall behavior. Especially in a new environment, the model can quickly adapt and maintain a high accuracy.

[0067] It can be understood that the CSI element signals collected in Step 1 are as follows:

[0068]

[0069] where p represents the data packet and s represents the subcarrier.

[0070] It can be understood that the CSI signal dimension reduction and feature extraction in Step 3 include:

[0071] (1) The original amplitude information obtained according to Step 2 is denoted as 500 is the number of data packets, 57 is the number of subcarriers received by the antenna under a 20Mhz bandwidth, and then use the Francis algorithm to solve the eigenvalues and eigenvectors required for singular value decomposition. The specific steps are as follows:

[0072] ① For the matrix Calculate Denote it as the symmetric positive definite matrix A, and use the Francis algorithm to perform iterative calculations on A to obtain a series of similar matrices A k .

[0073] ② Continuously iterate until the matrix A k converges to an upper triangular or diagonal matrix form. At this time, the elements on the main diagonal are the approximate eigenvalues λ of matrix A i .

[0074] ③For each eigenvalue λ i , by solving the linear equation system (A - λ i I)X = 0, the corresponding eigenvector X i can be obtained, where I is the identity matrix.

[0075] (2) Perform SVD decomposition:

[0076] ①For the obtained eigenvalues λ i and eigenvectors X i .

[0077] ②Sort the eigenvalues, λ1 ≥ λ2 ≥ Λ ≥ λ n ≥ 0, and the corresponding eigenvectors X1, X2, Λ, X n form the column vectors of matrix V.

[0078] ③Calculate the singular values and form the diagonal matrix ∑, whose main diagonal elements are σ i , and the remaining elements are 0.

[0079] ④Calculate the column vectors u i , when σ i = 0, u i can be obtained by solving the fundamental solution system of Au T = 0, thus obtaining matrix U, and the singular value decomposition of the amplitude information is obtained.

[0080] ⑤Take the product H Ad of the left singular values and the diagonal matrix, H

[0081] = U∑(:, 1:57).

[0082] It can be understood that calculating the matrix eigenvalues required for SVD decomposition indirectly through the Francis algorithm usually has high efficiency, especially for large matrices. In FPGA, its parallel computing ability and pipeline structure can be used to further accelerate the execution of the Francis algorithm, thereby improving the speed of the entire SVD calculation. For example, during the iterative process of calculating eigenvalues, operations such as matrix multiplication and QR decomposition can be performed in parallel to fully utilize the hardware advantages of FPGA. Ad

[0083] It can be understood that after calculating the Pearson correlation coefficient for H Ad obtained in step 3 in step 4, the dimension of each sample is reduced from 500x57 to 56x56.It can be understood that in step 5, various types of samples are mapped to the feature space through CNN technology. Experiments show that as the number of action categories significantly different from the falling action increases, the distinction between the falling action and other actions in the feature space gradually becomes significant, showing a more obvious differentiation trend. Finally, the ProtoNet model is used for classification, and a relatively high classification accuracy of falling behavior is obtained, with the prediction accuracy of falling reaching 98%.

[0084] It can be understood that the verification method of the falling behavior recognition method is as follows: collect the support set and query set in an environment and train and optimize the model to obtain optimized parameters; then, collect a small number of samples as the support set in the new environment, and good recognition results can be obtained without re-collecting a large number of samples and performing re-training or optimization.

[0085] It can be understood that in the ProtoNet model, the Euclidean distance is used to calculate the distance between the prototype representations of the support set and the query set.

[0086] Based on this, a method for recognizing the falling behavior of the elderly based on wireless sensing proposed by the present invention can collect the channel state information (CSI) of six behaviors such as human falling, walking, and standing still in two different simulated bathroom environments by using a software radio device as a WiFi CSI sensor, and each behavior lasts about 2 seconds. For the original CSI data, first extract the amplitude value, then eliminate the outliers, and then use the singular value decomposition (SVD) method to reduce the dimension of the normalized CSI data to reduce the number of data packets (time domain features). For the CSI data after dimension reduction, calculate the Pearson correlation coefficient as the model input to generate various behavior samples and labels. Input the divided data into the trained embedding network for training to obtain the prototype expressions of the support set and query set for each category, and realize cross-environment fall recognition by calculating the Euclidean distance between the fall behavior prototype and other behavior prototypes. Based on this, the embodiments of the present invention can effectively improve the recognition accuracy and reliability of the falling behavior of the elderly in public bathrooms in different environments.

