Campus safety monitoring method and system based on Wi-Fi perception and deep learning technology
Through Wi-Fi perception and deep learning technology, wireless transceiver devices are used to collect CSI data, build unsafe behavior data sets and conduct model training, solving the device deployment and privacy issues in campus security monitoring, and achieving low-cost and strong privacy Real-time security monitoring, suitable for non-line-of-sight environments.
Patent Information
- Application Number
- CN202411102100.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-12
- Publication Date
- 2024-12-06
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The prior art is difficult to achieve low-cost, highly private and real-time wireless perception solutions in campus security monitoring, especially in non-line-of-sight conditions and in under-light environments, and traditional methods have device deployment and privacy issues.
Adopting campus security monitoring methods based on Wi-Fi perception and deep learning technology, Wi-Fi CSI data is collected through wireless transceiver devices, unsafe behavior data sets are constructed with tags, and two-way long and short-term memory networks based on attention mechanisms Carry out model training to achieve real-time identification and early warning of human behavior.
It realizes low-cost and highly privacy campus security monitoring, improves the utilization rate of wireless network resources, can integrate perception and high-performance backend computing on low-performance front-end devices, and provides real-time high-risk risk behavior warnings, suitable for unsuitable A secret place for video surveillance.
Smart Images

Figure CN119095097A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of wireless sensing technology, and in particular to a campus safety monitoring method and system based on Wi-Fi sensing and deep learning technology. Background Art
[0002] Human behavioral activities are diverse. Analyzing daily human activities such as walking, running, falling, lying down and standing up can not only establish the interaction and linkage relationship between daily human activities and smart devices, but also judge safety hazards through behavioral recognition, such as judging that a fall has occurred within the monitoring range and triggering alarm measures.
[0003] Compared with traditional sensing technologies such as computer vision or wearable sensors, wireless sensing using Wi-Fi signals has more advantages. Wi-Fi technology has developed rapidly and matured, and is now widely used in daily life. It does not require the deployment of additional equipment and is low-cost. In addition, wireless sensing can be carried out simultaneously with network communication, which improves the utilization rate of wireless resources and is suitable for long-term deployment and use. Unlike computer vision, Wi-Fi signals are not affected by conditions such as light and can be used in dark or low-light environments. In non-line-of-sight conditions, Wi-Fi signals can pass through obstacles and obtain useful information. In addition, wireless signals do not need to collect visual information of users, so there is no privacy issue. Unlike wearable sensors, the objects to be identified do not need to wear any equipment or intentionally cooperate with behavior monitoring, so identification can be performed unconsciously. Summary of the invention
[0004] The purpose of the present invention is to provide a campus safety monitoring method and system based on Wi-Fi perception and deep learning technology, which fully utilizes the advantages of Wi-Fi signal perception and uses the rich information contained in Wi-Fi CSI to realize a low-cost, highly private and real-time campus safety monitoring solution.
[0005] In order to achieve the above tasks, the present invention adopts the following technical solutions:
[0006] Campus security monitoring methods based on Wi-Fi sensing and deep learning technology include:
[0007] Deploy wireless transceiver equipment in the monitored area, and obtain CSI data generated by the wireless transceiver equipment when there is human activity in the monitored area;
[0008] Arrange personnel to perform activities in the monitored area according to preset actions, wherein the preset actions are normal actions and unsafe behavior actions; use the wireless transceiver equipment in the wireless channel to collect CSI data when the preset actions occur, and receive and process the data according to the manually specified time length; classify the preset actions according to the danger level;
[0009] Preprocess the collected CSI data to construct a labeled unsafe behavior dataset;
[0010] The unsafe behavior dataset is divided into a training set and a validation set. The labeled training set is fed into a deep neural network for model training, and the validation set is used to test the model effect. The trained model is used as an unsafe behavior detection model.
[0011] The unsafe behavior detection model is used to predict the CSI data collected by the wireless transceiver equipment deployed in the monitored area and output the prediction results.
[0012] Furthermore, the unsafe behavior detection model is used to predict the CSI data collected by the wireless transceiver equipment deployed in the monitored area, including:
[0013] After deploying wireless channels in the monitored area, wireless transceiver equipment is used to collect real-time CSI data for the monitored area. The collected CSI data is pre-processed by equal-interval cutting, low-pass filtering, and dimensionality reduction, and then input into the trained unsafe behavior detection model. The unsafe behavior detection model outputs the prediction result, which is the probability of occurrence of various preset actions in the monitored area.
