Personnel behavior analysis management system and method based on AI vision

Abstract

The invention discloses a personnel behavior analysis management system and method based on AI vision, relates to the technical field of vision analysis, and solves the problem of imperfect definition of abnormal behaviors by introducing a dynamic behavior feature verification module, calculating an image verification factor Bcz and evaluating and extracting potential abnormal behavior features. A multi-target online analysis module is adopted, a convolutional neural network and a space-time diagram convolutional network are fused, a behavior abnormal factor Ax and a target tracking factor Tg are extracted through an attention mechanism, and Kalman filtering and an SORT algorithm are combined, so that accurate tracking of multi-target behaviors is realized, and the problem of difficulty in multi-target tracking is solved; the problem that the platform data fusion degree is low is solved through the self-adaptive comprehensive evaluation module; the abnormal behavior recognition, target tracking and multi-source data fusion analysis capabilities are effectively improved, and accurate monitoring and intelligent intervention of abnormal behaviors in a complex scene are realized.

Description

Technical Field

[0001] The present invention relates to the field of visual analysis technology, and specifically to a personnel behavior analysis management system and method based on AI vision. Background Art

[0002] With the rise of big data and artificial intelligence technology, the human behavior analysis and management system came into being. It originally originated from the military and security fields, used to identify potential threatening behaviors and strengthen monitoring. With the popularization of the Internet and the growth of corporate demand, the system has gradually expanded to areas such as employee performance appraisal, public place safety and commercial operation management. In the process of development, it has evolved from preliminary video surveillance and basic data collection to a multi-dimensional behavior insight platform that integrates deep learning algorithms, natural language processing and big data analysis. Through multi-source data and intelligent analysis, it not only improves security, but also lays the foundation for enterprise refined management and decision optimization.

[0003] The personnel behavior analysis and management system has a wide range of applications, especially in security monitoring; however, the personnel behavior analysis and management system currently used in security monitoring has the following technical disadvantages: 1. Incomplete definition of abnormal behavior: Since the behavior analysis algorithm relies on predefined behavior features, it is easy to have recognition bias or missed reports when faced with unlabeled abnormal behaviors or complex scenarios; 2. Difficulty in multi-target tracking: In complex scenarios, the tracking and identification of multiple actors may be confused.

[0004] 3. The platform’s data integration level is low, making it difficult to form a unified analytical ecosystem. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention provides an AI vision-based personnel behavior analysis management system and method, which solves the technical shortcomings mentioned in the background technology.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a personnel behavior analysis and management system based on AI vision, including a behavior image acquisition module, a dynamic behavior feature verification module, a behavior preprocessing module, a multi-target online analysis module and an adaptive comprehensive evaluation module; The behavior image acquisition module is used to collect personnel behavior image data in real time, including video clips and image frames; then collect and combine scene environment information to perform multi-dimensional annotation on the personnel behavior image data, including behavior subject identity information, scene type label, time series label and multi-target positioning information; The dynamic behavior feature verification module dynamically verifies the collected personnel behavior image data, calculates and evaluates the image verification factor Bcz by fitting the behavior image resolution Rf, frame rate Rz, illumination value Rg and background noise value Rb in the personnel behavior image data, and verifies the integrity and consistency of the image data; then, based on the personnel behavior image data, cluster analysis and pattern recognition are used to extract potential abnormal behavior features to form a dynamic behavior feature data set; The behavior preprocessing module optimizes the dynamic behavior feature data set, uses a multi-scale denoising algorithm to eliminate background interference, accurately extracts the behavior subject feature area through multi-target segmentation technology, realizes independent extraction and normalization of multi-target behavior features, and finally constructs a multi-target behavior feature data set and a scene adaptation data set respectively; The multi-target online analysis module is used to construct and use an improved deep learning model to perform real-time analysis on a multi-target behavior feature dataset and a scene adaptation dataset, and extract a behavior abnormality factor Ax and a target tracking factor Tg; The adaptive comprehensive evaluation module is used to construct the platform data fusion factor Pf, and associate the abnormal behavior factor Ax and the target tracking factor Tg with the platform data fusion factor Pf, calculate and evaluate the comprehensive behavior risk index Rz, and generate a behavior analysis and intervention strategy plan.

[0007] Preferably, the behavior image acquisition module continuously collects personnel behavior image data by deploying high-definition video equipment, infrared sensors and depth cameras in the monitoring area to monitor the target scene in real time. During the acquisition process, the video image data is first automatically framed and sampled to extract key behavior images. Subsequently, face recognition, posture estimation and target detection algorithms are used to identify and classify the behavior subject to obtain the identity information of the behavior subject. Combined with the data fed back by the scene environment sensor, the scene type is automatically identified and a scene type label is generated. Secondly, accurate time series tags are added to each frame of image data. At the same time, multi-target detection and SORT tracking algorithms are used to perform real-time positioning and trajectory tracking of multiple personnel behavior subjects in the same scene to obtain multi-target positioning information. Finally, the collected personnel behavior image data and annotation information are uniformly formatted and stored.

