Nursing home intelligent behavior recognition and anomaly detection method based on deep learning

Through deep learning technology combined with CNN and RNN, the video streams of nursing homes are analyzed, and the behavior of the elderly is identified and abnormal, which solves the problems of low monitoring efficiency and strong environmental dependence in the existing technology, and achieves efficient and accurate safety monitoring of the elderly and Emergency response.

CN119007278AInactive Publication Date: 2024-11-22FUSHOUKANG (SHANGHAI) FAMILY SERVICES CO LTD +2

Patent Information

Application Number
CN202410687493.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-05-30
Publication Date
2024-11-22
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing technology has problems such as low efficiency, strong environmental dependence, limited action recognition ability and low real-time performance in nursing homes, making it difficult to achieve all-weather and all-round safety monitoring and timely abnormal behavior detection.

Method used

A deep learning-based method is adopted, combining convolutional neural network (CNN) and recurrent neural network (RNN), feature extraction and timing analysis of monitoring video streams is performed, behavior patterns of elderly people are identified and abnormal behaviors are detected, and the alarm mechanism is triggered.

Benefits of technology

It realizes all-weather and all-round automated monitoring, improves abnormal detection efficiency and identification accuracy, reduces safety hazards and operating costs, promptly detects and responds to emergencies, and provides more effective safety guarantees for the elderly.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119007278A_ABST
    Figure CN119007278A_ABST
Patent Text Reader

Abstract

The invention provides a nursing home intelligent behavior recognition and anomaly detection method based on deep learning, and the method comprises the steps: carrying out the feature extraction of each frame of image in a monitoring video through a CNN, and recognizing the posture and motion of an old person; and performing time sequence analysis on the extracted features by using RNN, and establishing a behavior mode of the old. According to the method, the monitoring video stream in the nursing home is deeply analyzed by using the deep learning technology, especially the powerful image processing and time sequence analysis capabilities of CNN and RNN. By learning a large number of normal behavior modes, the system can accurately identify daily activities of the elderly and immediately give an alarm when abnormal behaviors such as falling down and long-time standing are detected, so that the safety of the elderly is guaranteed in real time; the technical problems that in the prior art, the labor efficiency is low, the model environment dependence is high, the action recognition accuracy is poor, and early warning is not timely are effectively solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of smart safety systems for the elderly, and in particular to a deep learning-based intelligent behavior recognition and anomaly detection method for nursing homes. Background Art

[0002] With the increasing aging of society, nursing homes, as important places for elderly care, have received high attention from all walks of life for their safety and quality of care. In the daily management of nursing homes, real-time monitoring of the elderly's behavior and timely identification of abnormal behavior are key links in ensuring the safety of the elderly. Traditional detection methods rely on manual observation, which is not only inefficient, but also difficult to achieve all-weather and all-round coverage, posing a major safety hazard; at the same time, there are also problems such as insufficient personnel, easy missed alarms and false alarms, and difficulty in accurately identifying the behavior patterns of the elderly. In addition, although some existing monitoring systems can provide a certain degree of monitoring, they are insufficient in recognition accuracy and real-time performance, and are difficult to meet the needs of nursing homes for real-time, efficient, and accurate monitoring. The main defects include the following aspects:

[0003] 1. Strong environmental dependence: Existing technologies are often limited by their operating environment, especially with high requirements for lighting conditions and background interference. Once the environmental conditions are not met, the system's recognition ability will drop significantly.

[0004] 2. Limited action recognition capabilities: Existing anomaly detection methods are mainly based on statistics or rule learning, which makes it difficult to accurately detect complex behavior patterns. For complex behavior patterns, especially abnormal behaviors in emergency situations, such as falls and fainting, the recognition accuracy of existing systems is not high, resulting in a high missed alarm rate.

[0005] 3. Low real-time performance and lack of sufficient early warning mechanisms: Unable to detect and report dangers in a timely manner. Existing technologies cannot accurately identify the specific behavior patterns of the elderly in a timely manner, have limited ability to detect abnormal behaviors in complex scenarios, and have slow response speeds, which cannot meet the nursing home's needs for rapid response to emergencies. Summary of the invention

[0006] The purpose of this invention is to propose a method for intelligent behavior recognition and anomaly detection in nursing homes based on deep learning. This method uses deep learning technology, combined with the powerful image processing and timing analysis capabilities of convolutional neural networks (CNN) and recurrent neural networks (RNN), to conduct in-depth analysis of the surveillance video stream in the nursing home, thereby achieving real-time protection of the safety of the elderly.