[0087] In addition, as Figure 4 shown, an embodiment of the present invention also discloses a device for recognizing the falling behavior of the elderly based on wireless sensing, and the device includes:

[0088] A collection module 110, configured to collect the original CSI data of different human behaviors in a simulated bathroom environment;

[0089] A processing module 120, configured to extract amplitude information from the original CSI data, and then perform outlier elimination processing and normalization processing on the amplitude information to obtain the normalized CSI data;

[0090] The dimensionality reduction module 130 is used to perform dimensionality reduction on the normalized CSI data by using the singular value decomposition method to obtain the CSI data after dimensionality reduction;

[0091] The calculation module 140 is used to calculate the Pearson correlation coefficient for the CSI data after dimensionality reduction to obtain the input for the fall behavior recognition model;

[0092] The construction module 150 is used to divide the sample data of each environment for each action into a support set and a query set;

[0093] The embedding module 160 is used to input the divided support set and query set into the trained embedding network for training to obtain the prototype expressions of each category of support set and query set;

[0094] The recognition module 170 is used to output the recognition result of the elderly fall behavior by calculating the Euclidean distance between the fall behavior prototype and other behavior prototypes.

[0095] The elderly fall behavior recognition device based on wireless sensing in the embodiments of the present invention is used to execute the elderly fall behavior recognition method based on wireless sensing in the above embodiments, and its specific processing process is the same as that of the elderly fall behavior recognition method based on wireless sensing in the above embodiments, which will not be elaborated here one by one.

[0096] In addition, as Figure 5 shown, an embodiment of the present invention also discloses an electronic device, including: at least one processor 210; at least one memory 220 for storing at least one program; when the at least one program is executed by the at least one processor 210, the elderly fall behavior recognition method based on wireless sensing in any of the previous embodiments is implemented.

[0097] In addition, an embodiment of the present invention also discloses a computer-readable storage medium, in which computer-executable instructions are stored, and the computer-executable instructions are used to execute the elderly fall behavior recognition method based on wireless sensing in any of the previous embodiments.

[0098] The system architecture and application scenarios described in the embodiments of the present invention are for more clearly explaining the technical solutions of the embodiments of the present invention, and do not constitute a limitation to the technical solutions provided by the embodiments of the present invention. Those skilled in the art know that with the evolution of the system architecture and the emergence of new application scenarios, the technical solutions provided by the embodiments of the present invention are equally applicable to similar technical problems.

[0099] Those of ordinary skill in the art can understand that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices can be implemented as software, firmware, hardware and their appropriate combinations.

[0100] In a hardware implementation, the division between the functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, one physical component can have multiple functions, or one function or step can be executed by several physical components in cooperation. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or as hardware, or as an integrated circuit, such as an application specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include a computer storage medium (or non-transitory medium) and a communication medium (or transitory medium). As is well known to those of ordinary skill in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information such as computer-readable instructions, data structures, program modules, or other data. Computer storage media includes but is not limited to RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disks (DVDs) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and that can be accessed by a computer. In addition, as is well known to those of ordinary skill in the art, communication media typically embodies computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism and includes any information delivery media.

[0101] The terms "component", "module", "system", etc. used in this specification are used to refer to computer-related entities, hardware, firmware, combinations of hardware and software, software, or software in execution. For example, a component can be, but is not limited to, a process running on a processor, a processor, an object, an executable, an execution thread, a program, or a computer. By way of illustration, an application running on a computing device and the computing device can both be components. One or more components can reside in a process or execution thread, and a component can be located on one computer or distributed between two or more computers. In addition, these components can execute from various computer-readable media having various data structures stored thereon. A component can communicate, for example, by signals with other systems through local or remote processes according to one or more data packets (e.g., data from two components interacting with each other from a local system, a distributed system, or a network, such as through the Internet interacting with other systems).

Claims

1. A method for identifying falling behavior of the elderly based on wireless sensing, characterized in that: include: Step 1: Collect raw CSI data of different human behaviors in a simulated bathroom environment; Step 2: Extract amplitude information from the original CSI data, then eliminate outliers and normalize the amplitude information to obtain normalized CSI data; Step 3: Use the singular value decomposition method to reduce the dimension of the normalized CSI data to obtain the reduced-dimensional CSI data; Step 4: Calculate the Pearson correlation coefficient for the CSI data after dimensionality reduction to obtain the input of the fall behavior recognition model; Step 5: Divide the sample data of each environment for each action into a support set and a query set; Step 6: Input the divided support set and query set into the trained embedding network training to obtain the prototype expression of each category support set and query set; Step 7: Output the recognition result of the falling behavior of the elderly by calculating the Euclidean distance between the falling behavior prototype and other behavior prototypes.

