Old people risk behavior identification and monitoring method and system based on visual identification

Abstract

The invention belongs to the technical field of behavior monitoring, and discloses an old people risk behavior identification and monitoring method and system based on visual identification, and the method comprises the steps: obtaining historical video data for monitoring the daily behaviors of old people; labeling and preprocessing the historical video data to obtain behavior video data with labels; taking the behavior video data with the label as a training set, training a pre-trained deep learning model so as to adjust model parameters of the pre-trained deep learning model, obtaining a trained risk behavior recognition model, and inputting the obtained real-time video data into the trained risk behavior recognition model for behavior recognition. And obtaining a behavior identification result, and carrying out risk early warning based on the identification result. According to the invention, a personalized recognition model can be provided for old people, the limitation of recognition is reduced, and the generalization ability of the model is improved; meanwhile, dangerous actions can be found in time.

Description

Technical Field

[0001] The present invention belongs to the technical field of behavior monitoring, and specifically relates to a method and system for identifying and monitoring risky behaviors of the elderly based on visual recognition. Background Art

[0002] As they age, the elderly may develop various health problems, including musculoskeletal system diseases, physical decline, and cognitive impairment, which may make them more susceptible to risky situations such as falls, fatigue, and weakness. Therefore, research on the identification and monitoring of risky behaviors in the elderly can help prevent risky behaviors in advance and protect the health and safety of the elderly.

[0003] However, existing identification methods only involve the identification of a small number of high-risk behaviors and have certain limitations in behavior identification, such as: Patent application number CN202310560220.9 discloses a radar-based continuous motion detection and behavior recognition method for smart elderly care applications. This method only recognizes a few specific continuous motion sequences, most of which are related to falls and seeking help after falls. The invention patent application number CN202110242379.7 discloses a monitoring system that uses a neural network to detect abnormal behavior in children and the elderly. The system also focuses on identifying whether the wearer has fallen, and the identification object is also limited by the wearer's wearable device; The invention patent with application number CN201911071994.5 uses digital twin technology to visualize the various actions and activity trajectories of the elderly, such as falls, long periods of sitting, long periods of squatting, and holding on to the wall, in the physical environment of community and home-based elderly care. It also issues alarms and warnings for abnormal high-risk states of the elderly. However, the false alarm rate is 4.3%, which shows that the system still has certain errors in distinguishing normal behavior from high-risk behavior.

[0004] The existing technology has problems such as high recognition limitations and high false alarm rate. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and system for identifying and monitoring risky behaviors of the elderly based on visual recognition, so as to solve the problems of high recognition limitations and high false alarm rate in the existing technology.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions: In a first aspect, the present invention provides a method for identifying and monitoring risky behaviors of the elderly based on visual recognition, the method comprising: Obtain historical video data to monitor the daily behavior of the elderly; Label and preprocess historical video data to obtain labeled behavioral video data; The pre-trained deep learning model is trained based on the labeled behavioral video data as a training set to adjust the model parameters of the pre-trained deep learning model to obtain a trained risk behavior recognition model; The acquired real-time video data is input into the trained risk behavior recognition model for behavior recognition, to obtain a behavior recognition result, and a risk warning is issued based on the recognition result.

[0007] Preferably, the deep learning model is pre-trained using a preset large-scale dataset.

[0008] Preferably, the acquired real-time video data is input into the trained risk behavior recognition model for behavior recognition, a behavior recognition result is obtained, and a risk warning is issued based on the recognition result, including: Acquire real-time video data for monitoring the daily behavior of the elderly, and pre-process the real-time video data to obtain pre-processed real-time video data; The pre-processed real-time video data is input into the trained risk behavior recognition model for behavior recognition to obtain the behavior recognition results; Based on the risk quantification table, the behavior recognition results are graded to obtain the risk level of the behavior recognition results; Generate early warning signals based on the risk level of behavior recognition results; Provide visual display of early warning signals.

[0009] Preferably, the deep learning model is a 3D convolutional neural network or a two-stream convolutional neural network.

[0010] Preferably, the historical video data is labeled and preprocessed to obtain labeled behavioral video data, including: Annotate historical video data based on a preset expert database to obtain annotation content, wherein the annotation content includes at least: behavior category, behavior start time, behavior end time, and behavior key features; Preprocessing the annotated historical video data to obtain processed historical video data; Construct labels for historical video data based on the annotation content; Based on the labels of the historical video data and the processed historical video data, the behavioral video data with labels is constructed.

