Chemical park safety assessment method and system combined with video diagnostic tools

Abstract

The present invention discloses a chemical park safety assessment method and system combined with a video diagnostic tool, relating to the field of safety management technology. The method comprises: determining multiple video monitoring areas of a target chemical park and their corresponding preset abnormal behavior sets; defining risk accompanying features and shielding features based on the abnormal behavior sets, and constructing a primary video frame screening tool; using the screening tool to preliminarily screen the video acquisition results to generate a screening frame set carrying a detection behavior identifier; determining whether the processing volume of the screening frame set exceeds a preset concurrency threshold; if so, performing a real-time risk analysis based on the detection behavior identifier, conducting real-time and delayed safety analysis according to risk priority, and outputting the regional safety analysis results. This achieves the technical effect of improving the efficiency of monitoring data processing and enhancing the accuracy and responsiveness of safety identification.

Description

Technical Field

[0001] The present invention relates to the technical field of safety management, and in particular to a chemical park safety assessment method and system combined with a video diagnostic tool. Background Art

[0002] Currently, various companies have invested in a large number of different types of video surveillance equipment on-site to promptly detect behavioral issues and unsafe conditions, enabling timely corrections and ensuring safety. However, with the continuous expansion of chemical parks and the increasing complexity of production activities, these traditional methods are gradually becoming inadequate in the face of massive amounts of video data and diverse safety hazards.

[0003] With the continuous expansion of the scale of chemical parks and the increasing complexity of production activities, the large amount of video footage and data has made it impossible for passers-by to view them in a timely manner. The existing video AI algorithms often encounter model upgrade problems, resulting in the inability to effectively evaluate historical data. Summary of the Invention

[0004] The present invention provides a chemical park safety assessment method and system combined with video diagnostic tools to solve the technical problems in the existing technology that the amount of monitoring data is too large, resulting in processing difficulties and the inability to perform safety identification in a timely and accurate manner, thereby achieving the technical effects of improving the efficiency of monitoring data processing and enhancing the accuracy and responsiveness of safety identification.

[0005] In a first aspect, the present invention provides a chemical park safety assessment method in combination with a video diagnostic tool, wherein the chemical park safety assessment method in combination with a video diagnostic tool comprises:

[0006] Identify multiple preset abnormal behavior sets for multiple video surveillance areas within the target chemical park.

[0007] The definition of risk accompanying features and shielding features in the video image is performed based on the multiple preset abnormal behavior sets, and multiple primary video frame screening tools are constructed based on the definition results.

[0008] The multiple primary video frame screening tools are used to perform preliminary risk frame screening of the video acquisition results of the multiple video monitoring areas to generate multiple screening frame sets, and any screening frame carries a detection behavior identifier.

[0009] Analyze whether the data processing amount of the plurality of screening frame sets is greater than a preset concurrent processing threshold.

[0010] If so, a real-time risk analysis of each area is performed based on the detection behavior identifier, and the multiple primary video frame screening tools are called with real-time risk priority to perform real-time security analysis and delayed security analysis of multiple screening frame sets to generate multiple regional security analysis results.

[0011] In a feasible implementation, after generating multiple regional safety analysis results, the following steps are further included:

[0012] Read the preset evaluation period.

[0013] The safety analysis results of multiple video monitoring areas are statistically analyzed in the preset evaluation period to generate an unsafe behavior statistical table.

[0014] A safety assessment of the plurality of video monitoring areas is performed based on the unsafe behavior statistical table to generate a plurality of safety assessment results.

[0015] In a feasible implementation, any preset abnormal behavior set among the multiple preset abnormal behavior sets includes unsafe behaviors of people and unsafe states of objects.

[0016] In a feasible implementation, based on the multiple preset abnormal behavior sets, risk accompanying features and shielding features in the video image are defined, and multiple primary video frame screening tools are constructed based on the definition results, including:

[0017] For each abnormal behavior in the plurality of preset abnormal behavior sets, an early warning shielding feature defined by a user is obtained, and a shielding feature definition result is generated.

[0018] For each abnormal behavior in the plurality of preset abnormal behavior sets, dynamic distinguishing features of the images of the warning triggering scene and the non-triggering scene are collected to generate a risk accompanying feature definition result.

[0019] A dynamic detection model is constructed based on the risk accompanying feature definition results, and is optimized based on the shielding feature definition results to generate the multiple primary video frame screening tools.

[0020] In a feasible implementation, based on the detection behavior identifier, a real-time risk analysis is performed on each area, and the multiple primary video frame screening tools are called according to the real-time risk priority to perform real-time security analysis and delayed security analysis on multiple screening frame sets, generating multiple regional security analysis results, including:

[0021] Based on the detected behavior identification, trigger risk assessment of unsafe behaviors in each area is performed to establish real-time risk priorities.

[0022] In combination with the preset concurrent processing threshold and the real-time priority of the risk, the multiple screening frame sets are matched in real-time security analysis mode and delayed security mode to generate real-time mode matching results and delayed mode matching results.

[0023] According to the real-time pattern matching results and the delayed pattern matching results, the multiple primary video frame screening tools are called to perform security analysis on multiple screening frame sets to generate the multiple regional security analysis results.

