Intelligent abnormal behavior early warning method for high-altitude operation based on human posture recognition and electronic device

Abstract

This invention provides an intelligent early warning method and electronic device for abnormal behavior in high-altitude operations based on human posture recognition, belonging to the field of intelligent early warning. This method can not only automatically extract the behavioral characteristics of workers, but also automatically identify and warn of abnormal behavior through a deep learning model, significantly improving the efficiency and accuracy of high-altitude operation safety monitoring. By introducing an improved HDBSCAN algorithm, the model can identify and cluster abnormal behaviors in real time in a dynamically changing work environment. Simultaneously, combined with human posture recognition technology, it can more accurately capture the dynamic posture changes of workers, thereby achieving early warning of potentially dangerous behaviors. This method overcomes the limitations of manual identification and slow response in traditional high-altitude operation safety monitoring methods, providing an efficient, accurate, and innovative solution for intelligent safety management of high-altitude operations.

Description

Technical Field

[0001] This invention relates to the field of intelligent early warning, specifically to an intelligent early warning method and electronic device for abnormal behavior in high-altitude operations based on human posture recognition. Background Technology

[0002] With the ever-increasing demands for safety management in high-altitude operations, traditional methods for monitoring and issuing early warnings of abnormal behavior are facing unprecedented challenges. In the field of modern safety management, the detection and early warning of abnormal behavior is not only an important means of ensuring operational safety but also a key factor in improving operational efficiency. However, current methods for detecting abnormal behavior in high-altitude operations mostly rely on manual monitoring or traditional rule-based methods, which are cumbersome and inefficient. Traditional methods for detecting abnormal behavior typically depend on the experience and observation of safety officers, using manual annotation or template-based approaches for abnormal behavior identification and early warning. While these methods can meet the needs in certain situations, their lack of real-time performance and automated processing often fails to meet the rapidly changing safety requirements in high-altitude work environments.

[0003] The limitation of traditional abnormal behavior detection lies in its slow response to behavior recognition and early warning. Existing methods often rely on fixed rules and manual intervention to identify and alert on abnormal behavior. This approach is not only time-consuming but also struggles to capture complex dynamic changes in the work environment in real time. When dealing with complex high-altitude work environments and changes in worker posture and behavior, traditional methods have very limited performance and adaptability. Especially given the complexity and diversity of high-altitude operations, manual monitoring and traditional methods lack efficient processing tools, resulting in inadequate accuracy and real-time performance in safety warnings.

[0004] Furthermore, traditional methods for detecting abnormal behavior often overlook the dynamic postures of workers and the complexity of the environment, making them difficult to adapt to different work scenarios and behavioral patterns. For example, in human posture recognition and behavior analysis, traditional methods cannot effectively capture action details or identify potential dangerous behaviors, leading to alarm triggering delays or even missed detections of dangerous behaviors. Even when machine learning techniques are used for behavior recognition, they often fail to fully utilize the potential information in multimodal data, causing the model to be unable to capture the complex behavioral patterns of workers and changes in the work environment, thus affecting the accuracy and response speed of the early warning system.

[0005] Therefore, how to achieve intelligent early warning of abnormal behavior in high-altitude operations based on modern deep learning technology, combined with isolated forest models and human posture recognition technology, has become an urgent problem to be solved in the field of high-altitude operation safety. Summary of the Invention

[0006] The purpose of this invention is to propose an intelligent early warning method and electronic device for abnormal behavior in high-altitude operations based on human posture recognition. By combining an improved HDBSCAN algorithm and an isolated forest model, it can more accurately detect and identify abnormal behaviors in high-altitude operations. This method can not only automatically extract the behavioral characteristics of workers, but also automatically identify and warn of abnormal behaviors through a deep learning model, significantly improving the efficiency and accuracy of high-altitude operation safety monitoring. By introducing the improved HDBSCAN algorithm, the model can identify and cluster abnormal behaviors in real time in a dynamically changing work environment. Simultaneously, combined with human posture recognition technology, it can more accurately capture the dynamic posture changes of workers, thereby achieving early warning of potentially dangerous behaviors. This method overcomes the limitations of manual identification and slow response in traditional high-altitude operation safety monitoring methods, providing an efficient, accurate, and innovative solution for intelligent safety management of high-altitude operations.

[0007] To achieve the above objectives, this invention proposes an intelligent early warning method for abnormal behavior in high-altitude operations based on human posture recognition, comprising the following steps:

[0008] S1. Collect images of the posture and behavior of workers at height and perform preprocessing. S2. Apply the HRNet algorithm to parse the preprocessed image data and extract high-precision pose information; S3. Construct an isolated forest model by inputting the training dataset and continuously training and iterating the isolated forest model. S4. Input high-precision attitude information into the isolated forest model, extract feature data for anomaly detection, and generate an abnormal behavior dataset. S5. Use the improved HDBSCAN algorithm to perform cluster analysis on the abnormal behavior dataset and extract the abnormal behavior cluster dataset. S6. Set safety standards for high-altitude operations, check abnormal behavior cluster datasets, extract dangerous behavior information, issue alarm signals, and upload dangerous behavior information to the remote dispatch center. S7. Based on information about hazardous behavior, the remote dispatch center formulates emergency measures, notifies on-site workers and on-site managers to implement emergency response, and monitors and adjusts the situation in real time until the risk is eliminated. S8. Record information on dangerous behaviors and effective handling measures, combine historical data analysis, continuously train and iterate the isolated forest model, update emergency response plans, and provide feedback to the field.