[0014] Furthermore, a classification scheme is performed based on the danger level, and when the prediction result reaches a preset danger level or above, an early warning message is issued.
[0015] Furthermore, the warning information is pushed to the campus security department, and the numbers of relevant monitoring equipment in and around the monitored area are also pushed.
[0016] Furthermore, the wireless transceiver device includes device A, device B and a router, and device A and device B are connected to the router;
[0017] Device A generates a Wi-Fi data packet and sends it to the router through the Wi-Fi transceiver module. After receiving the Wi-Fi data packet through the Wi-Fi transceiver module, the router obtains the channel state information CSI and returns it to device A. Device A continuously transmits the CSI data to device B, so as to obtain the CSI data when there is human activity in the detection area; the device B is used to pre-process the CSI data and train the unsafe behavior detection model.
[0018] Furthermore, the preprocessing of the collected CSI data to construct an unsafe behavior data set includes:
[0019] Step 1: Unpack the collected CSI data in .csi format and extract the amplitude information; label the CSI data accordingly according to the activity type, number of people and data collection time in the monitored area;
[0020] Step 2: Segment the CSI data obtained by different transceiver antenna combinations and frequency domain division of the wireless transceiver device. In the time domain, the CSI data needs to be cut at equal intervals to ensure that the data dimension is consistent.
[0021] Step 3: Use a Butterworth filter to low-pass filter the CSI data, and perform principal component analysis on the filtered data to reduce the data dimension; the data after dimensionality reduction constitutes an unsafe behavior data set.
[0022] Furthermore, the deep neural network adopts a bidirectional long short-term memory network based on an attention mechanism.
[0023] A campus security monitoring system based on Wi-Fi sensing and deep learning technology includes a device A deployed in a monitored area, a router, and a remote device B, wherein:
[0024] Device A and the router have Wi-Fi transceiver modules. Device A is used to send Wi-Fi data to the monitored area, and the router is used to receive Wi-Fi data and obtain corresponding CSI data, and then send it to device B according to the preset length;
[0025] Device B is used to construct an unsafe behavior data set based on the collected CSI data, and use the unsafe behavior data set and deep neural network training to obtain an unsafe behavior detection model. The unsafe behavior detection model is used to predict human behavior in the monitored area;
[0026] Among them, personnel are arranged to perform activities according to preset actions in the monitored area, and the preset actions are normal actions and unsafe behavior actions; device B is used to collect CSI data when the preset actions occur, and the data is received and processed according to the manually specified time length; the preset actions are divided according to the danger level; the collected CSI data are preprocessed to construct the unsafe behavior data set with labels; the unsafe behavior data set is divided into a training set and a verification set; the training set is sent to a deep neural network for model training, and the verification set is used to test the model effect, and the trained model is used as an unsafe behavior detection model.
[0027] A computer-readable storage medium stores a computer program; when the computer program is executed by a processor, the campus safety monitoring method based on Wi-Fi perception and deep learning technology is implemented.
[0028] Compared with the prior art, the present invention has the following technical features:
[0029] This invention innovatively applies Wi-Fi sensing to campus security monitoring, which can better protect privacy than video monitoring and improve the utilization rate of wireless network resources. The UDP transmission protocol is used to separate perception from calculation, and perception is performed on low-performance front-end devices, and high-performance background calculations are integrated, thereby achieving low-cost equipment deployment and real-time background monitoring. The system is designed with a front-end user interface for easy use by operators and can intuitively display monitoring results. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] Figure 1 It is Wi-Fi aware of the wireless channel environment;
[0031] Figure 2 It is a system function flow chart;
[0032] Figure 3 It is the automatic connection script flow chart;
[0033] Figure 4 It is a flowchart of the training data acquisition module;
[0034] Figure 5 It is a flow chart of data preprocessing procedure;
[0035] Figure 6 This is the BLSTM deep network architecture diagram;
[0036] Figure 7 It is the flow chart of the behavior action perception module;
[0037] Figure 8 It is a functional structure diagram of the user interface. DETAILED DESCRIPTION
[0038] The present invention aims to use Wi-Fi signals as a medium for wireless perception, explore the relationship between human behavior and CSI changes, and design a human behavior monitoring solution using a typical deep learning model. Through a self-designed interactive operation interface, the functions of data forwarding, data preprocessing, model training, and prediction verification are integrated together to facilitate operation and result display, thereby realizing a complete campus environment safety monitoring solution; the solution analyzes Wi-Fi channel state information (CSI) and uses deep learning technology to build the relationship between human motion and state information, thereby realizing real-time and accurate recognition of human behavior, and finally outputting the probability of high-risk behavior; the solution is particularly suitable for security monitoring of hidden places on campus that are not suitable for video surveillance.
