An intelligent early warning system and method based on programmable multi-algorithm cooperation
By constructing an event cognition coverage structure and coordinating multiple algorithms, the problems of incomplete event cognition and redundant analysis in existing intelligent early warning methods are solved, realizing unified organization of event cognition results and improving the accuracy and efficiency of intelligent early warning results.
Patent Information
- Application Number
- CN202611235703.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-14
- Publication Date
- 2026-11-10
AI Technical Summary
In existing intelligent early warning methods, each early warning analysis algorithm is executed independently, resulting in a lack of unified organization of event cognition results. This makes it impossible to accurately identify cognitive content of the target event that has not yet been analyzed, leading to incomplete event cognition, repeated execution of early warning analysis, and low accuracy and processing efficiency of intelligent early warning results.
By collecting video monitoring data and business-related data, a standardized event dataset is constructed. Multiple early warning analysis algorithms are used to generate event cognition results, and an event cognition coverage structure is built. Event cognition dimensions that are not fully covered are identified, and early warning analysis algorithms are dynamically matched for supplementary analysis. This achieves multi-algorithm collaborative orchestration and generates intelligent early warning results.
It achieves unified organization and completeness of event cognition results, accurately locates gaps in event cognition coverage, improves the pertinence of early warning analysis and overall processing efficiency, and ensures the accuracy and completeness of the generated intelligent early warning results.
Smart Images

Figure CN123071931A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent early warning, and in particular to an intelligent early warning system and method based on programmable multi-algorithm collaboration. Background Technology
[0002] With the development of video intelligent analysis technology, computer vision-based intelligent early warning technology has been widely used in scenarios such as park security, traffic monitoring, industrial inspection and public safety. Existing intelligent early warning methods usually analyze video monitoring data through early warning analysis algorithms such as target detection, target tracking, behavior recognition, identity recognition and relationship analysis, and combine early warning rules to complete the risk warning of target events.
[0003] However, in existing technologies, each early warning analysis algorithm usually executes independently according to a fixed process. The event cognition results generated by different algorithms lack a unified organization, making it difficult to accurately identify the cognitive content of the target event that has not yet been analyzed. This results in the inability to dynamically call the corresponding early warning analysis algorithm to supplement the analysis for the event cognition gap, which easily leads to problems such as incomplete event cognition, repeated execution of early warning analysis, and low accuracy and processing efficiency of intelligent early warning results. Summary of the Invention
[0004] One objective of this invention is to propose an intelligent early warning system and method based on programmable multi-algorithm collaboration. This invention, based on a programmable multi-algorithm collaboration method, realizes the supplementation of event cognition gaps and intelligent early warning, and has the advantages of complete cognition and efficient analysis.
[0005] An intelligent early warning system and method based on programmable multi-algorithm collaboration according to an embodiment of the present invention includes the following steps: Collect video monitoring data and business-related data, and preprocess them to generate standardized event datasets; Multiple early warning analysis algorithms are invoked to analyze the standardized event dataset, generate event cognition results, and construct an event cognition coverage structure according to the cognition dimensions of each event cognition result; Coverage analysis is performed on the event cognition dimensions in the event cognition coverage structure to identify the event cognition dimensions that are not fully covered and to generate event cognition coverage gaps; Based on the event cognition coverage gap matching early warning analysis algorithm, a multi-algorithm collaborative orchestration strategy is generated, and the early warning analysis algorithm is called according to the multi-algorithm collaborative orchestration strategy to supplement the event cognition coverage gap analysis; The event cognition coverage structure is updated using the newly added event cognition results obtained from the supplementary analysis, and the integrity of the updated event cognition coverage structure is checked to generate the event cognition completion result. Based on the results of event cognition, risk analysis is performed on the target event, and intelligent early warning results are generated.
[0006] Optionally, the video monitoring data specifically includes video image data, video time data, and camera device identification data; the business-related data specifically includes early warning rule data and scene area data; the preprocessing specifically includes time alignment, abnormal data removal, missing data completion, and data format unification processing.
[0007] Optionally, the construction of the event cognition coverage structure specifically includes: Read video image data from the standardized event dataset, divide the video time data into continuous event segments according to the time sequence of the video time data, and treat each continuous event segment as an independent event analysis object; The target detection algorithm, target tracking algorithm, behavior recognition algorithm, relationship analysis algorithm, and identity recognition algorithm are called respectively to analyze each independent event analysis object and generate spatial location recognition results, motion trajectory recognition results, behavior action recognition results, target relationship recognition results, and target identity recognition results. Extract the event identifier and cognitive dimension of each event cognitive result, merge them according to the same event identifier, and organize multiple event cognitive results belonging to the same event into an event cognitive result set; Traverse each event cognitive result set, count the cognitive dimensions contained in the event cognitive result set, construct event cognitive coverage units according to the correspondence between cognitive dimensions and event cognitive results, associate and organize all event cognitive coverage units of the same event to generate event cognitive coverage structure; The cognitive coverage structure of each event is saved based on the event identifier, forming a set of event cognitive coverage structures.
[0008] Optionally, the generation of the event cognition coverage gap specifically includes: Read the event cognition coverage structure of each event, traverse all event cognition dimensions in the event cognition coverage structure, and extract the event cognition results of each event cognition dimension; Based on the event cognition results of each event cognition dimension, the coverage status of each event cognition dimension is determined. Event cognition dimensions with event cognition results are determined as event cognition dimensions that have been fully covered, and event cognition dimensions without event cognition results are determined as event cognition dimensions that have not been fully covered. Based on all event cognitive dimensions for each event, the event cognitive dimensions that have been covered and those that have not been covered are linked and organized to form an event cognitive dimension coverage distribution; Traverse the coverage distribution of each event cognition dimension, extract all event cognition dimensions that are not fully covered, and merge them according to the event identifier to generate event cognition coverage gaps; Based on the event identifiers corresponding to the event cognition coverage gaps, the event cognition coverage gaps are classified and organized to form an event cognition coverage gap set.
[0009] Optionally, the supplementary analysis of the event cognition coverage gap specifically includes: Read the incomplete event cognition dimensions contained in the event cognition coverage gap, extract the items to be supplemented for analysis based on the cognition content corresponding to each event cognition dimension, and generate gap analysis results; Based on the gap analysis results, the set of early warning analysis algorithms is traversed, and early warning analysis algorithms that can complete the analysis items to be supplemented are selected. A mapping relationship between each analysis item to be supplemented and the corresponding early warning analysis algorithm is established, and candidate algorithm mapping results are generated. Based on the candidate algorithm mapping results, the data input dependency, data output dependency, and execution order dependency among the early warning analysis algorithms are analyzed, and the collaborative calling relationship among the early warning analysis algorithms is established according to the dependency relationship to generate the algorithm collaboration relationship; Based on the algorithm collaboration relationship, the calling order of the early warning analysis algorithms in the candidate algorithm mapping results is arranged, and a multi-algorithm collaborative arrangement strategy is generated by combining the early warning analysis algorithms corresponding to each supplementary analysis item. A multi-algorithm collaborative orchestration strategy is implemented, and the early warning analysis algorithm is called sequentially according to the calling order to perform supplementary analysis on each item to be supplemented, obtain the cognitive results of the new events, and associate and update the cognitive results of the new events with the corresponding cognitive dimensions in the event cognitive coverage gap to generate gap supplementation results.