[0008] Preferably, the dynamic behavior feature verification module includes a data integrity quantification unit, a data integrity evaluation unit and a dynamic feature analysis unit; The data integrity quantification unit is used to fit the behavioral image resolution Rf, frame rate Rz, illumination value Rg and background noise value Rb, and calculate the image verification factor Bcz; the specific calculation formula is as follows: .

[0009] Preferably, the data integrity evaluation unit is used to preset the image verification threshold Q, and compare and evaluate it with the image verification factor Bcz to verify the integrity and consistency of the image data. The specific evaluation content is as follows: When the image verification factor Bcz ≥ the image verification threshold Q, the personnel behavior image data is determined to be complete, indicating that the image resolution, frame rate, lighting conditions and background noise meet the system requirements; When the image verification factor Bcz is less than the image verification threshold Q, it is determined that the personnel behavior image data is incomplete, and there are problems such as image blur, low frame rate, insufficient lighting, and noise interference. At this time, the data re-collection or quality repair process is triggered.

[0010] Preferably, the dynamic feature analysis unit first preprocesses the complete personnel behavior image data, adjusts the image quality by using noise reduction and image enhancement algorithms, and then uses a multidimensional feature extraction algorithm to extract key behavior feature parameters from the personnel behavior image data, including posture changes, motion trajectories, speed changes, and behavior pattern sequences of the behavior subject; then, a clustering analysis algorithm is used to perform unsupervised classification on the extracted key behavior feature parameters, cluster similar behaviors into categories, and identify behavior patterns that deviate from or are undefined in the existing behavior definition database; the clustering results are further analyzed and matched by a pattern recognition algorithm, and suspected abnormal behaviors are supplemented with features and differentiated, and potential abnormal behavior features are extracted; finally, the extracted potential abnormal behavior features are classified and sorted according to behavior categories, time series, and scene types to construct a dynamic behavior feature data set.

[0011] Preferably, the behavior preprocessing module optimizes the dynamic behavior feature data set, firstly adopts a multi-scale denoising algorithm to adaptively eliminate background noise and environmental interference factors in the image data, including a joint filtering technology based on the spatial domain and the frequency domain, to suppress the interference of illumination fluctuations, dynamic background changes and random noise on the behavior feature extraction; After denoising, the multi-target segmentation technology is used to accurately segment multiple behavioral subjects in the image data. The deep learning segmentation model and target detection algorithm are used to segment the contours, postures and motion trajectories of people, and to distinguish different behavioral subjects in complex scenes. After the segmentation is completed, the dynamic parameters including the behavior amplitude, speed and angle of the behavior subject are standardized to eliminate the feature deviation caused by changes in camera angle, distance and lighting; finally, based on the individual characteristics of the behavior subject and the scene environment attributes, the processed data are constructed into a multi-target behavior feature dataset and a scene adaptation dataset.

[0012] Preferably, the multi-objective online analysis module constructs a behavior analysis framework based on an improved deep learning model; wherein the improved deep learning model jointly extracts the spatial and temporal dimension features of personnel behavior by fusing a convolutional neural network with a spatiotemporal graph convolutional network; During the analysis process, the input multi-target behavior feature dataset is first modeled into a behavior sequence to capture the dynamic changes of the behavior subject in a specific scenario. Then, the attention mechanism is used to enhance the model's sensitivity to the key behavior features in the multi-target behavior feature dataset, and deep learning analysis is performed on the behavior subject's posture changes, motion trajectories, and interaction relationships to extract the behavioral anomaly factor Ax related to the deviation from the behavior pattern. At the same time, a multi-target tracking algorithm and a global data association strategy are used to track and dynamically update the location information and trajectories of multiple behavior subjects in complex scenarios in real time to extract the target tracking factor Tg.

[0013] Preferably, the adaptive comprehensive assessment module includes a risk assessment unit and a strategy optimization unit; The risk assessment unit is used to calculate the comprehensive behavior risk index Rz; after extracting the abnormal behavior factor Ax, the target tracking factor Tg and the platform data fusion factor Pf and performing dimensionless processing, the comprehensive behavior risk index Rz is obtained using the following formula: .

[0014] Preferably, the strategy optimization unit is used to evaluate the comprehensive behavior risk index Rz and generate a behavior analysis and intervention strategy plan; by presetting the comprehensive first behavior risk threshold W1 and the second behavior risk threshold W2, the comprehensive behavior risk index Rz is compared, and the first behavior risk threshold W1 is less than the second behavior risk threshold W2, the specific contents are as follows: When the comprehensive behavior risk index Rz is less than the first behavior risk threshold W1, it is judged as normal behavior, and the system records the behavior data for model optimization without triggering any intervention measures; When the first behavior risk threshold W1≤comprehensive behavior risk index Rz<second behavior risk threshold W2, it is judged as a potential risk behavior, the system issues a risk warning, and focuses on monitoring the target; the specific behavior analysis and intervention strategy is: include the target behavior in the key monitoring object, adjust the camera angle or resolution; increase the behavior data sampling frequency, and track the behavior dynamics in real time; send risk warning information to the management terminal for reference by security personnel; When the comprehensive behavior risk index Rz ≥ the second behavior risk threshold W2, it is judged as a high-risk behavior, and the emergency intervention strategy is immediately implemented and the alarm system is linked; the specific behavior analysis and intervention strategy plan is: triggering the sound and light alarm device or broadcasting a warning; automatically linking the security system to physically isolate or block the behavior subject; sending emergency alarms and target positioning information to security personnel and the management center, and pushing on-site video images in real time; calling historical behavior data, analyzing the potential risk level of the behavior subject, and providing a basis for security decision-making.