[0007] To achieve the above objectives, the present invention proposes a method for intelligent behavior recognition and anomaly detection in nursing homes based on deep learning. CNN is used to extract features from each frame of the surveillance video to identify the posture and movements of the elderly. RNN is used to perform time series analysis on the extracted features to establish the behavior pattern of the elderly.

[0008] Further, the following steps are included:

[0009] Step 1: Collect video data;

[0010] Step 2: Preprocess the received video stream;

[0011] Step 3: The pre-processed video is input to the feature extraction module, and a convolutional neural network (CNN) is used to perform in-depth analysis on each frame of the image to identify the key parts and action features of the human body;

[0012] Step 4: Send the extracted feature information to the behavior analysis module, which uses a recurrent neural network (RNN) to analyze the behavior patterns of the elderly based on time series information;

[0013] Step 5: Based on the behavior pattern analysis, the system performs abnormal behavior detection;

[0014] Step 6: Detecting abnormal behavior triggers an alarm mechanism;

[0015] Step 7: Conduct user interaction and feedback.

[0016] Furthermore, in step 2, the preprocessing includes performing frame extraction, noise reduction, color correction and contrast enhancement processing on the video.

[0017] Furthermore, in step 3, the method of CNN feature extraction is as follows:

[0018] Preprocessing: The input video frames are preprocessed, including normalization and resizing, data augmentation, to adapt to the input requirements of the CNN model;

[0019] Convolutional layer: Use multiple convolutional layers to extract image features; each convolutional layer uses a set of learnable filters to slide over the input image to extract features;

[0020] The mathematical representation of the convolution operation is:

[0021]

[0022] in, is the feature map response of the (l)th layer at position (i, j);

[0023] σ represents the activation function, which is a nonlinear function. In convolutional neural networks, common activation functions include ReLU, Sigmoid, etc. It is applied to the result of the convolution operation here to introduce nonlinearity to the network and help the network learn complex patterns.

[0024] ∑ m ∑ n Indicates the accumulation of all 9m,n). Here, (m) and (n) are the indices of the convolution kernel moving in the vertical and horizontal directions, and their range is determined by the size of the convolution kernel;

[0025] is the weight of the convolution kernel or filter at position 9m,n) in the (l)th convolution layer. The weight is a learning parameter that is optimized through the training process;

[0026] I (i+m)(j+n) Represents the pixel value at position (i+m,j+n) of the input image or the feature map of the previous layer. When performing a convolution operation, the weight at each position of the convolution kernel is multiplied by the pixel value at the corresponding position in the input image;

[0027] b (l) : represents the bias term of the (l)th convolutional layer, which is a learnable parameter added after the convolution operation to increase the expressive power of the network;

[0028] Pooling layer: Pooling layers are added after consecutive convolutional layers to reduce feature dimensions and increase the robustness of the model;

[0029] Fully connected layer: The output of the convolution and pooling layers is flattened and passed through one or more fully connected layers to integrate the global features and output the final feature representation, including the Dropout layer to prevent overfitting.

[0030] Furthermore, in step 4, the method of using RNN to perform time series analysis on the feature sequence extracted from CNN is as follows:

[0031] Feature sequence input: The feature sequence extracted by the CNN module is used as input, and the sequence reflects the changes of the elderly's actions in the video over time;

[0032] Recurrent layer: Use multiple layers of stacked recurrent layers to process feature sequences and capture temporal dependencies. Each layer of the long short-term memory network is equipped with a Dropout mechanism to reduce overfitting. The update of the LSTM unit follows the following formula:

[0033] Forget gate: f t =σ(W f ·[h t-1 ,x t ]+b f )

[0034] Input gate: i t =σ(W i ·[h t-1 ,x t ]+b i )

[0035] Output gate: o t =σ(W o ·[h t-1 ,x t ]+b o )

[0036] New Memory Cells:

[0037] Final Memory Cell:

[0038] Final hidden state: h t =o t *tanh(C t )