[0024] In a feasible implementation, based on the detected behavior identification, trigger risk assessment of unsafe behaviors in each area is performed to establish real-time risk priorities, including:

[0025] Trigger feature fitting of unsafe behaviors in various areas is performed based on the detected behavior identifiers to construct multiple trigger features.

[0026] The trigger risk assessment is performed based on the multiple trigger features, including the risk of the behavior itself and the risk of linkage diffusion, to generate multiple risk assessment values.

[0027] The multiple risk evaluation values are arranged in descending order to generate the real-time risk priority.

[0028] In a feasible implementation, the method for determining the preset concurrent processing threshold includes:

[0029] Determine deployment environment information for multiple primary video frame filtering tools.

[0030] Perform twin simulation using the deployment environment information to determine the maximum stable concurrency.

[0031] The preset concurrent processing threshold is generated based on the maximum stable concurrent number.

[0032] In a feasible implementation, the maximum stable concurrency includes a frame rate that satisfies both a preset bottleneck utilization rate of the CPU / GPU and a preset instability condition.

[0033] In a feasible implementation, after generating multiple security assessment results, the following steps are further included:

[0034] The continuous safety assessment results under the preset assessment period are read to identify high-frequency unsafe behaviors and corresponding high-frequency unsafe areas in which the number of consecutive repetitions of unsafe behaviors is greater than a preset number.

[0035] The event video corresponding to the high-frequency unsafe behavior is retrieved and compared with the surveillance video corresponding to the low-frequency unsafe behavior with a continuous repetition number less than a preset number, the unsafe factor is determined, and sent to the management end for reminder.

[0036] In a second aspect, the present invention further provides a chemical park safety assessment system in combination with a video diagnostic tool, wherein the chemical park safety assessment system in combination with a video diagnostic tool comprises:

[0037] The behavior preset module is used to determine multiple preset abnormal behavior sets in multiple video monitoring areas within the target chemical park.

[0038] The screening tool construction module is used to define risk accompanying features and shielding features in the video image based on the multiple preset abnormal behavior sets, and to construct multiple primary video frame screening tools based on the definition results.

[0039] The preliminary risk frame screening module is used to perform preliminary risk frame screening of the video acquisition results of the multiple video monitoring areas through the multiple primary video frame screening tools, generate multiple screening frame sets, and any screening frame carries a detection behavior identifier.

[0040] The processing volume analysis module is used to analyze whether the data processing volume of the plurality of screening frame sets is greater than a preset concurrent processing threshold.

[0041] The security analysis module is used to perform real-time risk analysis of each area based on the detection behavior identifier, call the multiple primary video frame screening tools with real-time risk priority to perform real-time security analysis and delayed security analysis of multiple screening frame sets, and generate multiple regional security analysis results.

[0042] The present invention discloses a chemical park safety assessment method and system combined with a video diagnostic tool, including: determining multiple preset abnormal behavior sets corresponding to multiple video surveillance areas in a target chemical park; based on the multiple preset abnormal behavior sets, defining various risk accompanying features and shielding features, and constructing a number of primary video frame screening modules accordingly; using the multiple primary video frame screening modules to perform preliminary screening of the collected videos of each video surveillance area, extracting risk frame sets containing detection behavior identifiers, and forming multiple preliminary screening frame sets; analyzing the data processing volume required for the above-mentioned multiple screening frame sets to determine whether it exceeds a preset concurrent processing capacity threshold; if it exceeds, performing real-time risk assessment of each monitoring area based on the detection behavior identifier carried in the frame, calling and prioritizing the primary video frame screening modules according to the assessment results, performing real-time safety analysis and delayed safety analysis respectively, and outputting corresponding regional safety analysis results. The chemical park safety assessment method and system combined with a video diagnostic tool disclosed in the present invention solves the technical problem of excessive monitoring data volume leading to processing difficulties and inability to perform safety identification in a timely and accurate manner, and achieves the technical effect of improving monitoring data processing efficiency and enhancing safety identification accuracy and responsiveness. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 The figure is a flow chart of the chemical park safety assessment method combined with video diagnostic tools according to the present invention.

[0044] Figure 2This is a structural diagram of the chemical park safety assessment system combined with video diagnostic tools of the present invention.

[0045] Explanation of the accompanying drawings: behavior preset module 11, screening tool construction module 12, preliminary risk frame screening module 13, processing volume analysis module 14, safety analysis module 15. DETAILED DESCRIPTION

[0046] The above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods of the specification to better understand the above technical solution. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments of the present invention. It should be understood that the present invention is not limited to the example embodiments used only to explain the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. In addition, it should be noted that, for the convenience of description, only the parts related to the present invention, rather than all, are shown in the drawings.

[0047] Example 1, as Figure 1 The figure is a flow chart of a chemical park safety assessment method in combination with a video diagnostic tool according to the present invention, wherein the chemical park safety assessment method in combination with a video diagnostic tool comprises:

[0048] S100: Determine multiple preset abnormal behavior sets in multiple video monitoring areas within a target chemical park.

[0049] Specifically, the system first identifies abnormal behaviors based on the specific chemical production processes and types within the target chemical park. These behaviors are then output as multiple preset abnormal behavior sets. These sets can correspond to various video surveillance areas within the chemical park, such as production workshops, storage areas, and reactor perimeters, divided according to different security requirements. These preset abnormal behavior sets are determined by professional technicians based on the chemical park's production processes, historical accident data, and safety regulations.

