Method and system for analyzing internal potential safety hazards of industrial park based on generic intelligent video self-learning
By performing double-stream segmentation of the historical safety hazard incident video data of industrial parks, analyzing the hidden danger heat map score and movement posture risk coefficient, and training the safety hazard trend analysis model, the problem of insufficient safety hazard analysis capabilities in industrial parks is solved and improved Early warning capability and risk management.
Patent Information
- Application Number
- CN202411198662.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-29
- Publication Date
- 2024-11-12
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The prior art is difficult to conduct effective correlation analysis in complex environments of industrial parks, resulting in poor safety hazard analysis capabilities.
By performing dual-stream segmentation of the historical safety hazard event video data of industrial parks, the hidden danger heat map scores and motion posture risk coefficients in the spatial and temporal flow data samples are analyzed, and the safety hazard trend analysis model is trained to generate the monitoring target area Safety hazard score.
Accurate assessment of real-time monitoring of regional safety hazards has been achieved, the early warning capabilities of industrial parks have been improved, and potential risks have been reduced.
Smart Images

Figure CN118942015A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of machine learning technology, and in particular to a method and system for analyzing internal safety hazards of an industrial park based on pan-intelligent video self-learning. Background Art
[0002] Safety hazards within industrial parks refer to potential risks that may lead to accidents, injuries or losses within the park. These hazards may include equipment failure, improper operation, environmental pollution, fire hazards, and personal safety issues. Identifying and managing these hazards is crucial to protecting worker safety, ensuring production continuity, and avoiding property losses.
[0003] Although the existing technology uses intelligent monitoring for target areas in industrial parks, it all relies on static identification of matching people or objects in the target area. However, due to the complex environment of the target area in the industrial park, it is difficult for the existing monitoring methods to conduct effective correlation analysis, resulting in poor safety hazard analysis capabilities in the industrial park. Summary of the invention
[0004] In order to solve the above technical problems, a method and system for analyzing internal safety hazards in industrial parks based on pan-intelligent video self-learning is provided. This technical solution solves the above-mentioned problem that although in the prior art, intelligent monitoring is adopted for the target area of the industrial park, it is all through static identification of matching people or objects in the target area. However, due to the complex environment of the target area of the industrial park, it is difficult for the existing monitoring means to conduct effective correlation analysis, resulting in poor ability to analyze safety hazards in the industrial park.
[0005] In order to achieve the above purpose, the technical solution adopted by the present invention is: A method for analyzing internal safety hazards in an industrial park based on pan-intelligent video self-learning, comprising: Based on the historical video data of potential safety hazards in industrial parks, the associated factors that lead to potential safety hazards are marked, and a sample of historical video data of potential safety hazards in industrial parks is established; Based on the video data samples of historical safety hazard events in the industrial park, dual-stream segmentation is performed to obtain spatial stream data samples and temporal stream data samples of historical safety hazard events in the industrial park; According to the spatial flow data samples of historical safety hazard events in the industrial park, the heat map scores of the target area hazards in the spatial flow data samples of historical safety hazard events are analyzed; According to the historical safety hazard event time stream data samples of the industrial park, the risk coefficient of the target object motion posture in the historical safety hazard event time stream data samples is analyzed; The target object motion posture risk coefficient in the historical safety hazard event time stream data sample is synchronously compensated for the target object hazard heat map score in the historical safety hazard event spatial stream data sample, and the internal safety hazard trend analysis model of the industrial park is trained to generate the safety hazard score of the monitoring target area of the industrial park.
[0006] Preferably, according to the historical safety hazard event spatial stream data samples of the industrial park, analyzing the target area hazard heat map scores in the historical safety hazard event spatial stream data samples specifically includes: Based on the two-stream convolutional neural network, the spatial feature data of safety hazard events in the spatial stream data samples of historical safety hazard events in the industrial park are extracted; Based on the spatial characteristic data of safety hazard events, cluster spatial mapping is performed on the spatial characteristic data of safety hazard events to obtain a target area hazard map of the industrial park; Calculate the heat value of each spatial characteristic parameter in the target area hidden danger map of the industrial park to obtain the target area hidden danger heat map of the industrial park; Assign weights based on the intensity ratio of the hidden danger heat map in the target area of the industrial park; Based on each spatial characteristic parameter and the weight of each spatial characteristic parameter in the hidden danger heat map of the target area of the industrial park, the hidden danger score in each target area is calculated to obtain the hidden danger heat map score of the target area of the industrial park; The specific quantified value of the heat quantity of each characteristic parameter in the target area hidden danger map of the industrial park is: ; In the formula, Heat map of hidden dangers in target areas of industrial parks Coordinate heat value, is the i-th potential risk point coordinate, is the standard deviation of the Gaussian kernel function, For calculation The coordinates of The heat contribution value of the spatial characteristic parameters of the potential danger points, is the total number of characteristic parameters, exp is the exponential function; The calculation of the hidden danger score in each target area is specifically as follows: ; In the formula, Hidden danger scores of target areas in industrial parks, for Spatial feature parameter weights of coordinates.