[0039] Where: x t is the input feature vector at time t; h t is the output state at time t; C t is the state of the memory cell at time t; * represents element-by-element multiplication; σ is the sigmoid activation function; W and b are weight and bias parameters, respectively;

[0040] Behavior recognition: The output of the last recurrent layer can be processed through one or more fully connected layers to make the final behavior category prediction; use the softmax function as the last layer to obtain the probability distribution of different behavior categories; the hidden state of each time step can be used to predict the behavior of the current time step; this can be mapped to the output space using another weight matrix V:

[0041] y t =softmax(Vh t +c)

[0042] Where: y t is the output vector of the current time step; c is the bias term of the output layer; the function softmax is the activation function for multi-classification tasks, which is used to convert the output into a probability distribution.

[0043] Furthermore, in step 5, anomaly detection uses the validation set to determine the optimal threshold and determines whether the behavior pattern is abnormal based on the predefined threshold. If the probability of the detected behavior category exceeds the threshold and belongs to the abnormal category, an alarm is triggered.

[0044] Furthermore, performance indicators such as AUC-ROC curve are used to evaluate the accuracy of anomaly detection:

[0045] Define anomaly score: based on the output h of RNN t , and the degree of difference from the normal behavior dataset, define an anomaly score to measure whether the behavior at each time step is abnormal; the anomaly score is calculated by comparing the statistical measurement value with the typical pattern of the abnormal dataset;

[0046] Set a threshold. When the anomaly score exceeds this threshold, the behavior at this time step is considered abnormal.

[0047] Compared with the prior art, the advantages of the present invention are:

[0048] 1. After the video detection system of the present invention is introduced, the limitations of manual observation can be effectively overcome, all-weather and all-round automated monitoring and detection can be realized, efficiency can be improved, blind spots can be eliminated, and potential safety hazards can be reduced; the problem of low efficiency of manual observation can be solved.

[0049] 2. The present invention improves the efficiency of abnormal detection and reduces operating costs: After adopting this system, the nursing home has improved the efficiency of detecting abnormal activities of the elderly and reduced operating costs. Automated video analysis technology replaces the limited and costly manual monitoring and detection, ensuring real-time monitoring around the clock. This transformation not only reduces the monitoring omissions that may occur due to insufficient manpower or fatigue, but also greatly reduces labor costs, reduces nursing pressure, and improves nursing services.

[0050] 2. Through the powerful feature extraction capability of CNN, the system of the present invention can work stably under different lighting and complex backgrounds, significantly reducing the dependence on environmental conditions, improving the accuracy and reliability of recognition, and solving the problem of environmental dependence.

[0051] 3. By using RNN to perform time-series analysis on behavior patterns, the present invention can accurately identify complex abnormal behaviors and issue early warnings in a timely manner, solving the problem of insufficient recognition ability of the prior art in emergency situations, providing more effective protection for the safety of the elderly, and improving the accuracy of action recognition.

[0052] 4. The present invention utilizes convolutional neural network (CNN) and recurrent neural network (RNN) to perform real-time analysis on the surveillance video stream in the nursing home, combined with terminal warning, so as to realize timely detection of abnormal behavior, rapid response and rescue measures, bringing real-time and warning capabilities to the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a flow chart of a method for intelligent behavior recognition and anomaly detection in a nursing home based on deep learning in an embodiment of the present invention;

[0054] Figure 2 Schematic diagram of the application of the system in the embodiment of the present invention. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be further described below.

[0056] like Figure 1 As shown, the present invention proposes a method for intelligent behavior recognition and anomaly detection in nursing homes based on deep learning. The core of this method is to use convolutional neural network CNN for efficient image processing and feature extraction, and recurrent neural network RNN ​​for accurate temporal behavior analysis. By combining these two deep learning technologies, the system can perform real-time analysis of the surveillance video stream in the nursing home, thereby effectively identifying the daily behavior patterns of the elderly and detecting abnormal behaviors in a timely manner.

[0057] First, the system uses CNN to extract features from each frame of the surveillance video and identify the posture and movements of the elderly. CNN's multi-layer convolution and pooling operations can effectively capture key features in the image, including human body contours, movement postures, etc., and are highly robust to lighting changes and background interference, solving the problem of existing technologies' strong dependence on environmental conditions.