[0010] Optionally, the generation of the event cognition completion result specifically includes: Read the newly added event perception results, extract the event identifier, event perception dimension, and event perception result of the newly added event perception results, match the corresponding event perception coverage structure according to the event identifier, and locate the event perception dimension of the newly added event perception results; Write the newly added event cognition results into the event cognition dimension of the corresponding event cognition coverage structure to generate the updated event cognition coverage structure; Traverse all event cognition dimensions in the updated event cognition coverage structure, check whether each event cognition dimension corresponds to an event cognition result, and record the event cognition dimensions with event cognition results and the event cognition dimensions without event cognition results to form the event cognition coverage detection result; Based on the event cognition coverage detection results, the updated event cognition coverage structure is subjected to integrity detection. When all event cognition dimensions correspond to event cognition results, the updated event cognition coverage structure is determined to meet the integrity requirements; when there are event cognition dimensions that do not correspond to event cognition results, the updated event cognition coverage structure is determined to not meet the integrity requirements. For an event cognition coverage structure that meets the integrity requirements, all event cognition results are extracted and organized according to event identifiers to generate event cognition completion results.
[0011] Optionally, the generation of the intelligent early warning result specifically includes: Read the event cognition completion result, extract the event identifier and event cognition result of the event cognition completion result, and read the warning rule data and scene area data according to the event identifier; The target identity, behavior, movement trajectory, target relationship and time sequence changes in the event perception results are matched with the early warning rule data to determine the risk assessment results of each event perception result; The risk assessment results corresponding to the same event identifier are summarized and processed, and the risk type and risk level of the target event are determined by combining the scene area data; Based on event identifiers, scene area data, risk type, risk level, and event perception results, intelligent early warning results for target events are generated.
[0012] An intelligent early warning system based on programmable multi-algorithm collaboration according to an embodiment of the present invention includes: The data acquisition and preprocessing module is used to collect video monitoring data and business-related data, preprocess the video monitoring data and business-related data, and generate standardized event datasets. The event cognition module is used to call multiple early warning analysis algorithms to analyze standardized event datasets, generate event cognition results, and construct an event cognition coverage structure according to the cognition dimensions of each event cognition result. The event cognition coverage analysis module is used to perform coverage analysis on the event cognition dimensions in the event cognition coverage structure, identify the event cognition dimensions that are not fully covered, and generate event cognition coverage gaps. The multi-algorithm collaborative orchestration module is used to match early warning analysis algorithms based on the event cognition coverage gap, generate a multi-algorithm collaborative orchestration strategy, and call early warning analysis algorithms to supplement the event cognition coverage gap analysis according to the multi-algorithm collaborative orchestration strategy; The event cognition completion module is used to update the event cognition coverage structure using the new event cognition results obtained from the supplementary analysis, and to perform an integrity check on the updated event cognition coverage structure to generate the event cognition completion result. The intelligent early warning module is used to perform risk analysis on target events based on the results of event cognition and generate intelligent early warning results.
[0013] The beneficial effects of this invention are: This invention proposes an intelligent early warning method based on programmable multi-algorithm collaboration. By collecting video monitoring data and business-related data, a standardized event dataset is constructed, and multiple early warning analysis algorithms are used to generate event cognition results. Furthermore, an event cognition coverage structure is constructed, and the cognition results of the target event on different event cognition dimensions are uniformly organized. This enables the event cognition results originally generated by different early warning analysis algorithms to be associated and managed around the same event identifier, providing a unified data foundation for subsequent event cognition coverage analysis and risk analysis.
[0014] Furthermore, this invention identifies gaps in event cognition coverage through an event cognition coverage structure, accurately locating event cognition dimensions that have not yet been fully covered. Based on these gaps, it dynamically matches corresponding early warning analysis algorithms, generating a multi-algorithm collaborative orchestration strategy. This enables multiple early warning analysis algorithms to be collaboratively invoked according to data dependencies and execution order, performing supplementary analysis only on event cognition dimensions that have not yet been fully covered. This avoids repeatedly performing analysis on event cognition dimensions that have already been covered, improving the targeting of early warning analysis algorithm invocation and overall processing efficiency.
[0015] Furthermore, this invention continuously updates the event cognition coverage structure using newly added event cognition results, and combines this with integrity detection to confirm whether the target event has completed cognition coverage of all event cognition dimensions. This ensures that the generated event cognition completion results contain complete event cognition information. Based on this, risk assessment is performed according to the complete event cognition results, and the risk type and risk level of the target event are determined by combining scene area data, ultimately generating intelligent early warning results. Compared with existing technologies, this invention can achieve unified organization of event cognition results, precise location of event cognition coverage gaps, and dynamic collaborative arrangement of multiple algorithms, improving the completeness of event cognition, the accuracy of intelligent early warning results, and overall processing efficiency, and has good engineering application value. Attached Figure Description
[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 The flowchart shows an intelligent early warning system and method based on programmable multi-algorithm collaboration proposed in this invention. Figure 2 This is a schematic diagram illustrating the generation of a multi-algorithm collaborative orchestration strategy for an intelligent early warning system and method based on programmable multi-algorithm collaboration proposed in this invention. Detailed Implementation
[0017] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.
[0018] refer to Figures 1-2 A smart early warning method based on programmable multi-algorithm collaboration includes the following steps: Collect video monitoring data and business-related data, and preprocess them to generate standardized event datasets; Multiple early warning analysis algorithms are invoked to analyze the standardized event dataset, generate event cognition results, and construct an event cognition coverage structure according to the cognition dimensions of each event cognition result; Coverage analysis is performed on the event cognition dimensions in the event cognition coverage structure to identify the event cognition dimensions that are not fully covered and to generate event cognition coverage gaps; Based on the event cognition coverage gap matching early warning analysis algorithm, a multi-algorithm collaborative orchestration strategy is generated, and the early warning analysis algorithm is called according to the multi-algorithm collaborative orchestration strategy to supplement the event cognition coverage gap analysis; The event cognition coverage structure is updated using the newly added event cognition results obtained from the supplementary analysis, and the integrity of the updated event cognition coverage structure is checked to generate the event cognition completion result. Based on the results of event cognition, risk analysis is performed on the target event, and intelligent early warning results are generated.