[0009] As a preferred embodiment, step S1 specifically includes: S11. Use video surveillance equipment to collect real-time image data of workers at height; S12. Perform time synchronization processing on the image data to obtain time-synchronized image data; S13. Spatially align the time-synchronized image data to obtain spatially aligned image data; S14. Perform image enhancement processing on the spatially aligned image data, adjusting the brightness, contrast, and sharpness of the image to obtain the enhanced image data; S15. Cropping the enhanced image data retains the key areas of the workers and removes the background and irrelevant areas to obtain the preprocessed image data.

[0010] As a preferred embodiment, step S2 specifically includes: S21. Input the preprocessed image data into the HRNet model for feature extraction and image parsing. HRNet extracts features from the image through a multi-layer convolutional neural network, fuses and calculates information at different scales, obtains high-quality spatial features, and generates multi-scale feature maps. S22. Based on multi-scale feature maps, each key point in the image is detected through the terminal network layer of HRNet, and its coordinates in the image are regressed. The HRNet model outputs a probability map of each key point, indicating whether each pixel is a possible location of the key point. Then, the precise location is extracted through the non-maximum suppression algorithm, and the spatial location information of each part of the body is extracted to obtain the coordinates of the key point. S23. Based on the coordinates of key points, combine the coordinates of each key point, analyze the connection relationship of key parts and the angle changes between joints, construct a skeleton model, and obtain high-precision posture information of high-altitude workers. S24. Based on high-precision posture information of high-altitude workers, noise data is removed by using a sliding window algorithm, and different postures are classified to obtain a high-precision posture dataset.

[0011] As a preferred embodiment, step S3 specifically includes: S31. Prepare the training dataset, initialize the key parameters of the isolated forest, and set the initial values ​​for the number of random trees, the initial value for the data subset size, the initial value for the proportion of outlier data, and the initial value for the number of split features. S32. Based on the initial value of the data subset capacity, randomly select a data subset from the training dataset. Each time the tree splits, randomly select features from the dataset to reduce the amount of computation and introduce randomness. Based on the randomly selected features and the initial value of the number of split features, recursively split the data points until each data point is isolated to a leaf node. Repeat the above process until the initial value of the number of random trees is reached to build an isolated forest model. S33. Input the training dataset into the isolated forest model, calculate the path length and average path length of each data point in all isolated trees, compare the average path length with the expected path length of the tree, and calculate the anomaly score. S34. Based on the initial value of the abnormal data ratio, set an abnormal threshold and mark data points with abnormal scores exceeding the abnormal threshold as abnormal data; S35. Check the effectiveness of identifying abnormal data. Evaluate the performance of the isolated forest model based on accuracy, recall, and F1 score, and continuously train and iterate the isolated forest model.

[0012] As a preferred embodiment, step S4 specifically includes: S41. Analyze high-precision posture information, extract the coordinate information of each joint of the body from the posture data, calculate the angle of each joint of the body based on the key point coordinates, generate posture feature vectors based on the extracted joint coordinates and angle information, and integrate them to obtain the posture feature dataset. S42. Input the pose feature dataset into the isolated forest model, perform anomaly detection on the pose feature data, and calculate the anomaly score for each pose behavior. S43. Determine the abnormal score based on the abnormal threshold, identify postures and behaviors that do not meet the standards for high-altitude operations, integrate all non-standard postures and behaviors, and obtain an abnormal behavior dataset.

[0013] As a preferred embodiment, step S5 specifically includes: S51. Standardize and normalize the abnormal behavior dataset, and use the PCA method to reduce the dimensionality to obtain a dimensionality-reduced dataset. S52. Based on the dimensionality reduction dataset, the k value is adaptively selected through local density estimation, and the k-nearest neighbor distance of each point is calculated to obtain the k-distance map; S53. Based on the k-distance graph, calculate the core distance of each data point and construct the minimum spanning tree; S54. Based on the minimum spanning tree, calculate the stability value of each cluster; S55. Use stability values ​​to prune the minimum spanning tree, set a stability threshold, and remove clusters that are below the stability threshold; S56. Based on the minimum spanning tree, dynamically adjust the clustering parameters MinPts and ε, and use a weighted distance metric to optimize the clustering effect to obtain the clustering results; S57. Based on the clustering results, merge small clusters and split unstable clusters to obtain the final cluster labels and clustering results, and extract the abnormal behavior clustering dataset.

[0014] As a preferred embodiment, step S6 specifically includes: S61. Set the safety criteria for high-altitude work behavior, including whether the work behavior exceeds the set duration, whether the intensity of the work behavior exceeds the safety threshold, whether the work behavior enters a high-risk area, whether the work behavior involves improper use of equipment, and whether the work behavior occurs in an unsafe time window. S62. Use the safety regulations for high-altitude operations to check the abnormal behavior cluster dataset, and determine whether each abnormal behavior exceeds the preset standard range. If so, extract the dangerous behavior information. S63. Based on the dangerous behavior information, issue a warning signal and upload the dangerous behavior information to the remote dispatch center.

[0015] As a preferred embodiment, step S7 specifically includes: S71. The remote dispatch center determines the type of behavior based on the dangerous behavior information, assesses the risk level of the abnormal behavior based on the behavior type, and judges whether the risk threshold exceeds the safety standard threshold. If so, it is classified as a high-risk behavior. S72. When a dangerous behavior is determined to be a high-risk behavior, the remote dispatch center shall formulate emergency measures and notify on-site workers and on-site managers to implement emergency handling and evacuation and survival measures. S73. The remote dispatch center continuously monitors dangerous behaviors and adjusts emergency measures based on real-time changes in dangerous behavior information until the risk is eliminated.