[0050] In some embodiments, any preset abnormal behavior set among the multiple preset abnormal behavior sets includes unsafe behaviors of people and unsafe states of objects.

[0051] Specifically, the preset abnormal behavior set refers to a series of possible unsafe behavior templates set in advance based on the characteristics of different areas in the chemical park, including unsafe human behaviors (such as illegal operations, not wearing safety helmets, etc.) and unsafe conditions of objects (such as equipment leakage, improper material stacking, etc.).

[0052] The above process provides the foundation and direction for subsequent video diagnostic analysis, ensuring that video diagnostic tools can specifically identify and screen potential safety risks. For example, in a storage area containing flammable and explosive materials, the pre-set abnormal behavior set might include "illegal use of open flames," "explosion-proof equipment malfunction," and "material leakage." In an office area, the focus might be on behaviors such as "crowds blocking aisles" and "obstruction of firefighting equipment."

[0053] S200: defining risk accompanying features and shielding features within a video image based on the plurality of preset abnormal behavior sets, and constructing a plurality of primary video frame screening tools based on the definition results.

[0054] Specifically, risk-associated features refer to image features that are closely associated with pre-defined abnormal behaviors and can indicate potential safety risks. For example, for the behavior of "a person not wearing a helmet," features such as the helmet's shape and color are risk-associated features. Blocked features refer to image features that are unrelated to abnormal behavior or should not trigger an alert, such as fixed objects in the background or signs in specific areas.

[0055] The primary video frame screening tool is a preliminary screening mechanism built based on defined risk accompanying features and shielding features. It is used to quickly filter large amounts of video data and extract video frames that may contain abnormal behavior.

[0056] By defining the risk-associated features and shielding features mentioned above, we can effectively exclude invalid images for identification, thereby helping to improve overall response efficiency. At the same time, we can more accurately identify abnormal behaviors in videos and reduce false positives and missed negatives.

[0057] In some embodiments, based on the plurality of preset abnormal behavior sets, risk accompanying features and shielding features are defined within the video image, and a plurality of primary video frame screening tools are constructed based on the definition results, including:

[0058] For each abnormal behavior in the multiple preset abnormal behavior sets, the early warning shielding feature defined by the user is obtained, and a shielding feature definition result is generated; for each abnormal behavior in the multiple preset abnormal behavior sets, the dynamic distinguishing features of the images of the early warning triggering scene and the non-triggered scene are collected, and a risk accompanying feature definition result is generated; a dynamic detection model is constructed based on the risk accompanying feature definition result, and is optimized based on the shielding feature definition result to generate the multiple primary video frame screening tools.

[0059] Specifically, for each pre-defined abnormal behavior, the system first obtains user-defined warning shielding features and generates shielding feature definition results. Shielding features are user-defined image features that, despite dynamic changes, should not trigger warnings based on actual scenarios. These features are used to suppress false alarms, such as wind blowing leaves, lighting changes, scheduled equipment operation, and infrared interference at night.

[0060] Then, video clips of abnormal behavior and normal state clips are collected for comparative analysis. For example, the differences between the time periods when people are present and the time periods when no one is present are identified (such as by using spatiotemporal posture estimation and optical flow analysis), and the difference information is represented in vector form to generate risk-associated feature definition results.

[0061] Furthermore, we use risk-associated features as training targets to build lightweight models (such as MobileNet, LSTM+optical flow), and use masked features as negative samples or feature suppression factors to optimize the model's false alarm rate and generate multiple primary video frame screening tools.

[0062] Through these steps, the pre-defined set of abnormal behaviors is transformed into a specific, actionable detection model, providing a technical foundation for subsequent video frame screening. The primary screening tool, which can be run on the front-end or edge device, quickly removes low-risk images from a large number of video frames, retaining only suspected abnormal frames, reducing the back-end recognition pressure. Optimizing the model through user-defined masking features effectively eliminates common interference sources, reduces false positives, and improves system reliability.

[0063] S300: Preliminary risk frame screening of video acquisition results of the multiple video monitoring areas is performed using the multiple primary video frame screening tools to generate multiple screening frame sets, and any screening frame carries a detection behavior identifier.

[0064] Specifically, preliminary risk frame screening involves using the aforementioned primary video frame screening tool to perform an initial round of screening on the large amount of video data transmitted from each video monitoring area. Video frames identified by the screening tool as potentially exhibiting abnormal behavior are then classified and aggregated into several sets, each corresponding to a specific area, a specific type of abnormal behavior, etc. The video acquisition results are the video streams or stored video files for each monitoring area.

[0065] Specifically, any filtered frame carrying a detected behavior identifier means that each filtered video frame will be labeled with information identifying the type of abnormal behavior associated with the filtered frame, such as "running," "falling," or "trespassing," to facilitate rapid identification and classification during subsequent processing. Optionally, the detected behavior identifier can be a category label, behavior code, or confidence score.

[0066] In this way, potential risk points can be quickly located from massive video data, and different types of abnormal behaviors can be classified and summarized, providing an accurate and structured data foundation for subsequent further analysis and processing, thereby improving the intelligence level and processing efficiency of the entire security assessment process.