[0007] Preferably, according to the historical safety hazard event time stream data samples of the industrial park, analyzing the target object motion posture risk coefficient in the historical safety hazard event time stream data samples specifically includes: Based on the two-stream convolutional neural network, the time feature data of safety hazard events in the time stream data samples of historical safety hazard events in the industrial park are extracted; Based on the time characteristic data of potential safety hazards, the target object's running posture in the time characteristic data of potential safety hazards is extracted through time series analysis to obtain a time series motion posture characteristic data sample of potential safety hazards; Based on the time series motion posture data samples of the potential safety hazard events, calculating the correlation coefficient between the motion posture of each target object in the time series motion posture data samples per unit time and the potential safety hazard events; Based on the LSTM network, a risk assessment model for the target object's motion posture is constructed; Substitute the time series motion posture data samples of potential safety hazards into the target object motion posture risk assessment model, use the correlation coefficient between each target object's motion posture and potential safety hazards as the hidden layer attribute, and use the target object motion posture risk score per unit time as the output; Based on the risk score of the target object motion posture per unit time in the time series motion posture data sample of the potential safety hazard event, the risk coefficient of the target object motion posture per unit time is calculated through linear regression; The correlation coefficient between the target object's motion posture and the potential safety hazard event is specifically: ; In the formula, is the j-th motion posture risk coefficient of the target object under t unit time, is the characteristic parameter of the jth motion posture at the tth unit time, A binary variable indicating whether a hidden danger event occurs in the tth unit time. 1 means it happened, 0 means it didn't happen. is the standard deviation of the motion posture characteristic parameters, is the standard deviation of the indicator variable of hidden danger events, is the covariance function, is the total number of motion posture data samples; The calculation of the target object motion posture risk coefficient per unit time is specifically: ; In the formula, is the j-th motion posture risk coefficient of the target object at the t-th unit time, , , , are regression coefficients, is the error term at the tth unit time.
[0008] Preferably, the target object motion posture risk coefficient in the historical safety hazard event time stream data sample is synchronously compensated for the target object hazard heat map score in the historical safety hazard event spatial stream data sample, and the internal safety hazard trend analysis model of the industrial park is trained to generate the safety hazard score of the monitoring target area of the industrial park. Specifically, it includes: Obtain real-time video data of historical safety hazards in industrial parks; Determine the types of potential safety hazards in industrial parks and determine the tolerable threshold range of potential safety hazards; The target object motion posture risk coefficient in the historical safety hazard event time stream data sample is synchronously compensated for the target object hazard heat map score in the historical safety hazard event spatial stream data sample to obtain the safety hazard score of the monitoring target area of the industrial park; Screen out the target object motion posture risk coefficients in the historical safety hazard event time stream data samples and the target object hazard heat map scores in the historical safety hazard event spatial stream data samples, where the safety hazard scores of the monitoring target areas of the industrial park exceed the tolerable threshold interval of the safety hazard events, and record them as the optimal spatial and temporal feature data samples of the safety hazard events; Based on random forest, a decision tree structure that connects safety hazard events in series is established as a safety hazard decision model for industrial parks; Based on the industrial park safety hazard decision-making model, the optimal spatial and temporal feature data samples of each safety hazard event are used as the branch leaf nodes of the corresponding industrial park safety hazard decision-making model, the real-time historical safety hazard event video data of the industrial park is used as input, and the safety hazard score of the monitoring target area of the industrial park per unit time is used as output; Among them, the safety hazard scores of the monitoring target areas of the industrial park are specifically as follows: ; In the formula, Score the safety hazards of the monitored target areas of the industrial park.
[0009] Furthermore, a system for analyzing internal safety hazards in industrial parks based on pan-intelligent video self-learning is proposed, which is used to implement the method for analyzing internal safety hazards in industrial parks based on pan-intelligent video self-learning as described above, including: A data annotation module is used to annotate the associated factors leading to the occurrence of safety hazard events based on the historical safety hazard event video data of the industrial park, and to form a sample of the historical safety hazard event video data of the industrial park; A dual-stream module, the dual-stream module is electrically connected to the data annotation module, and the dual-stream module is used to perform dual-stream segmentation processing based on the video data samples of historical safety hazard events in the industrial park, so as to obtain spatial stream data samples and temporal stream data samples of historical safety hazard events in the industrial park; A hotspot analysis module, the hotspot analysis module is electrically connected to the dual-stream module, and the hotspot analysis module is used to analyze the target area hidden danger heat map score in the historical safety hidden danger event spatial flow data sample according to the historical safety hidden danger event spatial flow data sample of the industrial park; A risk analysis module, the risk analysis module is electrically connected to the dual-stream module, and the risk analysis module is used to analyze the risk coefficient of the target object motion posture in the historical safety hazard event time stream data sample according to the historical safety hazard event time stream data sample of the industrial park; The hidden danger assessment module is electrically connected to the hot spot analysis module and the risk analysis module. The hidden danger assessment module is used to synchronously compensate the target object motion posture risk coefficient in the historical safety hidden danger event time stream data sample for the target object hidden danger heat map score in the historical safety hidden danger event space stream data sample, train the internal safety hidden danger trend analysis model of the industrial park, and generate the safety hidden danger score of the monitoring target area of the industrial park.
[0010] Optionally, the hotspot analysis module includes: The spatial feature unit extracts the spatial feature data of potential safety hazards from the spatial stream data samples of historical potential safety hazards in industrial parks based on a two-stream convolutional neural network; The hidden danger map unit performs clustering spatial mapping on the spatial characteristic data of the safety hidden danger events based on the spatial characteristic data of the safety hidden danger events to obtain the hidden danger map of the target area of the industrial park; The hidden danger heat map unit calculates the heat value of each spatial characteristic parameter in the hidden danger map of the target area of the industrial park to obtain the hidden danger heat map of the target area of the industrial park; The weighted unit assigns weights based on the intensity ratio of the hidden danger heat map in the target area of the industrial park; The hidden danger heat map scoring unit calculates the hidden danger score in each target area based on each spatial characteristic parameter and the weight of each spatial characteristic parameter in the hidden danger heat map of the target area of the industrial park, and obtains the hidden danger heat map score of the target area of the industrial park.