[0011] Preferably, preprocessing the annotated historical video data to obtain processed historical video data includes: Extracting several image frames of the annotated historical video data; Perform light intensity analysis on each image frame to obtain light intensity detection results; Performing ambient light compensation on each image frame according to the light intensity detection result to obtain a number of compensated image frames; Performing color correction on each compensated image frame to obtain a plurality of corrected image frames; Perform background noise filtering on each corrected image frame to obtain a number of processed image frames; The processed image frames are fused with the initial image frames of the historical video data to obtain a number of fused image frames, and the video data composed of the fused image frames is used as the processed historical video data.

[0012] Preferably, the algorithm for filtering the background noise of each corrected image frame is one or more of mean filtering, median filtering, frequency domain filtering and low-pass filtering.

[0013] Preferably, the risk quantification table includes multiple assessment parts, posture codes of each assessment part, loads of the assessment parts, and load codes; the assessment parts include: upper limbs, lower limbs, and back.

[0014] In a second aspect, the present invention provides a system for identifying and monitoring risky behaviors of the elderly based on visual recognition, which is used to implement the above-mentioned method for identifying and monitoring risky behaviors of the elderly based on visual recognition. The system includes: A data acquisition module is used to obtain historical video data for monitoring the daily behavior of the elderly; The data processing module is used to label and preprocess historical video data to obtain labeled behavioral video data; A model training module is used to train a pre-trained deep learning model based on labeled behavioral video data as a training set, so as to adjust the model parameters of the pre-trained deep learning model and obtain a trained risk behavior recognition model; The behavior recognition module is used to input the acquired real-time video data into the trained risk behavior recognition model to perform behavior recognition, obtain behavior recognition results, and issue risk warnings based on the recognition results.

[0015] Beneficial effects: 1. This invention deploys a pre-trained deep learning model and collects historical video data of the elderly's daily behavior to train and deploy the pre-trained deep learning model. This can provide a personalized recognition model for the elderly, reduce recognition limitations, and improve the generalization ability of the model. 2. This invention can detect potential problems in a timely manner through continuous monitoring and analysis of daily movements, providing a basis for early intervention and treatment, helping to improve the health of the elderly and prevent the deterioration of diseases; 3. The present invention can timely and accurately identify dangerous actions of the elderly, such as falls and tumbles, and issue early warning information in a timely manner, so that family members or caregivers can take timely measures to prevent accidents and protect the lives of the elderly; 4. By monitoring and identifying the daily behaviors of the elderly, the present invention can improve the quality and efficiency of elderly care services, meet the growing health management needs of the elderly, promote the development of the elderly care industry, promote technological innovation and product upgrades in related industries, and provide new impetus for economic growth. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following detailed description, they are used to explain the embodiments of the present invention, but do not constitute a limitation of the embodiments of the present invention. In the accompanying drawings: Figure 1 This is a flow chart of a method for identifying and monitoring risky behaviors of the elderly based on visual recognition, provided by one embodiment of the present invention; Figure 2 This is an application scenario architecture diagram of a method for identifying and monitoring risky behaviors of the elderly based on visual recognition, provided in one embodiment of the present invention; Figure 3 is a schematic diagram of encoding of various postures provided by an embodiment of the present invention; Figure 4 is a schematic diagram of a risk quantification table provided in one embodiment of the present invention; Figure 5 This is a block diagram of a system for identifying and monitoring risky behaviors of the elderly based on visual recognition, provided in one embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the present invention will be briefly introduced below in conjunction with the drawings and the description of the embodiments or the prior art. Obviously, the following description of the structure of the drawings is only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work. It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention.

[0018] Example 1 Figure 1 This is a flow chart of a method for identifying and monitoring risky behaviors of the elderly based on visual recognition, provided by one embodiment of the present invention. Figure 1As shown, this embodiment provides a method for identifying and monitoring risky behaviors of the elderly based on visual recognition. The application scenario of this method includes: a cloud server, a local PC and several cameras. The camera establishes a communication connection with the local PC, and the local PC establishes a communication connection with the cloud server. The method runs on the local PC, as shown in FIG. Figure 2 As shown; the method includes: Step S10: Obtain historical video data for monitoring the daily behavior of the elderly; in this embodiment, the daily behavior data of the elderly is collected by deploying cameras, wherein the cameras are preferably high-definition cameras, which transmit the collected video data to the local PC, and the local PC stores the uploaded video data. At the same time, during the transmission process, an encryption algorithm is used for encrypted transmission; the data transmitted between the local PC and the cloud server is also encrypted to improve the security of data transmission and reduce the risk of privacy leakage.