[0016] As a preferred embodiment, step S8 specifically includes: S81. After the risk is eliminated, record the information on the dangerous behaviors of the on-site workers and the effective handling measures, and store them in the remote dispatch center; S82. Based on the hazardous behavior information of on-site workers and combined with historical data analysis, adjust the parameters of the isolated forest model and continuously train and iterate the isolated forest model. S83. Based on effective risk mitigation measures and historical data analysis, update best practice solutions for emergency response and provide feedback to on-site operators and managers.

[0017] Furthermore, the present invention also proposes an electronic device, which includes a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, it implements the above-mentioned intelligent early warning method for abnormal behavior in high-altitude operations based on human posture recognition.

[0018] Compared with the prior art, the present invention has at least the following technical effects: (1) This invention, by combining the isolated forest model and the improved HDBSCAN algorithm, successfully overcomes the shortcomings of traditional methods for detecting abnormal behavior in high-altitude operations when dealing with complex behavioral patterns. Traditional methods typically rely on rule-based behavior recognition techniques, which often perform poorly when handling complex actions and behaviors of workers. By introducing the isolated forest model, this invention can efficiently identify abnormal behavior and accurately distinguish between normal and abnormal behavior through the model's anomaly scoring mechanism. In particular, it significantly improves the accuracy of abnormal behavior detection when dealing with highly dynamic and complex high-altitude operation environments. The improved HDBSCAN algorithm optimizes the clustering effect for abnormal behavior by adaptively adjusting clustering parameters and conducting multi-scale stability assessments. This addresses the shortcomings of traditional methods in handling diverse work behaviors and areas of density variation, thereby improving the performance of the high-altitude operation safety early warning system. This method provides an efficient and accurate intelligent early warning solution for high-altitude operation safety, enabling real-time monitoring, identification, and timely response to dangerous behaviors in various work environments, thus promoting the advancement of high-altitude operation safety management technology.

[0019] (2) This invention achieves accurate detection and early warning of abnormal behavior of workers at heights by combining the isolated forest model, the improved HDBSCAN algorithm, and human posture recognition technology. The working environment at heights typically features complex spatial layouts and dynamic postures of workers, making it difficult for traditional methods to capture subtle changes in their behavior. Through human posture recognition technology, this invention can acquire key point location information of workers in real time and perform behavioral analysis using the isolated forest model, thereby accurately identifying dangerous behavior patterns. Furthermore, the improved HDBSCAN algorithm can effectively handle various behavioral expressions of workers, automatically identify abnormal clustering behaviors, and eliminate low-stability clusters, ensuring the accuracy and stability of the system. Through this method, abnormal behaviors in high-altitude operations can be identified and warned of in a timely manner, effectively protecting the safety of workers. Attached Figure Description

[0020] Figure 1 This is a flowchart of the intelligent early warning method for abnormal behavior in high-altitude operations based on human posture recognition, according to the present invention.

[0021] Figure 2 This is a flowchart illustrating the execution of the improved HDBSCAN algorithm in an embodiment of the present invention. Detailed Implementation

[0022] In the following description, numerous specific details are set forth in order to provide a more thorough understanding of the invention. However, it will be apparent to those skilled in the art that the invention can be practiced without one or more of these details. In other instances, certain technical features well-known in the art have not been described in order to avoid obscuring the invention.

[0023] The invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0024] refer to Figure 1 This embodiment discloses an intelligent early warning method for abnormal behavior in high-altitude operations based on human posture recognition, including the following steps: S1. Collect images of the posture and behavior of workers at height and perform preprocessing. S2. Apply the HRNet algorithm to parse the preprocessed image data and extract high-precision pose information; S3. Construct an isolated forest model by inputting the training dataset and continuously training and iterating the isolated forest model. S4. Input high-precision attitude information into the isolated forest model, extract feature data for anomaly detection, and generate an abnormal behavior dataset. S5. Use the improved HDBSCAN algorithm to perform cluster analysis on the abnormal behavior dataset and extract the abnormal behavior cluster dataset. S6. Set safety standards for high-altitude operations, check abnormal behavior cluster datasets, extract dangerous behavior information, issue alarm signals, and upload dangerous behavior information to the remote dispatch center. S7. Based on information about hazardous behavior, the remote dispatch center formulates emergency measures, notifies on-site workers and on-site managers to implement emergency response, and monitors and adjusts the situation in real time until the risk is eliminated. S8. Record information on dangerous behaviors and effective handling measures, combine historical data analysis, continuously train and iterate the isolated forest model, update emergency response plans, and provide feedback to the field.

[0025] In this embodiment, S1 specifically includes: S11. Use video surveillance equipment to collect real-time image data of workers at height; S12. Perform time synchronization processing on the image data to obtain time-synchronized image data; S13. Spatially align the time-synchronized image data to obtain spatially aligned image data; S14. Perform image enhancement processing on the spatially aligned image data, adjusting the brightness, contrast, and sharpness of the image to obtain the enhanced image data; S15. Cropping the enhanced image data retains the key areas of the workers and removes the background and irrelevant areas to obtain the preprocessed image data.

[0026] This implementation method employs multi-step preprocessing of real-time image data of high-altitude workers to ensure image quality and accurate capture of key body parts. Time synchronization processing ensures the temporal consistency of the image data, providing an accurate time reference for subsequent processing. Spatial alignment of the time-synchronized image data ensures consistent position and angle of the workers within the images, facilitating subsequent analysis. Image enhancement processing adjusts the brightness, contrast, and sharpness of the images, making details clearer and enhancing visual appeal. Cropping operations preserve key areas of the workers while removing background and irrelevant areas, resulting in the final preprocessed image data, providing high-quality input data for subsequent analysis and decision-making.