[0011] Optionally, the risk analysis module includes: The time feature unit extracts the time feature data of safety hazard events from the time stream data samples of historical safety hazard events in the industrial park based on a two-stream convolutional neural network; The posture feature unit, based on the time feature data of the potential safety hazard event, extracts the target object's running posture in the time feature data of the potential safety hazard event through time series analysis, and obtains a time series motion posture feature data sample of the potential safety hazard event; A correlation coefficient analysis unit, based on the time series motion posture data samples of the potential safety hazard events, calculates the correlation coefficient between the motion posture of each target object in the time series motion posture data samples under a unit time and the potential safety hazard events; The first model unit builds a target object motion posture risk assessment model based on the LSTM network; The scoring unit substitutes the time series motion posture data samples of the potential safety hazard event into the target object motion posture risk assessment model, takes the correlation coefficient between the motion posture of each target object and the potential safety hazard event as the hidden layer attribute, and takes the target object motion posture risk score per unit time as the output; The risk coefficient calculation unit calculates the risk coefficient of the target object motion posture per unit time through linear regression based on the risk score of the target object motion posture per unit time in the time series motion posture data sample of the potential safety hazard event.
[0012] Optionally, the hidden danger assessment module includes: Implement data collection units to obtain real-time video data of historical safety hazards in industrial parks; Threshold unit, which determines the type of potential safety hazard events in the industrial park and the tolerable threshold range of potential safety hazard events; A compensation unit synchronously compensates the target object motion posture risk coefficient in the historical safety hazard event time stream data sample for the target object hazard heat map score in the historical safety hazard event space stream data sample to obtain the safety hazard score of the monitoring target area of the industrial park; The optimal sample screening unit screens out the target object motion posture risk coefficient in the historical safety hazard event time stream data sample and the target object hazard heat map score in the historical safety hazard event spatial stream data sample whose safety hazard score of the monitoring target area of the industrial park exceeds the tolerable threshold interval of the safety hazard event, and records them as the optimal spatial and temporal feature data samples of the safety hazard event; The second model building unit, based on random forest, builds a decision tree structure that connects safety hazard events in series to form an industrial park safety hazard decision model; The safety hazard assessment unit is based on the safety hazard decision-making model of the industrial park. It uses the optimal spatial and temporal characteristic data samples of each safety hazard event as the branch leaf nodes of the corresponding industrial park safety hazard decision-making model, and the real-time historical safety hazard event video data of the industrial park as input, and the safety hazard score of the monitored target area of the industrial park per unit time as output.
[0013] Compared with the prior art, the present invention has the following beneficial effects: The present invention proposes a solution for analyzing the internal safety hazards of industrial parks based on pan-intelligent video self-learning. By performing dual-stream segmentation processing on the video data of historical safety hazard events in the industrial park, spatial stream and time stream data samples are obtained respectively. By analyzing the hazard heat map scores and motion posture risk coefficients in these data samples, combined with synchronous compensation, a safety hazard trend analysis model is trained to achieve accurate assessment of safety hazards in the real-time monitoring area, improve the early warning capability of the industrial park, and reduce potential risks. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] Figure 1 This is a flow chart of a method for analyzing internal safety hazards in industrial parks based on pan-intelligent video self-learning; Figure 2 A flow chart of the method for analyzing the heat map scoring of hidden dangers in the target area; Figure 3 A flow chart of the method for analyzing the risk factor of the target object's motion posture; Figure 4 To generate a flow chart of the safety hazard scoring method for the monitoring target area of the industrial park; Figure 5 This is a framework diagram of a system for analyzing internal safety hazards in industrial parks based on pan-intelligent video self-learning. DETAILED DESCRIPTION
[0015] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art may think of other obvious variations.
[0016] Reference Figure 1 As shown, a method for analyzing internal safety hazards in an industrial park based on pan-intelligent video self-learning includes: Based on the historical video data of potential safety hazards in industrial parks, the associated factors that lead to potential safety hazards are marked, and a sample of historical video data of potential safety hazards in industrial parks is established; Based on the video data samples of historical safety hazard events in the industrial park, dual-stream segmentation is performed to obtain spatial stream data samples and temporal stream data samples of historical safety hazard events in the industrial park; According to the spatial flow data samples of historical safety hazard events in the industrial park, the heat map scores of the target area hazards in the spatial flow data samples of historical safety hazard events are analyzed; According to the historical safety hazard event time stream data samples of the industrial park, the risk coefficient of the target object motion posture in the historical safety hazard event time stream data samples is analyzed; The target object motion posture risk coefficient in the historical safety hazard event time stream data sample is synchronously compensated for the target object hazard heat map score in the historical safety hazard event spatial stream data sample, and the internal safety hazard trend analysis model of the industrial park is trained to generate the safety hazard score of the monitoring target area of the industrial park.
[0017] This solution uses a dual-stream segmentation method to process the video data of historical safety hazards in industrial parks, and obtains spatial stream and temporal stream data samples respectively. By analyzing the hazard heat map scores and motion posture risk coefficients in these data samples, combined with synchronous compensation, a safety hazard trend analysis model is trained to achieve accurate assessment of safety hazards in real-time monitoring areas, improve the early warning capabilities of industrial parks, and reduce potential risks.