[0015] The personnel behavior analysis and management method based on AI vision includes the following steps: Step 1: Collect human behavior image data in real time, including video clips and image frames; then collect and combine scene environment information to perform multi-dimensional annotation on human behavior image data, including behavior subject identity information, scene type label, time series label and multi-target positioning information; Step 2: Dynamically verify the collected personnel behavior image data. By fitting the behavior image resolution Rf, frame rate Rz, illumination value Rg and background noise value Rb in the personnel behavior image data, calculate and evaluate the image verification factor Bcz to verify the integrity and consistency of the image data; then, based on the personnel behavior image data, cluster analysis and pattern recognition are used to extract potential abnormal behavior features to form a dynamic behavior feature data set; Step 3: Optimize the dynamic behavior feature dataset, use a multi-scale denoising algorithm to eliminate background interference, accurately extract the feature area of ​​the behavior subject through multi-target segmentation technology, realize the independent extraction and normalization of multi-target behavior features, and finally construct a multi-target behavior feature dataset and a scene adaptation dataset respectively; Step 4: Build and use an improved deep learning model to perform real-time analysis on multi-target behavior feature datasets and scene adaptation datasets, and extract behavior abnormality factors Ax and target tracking factors Tg; Step 5: Construct the platform data fusion factor Pf, and associate the abnormal behavior factor Ax and the target tracking factor Tg with the platform data fusion factor Pf, calculate and evaluate the comprehensive behavior risk index Rz, and generate a behavior analysis and intervention strategy plan.

[0016] The present invention provides a personnel behavior analysis management system and method based on AI vision. It has the following beneficial effects: (1) The AI ​​vision-based personnel behavior analysis management system and method effectively solves the problem of incomplete abnormal behavior definition by introducing a dynamic behavior feature verification module. The module uses cluster analysis and pattern recognition algorithms to dynamically verify the behavior image resolution Rf, frame rate Rz, illumination value Rg and background noise value Rb, calculates the image verification factor Bcz and compares it with the preset image verification threshold Q to achieve efficient verification of the integrity and consistency of the behavior data. Furthermore, through cluster analysis of the behavior subject's posture changes, motion trajectories, speed changes and behavior pattern sequences, it dynamically extracts potential abnormal behavior features and constructs a dynamic behavior feature data set, effectively expanding the definition scope of abnormal behavior and enhancing the system's ability to identify unlabeled abnormal behaviors and abnormal behaviors in complex scenarios. (2) This AI vision-based personnel behavior analysis management system and method solves the problem of multi-target tracking difficulties through a multi-target online analysis module. This module is based on an improved deep learning model that integrates convolutional neural networks and spatiotemporal graph convolutional networks. It performs real-time behavior sequence modeling on multi-target behavior feature datasets and scene adaptation datasets, accurately capturing the dynamic changes of behavior subjects in specific scenarios. At the same time, it combines the attention mechanism to improve the sensitivity to key behavior features, uses multi-target detection and global data association strategies to extract behavior anomaly factors Ax and target tracking factors Tg in real time, and optimizes multi-target real-time tracking through Kalman filtering and SORT algorithms, significantly improving the system's tracking accuracy and recognition stability for multi-target behavior subjects in complex environments. (3) This AI vision-based personnel behavior analysis management system and method addresses the problem of low platform data integration. The system achieves efficient integration and risk assessment of multi-source data through an adaptive comprehensive evaluation module. This module constructs a platform data integration factor Pf, and jointly calculates the comprehensive behavior risk index with the behavior abnormality factor Ax and the target tracking factor Tg. Based on the preset first behavior risk threshold W1 and second behavior risk threshold W2, the comprehensive behavior risk index Rz is dynamically graded and evaluated, and targeted behavior analysis and intervention strategy solutions are automatically generated. The scenario adaptation model is continuously optimized through multi-dimensional data fusion and feedback mechanism to ensure that the system achieves a unified analysis ecology in a multi-source data environment and comprehensively improves the detection and response capabilities of abnormal behaviors. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a schematic diagram of the framework structure of the personnel behavior analysis and management system based on AI vision of the present invention; Figure 2 The figure is a flowchart of the steps of the personnel behavior analysis and management method based on AI vision of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] Example 1 See also Figure 1 , a personnel behavior analysis and management system based on AI vision, including a behavior image acquisition module, a dynamic behavior feature verification module, a behavior preprocessing module, a multi-target online analysis module and an adaptive comprehensive evaluation module; The behavior image acquisition module is used to collect personnel behavior image data in real time, including video clips and image frames; then collect and combine scene environment information to perform multi-dimensional annotation on the personnel behavior image data, including behavior subject identity information, scene type label, time series label and multi-target positioning information; The dynamic behavior feature verification module dynamically verifies the collected personnel behavior image data. By fitting the behavior image resolution Rf, frame rate Rz, illumination value Rg and background noise value Rb in the personnel behavior image data, the image verification factor Bcz is calculated and evaluated to verify the integrity and consistency of the image data. Then, based on the personnel behavior image data, cluster analysis and pattern recognition are used to extract potential abnormal behavior features to form a dynamic behavior feature data set. The behavior preprocessing module optimizes the dynamic behavior feature data set, uses a multi-scale denoising algorithm to eliminate background interference, accurately extracts the feature area of ​​the behavior subject through multi-target segmentation technology, realizes the independent extraction and normalization of multi-target behavior features, and finally constructs a multi-target behavior feature data set and a scene adaptation data set respectively; The multi-target online analysis module is used to build and use an improved deep learning model to perform real-time analysis on multi-target behavior feature datasets and scene adaptation datasets, and to extract behavior abnormality factors Ax and target tracking factors Tg; The adaptive comprehensive evaluation module is used to construct the platform data fusion factor Pf, and associate the behavioral abnormality factor Ax and the target tracking factor Tg with the platform data fusion factor Pf, calculate and evaluate the comprehensive behavioral risk index Rz, and generate behavioral analysis and intervention strategy plans.