[0019] In this embodiment, camera layout and parameters must be optimized. By thoroughly studying the characteristics of the home environment, the optimal camera installation position and angle are determined to ensure comprehensive coverage of the elderly's primary activity areas, such as bedrooms, living rooms, kitchens, and bathrooms, while avoiding significant obstruction caused by furniture, doors, and windows. Camera parameters (such as exposure time and sensitivity) are optimized based on the lighting conditions in different rooms (e.g., a balcony with direct sunlight, a darkened bedroom corner, etc.) to obtain clear and stable video images.

[0020] In this embodiment, the captured historical video data includes both normal activities of the elderly (such as walking, sitting, standing, climbing stairs, dining, washing, etc.) and potentially high-risk behaviors (such as falls, prolonged stillness, abnormally rapid movement, and proximity to dangerous areas). Data collection also records relevant information such as the time and location of collection, as well as the elderly's basic health information (such as age, gender, and the presence of chronic diseases). Data collection lasts for at least 20 minutes to cover behavioral data from different activity scenarios and ensure data diversity and integrity.

[0021] Step S20: labeling and preprocessing the historical video data to obtain labeled behavior video data.

[0022] Specifically, historical video data is labeled and preprocessed to obtain labeled behavioral video data, including: Step a10: annotate the historical video data based on the preset expert database to obtain annotation content, which at least includes: behavior category, behavior start time, behavior end time and behavior key features; in this embodiment, professionals can be organized to annotate the collected video data in detail; for behaviors that are difficult to determine, geriatric medicine experts, rehabilitation experts, etc. are invited to jointly discuss and determine the annotations to form an expert database; among them, the key behavioral features include but are not limited to the spatiotemporal features of the elderly in daily activities such as walking, standing, sitting, lying, going up and down stairs, such as motion trajectory, speed change, joint angle change, etc., with a focus on features related to high-risk behaviors, such as changes in human posture at the moment of falling, body posture characteristics when standing still for a long time, abnormal wandering path patterns, etc.

[0023] Step a20: pre-process the annotated historical video data to obtain processed historical video data; wherein the pre-processing mainly includes image denoising, cropping, normalization and other operations to improve data quality and facilitate subsequent model training.

[0024] Step a30: Construct labels for historical video data based on the annotated content.

[0025] Step a40: constructing labeled behavior video data based on the labels of the historical video data and the processed historical video data.

[0026] In step a20, the specific steps of pre-processing the annotated historical video data are as follows: Step a201: extracting several image frames of the annotated historical video data.

[0027] Step a202: Perform light intensity analysis on each image frame to obtain light intensity detection results; that is, calculate the average brightness, contrast and other indicators of each area in the image through an algorithm to determine the intensity and distribution of the ambient light; for example: use image features and algorithms to identify the position and type of ambient light source, such as whether it is natural light or artificial light, as well as the direction and intensity distribution of the light source.

[0028] Step a203: performing ambient light compensation on each image frame according to the light intensity detection result to obtain a number of compensated image frames.

[0029] In this embodiment, the brightness value that needs to be adjusted is calculated based on the results of the light intensity detection. For areas that are too bright, the brightness is reduced; for areas that are too dark, the brightness is increased. The adjustment formula can be calculated based on the grayscale value of the image, for example: , or adjust according to the nonlinear form of the compensation curve correction: , where f(L,V) is a compensation function that combines light intensity and environmental variables.

[0030] Among them, I adjusted is the grayscale value after adjustment; I original The original grayscale value refers to the grayscale value of a pixel in an image before any adjustments are made. It typically ranges from 0 to 255 (for 8-bit images), where 0 represents black and 255 represents white. After light intensity detection, the original grayscale value serves as the basis for adjustment. α and β are adjustment factors, coefficients calculated based on the light intensity detection results. They are used to adjust the pixel grayscale value. These factors are determined by the ambient light intensity (L) and image characteristics (such as the environmental variable V). The adjustment factor is typically calculated inversely proportional to the light intensity. The environmental variable V may be calculated by segmenting the image, such as the ratio of the high-brightness area to the overall mean. The adjustment factor may be determined by a pre-defined mapping (such as interpolating a light intensity range with a compensation curve). For example, for overly bright areas, the adjustment factor can be less than 1 to reduce brightness; for overly dark areas, the adjustment factor can be greater than 1 to increase brightness.