[0018] Reference Figure 2 As shown in the figure, based on the historical safety hazard event spatial flow data samples of the industrial park, the target area hazard heat map scores in the historical safety hazard event spatial flow data samples are analyzed, including: Based on the two-stream convolutional neural network, the spatial feature data of safety hazard events in the spatial stream data samples of historical safety hazard events in the industrial park are extracted; Based on the spatial characteristic data of safety hazard events, cluster spatial mapping is performed on the spatial characteristic data of safety hazard events to obtain a target area hazard map of the industrial park; Calculate the heat value of each spatial characteristic parameter in the target area hidden danger map of the industrial park to obtain the target area hidden danger heat map of the industrial park; Assign weights based on the intensity ratio of the hidden danger heat map in the target area of the industrial park; Based on each spatial characteristic parameter and the weight of each spatial characteristic parameter in the hidden danger heat map of the target area of the industrial park, the hidden danger score in each target area is calculated to obtain the hidden danger heat map score of the target area of the industrial park; The specific quantified value of the heat quantity of each characteristic parameter in the target area hidden danger map of the industrial park is: ; In the formula, Heat map of hidden dangers in target areas of industrial parks Coordinate heat value, is the i-th potential risk point coordinate, is the standard deviation of the Gaussian kernel function, For calculation The coordinates of The heat contribution value of the spatial characteristic parameters of the potential danger points, is the total number of characteristic parameters, exp is the exponential function; The calculation of the hidden danger score in each target area is specifically as follows: ; In the formula, Hidden danger scores of target areas in industrial parks, for Spatial feature parameter weights of coordinates.
[0019] It can be understood that the spatial feature data of historical safety hazards in industrial parks are extracted through a two-stream convolutional neural network, and these feature data are used for clustering spatial mapping to generate a hazard map of the target area. Subsequently, each feature parameter is weighted according to the heat value and intensity ratio in the hazard map to calculate the hazard score of each target area, providing an accurate hazard score for each area.
[0020] It should be noted that the heat value in the plan represents the concentration of safety hazards in each area. The heat value reflects the risk level of a specific area by statistically analyzing the frequency and severity of hazard events. The larger the heat value, the greater the safety hazard in the area, and the smaller the heat value, the smaller the safety hazard in the area.
[0021] Reference Figure 3 As shown, according to the historical safety hazard event time stream data samples of the industrial park, the target object motion posture risk coefficient in the historical safety hazard event time stream data samples is analyzed, which specifically includes: Based on the two-stream convolutional neural network, the time feature data of safety hazard events in the time stream data samples of historical safety hazard events in the industrial park are extracted; Based on the time characteristic data of potential safety hazards, the target object's running posture in the time characteristic data of potential safety hazards is extracted through time series analysis to obtain a time series motion posture characteristic data sample of potential safety hazards; Based on the time series motion posture data samples of the potential safety hazard events, calculating the correlation coefficient between the motion posture of each target object in the time series motion posture data samples per unit time and the potential safety hazard events; Based on the LSTM network, a risk assessment model for the target object's motion posture is constructed; Substitute the time series motion posture data samples of potential safety hazards into the target object motion posture risk assessment model, use the correlation coefficient between each target object's motion posture and potential safety hazards as the hidden layer attribute, and use the target object motion posture risk score per unit time as the output; Based on the risk score of the target object motion posture per unit time in the time series motion posture data sample of the potential safety hazard event, the risk coefficient of the target object motion posture per unit time is calculated through linear regression; The correlation coefficient between the target object's motion posture and the potential safety hazard event is specifically: ; In the formula, is the j-th motion posture risk coefficient of the target object under t unit time, is the characteristic parameter of the jth motion posture at the tth unit time, A binary variable indicating whether a hidden danger event occurs in the tth unit time. 1 means it happened, 0 means it didn't happen. is the standard deviation of the motion posture characteristic parameters, is the standard deviation of the indicator variable of hidden danger events, is the covariance function, is the total number of motion posture data samples; The calculation of the target object motion posture risk coefficient per unit time is specifically: ; In the formula, is the j-th motion posture risk coefficient of the target object at the t-th unit time, , , , are regression coefficients, is the error term at the tth unit time.
[0022] This scheme uses a two-stream convolutional neural network to extract spatial features from the historical safety hazard event data of the industrial park, combines time series analysis to determine the motion posture risk, and calculates the correlation coefficient between motion posture and safety hazard events; secondly, the time series motion posture data samples of safety hazard events are input into the target object motion posture risk assessment model, the correlation coefficient is used as the hidden layer attribute, and the motion posture risk score per unit time is used as the output. Finally, the motion posture risk coefficient is calculated through linear regression, which is helpful to monitor and predict potential safety hazards in real time.
[0023] Reference Figure 4 As shown in the figure, the target object motion posture risk coefficient in the historical safety hazard event time stream data sample is synchronously compensated for the target object hazard heat map score in the historical safety hazard event spatial stream data sample, and the internal safety hazard trend analysis model of the industrial park is trained to generate the safety hazard score of the monitoring target area of the industrial park. Specifically, it includes: Obtain real-time video data of historical safety hazards in industrial parks; Determine the types of potential safety hazards in industrial parks and determine the tolerable threshold range of potential safety hazards; The target object motion posture risk coefficient in the historical safety hazard event time stream data sample is synchronously compensated for the target object hazard heat map score in the historical safety hazard event spatial stream data sample to obtain the safety hazard score of the monitoring target area of the industrial park; Screen out the target object motion posture risk coefficients in the historical safety hazard event time stream data samples and the target object hazard heat map scores in the historical safety hazard event spatial stream data samples, where the safety hazard scores of the monitoring target areas of the industrial park exceed the tolerable threshold interval of the safety hazard events, and record them as the optimal spatial and temporal feature data samples of the safety hazard events; Based on random forest, a decision tree structure that connects safety hazard events in series is established as a safety hazard decision model for industrial parks; Based on the industrial park safety hazard decision-making model, the optimal spatial and temporal feature data samples of each safety hazard event are used as the branch leaf nodes of the corresponding industrial park safety hazard decision-making model, the real-time historical safety hazard event video data of the industrial park is used as input, and the safety hazard score of the monitoring target area of the industrial park per unit time is used as output; Among them, the safety hazard scores of the monitoring target areas of the industrial park are specifically as follows: ; In the formula, Score the safety hazards of the monitored target areas of the industrial park.