[0020] In this embodiment, the behavior image acquisition module collects personnel behavior image data in real time and performs multi-dimensional annotation in combination with scene environment information, which can ensure the comprehensiveness and accuracy of the data and provide a high-quality data foundation for subsequent analysis; The dynamic behavior feature verification module calculates and evaluates the image verification factor Bcz to dynamically verify the integrity and consistency of image data. At the same time, it uses cluster analysis and pattern recognition algorithms to extract potential abnormal behavior features to form a dynamic behavior feature data set to improve the ability to identify abnormal behaviors. The behavior preprocessing module eliminates background interference through a multi-scale denoising algorithm, and uses multi-target segmentation technology to accurately extract the characteristic area of ​​the behavior subject. It independently extracts and normalizes the multi-target behavior feature data, and constructs a multi-target behavior feature data set and a scene adaptation data set respectively, to enhance the accuracy and adaptability of data processing; The multi-target online analysis module uses an improved deep learning model to analyze multi-target behavior feature data sets and scene adaptation data sets in real time, extracts behavior abnormality factors Ax and target tracking factors Tg, and effectively improves the real-time performance and accuracy of multi-target behavior analysis; The adaptive comprehensive assessment module calculates and evaluates the comprehensive behavioral risk index Rz, generates behavioral analysis and intervention strategy solutions, and finally continuously optimizes the scenario adaptation model through multi-dimensional data fusion and feedback mechanism, thereby improving the system's ability to identify, evaluate and intelligently respond to abnormal behaviors in complex scenarios.

[0021] Example 2 The behavior image acquisition module continuously collects personnel behavior image data by deploying high-definition video equipment, infrared sensors and depth cameras in the monitoring area to monitor the target scene in real time. During the acquisition process, the video image data is first automatically framed and sampled to extract key behavior images. Then, face recognition, posture estimation and target detection algorithms are used to identify and classify the behavior subject to obtain the identity information of the behavior subject. Combined with the data fed back by the scene environment sensor, the scene type is automatically identified and the scene type label is generated. Secondly, accurate time series tags are added to each frame of image data. At the same time, multi-target detection and SORT tracking algorithms are used to perform real-time positioning and trajectory tracking of multiple personnel behavior subjects in the same scene to obtain multi-target positioning information. Finally, the collected personnel behavior image data and annotation information are uniformly formatted and stored.

[0022] The dynamic behavior feature verification module includes a data integrity quantification unit, a data integrity assessment unit and a dynamic feature analysis unit; The data integrity quantification unit is used to fit the behavioral image resolution Rf, frame rate Rz, illumination value Rg and background noise value Rb, and calculate the image verification factor Bcz; the specific calculation formula is as follows: .

[0023] The data integrity assessment unit is used to preset the image verification threshold Q and compare and evaluate it with the image verification factor Bcz to verify the integrity and consistency of the image data. The specific assessment contents are as follows: When the image verification factor Bcz ≥ the image verification threshold Q, the personnel behavior image data is determined to be complete, indicating that the image resolution, frame rate, lighting conditions and background noise meet the system requirements; When the image verification factor Bcz is less than the image verification threshold Q, it is determined that the personnel behavior image data is incomplete, and there are problems such as image blur, low frame rate, insufficient lighting, and noise interference. At this time, the data re-collection or quality repair process is triggered.