[0027] In this embodiment, S2 specifically includes: S21. Input the preprocessed image data into the HRNet model for feature extraction and image parsing. HRNet extracts features from the image through a multi-layer convolutional neural network, fuses and calculates information at different scales, obtains high-quality spatial features, and generates multi-scale feature maps. S22. Based on multi-scale feature maps, each key point in the image is detected through the terminal network layer of HRNet, and its coordinates in the image are regressed. The HRNet model outputs a probability map of each key point, indicating whether each pixel is a possible location of the key point. Then, the precise location is extracted through the non-maximum suppression algorithm, and the spatial location information of each part of the body is extracted to obtain the coordinates of the key point. S23. Based on the coordinates of key points, combine the coordinates of each key point, analyze the connection relationship of key parts and the angle changes between joints, construct a skeleton model, and obtain high-precision posture information of high-altitude workers. S24. Based on high-precision posture information of high-altitude workers, noise data is removed by using a sliding window algorithm, and different postures are classified to obtain a high-precision posture dataset.

[0028] This implementation uses the HRNet model to process preprocessed image data in multiple steps to accurately extract the posture information of high-altitude workers. The preprocessed image data is input into the HRNet model for feature extraction and image parsing. HRNet uses a multi-layer convolutional neural network to extract features from the image, fusing and calculating information at different scales to obtain high-quality spatial features and generate multi-scale feature maps. Through multi-scale feature extraction and fusion, HRNet can accurately capture details and global information in the image at different levels, thus providing rich feature representations for subsequent posture estimation. Based on the multi-scale feature maps, each keypoint in the image is detected through the terminal network layer of HRNet, and its coordinates in the image are regressed. The HRNet model outputs a probability map for each keypoint, indicating whether each pixel is a possible location of that keypoint. Then, a non-maximum suppression algorithm is used to extract the precise location, extracting the spatial position information of each body part to obtain the keypoint coordinates. By accurately detecting and regressing keypoint coordinates, HRNet can efficiently locate the position of each keypoint, and the non-maximum suppression algorithm optimizes the positioning accuracy, ensuring the acquisition of high-precision spatial information of body parts. Based on key point coordinates, the coordinates of each key point are combined to analyze the connection relationships of key parts and the angular changes between joints, constructing a skeleton model to obtain high-precision posture information of aerial workers. Through the calculation of joint angles and the construction of the skeleton model, the worker's posture can be clearly depicted, providing high-precision posture information, which is helpful for the detection and analysis of abnormal behavior. Based on the high-precision posture information of aerial workers, a sliding window algorithm is used to remove noisy data and classify different postures to obtain a high-precision posture dataset. Through sliding window denoising and posture classification, invalid data is removed and different posture information of workers is accurately extracted, generating a high-precision posture dataset to ensure the accuracy and reliability of subsequent analysis and monitoring results.

[0029] In this embodiment, S3 specifically includes: S31. Prepare the training dataset, initialize the key parameters of the isolated forest, and set the initial values ​​for the number of random trees, the initial value for the data subset size, the initial value for the proportion of outlier data, and the initial value for the number of split features. S32. Based on the initial value of the data subset capacity, randomly select a data subset from the training dataset. Each time the tree splits, randomly select features from the dataset to reduce the amount of computation and introduce randomness. Based on the randomly selected features and the initial value of the number of split features, recursively split the data points until each data point is isolated to a leaf node. Repeat the above process until the initial value of the number of random trees is reached to construct the isolated forest model. S33. Input the training dataset into the isolated forest model, calculate the path length and average path length of each data point in all isolated trees, compare the average path length with the expected path length of the tree, and calculate the anomaly score. S34. Based on the initial value of the abnormal data ratio, set an abnormal threshold and mark data points with abnormal scores exceeding the abnormal threshold as abnormal data; S35. Check the effectiveness of identifying abnormal data. Evaluate the performance of the isolated forest model based on accuracy, recall, and F1 score, and continuously train and iterate the isolated forest model.

[0030] This implementation uses the Isolation Forest algorithm to detect abnormal behavior in a dataset, achieving efficient and accurate anomaly data identification. The training dataset is prepared, and key parameters of the Isolation Forest are initialized, including initial values ​​for the number of random trees, the size of the data subset, the proportion of abnormal data, and the number of splitting features. Initializing these key parameters provides a suitable starting configuration for subsequent training, ensuring the model can effectively process data and identify abnormal behavior. Based on the initial data subset size, a subset of data is randomly selected from the training dataset. Each time a tree splits, features are randomly selected from the dataset, reducing computation and introducing randomness. Data points are recursively split based on the randomly selected features and the initial number of splitting features until each data point is isolated to a leaf node. This process is repeated until the initial number of random trees is reached, constructing the Isolation Forest model. By introducing randomness and recursive splitting, the Isolation Forest effectively captures anomalies in the data while reducing computation, improving training efficiency and model robustness.

[0031] The training dataset is input into the Isolation Forest model. The path length and average path length of each data point across all isolated trees are calculated. The average path length is compared to the expected path length of the tree to calculate anomaly scores. By calculating the path length and average path length of each data point, the Isolation Forest can assess the degree of anomalousness of each point, thus accurately identifying anomalous behavior. Based on an initial value for the proportion of anomalous data, an anomaly threshold is set, and data points with anomaly scores exceeding the threshold are marked as anomalous data. By setting the anomaly threshold, points with high anomaly scores are identified as anomalous data, effectively filtering out potential anomalous behaviors and supporting subsequent processing and analysis. The effectiveness of anomalous data identification is checked. The performance of the Isolation Forest model is evaluated based on accuracy, recall, and F1 score, and the model is continuously trained and iterated. By using evaluation metrics such as accuracy, recall, and F1 score, the efficiency and accuracy of the Isolation Forest model in anomalous behavior identification are ensured, and model performance is optimized through iterative training.