[0024] This solution establishes a safety hazard decision model based on random forests by integrating the data of historical safety hazard events and real-time monitoring data. The specific implementation steps include: first, determine the types of safety hazards in the industrial park and their tolerance thresholds, then synchronously compensate the motion posture risk coefficients in the historical data with the hazard heat map scores to generate safety hazard scores for the monitoring target area. Next, by screening out the data whose scores exceed the tolerance threshold, the optimal spatial and temporal feature data samples are extracted. Using these feature data, a decision tree model is constructed based on the random forest algorithm. The model can use each sample as a branch leaf node of the decision tree, and use real-time video data as input to evaluate safety hazards. The main effect of implementing this solution is to improve the monitoring and early warning capabilities of safety hazards in industrial parks, and to achieve more accurate risk prediction and management through the combination of real-time data analysis and historical data.
[0025] Reference Figure 5 As shown, a system for analyzing internal safety hazards in industrial parks based on pan-intelligent video self-learning includes: A data annotation module is used to annotate the associated factors leading to the occurrence of safety hazard events based on the historical safety hazard event video data of the industrial park, and to form a sample of the historical safety hazard event video data of the industrial park; A dual-stream module, the dual-stream module is electrically connected to the data annotation module, and the dual-stream module is used to perform dual-stream segmentation processing based on the video data samples of historical safety hazard events in the industrial park, so as to obtain spatial stream data samples and temporal stream data samples of historical safety hazard events in the industrial park; A hotspot analysis module, the hotspot analysis module is electrically connected to the dual-stream module, and the hotspot analysis module is used to analyze the target area hidden danger heat map score in the historical safety hidden danger event spatial flow data sample according to the historical safety hidden danger event spatial flow data sample of the industrial park; A risk analysis module, the risk analysis module is electrically connected to the dual-stream module, and the risk analysis module is used to analyze the risk coefficient of the target object motion posture in the historical safety hazard event time stream data sample according to the historical safety hazard event time stream data sample of the industrial park; The hidden danger assessment module is electrically connected to the hot spot analysis module and the risk analysis module. The hidden danger assessment module is used to synchronously compensate the target object motion posture risk coefficient in the historical safety hidden danger event time stream data sample for the target object hidden danger heat map score in the historical safety hidden danger event space stream data sample, train the internal safety hidden danger trend analysis model of the industrial park, and generate the safety hidden danger score of the monitoring target area of the industrial park.
[0026] The hotspot analysis module includes: The spatial feature unit extracts the spatial feature data of potential safety hazards from the spatial stream data samples of historical potential safety hazards in industrial parks based on a two-stream convolutional neural network; The hidden danger map unit performs clustering spatial mapping on the spatial characteristic data of the safety hidden danger events based on the spatial characteristic data of the safety hidden danger events to obtain the hidden danger map of the target area of the industrial park; The hidden danger heat map unit calculates the heat value of each spatial characteristic parameter in the hidden danger map of the target area of the industrial park to obtain the hidden danger heat map of the target area of the industrial park; The weighted unit assigns weights based on the intensity ratio of the hidden danger heat map in the target area of the industrial park; The hidden danger heat map scoring unit calculates the hidden danger score in each target area based on each spatial characteristic parameter and the weight of each spatial characteristic parameter in the hidden danger heat map of the target area of the industrial park, and obtains the hidden danger heat map score of the target area of the industrial park.
[0027] The risk analysis module includes: The time feature unit extracts the time feature data of safety hazard events from the time stream data samples of historical safety hazard events in the industrial park based on a two-stream convolutional neural network; The posture feature unit, based on the time feature data of the potential safety hazard event, extracts the target object's running posture in the time feature data of the potential safety hazard event through time series analysis, and obtains a time series motion posture feature data sample of the potential safety hazard event; A correlation coefficient analysis unit, based on the time series motion posture data samples of the potential safety hazard events, calculates the correlation coefficient between the motion posture of each target object in the time series motion posture data samples under a unit time and the potential safety hazard events; The first model unit builds a target object motion posture risk assessment model based on the LSTM network; The scoring unit substitutes the time series motion posture data samples of the potential safety hazard event into the target object motion posture risk assessment model, takes the correlation coefficient between the motion posture of each target object and the potential safety hazard event as the hidden layer attribute, and takes the target object motion posture risk score per unit time as the output; The risk coefficient calculation unit calculates the risk coefficient of the target object motion posture per unit time through linear regression based on the risk score of the target object motion posture per unit time in the time series motion posture data sample of the potential safety hazard event.
[0028] The hidden danger assessment module includes: Implement data collection units to obtain real-time video data of historical safety hazards in industrial parks; Threshold unit, which determines the type of potential safety hazard events in the industrial park and the tolerable threshold range of potential safety hazard events; A compensation unit synchronously compensates the target object motion posture risk coefficient in the historical safety hazard event time stream data sample for the target object hazard heat map score in the historical safety hazard event space stream data sample to obtain the safety hazard score of the monitoring target area of the industrial park; The optimal sample screening unit screens out the target object motion posture risk coefficient in the historical safety hazard event time stream data sample and the target object hazard heat map score in the historical safety hazard event spatial stream data sample whose safety hazard score of the monitoring target area of the industrial park exceeds the tolerable threshold interval of the safety hazard event, and records them as the optimal spatial and temporal feature data samples of the safety hazard event; The second model building unit, based on random forest, builds a decision tree structure that connects safety hazard events in series to form an industrial park safety hazard decision model; The safety hazard assessment unit is based on the safety hazard decision-making model of the industrial park. It uses the optimal spatial and temporal characteristic data samples of each safety hazard event as the branch leaf nodes of the corresponding industrial park safety hazard decision-making model, and the real-time historical safety hazard event video data of the industrial park as input, and the safety hazard score of the monitored target area of the industrial park per unit time as output.