[0024] The dynamic feature analysis unit first preprocesses the complete personnel behavior image data, uses noise reduction and image enhancement algorithms to adjust the image quality, and then uses a multidimensional feature extraction algorithm to extract key behavior feature parameters from the personnel behavior image data, including the posture changes, motion trajectories, speed changes and behavior pattern sequences of the behavior subjects; then, a clustering analysis algorithm is used to perform unsupervised classification on the extracted key behavior feature parameters, cluster similar behaviors into categories, and identify behavior patterns that deviate from or are undefined in the existing behavior definition database; the clustering results are further analyzed and matched through a pattern recognition algorithm, and suspected abnormal behaviors are supplemented and differentiated to extract potential abnormal behavior features; finally, the extracted potential abnormal behavior features are classified and sorted according to behavior categories, time series and scene types to construct a dynamic behavior feature dataset.

[0025] In this embodiment, the behavior image acquisition module is used to realize real-time monitoring of the behavior of people in the monitoring area. The high-definition camera equipment, infrared sensor and depth camera are used to continuously collect the behavior image data of people, and the identity information of the behavior subject is obtained through face recognition, posture estimation and target detection algorithm. The scene type label is generated in combination with the feedback data of the scene environment sensor, and a time series tag is added to each frame of image data. The multi-target detection and SORT tracking algorithm are used to obtain multi-target positioning information in real time to ensure the comprehensiveness and timeliness of the data; the dynamic behavior feature verification module fits the behavior image resolution Rf, frame rate Rz, illumination value Rg and background noise value Rb through the data integrity quantization unit, calculates the image verification factor Bcz, and The data integrity assessment unit evaluates the image verification factor Bcz to accurately evaluate the integrity and consistency of the image data to ensure reliable data quality; the dynamic feature analysis unit uses noise reduction and image enhancement algorithms to pre-process the data, extract key behavioral features such as posture changes, motion trajectories, speed changes, and behavioral pattern sequences, and uses clustering analysis and pattern recognition algorithms to classify and compare behavioral features, extract potential abnormal behaviors and construct dynamic behavioral feature data sets to enhance the system's ability to identify abnormal behaviors; the various modules of the system work together to ensure that the entire process from data collection, feature verification, anomaly extraction to behavior analysis is accurate and efficient, and realize real-time monitoring, precise identification and intelligent intervention of multi-target behaviors in complex scenarios.

[0026] Example 3 The behavior preprocessing module optimizes the dynamic behavior feature data set. First, a multi-scale denoising algorithm is used to adaptively eliminate background noise and environmental interference factors in the image data, including a joint filtering technology based on the spatial domain and the frequency domain to suppress the interference of light fluctuations, dynamic background changes and random noise on the extraction of behavior features. After denoising, the multi-target segmentation technology is used to accurately segment multiple behavioral subjects in the image data. The deep learning segmentation model and target detection algorithm are used to segment the contours, postures and motion trajectories of people, and to distinguish different behavioral subjects in complex scenes. After the segmentation is completed, the dynamic parameters including the behavior amplitude, speed and angle of the behavior subject are standardized to eliminate the feature deviation caused by changes in camera angle, distance and lighting; finally, based on the individual characteristics of the behavior subject and the scene environment attributes, the processed data are constructed into a multi-target behavior feature dataset and a scene adaptation dataset.

[0027] The multi-target online analysis module builds a behavior analysis framework based on an improved deep learning model. The improved deep learning model jointly extracts the spatial and temporal dimension features of human behavior by integrating convolutional neural networks and spatiotemporal graph convolutional networks. During the analysis process, the input multi-target behavior feature dataset is first modeled into a behavior sequence to capture the dynamic changes of the behavior subject in a specific scenario. Then, the attention mechanism is used to enhance the model's sensitivity to the key behavior features in the multi-target behavior feature dataset, and deep learning analysis is performed on the behavior subject's posture changes, motion trajectories, and interaction relationships to extract the behavioral anomaly factor Ax related to the deviation from the behavior pattern. At the same time, a multi-target tracking algorithm and a global data association strategy are used to track and dynamically update the location information and trajectories of multiple behavior subjects in complex scenarios in real time to extract the target tracking factor Tg.

[0028] The adaptive comprehensive assessment module includes a risk assessment unit and a strategy optimization unit; The risk assessment unit is used to calculate the comprehensive behavior risk index Rz; after extracting the abnormal behavior factor Ax, the target tracking factor Tg and the platform data fusion factor Pf and performing dimensionless processing, the comprehensive behavior risk index Rz is obtained using the following formula: .