[0019] In this embodiment, the video monitoring data specifically includes video image data, video time data, and camera device identification data; the business-related data specifically includes early warning rule data and scene area data; the preprocessing specifically includes time alignment, abnormal data removal, missing data completion, and data format unification processing. The generation of standardized event datasets specifically includes: The system collects video image data, video frame identifier data, video time data, camera device identifier data, early warning rule data, and scene area data. Video image data is continuously acquired at a frequency of 25 to 30 frames per second. Video frame identifier data uses consecutive integer numbers, video time data is recorded using millisecond-level timestamps, and camera device identifier data is identified by unique device numbers. Data collected by different camera devices is categorized and organized based on the camera device identifier data. Video image data is sorted by time according to the video time data, and the video image data is then correlated with early warning rule data and scene area data based on the video time data. The system uses a continuous 1... The video data is organized as a time unit, with each unit containing 25 to 30 frames of video image data, arranged in ascending order of video frame identifier data. Each time unit is traversed, and abnormal data removal is performed on the video image data. When the number of consecutive abnormal video frames does not exceed 3 frames, missing data is filled in based on adjacent video frames. When the number of consecutive abnormal video frames exceeds 3 frames, the corresponding time unit is deleted. For missing warning rule data and scene area data, the most recent valid business-related data of the corresponding camera device is read to complete the filling. The video image data, video frame identifier data, video time data, camera device identifier data, warning rule data, and scene area data undergo unified data format processing. Video time data uniformly adopts the YYYY-MM-DD HH:MM:SS.mmm format, video frame identifier data uniformly adopts consecutive integer numbers, and camera device identifier data uniformly adopts a unique device number, thus completing the generation of a standardized event dataset.
[0020] In this embodiment, constructing the event cognition coverage structure specifically includes: Read video image data from the standardized event dataset, divide the video time data into continuous event segments according to the time sequence of the video time data, and treat each continuous event segment as an independent event analysis object; The target detection algorithm, target tracking algorithm, behavior recognition algorithm, relationship analysis algorithm, and identity recognition algorithm are called respectively to analyze each independent event analysis object and generate spatial location recognition results, motion trajectory recognition results, behavior action recognition results, target relationship recognition results, and target identity recognition results. Event cognition results include spatial location cognition results, motion trajectory cognition results, behavioral action cognition results, target relationship cognition results, and target identity cognition results; The process for generating event recognition results specifically includes: reading video image data, video time data, and scene region data from independent event analysis objects. Each independent event analysis object corresponds to a continuous video segment of 1 second to 10 seconds, containing 25 to 300 frames of video image data; calling a target detection algorithm to perform target detection frame by frame on the video image data, identifying targets in the video images, and extracting the spatial location information of the targets; when the number of targets identified in a single frame of video image is 1 to 50, assigning a unique target identifier to each target and generating a spatial location recognition result; associating targets with the same target identifier based on the video time data, calling a target tracking algorithm to generate continuous motion trajectories for each target, and generating motion trajectory recognition results when the number of continuously associated video frames is no less than 5 frames; combining the spatial location recognition results and motion trajectory recognition results, calling a behavior recognition algorithm to perform behavioral actions on each target. The process involves several steps: First, when the same action is continuously recognized for at least three frames, a behavioral action recognition result is generated. Second, based on the spatial location recognition result, motion trajectory recognition result, and behavioral action recognition result, a relationship analysis algorithm is used to analyze the target relationships between targets. When the target relationship is maintained continuously for at least three frames, a target relationship recognition result is generated. Third, based on video image data and warning rule data, an identity recognition algorithm is used to identify the identity information of each target, generating a target identity recognition result. Finally, the spatial location recognition result, motion trajectory recognition result, behavioral action recognition result, target relationship recognition result, and target identity recognition result are written with corresponding event identifiers and recognition dimensions, and then organized and processed according to the event identifiers to form the recognition result. The cognitive dimensions specifically include spatial location cognitive dimension, motion trajectory cognitive dimension, behavioral action cognitive dimension, target relationship cognitive dimension, and target identity cognitive dimension. The spatial location cognitive dimension is used to represent the spatial location information of the target in video images; the motion trajectory cognitive dimension is used to represent the motion trajectory information of the target in continuous video images; the behavioral action cognitive dimension is used to represent the behavioral action information of the target in video images; the target relationship cognitive dimension is used to represent the relationship information between targets; and the target identity cognitive dimension is used to represent the identity information of the target. Each cognitive dimension corresponds to a cognitive result of an event, and together they constitute the complete cognitive content of the same event. Extract the event identifier and cognitive dimension of each event cognitive result, merge them according to the same event identifier, and organize multiple event cognitive results belonging to the same event into an event cognitive result set; Traverse each event cognitive result set, count the cognitive dimensions contained in the event cognitive result set, construct event cognitive coverage units according to the correspondence between cognitive dimensions and event cognitive results, associate and organize all event cognitive coverage units of the same event to generate event cognitive coverage structure; The formation of an event cognition coverage structure specifically includes: Read all event cognition results from the event cognition result set, and extract the event identifier and cognition dimension corresponding to each event cognition result. Based on the cognition dimensions, establish event cognition coverage units corresponding to spatial location cognition dimension, motion trajectory cognition dimension, behavior action cognition dimension, target relationship cognition dimension, and target identity cognition dimension. Each event corresponds to 5 event cognition coverage units, and each event cognition coverage unit includes at least an event identifier, cognition dimension, and event cognition result. Write the event cognition results into the event cognition coverage units according to the cognition dimensions. When the number of event cognition results corresponding to the same event is 2 to 5, complete the data writing of 2 to 5 event cognition coverage units. For cognition dimensions that have not generated event cognition results, retain the event cognition coverage unit and set the event cognition result to empty. After completing the construction of all 5 event cognition coverage units for the same event, read all corresponding event cognition coverage units according to the same event identifier, arrange them in a fixed order of spatial location cognition dimension, motion trajectory cognition dimension, behavior action cognition dimension, target relationship cognition dimension, and target identity cognition dimension, and establish the corresponding relationship in sequence to form the event cognition coverage structure of the corresponding event. The event cognition coverage structure refers to a data structure that unifies and organizes the event cognition results corresponding to the spatial location cognition dimension, motion trajectory cognition dimension, behavioral action cognition dimension, target relationship cognition dimension, and target identity cognition dimension based on the same event identifier. It is used to uniformly describe the cognitive completion status of the same target event in each event cognition dimension. The event cognition coverage structure can not only centrally organize the event cognition results generated by multiple early warning analysis algorithms, but also clearly identify the event cognition dimensions that have not been fully covered. It provides a unified data organization foundation for event cognition coverage gap location, multi-algorithm collaborative orchestration, event cognition integrity detection, and event cognition completion result generation, thereby improving the completeness, continuity, and processing efficiency of event cognition analysis. The cognitive coverage structure of each event is saved based on the event identifier, forming a set of event cognitive coverage structures.