[0029] The use process of a system for analyzing internal safety hazards in industrial parks based on pan-intelligent video self-learning is as follows: Step 1: Based on the historical video data of potential safety hazards in industrial parks, mark the associated factors that lead to potential safety hazards and create a sample of historical video data of potential safety hazards in industrial parks; Step 2: Perform dual-stream segmentation processing based on the video data samples of historical safety hazard events in the industrial park to obtain spatial stream data samples and temporal stream data samples of historical safety hazard events in the industrial park; Step 3: Extract the spatial feature data of safety hazard events from the spatial stream data samples of historical safety hazard events in the industrial park based on the two-stream convolutional neural network; Step 4: Based on the spatial feature data of safety hazard events, cluster spatial mapping is performed on the spatial feature data of safety hazard events to obtain a target area hazard map of the industrial park; Step 5: Calculate the heat value of each spatial characteristic parameter in the target area hidden danger map of the industrial park to obtain the target area hidden danger heat map of the industrial park; Step 6: Assign weights based on the intensity ratio of the hidden danger heat map of the target area of the industrial park; Step 7: Based on each spatial characteristic parameter and the weight of each spatial characteristic parameter in the target area hidden danger heat map of the industrial park, the hidden danger score in each target area is calculated to obtain the hidden danger heat map score of the target area of the industrial park; Step 8: Extract the time feature data of safety hazard events from the time stream data samples of historical safety hazard events in the industrial park based on the two-stream convolutional neural network; Step 9: Based on the time characteristic data of the potential safety hazard event, extract the target object's running posture in the time characteristic data of the potential safety hazard event through time series analysis to obtain a time series motion posture characteristic data sample of the potential safety hazard event; Step 10: Based on the time series motion posture data samples of the potential safety hazard events, calculate the correlation coefficient between the motion posture of each target object in the time series motion posture data samples per unit time and the potential safety hazard events; Step 11: Based on the LSTM network, a target object motion posture risk assessment model is constructed; Step 12: Substitute the time series motion posture data samples of the potential safety hazard events into the target object motion posture risk assessment model, use the correlation coefficient between the motion posture of each target object and the potential safety hazard event as the hidden layer attribute, and use the target object motion posture risk score per unit time as the output; Step 13: Based on the risk score of the target object motion posture per unit time in the time series motion posture data sample of the potential safety hazard event, calculate the risk coefficient of the target object motion posture per unit time through linear regression; Step 14: Obtain real-time video data of historical safety hazard events in the industrial park; Step 15: Determine the types of safety hazard events in the industrial park and determine the tolerable threshold range of safety hazard events; Step 16: Synchronously compensate the target object motion posture risk coefficient in the historical safety hazard event time stream data sample for the target object hazard heat map score in the historical safety hazard event spatial stream data sample to obtain the safety hazard score of the monitoring target area of the industrial park; Step 17: Screen out the target object motion posture risk coefficient in the historical safety hazard event time stream data sample and the target object hazard heat map score in the historical safety hazard event spatial stream data sample for the monitoring target area of the industrial park that exceeds the tolerable threshold interval of the safety hazard event, and record them as the optimal spatial and temporal feature data samples of the safety hazard event; Step 18: Based on random forest, a decision tree structure of serial safety hazard events is established to form an industrial park safety hazard decision model; Step 19: Based on the industrial park safety hazard decision-making model, the optimal spatial and temporal feature data samples of each safety hazard event are used as the branch leaf nodes of the corresponding industrial park safety hazard decision-making model, the real-time historical safety hazard event video data of the industrial park is used as input, and the safety hazard score of the monitoring target area of the industrial park per unit time is used as output.
[0030] In summary, the advantages of the present invention are: by performing dual-stream segmentation processing on the video data of historical safety hazard events in the industrial park, spatial stream and time stream data samples are obtained respectively. By analyzing the hazard heat map scores and motion posture risk coefficients in these data samples, combined with synchronous compensation, a safety hazard trend analysis model is trained to achieve accurate assessment of safety hazards in the real-time monitoring area, improve the early warning capability of the industrial park, and reduce potential risks.
[0031] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions only describe the principles of the present invention. The present invention may be subject to various changes and improvements without departing from the spirit and scope of the present invention. These changes and improvements fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the attached claims and their equivalents.
Claims
1. A method for analyzing internal safety hazards in industrial parks based on pan-intelligent video self-learning, characterized in that: include: Based on the historical video data of potential safety hazards in industrial parks, the related factors that lead to potential safety hazards are marked, and a sample of historical video data of potential safety hazards in industrial parks is established; Based on the video data samples of historical safety hazard events in the industrial park, dual-stream segmentation is performed to obtain spatial stream data samples and temporal stream data samples of historical safety hazard events in the industrial park; According to the spatial flow data samples of historical safety hazard events in the industrial park, the heat map scores of the target area hazards in the spatial flow data samples of historical safety hazard events are analyzed; According to the historical safety hazard event time stream data samples of the industrial park, the risk coefficient of the target object motion posture in the historical safety hazard event time stream data samples is analyzed; The target object motion posture risk coefficient in the historical safety hazard event time stream data sample is synchronously compensated for the target object hazard heat map score in the historical safety hazard event spatial stream data sample, and the internal safety hazard trend analysis model of the industrial park is trained to generate the safety hazard score of the monitoring target area of the industrial park.