[0029] The strategy optimization unit is used to evaluate the comprehensive behavior risk index Rz and generate a behavior analysis and intervention strategy plan; by presetting the comprehensive first behavior risk threshold W1 and the second behavior risk threshold W2, the comprehensive behavior risk index Rz is compared, and the first behavior risk threshold W1 is less than the second behavior risk threshold W2. The specific contents are as follows: When the comprehensive behavior risk index Rz is less than the first behavior risk threshold W1, it is judged as normal behavior, and the system records the behavior data for model optimization without triggering any intervention measures; When the first behavior risk threshold W1≤comprehensive behavior risk index Rz<second behavior risk threshold W2, it is judged as a potential risk behavior, the system issues a risk warning, and focuses on monitoring the target; the specific behavior analysis and intervention strategy is: include the target behavior in the key monitoring object, adjust the camera angle or resolution; increase the behavior data sampling frequency, and track the behavior dynamics in real time; send risk warning information to the management terminal for reference by security personnel; When the comprehensive behavior risk index Rz ≥ the second behavior risk threshold W2, it is judged as a high-risk behavior, and the emergency intervention strategy is immediately implemented and the alarm system is linked; the specific behavior analysis and intervention strategy plan is: triggering the sound and light alarm device or broadcasting a warning; automatically linking the security system to physically isolate or block the behavior subject; sending emergency alarms and target positioning information to security personnel and the management center, and pushing on-site video images in real time; calling historical behavior data, analyzing the potential risk level of the behavior subject, and providing a basis for security decision-making.

[0030] In this embodiment, the dynamic behavior feature data set is optimized through the behavior preprocessing module, and the multi-scale denoising algorithm (based on the joint filtering technology of the spatial domain and the frequency domain) is used to effectively eliminate background noise and environmental interference factors, and suppress the interference of light fluctuations, dynamic background changes and random noise on the extraction of behavior features; then the multi-target segmentation technology is combined with the deep learning segmentation model and the target detection algorithm to accurately segment the personnel contours, postures and motion trajectories of multiple behavior subjects in the image data, so as to distinguish different behavior subjects in complex scenes; the segmented behavior subject data is standardized for dynamic parameters such as behavior amplitude, speed, angle, etc., to eliminate the feature deviation caused by camera angle, distance and lighting changes, and finally construct a multi-target behavior feature data set and a scene adaptation data set to provide accurate data support for subsequent analysis; the multi-target online analysis module is based on an improved deep learning model that integrates convolutional neural networks and spatiotemporal graph convolutional networks to jointly extract the behavior of personnel. The spatial and temporal dimension features capture the dynamic changes of the behavior subject in a specific scenario through behavior sequence modeling, use the attention mechanism to improve the sensitivity to key behavior features (including posture changes, motion trajectories and interaction relationships), extract the behavior abnormality factor Ax, and combine the multi-target tracking algorithm with the global data association strategy to track and dynamically update the location information and trajectory of the behavior subject in real time, and extract the target tracking factor Tg; the adaptive comprehensive evaluation module extracts the behavior abnormality factor Ax, the target tracking factor Tg and the platform data fusion factor Pf through the risk assessment unit, calculates the comprehensive behavior risk index Rz, and the strategy optimization unit performs a graded evaluation on the comprehensive behavior risk index Rz based on the preset first behavior risk threshold W1 and the second behavior risk threshold W2; through the collaborative work of various modules, the real-time monitoring of multi-target behaviors, the accurate identification of abnormal behaviors and the intelligent response capabilities in complex scenarios are effectively improved, and efficient and intelligent behavior analysis and risk intervention are realized.

[0031] Example 4 See also Figure 2 , the personnel behavior analysis management method based on AI vision includes the following steps: Step 1: Collect human behavior image data in real time, including video clips and image frames; then collect and combine scene environment information to perform multi-dimensional annotation on human behavior image data, including behavior subject identity information, scene type label, time series label and multi-target positioning information; Step 2: Dynamically verify the collected personnel behavior image data. By fitting the behavior image resolution Rf, frame rate Rz, illumination value Rg and background noise value Rb in the personnel behavior image data, calculate and evaluate the image verification factor Bcz to verify the integrity and consistency of the image data; then, based on the personnel behavior image data, cluster analysis and pattern recognition are used to extract potential abnormal behavior features to form a dynamic behavior feature data set; Step 3: Optimize the dynamic behavior feature dataset, use a multi-scale denoising algorithm to eliminate background interference, accurately extract the feature area of ​​the behavior subject through multi-target segmentation technology, realize the independent extraction and normalization of multi-target behavior features, and finally construct a multi-target behavior feature dataset and a scene adaptation dataset respectively; Step 4: Build and use an improved deep learning model to perform real-time analysis on multi-target behavior feature datasets and scene adaptation datasets, and extract behavior abnormality factors Ax and target tracking factors Tg; Step 5: Construct the platform data fusion factor Pf, and associate the abnormal behavior factor Ax and the target tracking factor Tg with the platform data fusion factor Pf, calculate and evaluate the comprehensive behavior risk index Rz, and generate a behavior analysis and intervention strategy plan.