[0021] In this embodiment, the generation of event cognition coverage gaps specifically includes: Read the event cognition coverage structure of each event, traverse all event cognition dimensions in the event cognition coverage structure, and extract the event cognition results of each event cognition dimension; Based on the event cognition results of each event cognition dimension, the coverage status of each event cognition dimension is determined. Event cognition dimensions with event cognition results are determined as event cognition dimensions that have been fully covered, and event cognition dimensions without event cognition results are determined as event cognition dimensions that have not been fully covered. Based on all event cognitive dimensions for each event, the event cognitive dimensions that have been covered and those that have not been covered are linked and organized to form an event cognitive dimension coverage distribution; The generation of the event cognition dimension coverage distribution specifically includes: The system reads the coverage status of the spatial location cognition dimension, motion trajectory cognition dimension, behavioral action cognition dimension, target relationship cognition dimension, and target identity cognition dimension corresponding to the same event. It counts the number of event cognition dimensions that are fully covered and the number that are not fully covered, with each event corresponding to 5 event cognition dimensions. Based on a fixed order of spatial location cognition dimension, motion trajectory cognition dimension, behavioral action cognition dimension, target relationship cognition dimension, and target identity cognition dimension, the coverage status of each event cognition dimension is sequentially arranged. When an event cognition dimension is fully covered, a "fully covered" flag is written to the corresponding event cognition dimension position; when an event cognition dimension is not fully covered, an "incomplete" flag is written to the corresponding event cognition dimension position. After writing the coverage status of all 5 event cognition dimensions, the system organizes the corresponding 5 event cognition dimension coverage statuses sequentially according to the event flags to form the event cognition dimension coverage distribution for the corresponding event. Traverse the coverage distribution of each event cognition dimension, extract all event cognition dimensions that are not fully covered, and merge them according to the event identifier to generate event cognition coverage gaps; The generation of event awareness coverage gaps specifically includes: Traverse all event cognitive dimension coverage distributions, and read the event identifier, event cognitive dimension, and coverage status corresponding to each event cognitive dimension coverage distribution; check the coverage status corresponding to each event cognitive dimension in turn. When the coverage status is an incomplete coverage identifier, extract the event identifier and event cognitive dimension, and generate an incomplete coverage record; when the number of incomplete coverage event cognitive dimensions corresponding to the same event is 1 to 5, retain all incomplete coverage records; merge all incomplete coverage records belonging to the same event according to the event identifier, and arrange them in a fixed order of spatial location cognitive dimension, motion trajectory cognitive dimension, behavioral action cognitive dimension, target relationship cognitive dimension, and target identity cognitive dimension to form the event cognitive coverage gap of the corresponding event; Event cognition coverage gaps refer to the set of event cognition dimensions in the event cognition coverage structure that do not correspond to event cognition results. They are used to describe the part of the current target event that has not yet been cognized. Event cognition coverage gaps can clearly identify event cognition dimensions that need to be further analyzed, providing supplementary analysis targets for multi-algorithm collaborative orchestration strategies. This allows the early warning analysis algorithm to only call on event cognition dimensions that have not been fully covered, avoiding repeated analysis of event cognition dimensions that have been fully covered, and improving the pertinence, completeness and processing efficiency of event cognition supplementary analysis. Based on the event identifiers corresponding to the event cognition coverage gaps, the event cognition coverage gaps are classified and organized to form an event cognition coverage gap set.
[0022] In this embodiment, the supplementary analysis of the event cognition coverage gap specifically includes: Read the incomplete event cognition dimensions contained in the event cognition coverage gap, extract the items to be supplemented for analysis based on the cognition content corresponding to each event cognition dimension, and generate gap analysis results; Based on the gap analysis results, the set of early warning analysis algorithms is traversed, and early warning analysis algorithms that can complete the analysis items to be supplemented are selected. A mapping relationship between each analysis item to be supplemented and the corresponding early warning analysis algorithm is established, and candidate algorithm mapping results are generated. The process of generating the candidate algorithm mapping results is as follows: Read all the missing analysis items from the gap analysis results, and extract the event identifier, event cognition dimension, cognition content, scene area data, and early warning rule data corresponding to each missing analysis item. Read the algorithm identifier, supported cognition dimension, input data type, output event cognition result type, supported video resolution, supported video frame rate, supported scene area, and supported target type for each early warning analysis algorithm in the early warning analysis algorithm set. The target detection algorithm supports the spatial location cognition dimension, with input data being video image data and output event cognition result type being spatial location cognition result. Supported video resolutions include 1280×720, 1920×1080, and 2560×1440, and a supported video frame rate of 25. The frame rate is fps to 30fps, supporting target types including people, vehicles, and objects. The target tracking algorithm supports motion trajectory recognition, with input data consisting of video image data and video time data, and outputting motion trajectory recognition results. The behavior recognition algorithm supports behavior action recognition, with input data consisting of spatial location recognition results and motion trajectory recognition results, and outputting behavior action recognition results, supporting behaviors including standing still, loitering, running, falling, and climbing. The relationship analysis algorithm supports target relationship recognition, with input data consisting of spatial location recognition results, motion trajectory recognition results, and behavior action recognition results, and outputting target relationship recognition results, supporting scene areas including passable areas, equipment areas, dangerous areas, and restricted areas. The identity recognition algorithm supports target identity recognition, with input data consisting of video image data and warning rule data, and outputting target identity recognition results. The system supports identity types including employees, visitors, construction workers, and unknown identities. It sequentially reads the event cognition dimensions corresponding to each item to be supplemented in the analysis and iterates through the set of early warning analysis algorithms. When the supported cognition dimension of the early warning analysis algorithm matches the event cognition dimension, the supported scene area includes the current scene area data, the supported video resolution matches the video image data resolution, the supported video frame rate covers the video image data acquisition frame rate, the supported target type meets the cognition content requirements, and the output event cognition result type corresponds to the cognition content, a mapping relationship is established between the item to be supplemented in the analysis and the early warning analysis algorithm. When establishing the mapping relationship, the event identifier, event cognition dimension, algorithm identifier, input data type, output event cognition result type, supported scene area, and supported target type are recorded. After establishing the mapping relationship for all items to be supplemented in the analysis, all mapping relationships belonging to the same event are organized and processed according to the event identifier, and candidate algorithm mapping results for the corresponding events are generated according to the order of the algorithm identifiers. The candidate algorithm mapping result refers to the data result formed by establishing a correspondence between each analysis item to be supplemented and the early warning analysis algorithm that can complete the corresponding analysis task based on the event cognition coverage gap. It is used to determine the candidate early warning analysis algorithm corresponding to each analysis item to be supplemented. The candidate algorithm mapping result can clarify the early warning analysis algorithm corresponding to different analysis items to be supplemented, and provide the algorithm selection basis for establishing algorithm collaboration relationship and generating multi-algorithm collaborative orchestration strategy. This enables the early warning analysis algorithm to supplement the analysis of event cognition coverage gap, and improves the targeting and processing efficiency of early warning analysis algorithm invocation. Based on the candidate algorithm mapping results, the data input dependency, data output dependency, and execution order dependency among the early warning analysis algorithms are analyzed, and the collaborative calling relationship among the early warning analysis algorithms is established according to the dependency relationship to generate the algorithm collaboration relationship; The generation of algorithmic collaborative relationships specifically includes: The algorithm reads the event identifier, algorithm identifier, corresponding cognitive dimension, input data type, output event cognitive result type, supported video resolution, supported video frame rate, supported scene region, and supported target type for each early warning analysis algorithm in the candidate algorithm mapping results. The target detection algorithm corresponds to the spatial location cognitive dimension, supports video resolutions including 1280×720, 1920×1080, and 2560×1440, supports video frame rates of 25 fps to 30 fps, supports target types including people, vehicles, and objects, and outputs spatial location cognitive results. The target tracking algorithm corresponds to the motion trajectory cognitive dimension, with a trajectory association window length of 8 to 32 frames, a trajectory buffer length of 128 frames, and a trajectory continuous holding time of 0.The motion trajectory recognition result is output within 5 to 10 seconds; the behavior recognition algorithm corresponds to the behavior action recognition dimension, with behavior analysis time windows including 1 second, 3 seconds, 5 seconds, and 10 seconds, and continuous behavior confirmation frames ranging from 3 to 15 frames. Corresponding behaviors include