2. According to claim 1, a method for analyzing internal safety hazards in industrial parks based on pan-intelligent video self-learning is characterized in that: According to the spatial flow data samples of historical safety hazard events in the industrial park, the heat map scores of the target area hazards in the spatial flow data samples of historical safety hazard events are analyzed, including: Based on the two-stream convolutional neural network, the spatial feature data of safety hazard events in the spatial stream data samples of historical safety hazard events in industrial parks are extracted; Based on the spatial characteristic data of safety hazard events, cluster spatial mapping is performed on the spatial characteristic data of safety hazard events to obtain a target area hazard map of the industrial park; Calculate the heat value of each spatial characteristic parameter in the target area hidden danger map of the industrial park to obtain the target area hidden danger heat map of the industrial park; Assign weights based on the intensity ratio of the hidden danger heat map in the target area of the industrial park; Based on each spatial characteristic parameter and the weight of each spatial characteristic parameter in the hidden danger heat map of the target area of the industrial park, the hidden danger score in each target area is calculated to obtain the hidden danger heat map score of the target area of the industrial park; The specific quantified value of the heat quantity of each characteristic parameter in the target area hidden danger map of the industrial park is: ; In the formula, Heat map of hidden dangers in target areas of industrial parks Coordinate heat value, is the i-th potential risk point coordinate, is the standard deviation of the Gaussian kernel function, For calculation The coordinates of The heat contribution value of the spatial characteristic parameter of each potential hazard point, is the total number of characteristic parameters, exp is the exponential function; The calculation of the hidden danger score in each target area is specifically as follows: ; In the formula, Hidden danger scores of target areas in industrial parks, for Spatial feature parameter weights of coordinates.
3. According to claim 2, a method for analyzing internal safety hazards in industrial parks based on pan-intelligent video self-learning is characterized in that: According to the historical safety hazard event time stream data samples of the industrial park, the risk coefficients of the target object motion posture in the historical safety hazard event time stream data samples are analyzed, including: Based on the two-stream convolutional neural network, the time feature data of safety hazard events in the time stream data samples of historical safety hazard events in the industrial park are extracted; Based on the time characteristic data of potential safety hazards, the target object's running posture in the time characteristic data of potential safety hazards is extracted through time series analysis to obtain a time series motion posture characteristic data sample of potential safety hazards; Based on the time series motion posture data samples of the potential safety hazard events, calculating the correlation coefficient between the motion posture of each target object in the time series motion posture data samples per unit time and the potential safety hazard events; Based on the LSTM network, a risk assessment model for the target object's motion posture is constructed; Substitute the time series motion posture data samples of potential safety hazards into the target object motion posture risk assessment model, use the correlation coefficient between each target object's motion posture and potential safety hazards as the hidden layer attribute, and use the target object motion posture risk score per unit time as the output; Based on the risk score of the target object motion posture per unit time in the time series motion posture data sample of the potential safety hazard event, the risk coefficient of the target object motion posture per unit time is calculated through linear regression; The correlation coefficient between the target object's motion posture and the potential safety hazard event is specifically: ; In the formula, is the j-th motion posture risk coefficient of the target object under t unit time, is the characteristic parameter of the jth motion posture at the tth unit time, A binary variable indicating whether a hidden danger event occurs in the tth unit time. 1 means it happened, 0 means it didn't happen. is the standard deviation of the motion posture characteristic parameters, is the standard deviation of the indicator variable of hidden danger events, is the covariance function, is the total number of motion posture data samples; The calculation of the target object motion posture risk coefficient per unit time is specifically: ; In the formula, is the j-th motion posture risk coefficient of the target object at the t-th unit time, , , , are regression coefficients, is the error term at the tth unit time.
4. According to claim 3, a method for analyzing internal safety hazards in industrial parks based on pan-intelligent video self-learning is characterized in that: The target object motion posture risk coefficient in the historical safety hazard event time stream data sample is synchronously compensated for the target object hazard heat map score in the historical safety hazard event spatial stream data sample, and the internal safety hazard trend analysis model of the industrial park is trained to generate the safety hazard score of the monitoring target area of the industrial park. Specifically, it includes: Obtain real-time video data of historical safety hazards in industrial parks; Determine the types of potential safety hazards in industrial parks and determine the tolerable threshold range of potential safety hazards; The target object motion posture risk coefficient in the historical safety hazard event time stream data sample is synchronously compensated for the target object hazard heat map score in the historical safety hazard event spatial stream data sample to obtain the safety hazard score of the monitoring target area of the industrial park; Screen out the target object motion posture risk coefficients in the historical safety hazard event time stream data samples and the target object hazard heat map scores in the historical safety hazard event spatial stream data samples, where the safety hazard scores of the monitoring target areas of the industrial park exceed the tolerable threshold interval of the safety hazard events, and record them as the optimal spatial and temporal feature data samples of the safety hazard events; Based on random forest, a decision tree structure that connects safety hazard events in series is established as a safety hazard decision model for industrial parks; Based on the industrial park safety hazard decision-making model, the optimal spatial and temporal feature data samples of each safety hazard event are used as the branch leaf nodes of the corresponding industrial park safety hazard decision-making model, the real-time historical safety hazard event video data of the industrial park is used as input, and the safety hazard score of the monitoring target area of the industrial park per unit time is used as output; Among them, the safety hazard scores of the monitoring target areas of the industrial park are specifically as follows: ; In the formula, Score the safety hazards of the monitored target areas of the industrial park.