[0039] Referring to the accompanying drawings, the present invention provides a campus safety monitoring method based on Wi-Fi perception and deep learning technology, comprising the following steps:
[0040] Step 1: Build a wireless channel
[0041] Wireless channels can be modeled as multipath channels, where the signal propagates from the transmitter to the receiver through multiple paths, each path having different delays, attenuations, and phase changes.
[0042] Wireless transceiver devices are deployed in the monitored area, where the wireless signals between the wireless transceiver devices should cover the entire monitored area; when there is human activity in the monitored area, the CSI data generated by the wireless transceiver devices is obtained.
[0043] Specifically, Figure 1 As shown, devices equipped with Wi-Fi transceiver modules (such as wireless network cards), such as computer A and computer B, are connected to a router. Computer A generates a continuous stream of Wi-Fi data packets through the ping command and sends them to the router through the Wi-Fi transceiver module. After the router receives the Wi-Fi data packets through the Wi-Fi transceiver module, it obtains the channel state information CSI and returns it to computer A. Computer A continuously transmits the CSI data to computer B through the UDP protocol, so that the CSI data when there is human activity in the detection area can be obtained.
[0044] Step 2: Collect unsafe behavior training data
[0045] Personnel entering the monitored area V n ={v1,v2,...,v n}According to the preset action A n ={a1,a2,...,a n}Activities, changes in human behavior, and the placement of objects in the monitored area will affect the status information of the constructed wireless channel.
[0046] Among them, the preset actions are normal actions and unsafe behavior actions, that is, the people in the monitored area affect the environmental information in the constructed wireless channel by simulating normal or unsafe behaviors; the wireless transceiver equipment in the wireless channel is used to collect CSI data when the preset action occurs, and the data is received and processed according to the manually specified time length.
[0047] In this solution, preset actions are classified according to their riskiness, for example, they can be divided into safe level, minor level, dangerous level, high-risk level, etc. Corresponding preset actions are designed for each level to make the collection of training data more comprehensive and complete, and also facilitate subsequent monitoring and early warning.
[0048] Step 3: CSI data preprocessing
[0049] The receiving device computer B needs to pre-process the received CSI data to construct a labeled unsafe behavior dataset for subsequent deep learning model training and behavior discrimination detection. The specific process is as follows:
[0050] Step 1: Unpack the collected CSI data in .csi format and extract the amplitude information; label the CSI data accordingly according to the activity type, number of people and data collection time of the people in the monitored area.
[0051] Step 2. This solution establishes a 2-transmit 2-receive Wi-Fi detection environment. It is necessary to divide the CSI data obtained by different transmit and receive antenna combinations and frequency domain division into four equal parts. In the time domain, the CSI data needs to be cut at equal intervals to ensure that the data dimension is consistent, which is convenient for deep learning model training.
[0052] Step 3: Use Butterworth filter to perform low-pass filtering on CSI data. The Butterworth filter is suitable for eliminating high-frequency noise and retaining low-frequency components of human activities due to its flat response and stable performance in the passband. Principal Components Analysis (PCA) is performed on the filtered data to reduce the data dimension, thereby achieving the functions of data noise reduction and eliminating redundant information; the data after dimensionality reduction constitutes the unsafe behavior data set.
[0053] Step 4: Unsafe behavior detection model training
[0054] The unsafe behavior dataset is divided into a training set and a validation set. The labeled training set is sent to a bidirectional long short-term memory network (BLSTM) based on an attention mechanism for model training. The training rounds and parameters such as the learning rate can be customized to store the network model for each round of training. The validation set is used to test the model effect. The trained model is used as an unsafe behavior detection model.