stopping, wandering, running, falling, and climbing, and the behavior action recognition result is output; the relationship analysis algorithm corresponds to the target relationship recognition dimension, with target distance calculation ranging from 0 pixels to 300 pixels. The target orientation angle ranges from 0° to 360°, with 2 to 32 associated targets, outputting target relationship recognition results. The identity recognition algorithm corresponds to the target identity recognition dimension, with a face image resolution of at least 112×112 pixels, an identity feature length of 256 dimensions, and an identity comparison scale supporting 1:1000. Corresponding identity types include employees, visitors, construction workers, and unknown identities, outputting target identity recognition results. All early warning analysis algorithms are traversed, and the input data type of each early warning analysis algorithm is matched item by item with the output event recognition result type of other early warning analysis algorithms. When the output event recognition result type matches the input data type, a corresponding data input dependency relationship is established. The spatial location recognition result output by the target detection algorithm serves as the input data for the behavior recognition algorithm and the relationship analysis algorithm, while the motion trajectory recognition result output by the target tracking algorithm serves as the input for the behavior recognition algorithm and the relationship analysis algorithm. The behavioral action recognition results output by the behavior recognition algorithm serve as input data for the relationship analysis algorithm. Data output dependencies are established based on input dependencies, and the execution order dependencies are determined accordingly. The target detection algorithm is designated as a pre-warning analysis algorithm for the target tracking algorithm, behavior recognition algorithm, and relationship analysis algorithm; the target tracking algorithm is designated as a pre-warning analysis algorithm for the behavior recognition algorithm and relationship analysis algorithm; and the behavior recognition algorithm is designated as a pre-warning analysis algorithm for the relationship analysis algorithm. The identity recognition algorithm executes independently based on video image data and warning rule data. When establishing algorithm collaboration relationships, the event identifier, pre-warning analysis algorithm, post-warning analysis algorithm, input data type, output event recognition result type, corresponding recognition dimension, and call sequence number are recorded. The call sequence number is consecutively numbered 01, 02, 03, 04, ultimately generating the algorithm collaboration relationship. Algorithm collaboration relationships refer to the collaborative calling relationships established based on the data input dependencies, data output dependencies, and execution order dependencies among early warning analysis algorithms. They are used to describe the collaborative cooperation methods among multiple early warning analysis algorithms. Algorithm collaboration relationships can clarify the data transmission order and calling sequence among various early warning analysis algorithms, providing an algorithm orchestration basis for multi-algorithm collaborative orchestration strategies. This ensures that multiple early warning analysis algorithms complete supplementary analysis in the correct collaborative order, improving the continuity, consistency, and processing efficiency of the generation of new event cognition results. Based on the algorithm collaboration relationship, the calling order of the early warning analysis algorithms in the candidate algorithm mapping results is arranged, and a multi-algorithm collaborative arrangement strategy is generated by combining the early warning analysis algorithms corresponding to each supplementary analysis item. The generation of multi-algorithm collaborative orchestration strategies specifically includes: The algorithm retrieves the event identifier, pre-warning analysis algorithm, post-warning analysis algorithm, input data type, output event cognition result type, corresponding cognition dimension, and call sequence number from the algorithm collaboration relationship. It also retrieves the pre-warning analysis algorithm corresponding to each supplementary analysis item in the candidate algorithm mapping results. Based on the call sequence number, the pre-warning analysis algorithms are arranged in a continuous order of 01, 02, 03, and 04. The event cognition result output by the pre-warning analysis algorithm is provided to the post-warning analysis algorithm as input data within 50ms. Data with the same event identifier is continuously transmitted between the two pre-warning analysis algorithms. For video image data with resolutions of 1280×720, 1920×1080, or 2560×1440 and a frame rate of 25 fps to 30 fps... For FPS event data, the call order of early warning analysis algorithms is organized according to the corresponding event identifier. When the same event corresponds to multiple items to be supplemented for analysis, the corresponding early warning analysis algorithms are inserted sequentially between the call order numbers based on the positions of the preceding and following early warning analysis algorithms in the algorithm collaboration relationship, ensuring that the output event cognition result type is consistent with the input data type of the next early warning analysis algorithm. For early warning analysis algorithms without data input dependencies, they are organized into the same call order according to the call order number. For early warning analysis algorithms with data input dependencies, they are called sequentially according to the call order number, ensuring that the output event cognition result type corresponds continuously with the input data type. A multi-algorithm collaborative orchestration strategy is generated for 16 to 64 events at a time, with each event corresponding to one call order orchestration result. The call order numbers of all early warning analysis algorithms for the same event are kept consecutive. After completing the call order orchestration of all early warning analysis algorithms, the event identifier, preceding early warning analysis algorithm, following early warning analysis algorithm, call order number, input data type, output event cognition result type, and corresponding cognition dimension are jointly organized to form the multi-algorithm collaborative orchestration strategy for the corresponding event. The multi-algorithm collaborative orchestration strategy refers to the strategy for organizing the invocation of early warning analysis algorithms based on the event cognition coverage gap, candidate algorithm mapping results, and algorithm collaboration relationships. It is used to determine the invocation order and collaboration method of each early warning analysis algorithm. The multi-algorithm collaborative orchestration strategy can dynamically select the early warning analysis algorithms to be invoked according to the event cognition dimension corresponding to the event cognition coverage gap, and complete the collaborative invocation according to the data dependency relationship and execution order between algorithms. This allows the newly added event cognition results to gradually supplement the event cognition coverage structure, avoid the repeated analysis of the event cognition dimensions that have already been covered, and improve the pertinence, collaboration, and processing efficiency of multi-algorithm collaborative analysis. A multi-algorithm collaborative orchestration strategy is implemented, and the early warning analysis algorithm is called sequentially according to the calling order to perform supplementary analysis on each item to be supplemented, obtain the cognitive results of the new events, and associate and update the cognitive results of the new events with the corresponding cognitive dimensions in the event cognitive coverage gap to generate gap supplementation results; The generation of gap filling results specifically includes: The algorithm acquires event identifiers, call sequence numbers, early warning analysis algorithms, event cognition dimensions corresponding to items to be supplemented in the multi-algorithm collaborative orchestration strategy, input data types, and output event cognition result types. It then sequentially calls the corresponding early warning analysis algorithms according to the call sequence number and reads corresponding data based on the input data type for supplementary analysis. The target detection algorithm reads video image data with resolutions of 1280×720, 1920×1080, and 2560×1440, and a frame rate of 25 fps to 30 fps, generating spatial location cognition results. The target tracking algorithm reads video image data and video time data, performs correlation analysis on target positions in continuous video images according to a trajectory association window length of 8 to 32 frames, and generates motion trajectory cognition results. The behavior recognition algorithm reads the spatial location cognition results and motion trajectory cognition results, analyzes target behavior within behavior analysis time windows of 1 s, 3 s, 5 s, and 10 s, and generates behavior action cognition results. The relationship analysis algorithm reads the spatial location cognition results, motion trajectory cognition results, and behavior action cognition results, and analyzes target behavior based on distances between targets ranging from 0 to 300 pixels. Relationship analysis is performed on pixels and target orientation angles from 0° to 360° to generate target relationship recognition results; the identity recognition algorithm reads video image data and early warning rule data, analyzes the identity information in the video image data, and generates target identity recognition results; the event recognition results output by each early warning analysis algorithm are organized and processed according to event identifiers, and the corresponding event recognition dimensions are extracted; based on the event recognition dimensions, the event recognition coverage gaps are matched with the incomplete event recognition dimensions, and the corresponding new event recognition results are written into the corresponding event recognition dimensions in the event recognition coverage gaps; when all the incomplete event recognition dimensions in the event recognition coverage gaps have generated corresponding new event recognition results, it is determined that the corresponding event recognition coverage gap has been supplemented; the event identifiers, event recognition coverage gaps, event recognition dimensions, and new event recognition results are organized and processed to form gap supplementation results; Gap filling results refer to the data results formed after the newly added event cognition results are associated with and updated with the event cognition coverage gaps. They are used to describe the completion status of filling the event cognition coverage gaps. Gap filling results can record the supplementation of the event cognition dimensions of the newly added event cognition results, providing a data basis for updating the event cognition coverage structure and subsequent event cognition integrity detection, so that the event cognition coverage structure can gradually complete the cognition coverage of all event cognition dimensions, and improve the integrity and continuity of event cognition results.