5. A system for analyzing internal safety hazards in industrial parks based on pan-intelligent video self-learning, characterized in that: A method for analyzing internal safety hazards of an industrial park based on pan-intelligent video self-learning as described in any one of claims 1 to 4, comprising: A data annotation module is used to annotate the associated factors leading to the occurrence of safety hazard events based on the historical safety hazard event video data of the industrial park, and to form a sample of the historical safety hazard event video data of the industrial park; A dual-stream module, the dual-stream module is electrically connected to the data annotation module, and the dual-stream module is used to perform dual-stream segmentation processing based on the video data samples of historical safety hazard events in the industrial park, so as to obtain spatial stream data samples and temporal stream data samples of historical safety hazard events in the industrial park; A hotspot analysis module, the hotspot analysis module is electrically connected to the dual-stream module, and the hotspot analysis module is used to analyze the target area hidden danger heat map score in the historical safety hidden danger event spatial flow data sample according to the historical safety hidden danger event spatial flow data sample of the industrial park; A risk analysis module, the risk analysis module is electrically connected to the dual-stream module, and the risk analysis module is used to analyze the risk coefficient of the target object motion posture in the historical safety hazard event time stream data sample according to the historical safety hazard event time stream data sample of the industrial park; The hidden danger assessment module is electrically connected to the hot spot analysis module and the risk analysis module. The hidden danger assessment module is used to synchronously compensate the target object motion posture risk coefficient in the historical safety hidden danger event time stream data sample for the target object hidden danger heat map score in the historical safety hidden danger event space stream data sample, train the internal safety hidden danger trend analysis model of the industrial park, and generate the safety hidden danger score of the monitoring target area of the industrial park.
6. According to claim 5, a system for analyzing internal safety hazards in industrial parks based on pan-intelligent video self-learning is characterized in that: The hotspot analysis module includes: The spatial feature unit extracts the spatial feature data of potential safety hazards from the spatial stream data samples of historical potential safety hazards in industrial parks based on a two-stream convolutional neural network; The hidden danger map unit performs clustering spatial mapping on the spatial characteristic data of the safety hidden danger events based on the spatial characteristic data of the safety hidden danger events to obtain the hidden danger map of the target area of the industrial park; The hidden danger heat map unit calculates the heat value of each spatial characteristic parameter in the hidden danger map of the target area of the industrial park to obtain the hidden danger heat map of the target area of the industrial park; The weighted unit assigns weights based on the intensity ratio of the hidden danger heat map in the target area of the industrial park; The hidden danger heat map scoring unit calculates the hidden danger score in each target area based on each spatial characteristic parameter and the weight of each spatial characteristic parameter in the hidden danger heat map of the target area of the industrial park, and obtains the hidden danger heat map score of the target area of the industrial park.
7. According to claim 5, a system for analyzing internal safety hazards in industrial parks based on pan-intelligent video self-learning is characterized in that: Inside the risk analysis module include: The time feature unit extracts the time feature data of safety hazard events from the time stream data samples of historical safety hazard events in the industrial park based on a two-stream convolutional neural network; The posture feature unit, based on the time feature data of the potential safety hazard event, extracts the target object's running posture in the time feature data of the potential safety hazard event through time series analysis, and obtains a time series motion posture feature data sample of the potential safety hazard event; A correlation coefficient analysis unit, based on the time series motion posture data samples of the potential safety hazard events, calculates the correlation coefficient between the motion posture of each target object in the time series motion posture data samples under a unit time and the potential safety hazard events; The first model unit builds a target object motion posture risk assessment model based on the LSTM network; The scoring unit substitutes the time series motion posture data samples of the potential safety hazard event into the target object motion posture risk assessment model, takes the correlation coefficient between the motion posture of each target object and the potential safety hazard event as the hidden layer attribute, and takes the target object motion posture risk score per unit time as the output; The risk coefficient calculation unit calculates the risk coefficient of the target object motion posture per unit time through linear regression based on the risk score of the target object motion posture per unit time in the time series motion posture data sample of the potential safety hazard event.
8. According to claim 5, a system for analyzing internal safety hazards in industrial parks based on pan-intelligent video self-learning is characterized in that: The hidden danger assessment module includes: Implement data collection units to obtain real-time video data of historical safety hazards in industrial parks; Threshold unit, which determines the type of potential safety hazard events in the industrial park and the tolerable threshold range of potential safety hazard events; A compensation unit synchronously compensates the target object motion posture risk coefficient in the historical safety hazard event time stream data sample for the target object hazard heat map score in the historical safety hazard event space stream data sample to obtain the safety hazard score of the monitoring target area of the industrial park; The optimal sample screening unit screens out the target object motion posture risk coefficient in the historical safety hazard event time stream data sample and the target object hazard heat map score in the historical safety hazard event spatial stream data sample whose safety hazard score of the monitoring target area of the industrial park exceeds the tolerable threshold interval of the safety hazard event, and records them as the optimal spatial and temporal feature data samples of the safety hazard event; The second model building unit, based on random forest, builds a decision tree structure that connects safety hazard events in series to form an industrial park safety hazard decision model; The safety hazard assessment unit is based on the safety hazard decision-making model of the industrial park. It uses the optimal spatial and temporal characteristic data samples of each safety hazard event as the branch leaf nodes of the corresponding industrial park safety hazard decision-making model, and the real-time historical safety hazard event video data of the industrial park as input, and the safety hazard score of the monitored target area of the industrial park per unit time as output.
Citation Information
Patent Citations
assessment method and system for Parkinson's disease gait dyskinesia severity, and equipment
CN111382679A
Parkinson's disease leg flexibility task evaluation method and system, storage medium and terminal
CN111539941A
Public digital life scene rule model prediction and early warning method based on deep Bayesian network
CN113010572A
Working site safety risk lean intelligent management and control auxiliary decision-making method
CN113506001A
Dynamic and static unsafe behavior identification method and system in an industrial environment
CN113657314A