[0067] S400: Analyze whether the data processing volume of the plurality of screening frame sets is greater than a preset concurrent processing threshold.

[0068] Furthermore, based on a threshold method, the data volume of the multiple filtered frame sets obtained in the previous step is determined to be excessive (this can be measured by indicators such as the number of frames, data size, and time span). If the data volume exceeds the preset concurrent processing threshold, it is considered that the current amount of data to be processed is too large. The preset concurrent processing threshold is an upper limit determined by a combination of factors such as the system's hardware configuration, software processing capabilities, and business needs. It is used to limit the amount of video frame data that can be analyzed simultaneously to ensure that the system can operate in a stable and efficient state.

[0069] During the implementation process, relevant data for all filtered frame sets is first collected, such as the number of video frames in each set, the resolution of each frame, and the data size. The total data processing volume is then calculated based on pre-set calculation rules (such as simple frame number addition, summation by data volume, or a comprehensive assessment combining time and space complexity). This data processing volume is then compared with a pre-set concurrent processing threshold. This ensures that the entire system's processing flow does not experience stalls or crashes due to data overload, ensuring the continuity and stability of security assessment work.

[0070] In some embodiments, the method for determining the preset concurrent processing threshold includes:

[0071] Determine the deployment environment information of multiple primary video frame screening tools; perform twin simulation based on the deployment environment information to determine the maximum stable concurrency; and generate the preset concurrent processing threshold based on the maximum stable concurrency.

[0072] In some implementations, the maximum stable concurrency includes a frame rate that simultaneously satisfies a preset bottleneck usage rate of the CPU / GPU and a preset instability condition.

[0073] Specifically, the preset concurrent processing threshold for scheduling control is determined through twin simulation to ensure system stability and real-time performance when multiple primary video frame screening tools are running concurrently. First, the hardware resource information of the current deployment platform is collected, including but not limited to the number of CPU cores, GPU type and video memory capacity, memory size, I / O bandwidth, etc.; then, based on the above environmental information, different numbers of concurrent running scenarios of screening tools are simulated in the digital twin environment to evaluate the operating stability of the system under different concurrency numbers; and according to the preset judgment conditions, the maximum stable concurrency number is determined, and the maximum stable concurrency number is used as the concurrent processing threshold in the current deployment environment to provide a scheduling control reference for the system during operation.

[0074] Specifically, the maximum stable concurrency is determined by satisfying the following two conditions: the system resource utilization does not exceed the preset bottleneck threshold (such as CPU utilization <85%, GPU utilization <90%); the frame processing frame rate does not fall below the preset instability condition (such as frame rate >15fps).

[0075] Through the above mechanism, the concurrency threshold can be automatically set according to the deployment platform to avoid overload operation, improve system stability and robustness, and at the same time help this method to be deployed and run on different platforms (such as edge devices and cloud platforms), that is, automatically adapt performance parameters.

[0076] S500: If yes, perform real-time risk analysis of each area based on the detection behavior identifier, call the multiple primary video frame screening tools with real-time risk priority to perform real-time security analysis and delayed security analysis of multiple screening frame sets, and generate multiple regional security analysis results.

[0077] Specifically, if the data processing volume of multiple filtered frame sets exceeds the preset concurrent processing threshold, the multiple filtered frame sets will need to be processed in a phased manner. This involves performing security analysis (real-time security analysis and delayed security analysis) based on the risk priority of each filtered frame set. Real-time security analysis is used to process filtered frames in high-risk or high-priority areas, while delayed security analysis is used to process filtered frames in low-priority areas.

[0078] This process ensures that high-risk areas receive timely attention and processing, effectively preventing accidents and safeguarding human life and property. It also helps rationally utilize system resources, avoiding the waste and processing delays associated with indiscriminate real-time analysis of all areas, and improving the efficiency and accuracy of overall safety assessments.

[0079] In some embodiments, based on the detected behavior identifier, a real-time risk analysis is performed on each area, and the multiple primary video frame screening tools are called according to the real-time risk priority to perform real-time security analysis and delayed security analysis on multiple screening frame sets, generating multiple regional security analysis results, including:

[0080] Based on the detection behavior identification, a trigger risk assessment of unsafe behaviors in each area is performed to construct a real-time risk priority; in combination with the preset concurrent processing threshold and the real-time risk priority, the multiple filtered frame sets are matched in real-time security analysis mode and delayed security mode to generate real-time mode matching results and delayed mode matching results; according to the real-time mode matching results and the delayed mode matching results, the multiple primary video frame screening tools are called to perform security analysis on the multiple filtered frame sets to generate the multiple regional security analysis results.

[0081] Specifically, the real-time risk priority is a priority that is comprehensively evaluated based on factors such as the risk level of the detected behavior, the importance of the occurrence area, and the historical behavior frequency, and is used to guide the scheduling order of analysis resources.

[0082] Specifically, each area is first assessed for risk based on a preset risk assessment model (based on factors such as the type of abnormal behavior, the weight of the area where it occurs, and the potential impact range). For example, this risk assessment model can be a rule-based regularized model. For example, the risk of "fighting" is higher than that of "wandering," for example, higher in the chemical plant area than in the office area. Next, the concurrent processing threshold of the current deployment environment is obtained (e.g., supporting a maximum of five real-time tasks), and the frame collection is filtered based on priority. High-priority tasks (the first five in the sorted results) are assigned to the real-time analysis mode, and low-priority tasks (the sixth and subsequent tasks) are assigned to the delayed analysis mode, forming two task assignment lists: the real-time mode matching results and the delayed mode matching results.