[0023] In this embodiment, the generation of the event cognition completion result specifically includes: Read the newly added event perception results, extract the event identifier, event perception dimension, and event perception result of the newly added event perception results, match the corresponding event perception coverage structure according to the event identifier, and locate the event perception dimension of the newly added event perception results; Write the newly added event cognition results into the event cognition dimension of the corresponding event cognition coverage structure to generate the updated event cognition coverage structure; Traverse all event cognition dimensions in the updated event cognition coverage structure, check whether each event cognition dimension corresponds to an event cognition result, and record the event cognition dimensions with event cognition results and the event cognition dimensions without event cognition results to form the event cognition coverage detection result; The generation of event cognitive coverage detection results specifically includes: The system reads the event identifier, event cognitive dimension, and event cognitive result from the updated event cognitive coverage structure. Following a fixed order of spatial location cognitive dimension, motion trajectory cognitive dimension, behavioral action cognitive dimension, target relationship cognitive dimension, and target identity cognitive dimension, it performs item-by-item detection on all event cognitive dimensions corresponding to the same event. When an event cognitive dimension corresponds to an event cognitive result, the corresponding event cognitive dimension is designated as having an event cognitive result, and the corresponding event identifier, event cognitive dimension, and event cognitive result are recorded. When an event cognitive dimension does not correspond to an event cognitive result, the corresponding event cognitive dimension is designated as having no event cognitive result, and the corresponding event identifier and event cognitive dimension are recorded. After completing the detection of all event cognitive dimensions for the same event, the system organizes and processes the event cognitive dimensions with and without event cognitive results according to the event identifier, forming the event cognitive coverage detection result for the corresponding event. The event cognitive coverage detection result includes the event identifier, the event cognitive dimension with an event cognitive result, the event cognitive dimension without an event cognitive result, and the event cognitive result. The event cognition coverage detection results are used to describe the data results of the coverage of each event cognition dimension in the updated event cognition coverage structure. The event cognition coverage detection results can uniformly record the coverage status of the event cognition results corresponding to each event cognition dimension, so as to directly determine whether the target event has completed the cognition coverage of all event cognition dimensions. This provides a basis for judging the integrity of the event cognition coverage structure and generating the event cognition completion results, avoids repeatedly traversing the entire event cognition coverage structure, and improves the efficiency of event cognition integrity detection. Based on the event cognition coverage detection results, the updated event cognition coverage structure is subjected to integrity detection. When all event cognition dimensions correspond to event cognition results, the updated event cognition coverage structure is determined to meet the integrity requirements; when there are event cognition dimensions that do not correspond to event cognition results, the updated event cognition coverage structure is determined to not meet the integrity requirements. Integrity checks specifically include: The event recognition coverage detection results are read from the event identifier, event recognition dimensions with event recognition results, event recognition dimensions without event recognition results, and corresponding event recognition results. The updated event recognition coverage structure is then matched based on the event identifier. Following the detection order of spatial location recognition dimension, motion trajectory recognition dimension, behavioral action recognition dimension, target relationship recognition dimension, and target identity recognition dimension, all event recognition dimensions in the updated event recognition coverage structure are detected item by item. During the detection process, the existence of event recognition results corresponding to each event recognition dimension is read, and the number of event recognition dimensions with and without event recognition results is counted. The event recognition coverage structure corresponding to the same event includes five event recognition dimensions: spatial location recognition dimension, motion trajectory recognition dimension, behavioral action recognition dimension, target relationship recognition dimension, and target identity recognition dimension. When the number of event cognition dimensions with corresponding event cognition results is 5 and the number of event cognition dimensions without corresponding event cognition results is 0, the updated event cognition coverage structure is determined to meet the integrity requirements. When any event cognition dimension is detected to have no corresponding event cognition result, the corresponding event identifier and the event cognition dimension without corresponding event cognition result are recorded, and the updated event cognition coverage structure is determined to not meet the integrity requirements. After completing the detection of all event cognition dimensions corresponding to the same event, the event cognition dimension detection results are organized according to the event identifier to form the integrity detection results of the corresponding event. The integrity detection results include the event identifier, the number of event cognition dimensions with event cognition results, the number of event cognition dimensions without event cognition results, and the integrity detection results of the corresponding event cognition coverage structure. The entire event cognition coverage detection results are traversed, and the processing procedure is repeated to complete the integrity detection of all updated event cognition coverage structures. Completeness detection refers to the process of determining whether the updated event cognition coverage structure has completed the coverage of all event cognition dimensions based on the event cognition coverage detection results. Completeness detection can confirm whether the target event has obtained complete event cognition results, prevent events with incomplete cognition coverage from directly entering the risk analysis process, provide a basis for judgment for the generation of event cognition completion results, and ensure that subsequent risk judgment results, risk types, risk levels and intelligent early warning results are based on complete and consistent event cognition, thereby improving the accuracy and reliability of intelligent early warning results. For an event cognition coverage structure that meets the integrity requirements, all event cognition results are extracted and organized according to event identifiers to generate event cognition completion results; The process by which the cognitive outcome of an event is formed is as follows: The system reads the event cognition coverage structure that meets the integrity requirements, extracts the corresponding event cognition results based on the event identifier, and reads the video time data, camera device identifier data, scene area data, and warning rule data corresponding to the event identifier. Video time data is recorded using millisecond-level timestamps; for example, the event start time is recorded as 09:18:23.125, and the event end time is recorded as 09:18:31.867. The time range of the corresponding event is determined based on the video time data. Camera device identifier data is identified using unique device numbers; for example, CAM-00128 and CAM-00129 correspond to data collected by different camera devices. Warning rule data is distinguished according to rule identifiers; for example, RULE-018, RULE-026, and RULE-031 correspond to different warning rules. The system extracts the spatial location cognition results, motion trajectory cognition results, behavioral action cognition results, target relationship cognition results, and target identity cognition results from the event cognition coverage structure and organizes them according to the event identifier. The spatial location cognition results retain the position change information of the corresponding target in the video image data; for example, target identifier T00045 in the continuous... The system records position changes from video frames 15263 to 15482; the motion trajectory recognition result retains the trajectory change information during the continuous movement of the corresponding target, for example, the corresponding target contains 218 consecutive trajectory position records; the behavior and action recognition result retains the behavior and action recognition