[0058] Next, the system uses RNN to perform time-series analysis on the extracted features and establish the behavior pattern of the elderly. RNN can process sequence data and capture the dynamic changes of behavior through the transmission of internal states, thereby identifying continuous behavior patterns. This step solves the limitations of existing technologies in motion recognition capabilities, especially in the recognition of abnormal behaviors in emergency situations, such as falls and fainting, and improves the recognition accuracy and early warning capabilities of the system.

[0059] Specifically, Figure 1 As shown, the following steps are included:

[0060] S1: Video data acquisition: The system first collects real-time monitoring video streams of the elderly’s daily activities through high-definition cameras installed at key locations in the nursing home. The video data is transmitted to the central processing unit via a wired or wireless network to provide raw data for subsequent analysis and processing.

[0061] S2. Video preprocessing: The received video stream first enters the preprocessing module. At this stage, the system performs frame extraction, noise reduction, color correction, contrast enhancement and other processing on the video to improve image quality and reduce interference factors in subsequent analysis.

[0062] S3. Feature extraction: The preprocessed video frames are then input into the feature extraction module. This module uses a convolutional neural network (CNN) to perform in-depth analysis on each frame to identify key parts of the human body and motion features. Through multiple convolutional and pooling layers, CNN automatically learns and extracts spatial features in the image, such as contours, shapes, textures, etc. These features are used to describe the posture, movements, and environmental context of the elderly in the scene, laying the foundation for behavior recognition.

[0063] The method of CNN feature extraction is as follows:

[0064] Preprocessing: The input video frames are preprocessed, including normalization and resizing, data augmentation, to adapt to the input requirements of the CNN model;

[0065] Convolutional layer: Use multiple convolutional layers to extract image features; each convolutional layer uses a set of learnable filters to slide over the input image to extract features;

[0066] The mathematical representation of the convolution operation is:

[0067]

[0068] in, is the feature map response of the (l)th layer at position (i, j);

[0069] σ represents the activation function, which is a nonlinear function. In convolutional neural networks, common activation functions include ReLU, Sigmoid, etc. It is applied to the result of the convolution operation here to introduce nonlinearity to the network and help the network learn complex patterns.

[0070] ∑ m ∑ n Indicates the accumulation of all (m,n). Here, (m) and (n) are the indices of the convolution kernel moving in the vertical and horizontal directions, and their range is determined by the size of the convolution kernel;

[0071] is the convolution kernel of the (l)th convolutional layer or the weight of the filter at position (m,n). The weight is a learning parameter that is optimized through the training process;

[0072] I (i+m)(j+n) Represents the pixel value at position (i+m,j+n) of the input image or the feature map of the previous layer. When performing a convolution operation, the weight at each position of the convolution kernel is multiplied by the pixel value at the corresponding position in the input image;

[0073] b (l) : represents the bias term of the (l)th convolutional layer, which is a learnable parameter added after the convolution operation to increase the expressive power of the network;

[0074] Pooling layer: Pooling layers are added after consecutive convolutional layers to reduce feature dimensions and increase the robustness of the model;

[0075] Fully connected layer: The output of the convolution and pooling layers is flattened and passed through one or more fully connected layers to integrate the global features and output the final feature representation, including the Dropout layer to prevent overfitting.

[0076] S4. Behavior pattern analysis: The extracted features are then fed into the behavior analysis module. This module uses a recurrent neural network (RNN) to analyze the behavior patterns of the elderly based on time series information. RNN can process the temporal relationship between video frames and identify continuous behavioral actions such as walking, standing, sitting, etc.

[0077] S5. Abnormal behavior detection: Based on the behavior pattern analysis, the system enters the abnormal behavior detection stage. In this stage, the system compares the current behavior pattern with the pre-trained normal behavior model. By setting thresholds and using classification algorithms, the system can determine whether the current behavior deviates from the normal pattern, thereby identifying possible abnormal behaviors such as falling, long periods of inactivity, etc.