[0083] Furthermore, based on the matching results, multiple primary video frame screening tools are dispatched to perform real-time analysis on high-priority areas and delayed analysis on low-priority areas. Ultimately, all analysis results are aggregated to generate safety analysis results for multiple areas. For example, in a chemical park, if the detection behavior indicator for an area indicates "flammable liquid leakage," the real-time risk priority of that area will be higher than that of another area with only "personnel not wearing safety helmets." The system will prioritize the screening tool for real-time analysis of the "flammable liquid leakage" area and delayed analysis of the "personnel not wearing safety helmets" area.

[0084] Optionally, the principle of integrated learning method is adopted, with multiple primary video frame screening tools as weak learners, to build a corresponding fusion security analysis model to improve the security analysis performance and enhance the accuracy and robustness of risk identification.

[0085] Through the above process, risk-driven analysis resource scheduling can be achieved to ensure faster responses in high-risk areas. At the same time, low-priority tasks can be queued in an orderly manner through the delayed analysis mode to avoid system crashes due to task overload and improve overall stability.

[0086] In some implementations, performing trigger risk assessments of unsafe behaviors in various areas based on the detected behavior identifiers and establishing real-time risk priorities includes:

[0087] Based on the detected behavior identification, trigger feature fitting of unsafe behaviors in each area is performed to construct multiple trigger features; trigger risk evaluation is performed using the multiple trigger features, including the risk of the behavior itself and the risk of linked diffusion, to generate multiple risk evaluation values; the multiple risk evaluation values are arranged in order from large to small to generate the real-time priority of the risk.

[0088] Specifically, for each detected behavior identifier, a feature fit is performed based on the operating status of the area to which it belongs and the environmental background information. This generates multiple trigger features, including but not limited to the following information: behavior attributes (such as intrusion, obstruction, fall, crowding, etc.); the association between trigger time and periodicity; population density and activity frequency; the geographic location of the area to which it belongs and the connectivity with surrounding areas; and the associated paths corresponding to historical risk events. These multiple trigger features are then subjected to a dual-channel risk assessment. The behavior risk channel scores dimensions such as behavior intensity, historical hazard index, and violation level; while the linkage diffusion risk channel assesses the likelihood that the behavior will cause secondary risks to adjacent areas within the current spatial structure (for example, a stampede may trigger chaos in a chain reaction). The risk values of the two channels are then merged to produce multiple risk assessment values.

[0089] Furthermore, multiple risk assessment values are arranged in descending order to form a real-time updated risk priority queue. Through the above steps, dynamic perception and risk ranking of unsafe behaviors in each area are realized. The risk propagation model and weight parameters can be dynamically adjusted according to different park structures and regional functions, thereby improving the environmental adaptability of this method.

[0090] In some embodiments, after generating multiple regional safety analysis results, the method further includes:

[0091] Reading a preset evaluation period; performing safety analysis statistics on multiple video monitoring areas using the preset evaluation period to generate an unsafe behavior statistics table; performing safety assessments on the multiple video monitoring areas based on the unsafe behavior statistics table to generate multiple safety assessment results.

[0092] Specifically, first, multiple safety assessment result data within a preset assessment cycle are obtained as the basis for statistical analysis; then, within a set time period, the safety analysis results of all monitoring areas are summarized and organized to form a detailed unsafe behavior statistical table, which records the number, type and other information of various unsafe behaviors that occurred in each area during the period; through the data in the statistical table, the safety status of each monitoring area can be comprehensively evaluated in combination with certain assessment rules and standards, and finally multiple safety assessment results are generated to reflect the safety management status and existing problems of each area, providing a basis for subsequent improvement measures.

[0093] For example, a safety assessment was conducted on the liquid ammonia and liquid chlorine areas of a certain industrial park to monitor risks such as single-person inspections, liquid chlorine storage doors frequently left open, single-person operations, operations without supervisors, and incomplete PPE wear during inspections. The corresponding statistical table for unsafe behaviors is as follows:

[0094] Table 1 Statistics of unsafe behaviors in the liquid ammonia and liquid chlorine areas of a certain park

[0095] Place Single-person inspection Warehouse door open / always open Single-person work Homework without a supervisor Incomplete PPE wearing during inspections Liquid ammonia station 2 1 Liquid ammonia station 4 4 1 4 Liquid ammonia gasification area 3 4 Liquid ammonia loading and unloading area 1 2

[0096] In some implementations, after generating multiple security assessment results, the method further includes:

[0097] Read the continuous safety assessment results under the preset assessment period, identify high-frequency unsafe behaviors and corresponding high-frequency unsafe areas whose consecutive repetition times of unsafe behaviors are greater than the preset number; retrieve the event video corresponding to the high-frequency unsafe behavior, and compare it with the surveillance video corresponding to the low-frequency unsafe behavior whose consecutive repetition times are less than the preset number, determine the unsafe factors, and send them to the management end for reminder.