result of the corresponding target; the target relationship recognition result retains the relationship analysis result between the corresponding targets; the target identity recognition result retains the target identity recognition result; according to the time order of the video time data, all event recognition results corresponding to the same event are organized and processed, and the correspondence between event identifiers, camera device identifier data, scene area data, warning rule data and event recognition results is maintained; when multiple events correspond to the same camera device identifier data, the corresponding event recognition results are organized separately according to the event identifier to avoid confusion between event recognition results corresponding to different events; after completing the organization of all event recognition results, the event identifiers, video time data, camera device identifier data, scene area data, warning rule data, as well as the spatial position recognition result, motion trajectory recognition result, behavior and action recognition result, target relationship recognition result and target identity recognition result are jointly organized to form the event recognition completion result; Event cognition completion results refer to the data results formed by organizing all event cognition results in the event cognition coverage structure that meets the integrity requirements according to the event identifier. It is used to uniformly describe the complete event cognition information of the target event. Event cognition completion results can centrally organize spatial location cognition results, motion trajectory cognition results, behavioral action cognition results, target relationship cognition results, and target identity cognition results. It provides a complete and consistent data foundation for the generation of risk assessment results, the determination of risk type and risk level, and the generation of intelligent early warning results, thereby improving the accuracy and completeness of target event risk analysis.
[0024] In this embodiment, the generation of intelligent early warning results specifically includes: Read the event cognition completion result, extract the event identifier and event cognition result of the event cognition completion result, and read the warning rule data and scene area data according to the event identifier; The target identity, behavior, movement trajectory, target relationship and time sequence changes in the event perception results are matched with the early warning rule data to determine the risk assessment results of each event perception result; The generation of risk assessment results specifically includes: The system reads the event identifier and event recognition result from the event recognition completion result, and then reads the corresponding early warning rule data and scene area data based on the event identifier. Early warning rule data is distinguished by rule identifiers; for example, RULE-018, RULE-026, and RULE-031 correspond to different early warning rule data. Scene area data is distinguished by scene area identifiers; for example, AREA-001, AREA-003, and AREA-008 correspond to different scene areas. A correspondence is established between the event recognition result and the early warning rule data based on the event identifier. The system then extracts the target identity, behavioral actions, movement trajectory, target relationship, and time sequence changes from the event recognition result and matches them with the corresponding early warning rule data. Target identity is matched with the identity information in the target identity recognition result; behavioral actions are matched with the behavioral action information in the behavioral action recognition result; movement trajectory is matched with the trajectory change information in the movement trajectory recognition result; and target relationship is matched with the target relationship information in the target relationship recognition result. The target association information is matched with the early warning rule data. Temporal changes are matched with the time sequence information in the video time data. During the matching process, video time data, camera device identification data, and scene area data corresponding to the event identifier are read. The video time data is recorded using millisecond-level timestamps; for example, the video time data range for an event is 09:18:23.125 to 09:18:31.867. Camera device identification data is identified using a unique device number; for example, CAM-00128 corresponds to the video monitoring data collected for this event. When the event perception result meets the matching requirements of the corresponding early warning rule data, the corresponding event perception result is determined to meet the risk assessment conditions. When the event perception result does not meet the matching requirements of the corresponding early warning rule data, the corresponding event perception result is determined to not meet the risk assessment conditions. After completing the matching of all event perception results with the early warning rule data, the corresponding early warning rule data, event perception results, and matching results are organized according to the event identifier to generate a risk assessment result. The risk assessment results corresponding to the same event identifier are summarized and processed, and the risk type and risk level of the target event are determined by combining the scene area data; The generation of risk types and risk levels specifically includes: The system reads all risk assessment results corresponding to the same event identifier, and then reads the corresponding scene area data, warning rule data, video time data, and camera device identifier data based on the event identifier. Scene area data is distinguished by scene area identifiers; for example, AREA-001, AREA-003, and AREA-008 correspond to different scene areas. Warning rule data is distinguished by rule identifiers; for example, RULE-018, RULE-026, and RULE-031 correspond to different warning rule data. Camera device identifier data is identified by a unique device number; for example, CAM-00128 and CAM-00129 correspond to different camera devices. Video time data is recorded using millisecond-level timestamps; for example, the video time data range for the event is 09:18:23.125 to 09:18:31.867. The risk assessment results are merged according to the event identifier, and a corresponding relationship is established based on the event cognition results, early warning rule data, and scene area data corresponding to the risk assessment results. When the same event identifier corresponds to multiple risk assessment results, the risk assessment results are arranged according to the chronological order of the video time data. For example, the risk assessment results corresponding to 09:18:23.125, 09:18:24.080, and 09:18:25.615 are organized in chronological order. The risk type of the corresponding target event is determined based on the merged risk assessment results, and the risk level of the target event is determined by combining the scene area data and early warning rule data. After the risk type and risk level are determined, the event identifier, scene area data, risk type, risk level, and corresponding risk assessment results are linked and organized to form the risk type and risk level corresponding to the target event. Based on event identifiers, scene area data, risk type, risk level, and event perception results, intelligent early warning results for target events are generated.
[0025] An intelligent early warning system based on programmable multi-algorithm collaboration includes: The data acquisition and preprocessing module is used to collect video monitoring data and business-related data, preprocess the video monitoring data and business-related data, and generate standardized event datasets. The event cognition module is used to call multiple early warning analysis algorithms to analyze standardized event datasets, generate event cognition results, and construct an event cognition coverage structure according to the cognition dimensions of each event cognition result. The event cognition coverage analysis module is used to perform coverage analysis on the event cognition dimensions in the event cognition coverage structure, identify the event cognition dimensions that are not fully covered, and generate event cognition coverage gaps. The multi-algorithm collaborative orchestration module is used to match early warning analysis algorithms based on the event cognition coverage gap, generate a multi-algorithm collaborative orchestration strategy, and call early warning analysis algorithms to supplement the event cognition coverage gap analysis according to the multi-algorithm collaborative orchestration strategy; The event cognition completion module is used to update the event cognition coverage structure using the new event cognition results obtained from the supplementary analysis, and to perform an integrity check on the updated event cognition coverage structure to generate the event cognition completion result. The intelligent early warning module is used to perform risk analysis on target events based on the results of event cognition and generate intelligent early warning results.