[0055] Step 5: Behavior decision detection
[0056] In actual application, after the wireless channel is deployed in the monitored area, the wireless transceiver equipment is used to collect real-time CSI data for the monitored area. The collected CSI data is pre-processed by equal-interval cutting, low-pass filtering, and dimensionality reduction, and then input into the trained unsafe behavior detection model. The prediction result is output by the unsafe behavior detection model, and the prediction result is the probability of occurrence of various preset actions in the monitored area; according to the previous hazard classification scheme, when the prediction result reaches the preset hazard level or above, an early warning information is issued; the early warning information can be pushed to the campus security department in real time, and the numbers of relevant monitoring equipment in the monitored area and its vicinity can also be pushed to facilitate the campus security department to check and prevent the occurrence of dangerous behaviors.
[0057] This solution uses PyQt5 for interface design, adopts a master-slave structure, and designs a master control interface to call each sub-function. The system can automatically receive, forward and parse Wi-Fi CSI data, and perform preprocessing in the time and frequency domains. In training mode, the system uses the collected rich data set to train the prediction model; in prediction mode, real-time data is used to predict the results.
[0058] Example:
[0059] This solution is applied in Figure 1 In the Wi-Fi-aware wireless channel environment shown in the figure. The present invention establishes a 2-transmit 2-receive Wi-Fi detection environment, which consists of a computer A equipped with an AX210 network card, a computer B equipped with an AX210 network card, and a commercial router; Computer A responsible for executing data packet transmission with the router does not have high performance requirements, while Computer B used to receive UDP transmission data and train unsafe behavior detection models can be configured with a high-performance processor and can be located remotely. The operating system uses Ubuntu 20.04, the programming language is Python, and the CSI extraction uses the PicoScenes platform.
[0060] The area between computer A and the router is the human activity area, that is, the monitored area; Figure 2As shown in the figure, the system includes two modes, namely training mode and prediction mode. The router sends signals to sense human activities and collect CSI sensing data; computer A automatically connects to Wi-Fi through a script and receives CSI sensing information; through the UDP transmission protocol, computer A continuously sends CSI sensing information to computer B, and computer B pre-processes the received data and performs further analysis and identification.
[0061] Run the automatic connection script on computer A, such as Figure 3 As shown in the figure, the automatic connection of the specified Wi-Fi signal, the ping command to send packets to the target router at a specified interval, and the command to forward the received PicoScenes Rx frame to computer B through the UDP-Forwarder plug-in can be realized in sequence. Before computer A finishes running the script, it will continuously obtain the channel CSI information and send it to the specified receiving end.
[0062] Taking the perception of unsafe crowd gathering as an example, the scene is set to 2-3 people, and the behavior status of people is set to two types, namely unsafe gathering and crowd dispersion. The specific action labels are set to a0, a1, a2, and a3, where a0 represents the action of three people not showing unsafe behavior, a1 represents the action of three people showing unsafe behavior, a2 represents the action of two people not showing unsafe behavior, and a3 represents the action of two people showing unsafe behavior. Changes in human behavior and the placement of objects in the monitored area will affect the state information of the constructed wireless channel. In addition to this example, unsafe behavior actions and normal actions under other numbers and situations can also be collected for model training.
[0063] This solution includes two modes: training mode and prediction mode. First, you need to enter the training mode, use UDPRemoteLogger to monitor the port, and run it in the background for a specified time interval (assuming 60 seconds), then stop the UDPRemoteLogger process, and save the generated .csi file to the specified path; the length of the receiving time can be manually specified, and the length of time will affect the size of the data.
[0064] Preprocess the received CSI data:
[0065] First, get the latest .csi file from the folder where the .csi file is stored in the order of creation time, extract the CSI amplitude data, and segment it by time, volunteer and action identifier, and save it as multiple .npy files; the matrix data is divided into four parts according to the antenna combination, and the four antenna combinations are distinguished by "11, 12, 21, 22", and saved in the specified output directory; after the processing is completed, the system will ask the user whether to continue training. If you choose to continue, the system will re-extract the data; if you choose to exit, the system will terminate the training data acquisition. The flowchart of the training data acquisition module is as follows Figure 4 shown.