[0031] Therefore, by adjusting the brightness, the image can maintain a good visual effect in different lighting environments, reduce the impact of excessive or dark light on target recognition, and improve the quality and readability of the image.

[0032] Step a204: performing color correction on each compensated image frame to obtain a plurality of corrected image frames.

[0033] In this embodiment, the effect of ambient light on image color is analyzed, and image color correction is performed based on information such as the light source's color temperature. For example, if the light source is yellowish, the blue component in the image can be increased; if the light source is bluish, the yellow component can be increased. The color correction formula can be implemented based on color space conversion and adjustment, such as converting RGB color space to XYZ color space, adjusting the XYZ color values, and then converting back to RGB color space.

[0034] By correcting the color distortion caused by ambient light, the color of the target object in the image is made more realistic, which helps improve the accuracy of subsequent feature extraction and recognition.

[0035] Step a205: Background noise filtering is performed on each corrected image frame to obtain a number of processed image frames. In this embodiment, the algorithm for background noise filtering on each corrected image frame is one or more of mean filtering, median filtering, frequency domain filtering and low-pass filtering.

[0036] Among them, mean filtering: for each pixel in the image, calculate the average value of the pixels in its neighborhood and use the average value as the new value of the current pixel. Mean filtering can effectively remove random noise such as salt and pepper noise in the image; the calculation formula of mean filtering is: , where f(i,j) is the original pixel value in the neighborhood, usually the pixel value in an n×n area centered at (x,y); g(x,y) is the filtered pixel value, that is, the arithmetic mean of all pixels in the neighborhood; N is the total number of pixels in the neighborhood, for example, for a 3×3 neighborhood, N=9. Among them, median filtering: sort the pixel values ​​in the neighborhood and take the middle value as the new value of the current pixel; median filtering is effective in removing impulse noise and salt and pepper noise, and can better preserve the edge information of the image. By removing noise points in the image, the image will be smoother, which can reduce the interference of noise on target recognition.

[0037] Frequency domain filtering uses the Fourier transform algorithm, which converts the image from the spatial domain to the frequency domain. In the frequency domain, the image's frequency components can be analyzed, including low-frequency components (representing the overall image characteristics) and high-frequency components (representing the image's detailed features). Frequency domain filtering can effectively preserve important image features while removing noise, thus preventing significant interference with object recognition.

[0038] Low-pass filtering involves selecting an appropriate low-pass filter, such as a Butterworth low-pass filter or a Gaussian low-pass filter, to attenuate high-frequency components in the frequency domain while retaining low-frequency components. Low-pass filtering can remove high-frequency noise from an image while preserving its basic outline and features.

[0039] Step a206: Fusing the processed image frames with the initial image frames of the historical video data to obtain a number of fused image frames, and using the video data composed of the fused image frames as the processed historical video data.

[0040] This embodiment fully utilizes the advantages of ambient light compensation and background noise filtering through image fusion, while retaining part of the information of the original image, improving the quality and reliability of the image, and providing a better foundation for subsequent feature extraction and recognition.

[0041] Step S30: training the pre-trained deep learning model based on the labeled behavior video data as a training set to adjust the model parameters of the pre-trained deep learning model to obtain a trained risk behavior recognition model; Step S40: inputting the acquired real-time video data into the trained risk behavior recognition model to perform behavior recognition, obtain behavior recognition results, and perform risk warning based on the recognition results.

[0042] In this embodiment, before using the training set to train the pre-trained deep learning model, the method also includes: collecting basic information of the elderly, including height, weight, age, gender, etc., which will be used for subsequent model training and personalized adjustment; at the same time, dividing the labeled behavioral video data into a training set, a validation set, and a test set according to a certain ratio (such as 70%:20%:10%); and using a suitable loss function (such as a cross-entropy loss function) and an optimization algorithm (such as an Adam optimizer) to train the model.