[0098] Specifically, by analyzing the safety assessment results of consecutive cycles, we can find out the unsafe behaviors that have been repeated more than the set threshold in different cycles, as well as the specific areas where these behaviors occur. For example, if the preset number of times is 3 times, and the "illegal operation of equipment" behavior occurs 4 times in a certain area within three consecutive assessment cycles, then the behavior is identified as a high-frequency unsafe behavior, and the corresponding area is a high-frequency unsafe area. Then, the original video clips corresponding to the high-frequency unsafe behavior are retrieved and compared with the videos of low-frequency unsafe behaviors for analysis. By comparing the details of the video images (such as the operating methods of the personnel, the status of the equipment, the environmental conditions, etc.), the specific factors that lead to the high-frequency unsafe behavior are determined, and then this information is sent to the management end so that the management personnel can understand the situation in a timely manner and take appropriate measures to improve it.

[0099] Through the above process, comprehensive, in-depth, and dynamic monitoring and assessment of the safety status of chemical parks can be achieved. On the one hand, the periodic statistics and evaluation of safety analysis results can help managers promptly understand the overall safety trends of the park and the safety management status of each area, identify potential safety hazards and management loopholes, and thus formulate targeted improvement plans and training programs to improve the park's safety management level. On the other hand, the identification and analysis of high-frequency unsafe behaviors can accurately locate typical problem areas and types, allowing managers to take more targeted measures to resolve recurring safety issues, such as strengthening supervision of specific areas, improving operating procedures, or retraining relevant personnel.

[0100] In summary, the chemical park safety assessment method combined with video diagnostic tools provided by the present invention has the following technical effects:

[0101] By determining multiple preset abnormal behavior sets corresponding to multiple video surveillance areas in the target chemical park; based on the multiple preset abnormal behavior sets, defining various risk accompanying features and shielding features, and constructing a number of primary video frame screening modules accordingly; using the multiple primary video frame screening modules to perform preliminary screening of the collected videos of each video surveillance area, extracting risk frame sets containing detection behavior identifiers, and forming multiple preliminary screening frame sets; analyzing the data processing volume required for the above-mentioned multiple screening frame sets to determine whether it exceeds the preset concurrent processing capacity threshold; if it exceeds, performing real-time risk assessment of each monitoring area based on the detection behavior identifier carried in the frame, and calling the primary video frame screening module according to the assessment results. Priority sorting, respectively perform real-time safety analysis and delayed safety analysis, and output corresponding regional safety analysis results, thereby achieving the technical effect of improving monitoring data processing efficiency, enhancing safety identification accuracy and responsiveness.

[0102] Example 2, as Figure 2 This is a schematic diagram of the structure of the chemical park safety assessment system combined with the video diagnostic tool of the present invention. For example, Figure 1 The flow chart of the chemical park safety assessment method combined with the video diagnostic tool of the present invention can be shown as follows: Figure 2 The structure shown is implemented.

[0103] Based on the same concept as the chemical park safety assessment method combined with the video diagnostic tool in the embodiment, the present invention also provides a chemical park safety assessment system combined with the video diagnostic tool, including:

[0104] The behavior preset module 11 is used to determine multiple preset abnormal behavior sets in multiple video monitoring areas within the target chemical park.

[0105] The screening tool construction module 12 is used to define risk accompanying features and shielding features in the video image based on the multiple preset abnormal behavior sets, and to construct multiple primary video frame screening tools based on the definition results.

[0106] The preliminary risk frame screening module 13 is used to perform preliminary risk frame screening of the video acquisition results of the multiple video monitoring areas through the multiple primary video frame screening tools to generate multiple screening frame sets, and any screening frame carries a detection behavior identifier.

[0107] The processing volume analysis module 14 is configured to analyze whether the data processing volume of the plurality of screening frame sets is greater than a preset concurrent processing threshold.

[0108] The security analysis module 15 is used to perform real-time risk analysis of each area based on the detection behavior identifier, call the multiple primary video frame screening tools with real-time risk priority to perform real-time security analysis and delayed security analysis of multiple screening frame sets, and generate multiple regional security analysis results.

[0109] Furthermore, in the behavior preset module 11, any preset abnormal behavior set among the multiple preset abnormal behavior sets includes unsafe behaviors of people and unsafe states of objects.

[0110] In some embodiments, the screening tool construction module 12 includes:

[0111] The early warning shielding feature acquisition and definition unit is used to acquire the early warning shielding feature defined by the user for each abnormal behavior in the plurality of preset abnormal behavior sets, and generate a shielding feature definition result.

[0112] The risk accompanying feature collection and definition unit is used to collect dynamic distinguishing features of the images of the warning triggering scene and the non-triggered scene for each abnormal behavior in the multiple preset abnormal behavior sets, and generate a risk accompanying feature definition result.

[0113] The primary video frame screening tool generating unit is used to construct a dynamic detection model based on the risk accompanying feature definition result, and optimize it based on the shielding feature definition result to generate the multiple primary video frame screening tools.

[0114] In some embodiments, the security analysis module 15 includes:

[0115] The real-time risk priority building unit is used to perform trigger risk evaluation of unsafe behaviors in each area based on the detection behavior identifier and build the real-time risk priority.