[0032] In this embodiment, S4 specifically includes: S41. Analyze high-precision posture information, extract the coordinate information of each joint of the body from the posture data, calculate the angle of each joint of the body based on the key point coordinates, generate posture feature vectors based on the extracted joint coordinates and angle information, and integrate them to obtain the posture feature dataset. S42. Input the pose feature dataset into the isolated forest model, perform anomaly detection on the pose feature data, and calculate the anomaly score for each pose behavior. S43. Determine the abnormal score based on the abnormal threshold, identify postures and behaviors that do not meet the standards for high-altitude operations, integrate all non-standard postures and behaviors, and obtain an abnormal behavior dataset.

[0033] This implementation achieves intelligent detection of abnormal behavior in high-altitude workers by analyzing high-precision posture information and applying an isolated forest model. High-precision posture information is analyzed to extract the coordinates of each joint in the body. Based on keypoint coordinates, the angles of each joint are calculated. Posture feature vectors are generated based on the extracted joint coordinates and angles, and these are integrated to obtain a posture feature dataset. By extracting and integrating keypoint coordinates and joint angle information, a complete posture feature vector is constructed, ensuring accurate representation of the high-altitude worker's posture and providing foundational data for subsequent behavior analysis. The posture feature dataset is then input into an isolated forest model to perform anomaly detection, calculating anomaly scores for each posture behavior.

[0034] Anomaly detection of posture features is performed using an isolated forest model, and anomaly scores are evaluated for each posture, effectively identifying behaviors that significantly differ from normal postures. Anomaly scores are determined based on anomaly thresholds to identify postures and behaviors that do not meet high-altitude operation standards. All non-compliant postures and behaviors are then integrated to obtain an abnormal behavior dataset. By setting anomaly score threshold, postures and behaviors that do not meet safety operation standards are precisely filtered out, forming an abnormal behavior dataset that facilitates further analysis and safety early warning.

[0035] In this embodiment, see Figure 2 The improved HDBSCAN algorithm specifically includes: The abnormal behavior dataset is standardized and normalized, and the PCA method is used to reduce the dimensionality to obtain a dimensionality-reduced dataset. Based on the dimensionality reduction dataset, the k value is adaptively selected through local density estimation, and the k-nearest neighbor distance of each point is calculated to obtain the k-distance map; Based on the k-distance graph, the core distance of each data point is calculated, and a minimum spanning tree is constructed. Based on the minimum spanning tree, the stability value of each cluster is calculated:

[0036] in, For the stability of cluster k, Let k be the minimum spanning tree of cluster k. For point and points The boundary weight between; Use stability values ​​to prune the minimum spanning tree, set a stability threshold, and remove clusters that are below the stability threshold; Based on the minimum spanning tree, the clustering parameters MinPts and ε are dynamically adjusted, and a weighted distance metric is used to optimize the clustering effect, yielding the following clustering results:

[0037] in, For data points and The weighted distance between them The weight of the k-th feature. and For data points and The value at the k-th feature, where d is the dimension of the feature; Based on the clustering results, small clusters are merged and unstable clusters are split to obtain the final cluster labels and clustering results, and the abnormal behavior clustering dataset is extracted.

[0038] This implementation uses an improved HDBSCAN algorithm to effectively process and identify anomalous behavior datasets. The anomalous behavior dataset is standardized and normalized, and dimensionality is reduced using PCA to obtain a dimensionality-reduced dataset. Standardization, normalization, and PCA dimensionality reduction simplify the data structure, preserve the main features of the data, and provide high-quality input data for subsequent clustering analysis. Based on the dimensionality-reduced dataset, a k-value is adaptively selected through local density estimation, and the k-nearest neighbor distance for each point is calculated to obtain a k-distance map. Adaptive selection of the k-value and calculation of the k-nearest neighbor distance ensure accurate construction of similarity maps under different local densities, providing a good data foundation for anomaly detection. Based on the k-distance map, the core distance for each data point is calculated, and a minimum spanning tree is constructed. By calculating the core distance and constructing the minimum spanning tree, the local structural relationships of the data are captured, optimizing the connections between data points and enhancing the effectiveness and accuracy of clustering. Based on the minimum spanning tree, the stability value of each cluster is calculated. Through multi-scale stability evaluation, each cluster is analyzed in depth to identify clusters with strong stability, ensuring accurate identification of anomalous behavior. The minimum spanning tree is pruned using stability values. A stability threshold is set, and clusters below the threshold are removed. Pruning and noise optimization remove irrelevant or low-quality clusters, improving the accuracy of anomaly detection and reducing false positives. Based on the minimum spanning tree, the clustering parameters MinPts and ε are dynamically adjusted, and a weighted distance metric is used to optimize the clustering effect, yielding the final clustering results. Dynamically adjusting the clustering parameters and weighted distance metric allows the model to adapt more flexibly to data changes, optimizing clustering performance and improving the ability to identify anomalies. Based on the clustering results, small clusters are merged and unstable clusters are split to obtain the final cluster labels and clustering results, extracting the anomaly behavior clustering dataset. Merging and splitting operations fine-tune the clustering results, ensuring the accuracy of the final anomaly behavior dataset and providing reliable data support for further analysis and applications.