[0032] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. AI vision-based personnel behavior analysis and management system, characterized by: It includes a behavior image acquisition module, a dynamic behavior feature verification module, a behavior preprocessing module, a multi-target online analysis module and an adaptive comprehensive evaluation module; The behavior image acquisition module is used to collect personnel behavior image data in real time, including video clips and image frames; then collect and combine scene environment information to perform multi-dimensional annotation on the personnel behavior image data, including behavior subject identity information, scene type label, time series label and multi-target positioning information; The dynamic behavior feature verification module dynamically verifies the collected personnel behavior image data, calculates and evaluates the image verification factor Bcz by fitting the behavior image resolution Rf, frame rate Rz, illumination value Rg and background noise value Rb in the personnel behavior image data; then, based on the personnel behavior image data, cluster analysis and pattern recognition are used to extract potential abnormal behavior features to form a dynamic behavior feature data set; The behavior preprocessing module optimizes the dynamic behavior feature data set, accurately extracts the behavior subject feature area through the multi-target segmentation technology, and constructs the multi-target behavior feature data set and the scene adaptation data set respectively; The multi-target online analysis module is used to construct and use an improved deep learning model to perform real-time analysis on a multi-target behavior feature dataset and a scene adaptation dataset, and extract a behavior abnormality factor Ax and a target tracking factor Tg; The adaptive comprehensive evaluation module is used to construct the platform data fusion factor Pf, and associate the abnormal behavior factor Ax and the target tracking factor Tg with the platform data fusion factor Pf, calculate and evaluate the comprehensive behavior risk index Rz, and generate a behavior analysis and intervention strategy plan.

2. The AI ​​vision-based personnel behavior analysis and management system according to claim 1, characterized in that: The behavior image acquisition module monitors the target scene in real time by deploying high-definition video equipment, infrared sensors and depth cameras in the monitoring area, and continuously collects personnel behavior image data; during the acquisition process, the video image data is first automatically framed and sampled to extract key behavior images; Then, face recognition, posture estimation and target detection algorithms are used to identify and classify the subject and obtain the subject's identity information. Combined with the data fed back by the scene environment sensors, the scene type is automatically identified and the scene type label is generated. Secondly, accurate time series tags are added to each frame of image data. At the same time, multi-target detection and SORT tracking algorithms are used to perform real-time positioning and trajectory tracking of multiple human behavior subjects in the same scene to obtain multi-target positioning information. Finally, the collected human behavior image data and annotation information are uniformly formatted and stored.

3. The AI ​​vision-based personnel behavior analysis and management system according to claim 2 is characterized in that: The dynamic behavior feature verification module includes a data integrity quantification unit, a data integrity evaluation unit and a dynamic feature analysis unit; The data integrity quantification unit is used to fit the behavioral image resolution Rf, frame rate Rz, illumination value Rg and background noise value Rb, and calculate the image verification factor Bcz; the specific calculation formula is as follows: 。 4. The AI ​​vision-based personnel behavior analysis and management system according to claim 3 is characterized in that: The data integrity evaluation unit is used to preset the image verification threshold Q and compare and evaluate it with the image verification factor Bcz to verify the integrity and consistency of the image data. The specific evaluation contents are as follows: When the image verification factor Bcz ≥ the image verification threshold Q, the personnel behavior image data is determined to be complete, indicating that the image resolution, frame rate, lighting conditions and background noise meet the system requirements; When the image verification factor Bcz is less than the image verification threshold Q, it is determined that the personnel behavior image data is incomplete, and there are problems such as image blur, low frame rate, insufficient lighting, and noise interference. At this time, the data re-collection or quality repair process is triggered.

5. The AI ​​vision-based personnel behavior analysis and management system according to claim 4 is characterized in that: The dynamic feature analysis unit first pre-processes the complete personnel behavior image data, uses noise reduction and image enhancement algorithms to adjust the image quality, and then uses a multi-dimensional feature extraction algorithm to extract key behavior feature parameters from the personnel behavior image data, including posture changes, motion trajectories, speed changes, and behavior pattern sequences of the behavior subject; then, a clustering analysis algorithm is used to perform unsupervised classification on the extracted key behavior feature parameters, cluster similar behaviors into categories, and identify behavior patterns that deviate from or are undefined in the existing behavior definition database; the clustering results are further analyzed and matched by a pattern recognition algorithm, and feature supplementation and differentiation analysis are performed on suspected abnormal behaviors to extract potential abnormal behavior features; Finally, the extracted potential abnormal behavior features are classified and organized according to behavior category, time series and scenario type to construct a dynamic behavior feature dataset.

6. The AI ​​vision-based personnel behavior analysis and management system according to claim 5, characterized in that: The behavior preprocessing module optimizes the dynamic behavior feature data set, firstly adopts a multi-scale denoising algorithm to adaptively eliminate background noise and environmental interference factors in the image data, including a joint filtering technology based on the spatial domain and the frequency domain, to suppress the interference of light fluctuations, dynamic background changes and random noise on the behavior feature extraction; After denoising, the multi-target segmentation technology is used to accurately segment multiple behavioral subjects in the image data. The deep learning segmentation model and target detection algorithm are used to segment the contours, postures and motion trajectories of people, and to distinguish different behavioral subjects in complex scenes. After the segmentation is completed, the dynamic parameters including the behavior amplitude, speed and angle of the behavior subject are standardized to eliminate the feature deviation caused by changes in camera angle, distance and lighting; finally, based on the individual characteristics of the behavior subject and the scene environment attributes, the processed data are constructed into a multi-target behavior feature dataset and a scene adaptation dataset.