[0116] The security analysis pattern matching unit is used to match the real-time security analysis mode and the delayed security mode on the multiple screening frame sets in combination with the preset concurrent processing threshold and the real-time risk priority, and generate a real-time pattern matching result and a delayed pattern matching result.

[0117] The regional security analysis result generating unit is used to call the multiple primary video frame screening tools to perform security analysis on multiple screening frame sets according to the real-time pattern matching results and the delayed pattern matching results, and generate the multiple regional security analysis results.

[0118] In some implementations, the real-time risk priority building unit in the security analysis module 15 includes:

[0119] The trigger feature fitting subunit is used to perform trigger feature fitting of unsafe behaviors in each area based on the detected behavior identifier to construct multiple trigger features.

[0120] The trigger risk assessment subunit is used to perform trigger risk assessment based on the multiple trigger features, including the risk of the behavior itself and the risk of linkage diffusion, and generate multiple risk assessment values.

[0121] The risk real-time priority generation subunit is used to arrange the multiple risk evaluation values in descending order to generate the risk real-time priority.

[0122] In some embodiments, the throughput analysis module 14 includes:

[0123] The deployment environment information determining unit is used to determine the deployment environment information of multiple primary video frame screening tools.

[0124] The twin simulation and maximum stable concurrency determination unit is used to perform twin simulation based on the deployment environment information and determine the maximum stable concurrency.

[0125] The preset concurrent processing threshold generating unit is used to generate the preset concurrent processing threshold based on the maximum stable concurrent number.

[0126] Furthermore, the maximum stable concurrency includes a frame rate that satisfies both a preset bottleneck utilization rate of the CPU / GPU and a preset instability condition.

[0127] In some embodiments, the chemical park safety assessment system combined with the video diagnostic tool further includes:

[0128] The preset evaluation period reading unit is used to read the preset evaluation period.

[0129] The safety analysis result statistics unit is used to perform safety analysis result statistics of multiple video monitoring areas in the preset evaluation period to generate an unsafe behavior statistics table.

[0130] The safety assessment result generating unit is used to perform safety assessment of the multiple video monitoring areas based on the unsafe behavior statistical table to generate multiple safety assessment results.

[0131] In some implementations, the security assessment result generating unit includes:

[0132] The high-frequency unsafe behavior identification subunit is used to read the continuous safety assessment results under the preset assessment period, and identify high-frequency unsafe behaviors and corresponding high-frequency unsafe areas whose continuous repetition times of unsafe behaviors are greater than the preset number.

[0133] The unsafe factor determination and reminder subunit is used to retrieve the event video corresponding to the high-frequency unsafe behavior, and compare it with the surveillance video corresponding to the low-frequency unsafe behavior with a continuous repetition number less than a preset number, determine the unsafe factor, and send it to the management end for reminder.

[0134] It should be understood that the embodiments mentioned in this specification focus on their differences from other embodiments. The specific embodiments in the aforementioned embodiment one are also applicable to the chemical park safety assessment system combined with video diagnostic tools described in embodiment two. For the sake of brevity of the specification, they will not be further elaborated here.

[0135] It should be understood that the embodiments disclosed in the present invention and the above description can enable those skilled in the art to use the present invention to implement the present invention. At the same time, the present invention is not limited to the embodiments mentioned above. It should be understood that those skilled in the art can still modify the technical solutions described in the above embodiments or replace some of the technical features therein with equivalents; and such modifications or replacements do not deviate from the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention and are all included in the scope of protection of the present invention.

Claims

1. A chemical park safety assessment method combined with a video diagnostic tool, characterized in that: include: Identify multiple preset abnormal behavior sets across multiple video surveillance areas within a target chemical park; Based on the multiple preset abnormal behavior sets, the definition of risk-accompanying features and shielding features in the video image is executed, and multiple primary video frame screening tools are constructed based on the definition results, wherein the risk-accompanying features refer to image features that are closely related to the preset abnormal behavior and can indicate potential security risks, and the shielding features refer to image features that are unrelated to the abnormal behavior or should not trigger an early warning. The risk-accompanying features are used as training targets to build a lightweight model, and the shielding features are used as negative samples or feature suppression factors to optimize the model false alarm rate. The primary screening tool can be run on the front-end or edge device to quickly eliminate low-risk images from a large number of video frames, retaining only suspected abnormal frames, thereby reducing the back-end recognition pressure; Performing preliminary risk frame screening of video acquisition results of the multiple video monitoring areas using the multiple primary video frame screening tools to generate multiple screening frame sets, wherein any screening frame carries a detection behavior identifier; analyzing whether a data processing amount of the plurality of screening frame sets is greater than a preset concurrent processing threshold; If so, a real-time risk analysis of each area is performed based on the detection behavior identifier, and the multiple primary video frame screening tools are called with real-time risk priority to perform real-time security analysis and delayed security analysis of multiple screening frame sets to generate multiple regional security analysis results.

2. The chemical park safety assessment method combined with a video diagnostic tool according to claim 1, characterized in that: After generating multiple regional safety analysis results, it also includes: Read the preset evaluation period; Performing safety analysis statistics for multiple video monitoring areas in the preset evaluation period to generate a statistical table of unsafe behaviors; A safety assessment of the plurality of video monitoring areas is performed based on the unsafe behavior statistical table to generate a plurality of safety assessment results.