[0043] In this embodiment, the deep learning model uses a 3D convolutional neural network (3D CNN) or a two-stream convolutional neural network (Two-Stream CNN). The selected model is improved to address the diversity and complexity of elderly behaviors in complex home settings. For example, the 3D CNN model optimizes the convolution kernel size and number of layers to better extract the spatiotemporal features of the video. The Two-Stream CNN model adjusts the network structure of the spatial and temporal streams to enhance its understanding of behavioral movements and motion trends. An attention mechanism is introduced to enable the model to automatically focus on key areas and features in the video relevant to behavior recognition, improving recognition accuracy in complex backgrounds and occlusions.

[0044] In this embodiment, the deep learning model is pre-trained using a preset large-scale data set. The pre-training process can be performed on a cloud server, and the large-scale data set is also stored on the cloud server. After the pre-training is completed, the model parameters obtained by pre-training are sent to a local PC. The local PC uses the local training set to train the model again. This method can accelerate the model convergence speed; at the same time, it can provide personalized recognition models for the elderly, reduce recognition limitations, and improve the generalization ability of the model.

[0045] For the training set, we use data augmentation techniques such as random cropping, flipping, and rotation to expand the video data, increasing data diversity and improving the model's generalization ability. During training, we adjust training parameters such as the learning rate and number of iterations based on the performance of the validation set. We use early stopping to prevent overfitting and ensure that the model achieves good performance on both the training and validation sets. For example, depending on the training progress and model performance, we use a higher learning rate in the early stages of training to quickly converge to the optimal solution; and a lower learning rate in the later stages of training to avoid overfitting.

[0046] As a further optimization of this embodiment, the acquired real-time video data is input into the trained risk behavior recognition model to perform behavior recognition, obtain behavior recognition results, and issue risk warnings based on the recognition results, including: Step b10: Acquire real-time video data for monitoring the daily behavior of the elderly, and preprocess the real-time video data to obtain preprocessed real-time video data; wherein, the step of preprocessing the real-time video data adopts the method of steps a201 to a206 for processing.

[0047] Step b20: Input the pre-processed real-time video data into the trained risk behavior recognition model for behavior recognition to obtain behavior recognition results; wherein the behavior recognition results include: whether the elderly person falls, the elderly person's movement posture, etc.

[0048] Step b30: Risk grading the behavior recognition result based on the risk quantification table to obtain the risk level of the behavior recognition result.

[0049] In this embodiment, after obtaining the elderly's posture, it is necessary to conduct a risk assessment on the posture to determine whether the elderly's physical condition is at risk. In this embodiment, the risk quantification table includes multiple assessment parts, posture codes of each assessment part, loads (weight and force) of the assessment parts, and load codes; the assessment parts include: upper limbs (arms), lower limbs (legs), and back (back); Figure 3 and Figure 4 As shown, Figure 3 Each posture is encoded in the , and the corresponding posture code is obtained. Then, according to the evaluation part, the posture code of each evaluation part, the load (weight and force) of the evaluation part and the load code, the following is established: Figure 4 Risk quantification table shown.

[0050] In the risk quantification table, "1" indicates normal posture, "2" indicates body posture causing fatigue, "3" indicates body posture obviously causing fatigue, and "4" indicates body posture obviously causing high fatigue; among them, normal posture corresponds to the lowest risk level; body posture obviously causing high fatigue corresponds to the highest risk level.

[0051] Step b40: Generate an early warning signal based on the risk level of the behavior recognition result.

[0052] Step b50: Visually display the warning signal.

[0053] In this embodiment, the local PC terminal locally visualizes the warning information and uploads the warning information to the cloud server. The cloud server then sends the warning information to the user terminals of other users (family members or guardians), such as mobile phones and other devices; the warning information can be sent by text messages, mobile phone APP push, smart speaker voice prompts, etc.

[0054] The present invention has the following advantages: 1. Health management 1. Early risk identification: This system can identify the risk of musculoskeletal strain and disease in the elderly before they experience obvious symptoms. By continuously monitoring and analyzing daily movements, potential problems can be identified in a timely manner, providing a basis for early intervention and treatment, helping to improve the health of the elderly and prevent the worsening of diseases. 2. Personalized health management: Adapting to individual differences, taking into account individual differences among seniors, such as body shape and movement habits, through personalized training and algorithm adjustments, it can more accurately identify the risk characteristics of each senior and provide targeted health management recommendations. Based on the senior's health status and risk assessment results, a personalized monitoring plan is customized, including monitoring frequency and monitoring indicators, to improve monitoring efficiency and accuracy; 3. Reduce medical costs: Early detection and intervention can reduce the progression of the disease and the occurrence of complications, thereby reducing medical costs. It avoids the high costs of late-stage treatment of the disease and reduces the economic burden on the elderly and society.