7. The AI ​​vision-based personnel behavior analysis and management system according to claim 6, characterized in that: The multi-target online analysis module builds a behavior analysis framework based on an improved deep learning model; wherein the improved deep learning model jointly extracts the spatial and temporal dimension features of personnel behavior by fusing a convolutional neural network with a spatiotemporal graph convolutional network; During the analysis process, the input multi-target behavior feature dataset is first modeled into a behavior sequence to capture the dynamic changes of the behavior subject in a specific scenario. Then, the attention mechanism is used to enhance the model's sensitivity to the key behavior features in the multi-target behavior feature dataset, and deep learning analysis is performed on the behavior subject's posture changes, motion trajectories, and interaction relationships to extract the behavioral anomaly factor Ax related to the deviation from the behavior pattern. At the same time, a multi-target tracking algorithm and a global data association strategy are used to track and dynamically update the location information and trajectories of multiple behavior subjects in complex scenarios in real time to extract the target tracking factor Tg.

8. The AI ​​vision-based personnel behavior analysis and management system according to claim 7, characterized in that: The adaptive comprehensive assessment module includes a risk assessment unit and a strategy optimization unit; The risk assessment unit is used to calculate the comprehensive behavior risk index Rz; after extracting the abnormal behavior factor Ax, the target tracking factor Tg and the platform data fusion factor Pf and performing dimensionless processing, the comprehensive behavior risk index Rz is obtained using the following formula: 。 9. The AI ​​vision-based personnel behavior analysis and management system according to claim 8, characterized in that: The strategy optimization unit is used to evaluate the comprehensive behavior risk index Rz and generate a behavior analysis and intervention strategy plan; By presetting the comprehensive first behavior risk threshold W1 and the second behavior risk threshold W2, the comprehensive behavior risk index Rz is compared, and the first behavior risk threshold W1 is less than the second behavior risk threshold W2. The specific contents are as follows: When the comprehensive behavior risk index Rz is less than the first behavior risk threshold W1, it is judged as normal behavior, and the system records the behavior data for model optimization without triggering any intervention measures; When the first behavior risk threshold W1≤comprehensive behavior risk index Rz<second behavior risk threshold W2, it is judged as a potential risk behavior, the system issues a risk warning, and focuses on monitoring the target; The specific behavior analysis and intervention strategy plan is: include the target behavior in the key monitoring objects, adjust the camera angle or resolution; increase the sampling frequency of behavior data, and track the behavior dynamics in real time; Send risk warning information to the management terminal for reference by security personnel; When the comprehensive behavior risk index Rz ≥ the second behavior risk threshold W2, it is judged as a high-risk behavior, and the emergency intervention strategy is immediately implemented and the alarm system is linked; The specific behavior analysis and intervention strategy is as follows: triggering sound and light alarm devices or broadcasting warnings; automatically linking the security system to physically isolate or block the behavior subject; sending emergency alarms and target location information to security personnel and management centers, and pushing live video images in real time; Call up historical behavior data to analyze the potential risk level of the behavior subject and provide a basis for security decision-making.

10. A personnel behavior analysis and management method based on AI vision, applied to a personnel behavior analysis and management system based on AI vision according to any one of claims 1 to 9, characterized in that: The following steps are involved: Step 1: Collect human behavior image data in real time, including video clips and image frames; then collect and combine scene environment information to perform multi-dimensional annotation on human behavior image data, including behavior subject identity information, scene type label, time series label and multi-target positioning information; Step 2: Dynamically verify the collected personnel behavior image data. By fitting the behavior image resolution Rf, frame rate Rz, illumination value Rg and background noise value Rb in the personnel behavior image data, calculate and evaluate the image verification factor Bcz to verify the integrity and consistency of the image data; then, based on the personnel behavior image data, cluster analysis and pattern recognition are used to extract potential abnormal behavior features to form a dynamic behavior feature data set; Step 3: Optimize the dynamic behavior feature dataset, use a multi-scale denoising algorithm to eliminate background interference, accurately extract the behavior subject feature area through multi-target segmentation technology, realize independent extraction and normalization of multi-target behavior features, and finally construct a multi-target behavior feature dataset and a scene adaptation dataset respectively; Step 4: Build and use an improved deep learning model to perform real-time analysis on multi-target behavior feature datasets and scene adaptation datasets, and extract behavior abnormality factors Ax and target tracking factors Tg; Step 5: Construct the platform data fusion factor Pf, and associate the abnormal behavior factor Ax and the target tracking factor Tg with the platform data fusion factor Pf, calculate and evaluate the comprehensive behavior risk index Rz, and generate a behavior analysis and intervention strategy plan.

Images