3. The chemical park safety assessment method combined with a video diagnostic tool according to claim 1, characterized in that: Any preset abnormal behavior set among the multiple preset abnormal behavior sets includes unsafe behaviors of people and unsafe states of objects.

4. The chemical park safety assessment method combined with a video diagnostic tool according to claim 1, characterized in that: Based on the plurality of preset abnormal behavior sets, risk accompanying features and shielding features in the video image are defined, and a plurality of primary video frame screening tools are constructed based on the definition results, including: For each abnormal behavior in the plurality of preset abnormal behavior sets, obtaining a warning shielding feature defined by a user, and generating a shielding feature definition result; For each abnormal behavior in the plurality of preset abnormal behavior sets, dynamic distinguishing features of the images of the warning triggering scene and the non-triggering scene are collected to generate a risk accompanying feature definition result; A dynamic detection model is constructed based on the risk accompanying feature definition results, and is optimized based on the shielding feature definition results to generate the multiple primary video frame screening tools.

5. The chemical park safety assessment method combined with a video diagnostic tool according to claim 1, characterized in that: Based on the detection behavior identifier, a real-time risk analysis is performed on each area. The multiple primary video frame screening tools are called according to the real-time risk priority to perform real-time security analysis and delayed security analysis on multiple screening frame sets, and multiple regional security analysis results are generated, including: Based on the detected behavior identification, trigger risk assessment of unsafe behaviors in each area is performed to establish real-time risk priority; In combination with the preset concurrent processing threshold and the real-time risk priority, matching the multiple screening frame sets in a real-time security analysis mode and a delayed security mode is performed to generate a real-time mode matching result and a delayed mode matching result; According to the real-time pattern matching results and the delayed pattern matching results, the multiple primary video frame screening tools are called to perform security analysis on multiple screening frame sets to generate the multiple regional security analysis results.

6. The chemical park safety assessment method combined with a video diagnostic tool according to claim 5, characterized in that: Based on the detected behavior identification, trigger risk assessment of unsafe behaviors in each area is performed to establish real-time risk priorities, including: Perform trigger feature fitting of unsafe behaviors in each area based on the detected behavior identifier to construct multiple trigger features; Perform trigger risk assessment based on the multiple trigger features, including the risk of the behavior itself and the risk of linkage diffusion, to generate multiple risk assessment values; The multiple risk evaluation values are arranged in descending order to generate the real-time risk priority.

7. The chemical park safety assessment method combined with a video diagnostic tool according to claim 1, characterized in that: The method for determining the preset concurrent processing threshold includes: Determine deployment environment information for multiple primary video frame screening tools; Perform twin simulation using the deployment environment information to determine the maximum stable concurrency; The preset concurrent processing threshold is generated based on the maximum stable concurrent number.

8. The chemical park safety assessment method combined with a video diagnostic tool according to claim 7, characterized in that: The maximum stable concurrency includes a frame rate that satisfies both a preset bottleneck utilization rate of the CPU / GPU and a preset instability condition.

9. The chemical park safety assessment method combined with a video diagnostic tool as claimed in claim 2, characterized in that: After generating multiple security assessment results, it also includes: Reading the continuous safety assessment results under the preset assessment period, identifying high-frequency unsafe behaviors and corresponding high-frequency unsafe areas in which the number of consecutive repetitions of unsafe behaviors is greater than a preset number; The event video corresponding to the high-frequency unsafe behavior is retrieved and compared with the surveillance video corresponding to the low-frequency unsafe behavior with a continuous repetition number less than a preset number, the unsafe factor is determined, and sent to the management end for reminder.

10. A chemical park safety assessment system combined with video diagnostic tools is characterized by: The method for implementing the chemical park safety assessment method in combination with a video diagnostic tool as described in any one of claims 1 to 9 comprises: A behavior preset module is used to determine multiple preset abnormal behavior sets in multiple video monitoring areas within the target chemical park; A screening tool construction module is used to define risk-associated features and shielding features within video images based on the multiple preset abnormal behavior sets, and to construct multiple primary video frame screening tools based on the definition results. Risk-associated features refer to image features that are closely related to the preset abnormal behavior and can indicate potential security risks, while shielding features refer to image features that are unrelated to abnormal behavior or should not trigger an early warning. A lightweight model is constructed using risk-associated features as training targets, and shielding features are used as negative samples or feature suppression factors to optimize the model's false alarm rate. The primary screening tool can run on front-end or edge devices, quickly eliminating low-risk images from a large number of video frames, retaining only suspected abnormal frames, and reducing back-end recognition pressure. a preliminary risk frame screening module, configured to perform preliminary risk frame screening of the video acquisition results of the multiple video monitoring areas using the multiple primary video frame screening tools, and generate multiple screening frame sets, wherein any screening frame carries a detection behavior identifier; a processing volume analysis module, configured to analyze whether the data processing volume of the plurality of screening frame sets is greater than a preset concurrent processing threshold; The security analysis module is used to perform real-time risk analysis of each area based on the detection behavior identifier, call the multiple primary video frame screening tools with real-time risk priority to perform real-time security analysis and delayed security analysis of multiple screening frame sets, and generate multiple regional security analysis results.

Images