2. The method for identifying the falling behavior of the elderly based on wireless sensing according to claim 1 is characterized in that: The original CSI data is collected using two USRP devices, one as a transmitter and one as a receiver. The original CSI data collected using a 1x1 antenna transceiver system is expressed as follows: Wherein, p represents a data packet, and s represents a subcarrier.

3. The method for identifying the falling behavior of the elderly based on wireless sensing according to claim 2 is characterized in that: The step 3 comprises: (1) The original amplitude information obtained according to step 2 is recorded as 500 is the number of packets, 57 is the number of subcarriers received by the antenna at 20Mhz bandwidth, and then the Francis algorithm is used to solve The eigenvalues ​​and eigenvectors required for singular value decomposition: ①For the matrix calculate Denoted as a symmetric positive definite matrix A, A is iterated using the Francis algorithm to obtain a series of similar matrices A k ; ②Continue iterating until the matrix A k Converges to an upper triangular or diagonal matrix form, where the elements on the main diagonal are the approximate eigenvalues ​​λ of the matrix A. i ; ③For each eigenvalue λ i , by solving the linear equation system (A-λ i I) X = 0, we can get the corresponding eigenvector X i , where I is the identity matrix; (2) Perform SVD decomposition: ① For the obtained The eigenvalue λ i and the eigenvector X i ; ② Sort the eigenvalues, λ1≥λ2≥Λ≥λ n ≥0, the corresponding eigenvectors X1,X2,Λ,X n The column vectors that make up the matrix V; ③Calculate singular values And form a diagonal matrix Σ, whose main diagonal elements are σ i , the rest of the elements are 0; ④ Calculate the column vector u of U i , When σ i = 0, u i By solving A T The basic solution system of u=0 is obtained, thus obtaining the matrix U, from which the amplitude information is obtained The singular value decomposition of ; ⑤ Take the product of the left singular value and the diagonal matrix H Ad =UΣ(:,1:57); (3) The matrix eigenvalues ​​required for SVD decomposition are indirectly calculated through the Francis algorithm. In the iterative process of calculating the matrix eigenvalues, FPGA is used to perform matrix multiplication and QR decomposition operations in parallel.

4. The method for identifying the falling behavior of the elderly based on wireless sensing according to claim 3 is characterized in that: In step 3, the CSI data is reduced in dimension and features are extracted to obtain H Ad After calculating the Pearson correlation coefficient, the dimension of each sample is reduced from 500x57 to 56x56.

5. The method for identifying the falling behavior of the elderly based on wireless sensing according to claim 1 is characterized in that: In the fall behavior recognition model, various samples are mapped to the feature space through CNN technology, and the ProtoNet model is used to classify them according to the distinction between the fall action and other actions in the feature space to obtain the fall behavior recognition results of the elderly.

6. The method for identifying elderly people's falling behavior based on wireless sensing according to claim 1 is characterized in that: The verification method of the elderly fall behavior recognition method is as follows: a support set and a query set are collected in one environment and the model is trained and tuned to obtain optimized parameters; then, a small number of samples are collected as a support set in a new environment to obtain a recognition effect without the need to collect a large number of samples again and perform retraining or tuning.

7. The method for identifying elderly people's falling behavior based on wireless sensing according to claim 5 is characterized in that: In the ProtoNet model, the distance between the prototype representations of the support set and the query set is calculated using the Euclidean distance.

8. A device for identifying elderly people's falling behavior based on wireless sensing, characterized in that: The device comprises: The acquisition module is used to collect the raw CSI data of different human behaviors in the simulated bathroom environment; A processing module is used to extract amplitude information from the original CSI data, and then perform outlier elimination and normalization processing on the amplitude information to obtain normalized CSI data; A dimension reduction module, used for reducing the dimension of the normalized CSI data by using a singular value decomposition method to obtain CSI data after dimension reduction; A calculation module is used to calculate the Pearson correlation coefficient of the CSI data after dimensionality reduction to obtain the input of the fall behavior recognition model; Building a module for partitioning the sample data of each environment for each action into a support set and a query set; The embedding module is used to input the divided support set and query set into the trained embedding network training to obtain the prototype expression of the support set and query set of each category; The recognition module is used to output the recognition result of the falling behavior of the elderly by calculating the Euclidean distance between the falling behavior prototype and other behavior prototypes.

9. An electronic device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for identifying the falling behavior of the elderly based on wireless sensing as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions are used to execute the method for identifying falling behavior of the elderly based on wireless sensing as described in any one of claims 1 to 7.

Images