[0066] After collecting enough training data, enter the data preprocessing procedure. Figure 5 As shown, the preprocessor will load the .npy file from the specified directory, use the Butterworth filter for low-pass filtering, perform principal component analysis on the filtered data, reduce the data dimension, thereby achieving data noise reduction and eliminating redundant information. The filtered data of each file is then segmented by time period, each segment containing data of a specified dimension. Finally, the processed data is transposed and saved to the specified storage directory. Through data preprocessing, a number of experimental data are obtained in the format of {hours}_{minutes}_{volunteer number}_{action label}_{csi}_{number of antenna combinations}_{group number after equal division}.npy.
[0067] After the data set preprocessing is completed, the deep learning model training phase begins.
[0068] Long short-term memory network (LSTM) is a variant of recurrent neural network (RNN) and is widely used to process and predict time series data. However, the traditional LSTM network only analyzes CSI data in one direction, which means that the current hidden state only considers the past CSI information. The present invention adopts a bidirectional long short-term memory network (BLSTM) based on the attention mechanism to analyze the past and future information, thereby overcoming the long-term dependency problem. Figure 6 As shown in the figure, the BLSTM deep network architecture contains a forward layer and a backward layer to extract information from two directions. The data set is divided into 80% training set and 20% validation set, and the network training rounds and initial learning rate are set manually. The cross entropy loss function and Adam optimizer are used in the training process, the learning rate is adjusted dynamically, and the model is saved regularly.
[0069] After obtaining the trained network model, the system can switch to prediction mode.
[0070] In this mode, computer A continues to obtain CSI data and forward it, and computer B obtains test set data by receiving data; the collection of test set data can be completed in a short time (assuming 5 seconds), and by obtaining channel status information in a short time, real-time monitoring of the environmental status can be achieved. After the data preprocessing is completed, the data is directly input into the trained network model for result prediction, outputting the probability of occurrence of various actions, and issuing warnings for high-risk behaviors.
[0071] The present invention designs the interface based on PyQt5, integrates the above functions, and converts the complex Linux commands into buttons and input boxes that are easy to operate. The control interface adopts a general-specific structure, calling each sub-function through a master control interface. The functional structure of the user interface is as follows: Figure 8 The usage process is as follows:
[0072] Step 1: First run the training data acquisition module. The user can enter the expected collection time for each action and complete the data labeling work.
[0073] Step 2: Run the data processing and model training module. In this module, data preprocessing and model training are performed. The user needs to set the deep learning rounds and initial learning rate. The accuracy during the training process will be displayed in the specified position of the interface, and the user can stop the deep learning process at any time.
[0074] Step 3: When a satisfactory model is obtained, the behavior perception module can be run. This module integrates data collection, label marking, and data preprocessing functions, predicts the data obtained within a specified time interval, and finally displays the results at a specified location on the interface, and issues an alarm for high-risk behaviors.
[0075] After software testing, the user interface can automatically receive, forward, parse and label Wi-Fi CSI data, and perform preprocessing in the time and frequency domains. In training mode, the system uses the collected rich data set to train the prediction model; in prediction mode, real-time data is used to predict results.
[0076] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.
Claims
1. A campus security monitoring method based on Wi-Fi sensing and deep learning technology, characterized in that: include: Deploy wireless transceiver equipment in the monitored area, and obtain CSI data generated by the wireless transceiver equipment when there is human activity in the monitored area; Arrange personnel to perform activities in the monitored area according to preset actions, wherein the preset actions are normal actions and unsafe behavior actions; The wireless transceiver in the wireless channel collects CSI data when the preset action occurs, and receives and processes the data according to the manually specified time length; Classifying the preset actions according to risk levels; Preprocess the collected CSI data to construct a labeled unsafe behavior dataset; The unsafe behavior dataset is divided into a training set and a validation set. The labeled training set is fed into a deep neural network for model training, and the validation set is used to test the model effect. The trained model is used as an unsafe behavior detection model. The unsafe behavior detection model is used to predict the CSI data collected by the wireless transceiver equipment deployed in the monitored area and output the prediction results.