[0055] 2. Security 1. Accident prevention: Timely and accurate identification of dangerous actions of the elderly, such as falls and tumbles, and timely issuance of early warning information so that family members or caregivers can take timely measures to prevent accidents and protect the lives of the elderly; 2. Provide safety guidance: Provide safety guidance to the elderly to help them improve their posture, enhance their body stability and balance, and reduce the risk of accidents such as falls; 3. Improve quality of life: Reduce pain and discomfort caused by musculoskeletal problems, improve the elderly's ability to care for themselves and their quality of life. The elderly can carry out daily activities more freely, reduce dependence on others, and enhance their self-confidence and sense of well-being.

[0056] 3. Social and Economic Aspects 1. Promote the development of the elderly care industry: The application of this technology can improve the quality and efficiency of elderly care services, meet the growing health management needs of the elderly, and promote the development of the elderly care industry. It will also promote technological innovation and product upgrades in related industries and provide new impetus for economic growth. 2. Reduce social burden: By identifying and preventing musculoskeletal diseases at an early stage, the long-term care needs of the elderly due to diseases can be reduced, the social burden of pension can be reduced, and the overall welfare level of the society can be improved.

[0057] 4. Technology Application 1. Promote technological innovation: This technology involves the integration of technologies from multiple fields, including computer vision, deep learning, and image processing. By continuously improving and optimizing algorithms, it promotes the development and innovation of related technologies. It also provides reference for intelligent monitoring and diagnostic technologies in other fields. 2. Improve the value of data utilization: By collecting and analyzing the exercise data of the elderly, we can provide more data support for medical research, gain a deeper understanding of the pathogenesis and development of musculoskeletal diseases, and provide new ideas and methods for medical research.

[0058] In summary, through real-time monitoring and analysis of the daily behaviors of the elderly, high-risk behaviors that lead to musculoskeletal strain can be identified promptly and accurately, and early warning information can be quickly issued so that family members or relevant caregivers can take timely measures to ensure the safety of the elderly at home and improve their quality of life. At the same time, strong technical support can be provided for home-based care, thereby reducing the burden on families and society and promoting the realization of the Healthy China strategy.

[0059] Example 2 Figure 5 This is a block diagram of a system for identifying and monitoring risky behaviors of the elderly based on visual recognition, provided by one embodiment of the present invention. Figure 5 As shown, this embodiment provides a system for identifying and monitoring risky behaviors of the elderly based on visual recognition, which is used to implement the method for identifying and monitoring risky behaviors of the elderly based on visual recognition in Example 1. The system includes: A data acquisition module is used to obtain historical video data for monitoring the daily behavior of the elderly; The data processing module is used to label and preprocess historical video data to obtain labeled behavioral video data; A model training module is used to train a pre-trained deep learning model based on labeled behavioral video data as a training set, so as to adjust the model parameters of the pre-trained deep learning model and obtain a trained risk behavior recognition model; The behavior recognition module is used to input the acquired real-time video data into the trained risk behavior recognition model to perform behavior recognition, obtain behavior recognition results, and issue risk warnings based on the recognition results.

[0060] This embodiment also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method for identifying and monitoring risk behaviors of the elderly based on visual recognition in the first embodiment is implemented.

[0061] This embodiment also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method for identifying and monitoring risky behaviors of the elderly based on visual recognition in embodiment one is implemented.

[0062] The present invention deploys a pre-trained deep learning model and then collects historical video data of the elderly's daily behavior to train and deploy the pre-trained deep learning model, which can provide personalized recognition models for the elderly, reduce recognition limitations, and improve the generalization ability of the model; the present invention continuously monitors and analyzes daily movements to timely capture potential problems, provide a basis for early intervention and treatment, and help improve the health of the elderly and prevent the deterioration of diseases; the present invention can timely and accurately identify dangerous movements of the elderly, such as falls and falls, and issue early warning information in time, so that family members or caregivers can take timely measures to prevent accidents and protect the lives of the elderly; the present invention monitors the daily behavior of the elderly and identifies behavior, which can improve the quality and efficiency of elderly care services, meet the growing health management needs of the elderly, promote the development of the elderly care industry, promote technological innovation and product upgrades in related industries, and provide new impetus for economic growth.