[0039] In this embodiment, S6 specifically includes: S61. Set the safety criteria for high-altitude work behavior, including whether the work behavior exceeds the set duration, whether the intensity of the work behavior exceeds the safety threshold, whether the work behavior enters a high-risk area, whether the work behavior involves improper use of equipment, and whether the work behavior occurs in an unsafe time window. S62. Use the safety regulations for high-altitude operations to check the abnormal behavior cluster dataset, and determine whether each abnormal behavior exceeds the preset standard range. If so, extract the dangerous behavior information. S63. Based on the dangerous behavior information, issue a warning signal and upload the dangerous behavior information to the remote dispatch center.

[0040] This implementation method establishes safety standards for high-altitude operations and detects abnormal behaviors to promptly identify potential safety risks and ensure the safety of workers. The system sets judgment conditions for high-altitude operation safety standards, including whether the operation exceeds a set duration, whether the intensity of the action exceeds a safety threshold, whether the operation enters a high-risk area, whether there is improper use of equipment, and whether the operation occurs within an unsafe time window. By setting detailed safety standards, high-altitude operations are comprehensively monitored, ensuring accurate assessment of the safety of each operation. Using these judgment conditions, the system checks the abnormal behavior cluster dataset, determining whether each abnormal behavior exceeds a preset standard range; if so, hazardous behavior information is extracted. The system compares the abnormal behavior cluster dataset against the standard conditions, automatically identifying behaviors exceeding safety standards and extracting potential hazard information, providing data support for timely response. Based on the hazardous behavior information, a warning signal is issued, and the hazardous behavior information is uploaded to a remote dispatch center.

[0041] After identifying dangerous behavior, a warning signal is issued through the early warning system, and detailed information on the dangerous behavior is uploaded to the remote dispatch center to ensure that remote personnel can take timely intervention measures.

[0042] In this embodiment, S7 specifically includes: S71. The remote dispatch center determines the type of behavior based on the dangerous behavior information, assesses the risk level of the abnormal behavior based on the behavior type, and judges whether the risk threshold exceeds the safety standard threshold. If so, it is classified as a high-risk behavior. S72. When a dangerous behavior is determined to be a high-risk behavior, the remote dispatch center shall formulate emergency measures and notify on-site workers and on-site managers to implement emergency handling and evacuation and survival measures. S73. The remote dispatch center continuously monitors dangerous behaviors and adjusts emergency measures based on real-time changes in dangerous behavior information until the risk is eliminated.

[0043] This implementation method ensures safety and effectively addresses potential risks during high-altitude operations through real-time monitoring and emergency response of hazardous behaviors via a remote dispatch center. The remote dispatch center determines the type of hazardous behavior based on its information, assesses the risk level of the abnormal behavior based on the type, and determines whether the risk threshold exceeds the safety standard threshold; if so, it is classified as a high-risk behavior. By analyzing the type of hazardous behavior and assessing its risk level, the remote dispatch center can accurately determine whether a high-risk behavior exists and ensure timely implementation of appropriate safety measures. When a hazardous behavior is determined to be high-risk, the remote dispatch center formulates emergency measures and notifies on-site workers and management personnel to implement emergency handling and evacuation procedures. Upon confirmation of a high-risk behavior, the remote dispatch center can immediately formulate an emergency response plan and notify relevant personnel to ensure the safety of on-site workers and prevent further danger. The remote dispatch center continuously monitors hazardous behaviors and adjusts emergency measures based on real-time changes in hazardous behavior information until the risk is eliminated. Through real-time monitoring of hazardous behaviors and dynamic adjustment of emergency responses, the remote dispatch center can flexibly adjust emergency measures according to changes in the work site until the risk is completely eliminated.

[0044] In this embodiment, S8 specifically includes: S81. After the risk is eliminated, record the information on the dangerous behaviors of the on-site workers and the effective handling measures, and store them in the remote dispatch center; S82. Based on the hazardous behavior information of on-site workers and combined with historical data analysis, adjust the parameters of the isolated forest model and continuously train and iterate the isolated forest model. S83. Based on effective risk mitigation measures and historical data analysis, update best practice solutions for emergency response and provide feedback to on-site operators and managers.

[0045] This implementation method ensures continuous safety management and improvement of high-altitude operations by recording hazardous behavior information, optimizing the isolated forest model, and updating emergency response plans. After the risk is eliminated, the hazardous behavior information and effective handling measures of on-site workers are recorded and stored in a remote dispatch center. By recording and storing the handling information, the safety data of on-site operations is preserved, providing a basis for subsequent improvements and decisions. Based on the hazardous behavior information of on-site workers and combined with historical data analysis, the parameters of the isolated forest model are adjusted, and the isolated forest model is continuously trained and iterated. By continuously optimizing the isolated forest model, the ability to detect abnormal behaviors is improved, ensuring that the model maintains efficient identification in changing work environments. Based on the effective measures for risk elimination and combined with historical data analysis, the best practice emergency response plan is updated and fed back to on-site workers and on-site managers. By updating the emergency response plan and feedback mechanism, it is ensured that on-site personnel can quickly respond to potential future hazards, improving the effectiveness of operational safety management.

[0046] To verify the feasibility of this invention in high-altitude operation safety management, it was applied to an intelligent early warning system for abnormal behavior in high-altitude operations on a high-altitude operation safety monitoring platform (hereinafter referred to as "Platform A"). In traditional high-altitude operation safety monitoring systems, the system typically relies on rule-based behavior detection methods or traditional machine learning algorithms to identify abnormal behavior. These methods are not only inefficient but also lack accuracy when dealing with complex operating environments and diverse operational behaviors, making it difficult to respond to various emergencies in real time. To address these issues, Platform A decided to adopt the intelligent early warning method for abnormal behavior in high-altitude operations based on human posture recognition proposed in this invention.