2. The campus safety monitoring method based on Wi-Fi perception and deep learning technology according to claim 1 is characterized in that: The unsafe behavior detection model is used to predict the CSI data collected by the wireless transceiver equipment deployed in the monitored area, including: After deploying wireless channels in the monitored area, wireless transceiver equipment is used to collect real-time CSI data for the monitored area. The collected CSI data is pre-processed by equal-interval cutting, low-pass filtering, and dimensionality reduction, and then input into the trained unsafe behavior detection model. The unsafe behavior detection model outputs the prediction result, which is the probability of occurrence of various preset actions in the monitored area.
3. The campus safety monitoring method based on Wi-Fi perception and deep learning technology according to claim 1 is characterized in that: A classification scheme is performed based on the danger level, and when the prediction result reaches a preset danger level or above, an early warning message is issued.
4. The campus safety monitoring method based on Wi-Fi perception and deep learning technology according to claim 1 is characterized in that: The warning information is pushed to the campus security department, and the numbers of relevant monitoring equipment in the monitored area and its vicinity are also pushed.
5. The campus safety monitoring method based on Wi-Fi perception and deep learning technology according to claim 1 is characterized in that: The wireless transceiver device includes device A, device B and a router, and device A and device B are connected to the router; Device A generates a Wi-Fi data packet and sends it to the router through the Wi-Fi transceiver module. After receiving the Wi-Fi data packet through the Wi-Fi transceiver module, the router obtains the channel state information CSI and returns it to device A. Device A continuously transmits the CSI data to device B, so as to obtain the CSI data when there is human activity in the detection area; the device B is used to pre-process the CSI data and train the unsafe behavior detection model.
6. The campus safety monitoring method based on Wi-Fi perception and deep learning technology according to claim 1 is characterized in that: The preprocessing of the collected CSI data to construct an unsafe behavior data set includes: Step 1: Unpack the collected CSI data in .csi format and extract the amplitude information; label the CSI data accordingly according to the activity type, number of people and data collection time in the monitored area; Step 2: Segment the CSI data obtained by different transceiver antenna combinations and frequency domain division of the wireless transceiver device. In the time domain, the CSI data needs to be cut at equal intervals to ensure that the data dimension is consistent. Step 3: Use a Butterworth filter to low-pass filter the CSI data, and perform principal component analysis on the filtered data to reduce the data dimension; the data after dimensionality reduction constitutes an unsafe behavior data set.
7. The campus safety monitoring method based on Wi-Fi perception and deep learning technology according to claim 1 is characterized in that: The deep neural network adopts a bidirectional long short-term memory network based on the attention mechanism.
8. A campus security monitoring system based on Wi-Fi perception and deep learning technology, characterized in that: It includes device A and router deployed in the monitored area and device B located at a remote end, where: Device A and the router have Wi-Fi transceiver modules. Device A is used to send Wi-Fi data to the monitored area, and the router is used to receive Wi-Fi data and obtain corresponding CSI data, and then send it to device B according to the preset length; Device B is used to construct an unsafe behavior data set based on the collected CSI data, and use the unsafe behavior data set and deep neural network training to obtain an unsafe behavior detection model. The unsafe behavior detection model is used to predict human behavior in the monitored area; Among them, personnel are arranged to perform activities according to preset actions in the monitored area, and the preset actions are normal actions and unsafe behavior actions; device B is used to collect CSI data when the preset actions occur, and the data is received and processed according to the manually specified time length; the preset actions are divided according to the danger level; the collected CSI data are preprocessed to construct the unsafe behavior data set with labels; the unsafe behavior data set is divided into a training set and a verification set; the training set is sent to a deep neural network for model training, and the verification set is used to test the model effect, and the trained model is used as an unsafe behavior detection model.
9. A computer-readable storage medium, wherein a computer program is stored in the medium; characterized in that: When the computer program is executed by a processor, the campus safety monitoring method based on Wi-Fi perception and deep learning technology according to any one of claims 1 to 7 is implemented.
Citation Information
Patent Citations
Channel state information human activity identification method and system based on deep learning
CN111797804A
Internet of Things home intelligent protection system and method based on WiFi signal perception
CN113965362A
Human body behavior data labeling and recognition system and method based on cross-modal migration
CN114781431A
Real-time monitoring and early warning system and method for indoor personnel activity of solitary group
CN116416754A