[0078] The method of using RNN to perform time series analysis on the feature sequence extracted from CNN is as follows:

[0079] Feature sequence input: The feature sequence extracted by the CNN module is used as input, and the sequence reflects the changes of the elderly's actions in the video over time;

[0080] Recurrent layer: Use multiple layers of stacked recurrent layers to process feature sequences and capture temporal dependencies. Each layer of the long short-term memory network is equipped with a Dropout mechanism to reduce overfitting. The update of the LSTM unit follows the following formula:

[0081] Forget gate: f t =σ(W f ·[h t-1 ,x t ]+b f )

[0082] Input gate: i t =σ(W i ·[h t-1 ,x t ]+b i )

[0083] Output gate: o t =σ(W o ·[h t-1 ,x t ]+b o )

[0084] New Memory Cells:

[0085] Final Memory Cell:

[0086] Final hidden state: h t =o t *tanh(C t )

[0087] Where: x t is the input feature vector at time t; h t is the output state at time t; C t is the state of the memory cell at time t; * represents element-by-element multiplication; σ is the sigmoid activation function; W and b are weight and bias parameters, respectively;

[0088] Behavior recognition: The output of the last recurrent layer can be processed through one or more fully connected layers to make the final behavior category prediction; use the softmax function as the last layer to obtain the probability distribution of different behavior categories; the hidden state of each time step can be used to predict the behavior of the current time step; this can be mapped to the output space using another weight matrix V:

[0089] y t =softmax(Vh t +c)

[0090] Where: y t is the output vector of the current time step; c is the bias term of the output layer; the function softmax is the activation function for multi-classification tasks, which is used to convert the output into a probability distribution.

[0091] S6. Alarm and response: Once the system detects abnormal behavior, the alarm mechanism is triggered immediately. The alarm information is displayed through the system interface and can be sent to the caregiver's mobile device for quick response. At the same time, the system automatically records the time, location and related video clips of the abnormal behavior for subsequent analysis and processing by the caregiver.

[0092] S7. User interaction and feedback: Figure 2 As shown in Figure 1, the system provides a user interaction interface that allows caregivers and managers to view real-time monitoring images, historical event records, and statistical reports on the terminal. Managers can also provide information to the system through a feedback mechanism, such as correcting incorrect recognition results, and these feedbacks will be used to improve the system.

[0093] The above is only a preferred embodiment of the present invention and does not limit the present invention in any way. Any technician in the relevant technical field, without departing from the scope of the technical solution of the present invention, makes any form of equivalent replacement or modification to the technical solution and technical content disclosed in the present invention, which does not depart from the content of the technical solution of the present invention and still falls within the protection scope of the present invention.

Claims

1. A method for intelligent behavior recognition and anomaly detection in nursing homes based on deep learning, It is characterized in that CNN is used to extract features from each frame of the surveillance video to identify the posture and movements of the elderly. RNN is used to perform time series analysis on the extracted features to establish the behavioral pattern of the elderly.

2. According to the deep learning-based intelligent behavior recognition and anomaly detection method for nursing homes according to claim 1, It is characterized in that The following steps are involved: Step 1: Collect video data; Step 2: Preprocess the received video stream; Step 3: The pre-processed video is input to the feature extraction module, and a convolutional neural network (CNN) is used to perform in-depth analysis on each frame of the image to identify the key parts and action features of the human body; Step 4: Send the extracted feature information to the behavior analysis module, which uses a recurrent neural network (RNN) to analyze the behavior patterns of the elderly based on time series information; Step 5: Based on the behavior pattern analysis, the system performs abnormal behavior detection; Step 6: Detecting abnormal behavior triggers an alarm mechanism; Step 7: Conduct user interaction and feedback.

3. According to claim 2, the deep learning-based intelligent behavior recognition and anomaly detection method for nursing homes, It is characterized in that In step 2, the preprocessing includes performing frame extraction, noise reduction, color correction and contrast enhancement on the video.