[0047] During implementation, Platform A first uses human pose recognition technology to acquire real-time pose data of workers at height. By introducing an HRNet model for image feature extraction, Platform A can accurately extract the spatial position information of each joint of the worker and construct a skeleton model of the worker, generating high-precision pose information. This pose information is then input into an isolated forest model, and by calculating anomaly scores for each pose, possible abnormal behaviors are identified.

[0048] Platform A successfully implemented intelligent early warning for abnormal behaviors in high-altitude operations by combining the Isolation Forest model and the improved HDBSCAN algorithm. The Isolation Forest model can efficiently identify abnormal behaviors and distinguish between normal and abnormal behaviors by calculating anomaly scores. The improved HDBSCAN algorithm further optimizes its adaptability to density changes in the working environment, enabling it to accurately identify abnormal patterns in diverse operational behaviors.

[0049] During implementation, the technical team of Platform A discovered that, compared to traditional rule-based or simple deep learning-based abnormal behavior recognition systems, the method of this invention, based on human posture recognition, an isolated forest model, and an improved HDBSCAN algorithm, significantly improves the accuracy and efficiency of abnormal behavior recognition. Traditional methods typically rely on fixed rules for behavior determination, making them difficult to adapt to complex working environments. In contrast, the method of this invention ensures efficient execution of abnormal behavior recognition and early warning by adjusting the early warning strategy in real time. Through human posture recognition, Platform A can accurately capture changes in the posture of workers, identify potential dangerous behaviors, and issue timely warnings.

[0050] To further verify the effectiveness of this method, Platform A compared the method of this invention with traditional abnormal behavior detection methods. The comparison table is as follows: Table 1 Comparison of Intelligent Early Warning Methods for Abnormal High-Altitude Operation Behavior on Platform A

[0051] As shown in Table 1, with the application of the method of this invention, the abnormal behavior recognition accuracy of Platform A has increased from 80% to 94% compared to the traditional method, the abnormal behavior recognition efficiency has been greatly improved, and the recognition time has been shortened from 120 seconds to 45 seconds. The system response time has also been reduced by 60%, and the false recognition rate has been significantly reduced to only 5%. In addition, the method of this invention has significantly improved processing power, increasing the number of abnormal behaviors processed per second by 200%, and improving multi-task parallel processing capability by 21.4%.

[0052] Using the method of this invention, Platform A can more efficiently identify abnormal behaviors during high-altitude operations, issue timely safety warnings, reduce safety risks, and improve the accuracy of operational safety management. This method not only enhances the system's automation level and reduces manual intervention, but also significantly improves the stability and robustness of the abnormal behavior warning system, providing efficient technical support for large-scale high-altitude operation safety management.

[0053] The technical processes of the methods disclosed in the above embodiments can be implemented, in whole or in part, through software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another.

[0054] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0055] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for intelligent early warning of abnormal behavior in high-altitude operations based on human posture recognition, characterized in that, Includes the following steps: S1. Collect images of the posture and behavior of workers at height and perform preprocessing. S2. Use the HRNet model to parse the preprocessed image data and extract high-precision pose information; S3. Construct an isolated forest model by inputting the training dataset and continuously training and iterating the isolated forest model. S4. Input high-precision attitude information into the isolated forest model, extract feature data for anomaly detection, and generate an abnormal behavior dataset. S5. Use the HDBSCAN algorithm to perform cluster analysis on the abnormal behavior dataset and extract the abnormal behavior cluster dataset. S6. Set safety standards for high-altitude operations, check abnormal behavior cluster datasets, extract dangerous behavior information, issue alarm signals, and upload dangerous behavior information to the remote dispatch center. S7. Based on information about hazardous behavior, the remote dispatch center formulates emergency measures, notifies on-site workers and on-site managers to implement emergency response, and monitors and adjusts the situation in real time until the risk is eliminated. S8. Record information on dangerous behaviors and effective handling measures, combine historical data analysis, continuously train and iterate the isolated forest model, update emergency response plans, and provide feedback to the field.

2. The intelligent early warning method for abnormal behavior in high-altitude operations based on human posture recognition according to claim 1, characterized in that, Step S1 specifically includes: S11. Use video surveillance equipment to collect real-time image data of workers at height; S12. Perform time synchronization processing on the image data to obtain time-synchronized image data; S13. Spatially align the time-synchronized image data to obtain spatially aligned image data; S14. Perform image enhancement processing on the spatially aligned image data, adjusting the brightness, contrast, and sharpness of the image to obtain the enhanced image data; S15. Cropping the enhanced image data retains the key areas of the workers and removes the background and irrelevant areas to obtain the preprocessed image data.

3. The intelligent early warning method for abnormal behavior in high-altitude operations based on human posture recognition according to claim 1, characterized in that, Step S2 specifically includes: S21. Input the preprocessed image data into the HRNet model for feature extraction and image parsing. HRNet extracts features from the image through a multi-layer convolutional neural network, fuses and calculates information at different scales, obtains high-quality spatial features, and generates multi-scale feature maps. S22. Based on multi-scale feature maps, each key point in the image is detected through the terminal network layer of HRNet, and its coordinates in the image are regressed. The HRNet model outputs a probability map of each key point, indicating whether each pixel is a possible location of the key point. Then, the precise location is extracted through the non-maximum suppression algorithm, and the spatial location information of each part of the body is extracted to obtain the coordinates of the key point. S23. Based on the coordinates of key points, combine the coordinates of each key point, analyze the connection relationship of key parts and the angle changes between joints, construct a skeleton model, and obtain high-precision posture information of high-altitude workers. S24. Based on high-precision posture information of high-altitude workers, noise data is removed by using a sliding window algorithm, and different postures are classified to obtain a high-precision posture dataset.