[0063] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0064] The present application is described in terms of flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A system that specifies the functions of a box or boxes.

[0065] The above are merely embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A method for identifying and monitoring risky behaviors of the elderly based on visual recognition, characterized in that: The method comprises: Obtain historical video data to monitor the daily behavior of the elderly; Label and preprocess historical video data to obtain labeled behavioral video data; The pre-trained deep learning model is trained based on the labeled behavioral video data as a training set to adjust the model parameters of the pre-trained deep learning model to obtain a trained risk behavior recognition model; The acquired real-time video data is input into the trained risk behavior recognition model for behavior recognition, to obtain a behavior recognition result, and a risk warning is issued based on the recognition result.

2. The method for identifying and monitoring risky behaviors of the elderly based on visual recognition according to claim 1 is characterized in that: The deep learning model is pre-trained using a preset large-scale dataset.

3. The method for identifying and monitoring risky behaviors of the elderly based on visual recognition according to claim 1 is characterized in that: Input the acquired real-time video data into the trained risk behavior recognition model to perform behavior recognition, obtain behavior recognition results, and issue risk warnings based on the recognition results, including: Acquire real-time video data for monitoring the daily behavior of the elderly, and pre-process the real-time video data to obtain pre-processed real-time video data; The pre-processed real-time video data is input into the trained risk behavior recognition model for behavior recognition to obtain the behavior recognition results; Based on the risk quantification table, the behavior recognition results are graded to obtain the risk level of the behavior recognition results; Generate early warning signals based on the risk level of behavior recognition results; Provide visual display of early warning signals.

4. The method for identifying and monitoring risky behaviors of the elderly based on visual recognition according to any one of claims 1 to 3, characterized in that: The deep learning model is a 3D convolutional neural network or a two-stream convolutional neural network.

5. The method for identifying and monitoring risky behaviors of the elderly based on visual recognition according to claim 3 is characterized in that: Label and preprocess historical video data to obtain labeled behavioral video data, including: Annotate historical video data based on a preset expert database to obtain annotation content, wherein the annotation content includes at least: behavior category, behavior start time, behavior end time, and behavior key features; Preprocessing the annotated historical video data to obtain processed historical video data; Construct labels for historical video data based on the annotation content; Based on the labels of the historical video data and the processed historical video data, the behavioral video data with labels is constructed.

6. The method for identifying and monitoring risky behaviors of the elderly based on visual recognition according to claim 5 is characterized in that: The annotated historical video data is preprocessed to obtain processed historical video data, including: Extracting several image frames of the annotated historical video data; Perform light intensity analysis on each image frame to obtain light intensity detection results; Performing ambient light compensation on each image frame according to the light intensity detection result to obtain a number of compensated image frames; Performing color correction on each compensated image frame to obtain a plurality of corrected image frames; Perform background noise filtering on each corrected image frame to obtain a number of processed image frames; The processed image frames are fused with the initial image frames of the historical video data to obtain a number of fused image frames, and the video data composed of the fused image frames is used as the processed historical video data.

7. The method for identifying and monitoring risky behaviors of the elderly based on visual recognition according to claim 6 is characterized in that: The algorithm for filtering the background noise of each corrected image frame is one or more of mean filtering, median filtering, frequency domain filtering and low-pass filtering.

8. The method for identifying and monitoring risky behaviors of the elderly based on visual recognition according to claim 3 is characterized in that: The risk quantification table includes multiple assessment parts, posture codes of each assessment part, loads of each assessment part, and load codes; The assessment areas include upper limbs, lower limbs and back.

9. A system for identifying and monitoring risky behaviors of the elderly based on visual recognition, used to implement the method for identifying and monitoring risky behaviors of the elderly based on visual recognition as claimed in any one of claims 1 to 8, characterized in that: The system comprises: A data acquisition module is used to obtain historical video data for monitoring the daily behavior of the elderly; The data processing module is used to label and preprocess historical video data to obtain labeled behavioral video data; A model training module is used to train a pre-trained deep learning model based on labeled behavioral video data as a training set, so as to adjust the model parameters of the pre-trained deep learning model and obtain a trained risk behavior recognition model; The behavior recognition module is used to input the acquired real-time video data into the trained risk behavior recognition model to perform behavior recognition, obtain behavior recognition results, and issue risk warnings based on the recognition results.

Images