4. According to claim 2, the deep learning-based intelligent behavior recognition and anomaly detection method for nursing homes, It is characterized in that In step 3, the method of CNN feature extraction is as follows: Preprocessing: The input video frames are preprocessed, including normalization and resizing, data augmentation, to adapt to the input requirements of the CNN model; Convolutional layer: Use multiple convolutional layers to extract image features; each convolutional layer uses a set of learnable filters to slide over the input image to extract features; The mathematical representation of the convolution operation is: in, is the feature map response of the (l)th layer at position (i, j); σ represents the activation function, which is a nonlinear function. In convolutional neural networks, common activation functions include ReLU, Sigmoid, etc. It is applied to the result of the convolution operation here to introduce nonlinearity to the network and help the network learn complex patterns. ∑ m ∑ n Indicates the accumulation of all (m,n). Here, (m) and (n) are the indices of the convolution kernel moving in the vertical and horizontal directions, and their range is determined by the size of the convolution kernel; is the convolution kernel of the (l)th convolutional layer or the weight of the filter at position (m,n). The weight is a learning parameter that is optimized through the training process; I (i+m)(j+n) Represents the pixel value at position (i+m,j+n) of the input image or the feature map of the previous layer. When performing a convolution operation, the weight at each position of the convolution kernel is multiplied by the pixel value at the corresponding position in the input image; b (l) : represents the bias term of the (l)th convolutional layer, which is a learnable parameter added after the convolution operation to increase the expressive power of the network; Pooling layer: Pooling layers are added after consecutive convolutional layers to reduce feature dimensions and increase the robustness of the model; Fully connected layer: The output of the convolution and pooling layers is flattened and passed through one or more fully connected layers to integrate the global features and output the final feature representation, including the Dropout layer to prevent overfitting.

5. According to claim 2, the deep learning-based intelligent behavior recognition and anomaly detection method for nursing homes, It is characterized in that In step 4, the method of using RNN to perform time series analysis on the feature sequence extracted from CNN is as follows: Feature sequence input: The feature sequence extracted by the CNN module is used as input, and the sequence reflects the changes of the elderly's actions in the video over time; Recurrent layer: Use multiple layers of stacked recurrent layers to process feature sequences and capture temporal dependencies. Each layer of the long short-term memory network is equipped with a Dropout mechanism to reduce overfitting. The update of the LSTM unit follows the following formula: Forget gate: f t =σ(W f ·[h t-1 ,x t ]+b f ) Input gate: i t =σ(W i ·[h t-1 ,x t ]+b i ) Output gate: o t =σ(W o ·[h t-1 ,x t ]·+b o ) New Memory Cells: Final Memory Cell: Final hidden state: h t =o t *tanh(C t ) Where: x t is the input feature vector at time t; h t is the output state at time t; C t is the state of the memory cell at time t; * represents element-by-element multiplication; σ is the sigmoid activation function; W and b are weight and bias parameters, respectively; Behavior recognition: The output of the last recurrent layer can be processed through one or more fully connected layers to make the final behavior category prediction; use the softmax function as the last layer to obtain the probability distribution of different behavior categories; the hidden state of each time step can be used to predict the behavior of the current time step; this can be mapped to the output space using another weight matrix V: y t =softmax(Vh t +c) Where: y t is the output vector of the current time step; c is the bias term of the output layer; the function softmax is the activation function for multi-classification tasks, which is used to convert the output into a probability distribution.

6. The method for intelligent behavior recognition and anomaly detection in nursing homes based on deep learning according to claim 2, It is characterized in that In step 5, anomaly detection uses the validation set to determine the optimal threshold and determines whether the behavior pattern is abnormal based on the predefined threshold. If the probability of the detected behavior category exceeds the threshold and belongs to the abnormal category, an alarm is triggered.

7. According to claim 5, the deep learning-based intelligent behavior recognition and anomaly detection method for nursing homes, It is characterized in that Use performance metrics such as AUC-ROC curve to evaluate the accuracy of anomaly detection: Define anomaly score: based on the output h of RNN t , and the degree of difference from the normal behavior dataset, define an anomaly score to measure whether the behavior at each time step is abnormal; the anomaly score is calculated by comparing the statistical measurement value with the typical pattern of the abnormal dataset; Set a threshold. When the anomaly score exceeds this threshold, the behavior at this time step is considered abnormal.

Citation Information

Patent Citations

  • Method for identifying fall behavior of old people based on deep learning

    CN108549841A

  • Artificial intelligence CNN and LSTM neural network dynamic identification system

    CN110110707A

  • Real-time old man falling detection method based on anomaly detection

    CN114842397A

Cited By

  • Computer image recognition method

    CN119723646A

  • Human body tumble detection method based on uncooled thermal infrared imager and application of human body tumble detection method

    CN122416536A