4. The intelligent early warning method for abnormal behavior in high-altitude operations based on human posture recognition according to claim 1, characterized in that, Step S3 specifically includes: S31. Prepare the training dataset, initialize the key parameters of the isolated forest, and set the initial values ​​for the number of random trees, the initial value for the data subset size, the initial value for the proportion of outlier data, and the initial value for the number of split features. S32. Based on the initial value of the data subset capacity, randomly select a data subset from the training dataset. Each time the tree splits, randomly select features from the dataset to reduce the amount of computation and introduce randomness. Based on the randomly selected features and the initial value of the number of split features, recursively split the data points until each data point is isolated to a leaf node. Repeat the above process until the initial value of the number of random trees is reached to build an isolated forest model. S33. Input the training dataset into the isolated forest model, calculate the path length and average path length of each data point in all isolated trees, compare the average path length with the expected path length of the tree, and calculate the anomaly score. S34. Based on the initial value of the abnormal data ratio, set an abnormal threshold and mark data points with abnormal scores exceeding the abnormal threshold as abnormal data; S35. Check the effectiveness of identifying abnormal data. Evaluate the performance of the isolated forest model based on accuracy, recall, and F1 score, and continuously train and iterate the isolated forest model.

5. The intelligent early warning method for abnormal behavior in high-altitude operations based on human posture recognition according to claim 1, characterized in that, Step S4 specifically includes: S41. Analyze high-precision posture information, extract the coordinate information of each joint of the body from the posture data, calculate the angle of each joint of the body based on the key point coordinates, generate posture feature vectors based on the extracted joint coordinates and angle information, and integrate them to obtain the posture feature dataset. S42. Input the pose feature dataset into the isolated forest model, perform anomaly detection on the pose feature data, and calculate the anomaly score for each pose behavior. S43. Determine the abnormal score based on the abnormal threshold, identify postures and behaviors that do not meet the standards for high-altitude operations, integrate all non-standard postures and behaviors, and obtain an abnormal behavior dataset.

6. The intelligent early warning method for abnormal behavior in high-altitude operations based on human posture recognition according to claim 1, characterized in that, Step S5 specifically includes: S51. Standardize and normalize the abnormal behavior dataset, and use the PCA method to reduce the dimensionality to obtain a dimensionality-reduced dataset. S52. Based on the dimensionality reduction dataset, the k value is adaptively selected through local density estimation, and the k-nearest neighbor distance of each point is calculated to obtain the k-distance map; S53. Based on the k-distance graph, calculate the core distance of each data point and construct the minimum spanning tree; S54. Based on the minimum spanning tree, calculate the stability value of each cluster; S55. Use stability values ​​to prune the minimum spanning tree, set a stability threshold, and remove clusters that are below the stability threshold; S56. Based on the minimum spanning tree, dynamically adjust the clustering parameters MinPts and ε, and use a weighted distance metric to optimize the clustering effect to obtain the clustering results; S57. Based on the clustering results, merge small clusters and split unstable clusters to obtain the final cluster labels and clustering results, and extract the abnormal behavior clustering dataset.

7. The intelligent early warning method for abnormal behavior in high-altitude operations based on human posture recognition according to claim 1, characterized in that, Step S6 specifically includes: S61. Set the safety criteria for high-altitude work behavior, including whether the work behavior exceeds the set duration, whether the intensity of the work behavior exceeds the safety threshold, whether the work behavior enters a high-risk area, whether the work behavior involves improper use of equipment, and whether the work behavior occurs in an unsafe time window. S62. Use the safety regulations for high-altitude operations to check the abnormal behavior cluster dataset, and determine whether each abnormal behavior exceeds the preset standard range. If so, extract the dangerous behavior information. S63. Based on the dangerous behavior information, issue a warning signal and upload the dangerous behavior information to the remote dispatch center.

8. The intelligent early warning method for abnormal behavior in high-altitude operations based on human posture recognition according to claim 1, characterized in that, Step S7 specifically includes: S71. The remote dispatch center determines the type of behavior based on the dangerous behavior information, assesses the risk level of the abnormal behavior based on the behavior type, and judges whether the risk threshold exceeds the safety standard threshold. If so, it is classified as a high-risk behavior. S72. When a dangerous behavior is determined to be a high-risk behavior, the remote dispatch center shall formulate emergency measures and notify on-site workers and on-site managers to implement emergency handling and evacuation and survival measures. S73. The remote dispatch center continuously monitors dangerous behaviors and adjusts emergency measures based on real-time changes in dangerous behavior information until the risk is eliminated.

9. The intelligent early warning method for abnormal behavior in high-altitude operations based on human posture recognition according to claim 1, characterized in that, Step S8 specifically includes: S81. After the risk is eliminated, record the information on the dangerous behaviors of the on-site workers and the effective handling measures, and store them in the remote dispatch center; S82. Based on the hazardous behavior information of on-site workers and combined with historical data analysis, adjust the parameters of the isolated forest model and continuously train and iterate the isolated forest model. S83. Based on effective risk mitigation measures and historical data analysis, update best practice solutions for emergency response and provide feedback to on-site operators and managers.

10. An electronic device, characterized in that, include: Processor and memory storing computer program instructions; When the processor executes the computer program instructions, it implements the intelligent early warning method for abnormal behavior in high-altitude operations based on human posture recognition as described in any one of claims 1 to 9.

Images