[0026] Example 1: To verify the feasibility of the present invention in practice, it was applied to a video-based intelligent early warning scenario in a large smart park. The park includes multiple monitoring areas such as personnel entrances and exits, office areas, equipment areas, storage areas, and public passages. Each monitoring area is equipped with video monitoring equipment, which continuously collects video monitoring data through a video monitoring platform. The video monitoring equipment continuously acquires video image data at a fixed acquisition frequency, and simultaneously generates corresponding video time data and camera identification data. The management platform synchronously maintains the early warning rule data and scene area data corresponding to each monitoring area. Due to frequent personnel activity within the park, various events may occur in different scenarios, such as personnel gathering, abnormal stays, unauthorized entry, and abnormal behavior. Different early warning analysis algorithms are responsible for target detection, target tracking, behavior recognition, relationship analysis, and identity recognition, respectively. Each algorithm independently outputs corresponding event recognition results. Traditional processing methods typically call all early warning analysis algorithms sequentially according to a fixed process. When the event recognition results of some events can meet the analysis requirements, the remaining early warning analysis algorithms are still called for repeated analysis. However, when some early warning analysis algorithms do not obtain effective analysis results, they cannot automatically identify the event recognition dimensions that have not yet been recognized, which can easily lead to incomplete target event recognition information and affect the accuracy of subsequent risk analysis.
[0027] When applied to the aforementioned scenarios, this invention first continuously collects video monitoring data and business-related data, and completes standardization processing. Video image data is collected using two resolutions: 1920×1080 pixels and 1280×720 pixels. The video acquisition frequency is maintained at 25 fps to 30 fps. Video time data is recorded using millisecond-level timestamps. Camera device identification data is identified using unique device numbers, such as CAM-00128, CAM-00129, and CAM-00135, which correspond to different monitoring locations. Business-related data includes early warning rule data and scene area data. Early warning rule data is organized using rule identifiers such as RULE-018, RULE-026, and RULE-031, while scene area data is organized using scene area identifiers such as AREA-001, AREA-003, and AREA-008. The system performs time alignment based on the video time data and organizes the data into a standardized event dataset according to continuous video frames, providing unified input data for subsequent multiple early warning analysis algorithms.
[0028] Subsequently, the system calls target detection, target tracking, behavior recognition, relationship analysis, and identity recognition algorithms to analyze the standardized event dataset. The target detection algorithm first identifies target objects in the video image data. The target tracking algorithm generates continuous motion trajectories of the target based on continuous video frames. The behavior recognition algorithm identifies the behavior actions corresponding to the target. The relationship analysis algorithm analyzes the target relationships between different targets. The identity recognition algorithm identifies the target identity, generating spatial location recognition results, motion trajectory recognition results, behavior action recognition results, target relationship recognition results, and target identity recognition results, respectively. The system organizes the above event recognition results uniformly based on the same event identifier to form an event recognition coverage structure, so that the event recognition results generated by multiple early warning analysis algorithms are associated around the same target event, instead of being scattered and stored in each early warning analysis algorithm.
[0029] After the event cognition coverage structure is formed, the system further traverses each event cognition dimension to analyze whether each event cognition dimension of the current target event has corresponding event cognition results. When some events have only completed spatial location cognition results and motion trajectory cognition results, but behavioral action cognition results or target identity cognition results have not yet been formed, the system automatically identifies that the corresponding event cognition dimension has not been fully covered and generates an event cognition coverage gap. Subsequently, based on the event cognition coverage gap, it extracts the items to be supplemented for analysis, traverses the set of early warning analysis algorithms, generates candidate algorithm mapping results based on the correspondence between the items to be supplemented for analysis and the early warning analysis algorithms, and establishes algorithm collaboration relationships based on the data input dependencies, data output dependencies, and execution order dependencies among the early warning analysis algorithms, and generates a multi-algorithm collaborative orchestration strategy. For example, when the target tracking algorithm has already generated a complete motion trajectory cognition result, the system no longer calls the target tracking algorithm repeatedly, but directly calls the behavior recognition algorithm or identity recognition algorithm based on the event cognition coverage gap to complete the supplementary analysis of the corresponding event cognition dimension. This allows multiple early warning analysis algorithms to conduct collaborative analysis around the event cognition coverage gap, avoiding repeated calls to early warning analysis algorithms that have already completed their analysis tasks, and improving the targeting and collaboration of the entire event cognition analysis process.
[0030] After completing the multi-algorithm collaborative orchestration strategy, the system sequentially calls the corresponding early warning analysis algorithms according to the generated strategy to continue supplementing the analysis items to be supplemented for the event cognition coverage gaps. During the supplementary analysis, the newly generated spatial location cognition results, motion trajectory cognition results, behavioral action cognition results, target relationship cognition results, or target identity cognition results are written into the previously uncovered event cognition dimensions in the corresponding event cognition coverage structure according to the event identifier, forming an updated event cognition coverage structure. For example, in a target event segment with video time data from 09:18:23.125 to 09:18:31.867, the system first completes target detection and target tracking analysis, and then continues to call the behavior recognition algorithm and identity recognition algorithm to generate new event cognition results based on the event cognition coverage gaps, and automatically updates the corresponding event cognition dimensions; after the supplementary analysis is completed for the area corresponding to camera device identifier data CAM-00128, the system continues to process the video monitoring data corresponding to camera device identifier data CAM-00129 and CAM-00135, so that the data collected by each camera device completes the event cognition coverage structure update according to the same processing flow.
[0031] After the updated event cognition coverage structure is formed, the system further generates event cognition coverage detection results. These results uniformly record the coverage status of event cognition results corresponding to each event cognition dimension, and perform integrity checks based on these results. When all event cognition dimensions have corresponding event cognition results, the system determines that the target event has achieved full cognition coverage across all event cognition dimensions and generates the corresponding event cognition completion result. If there are still incompletely covered event cognition dimensions, the system maintains the current event cognition coverage structure and continues to regenerate multi-algorithm collaborative orchestration strategies based on the event cognition coverage gaps until the event cognition coverage structure meets the integrity requirements. Through this process, the system always ensures that the data entering the risk analysis stage are all event cognition completion results after achieving full coverage of all event cognition dimensions, avoiding omissions in subsequent risk analysis due to missing event cognition information.
[0032] Subsequently, the system conducts risk analysis of the target event based on the event cognition completion results. The system first reads the event cognition results corresponding to the event identifiers, and simultaneously reads the corresponding early warning rule data and scene area data. The early warning rule data is organized using rule identifiers such as RULE-018, RULE-026, and RULE-031, while the scene area data is organized using scene area identifiers such as AREA-001, AREA-003, and AREA-008. The system matches the target identity, behavior, movement trajectory, target relationship, and time-series changes with the corresponding early warning rule data to form a risk assessment result. Then, all risk assessment results are uniformly organized according to the event identifiers, and the risk type and risk level are determined in conjunction with the scene area data, ultimately generating an intelligent early warning result for the corresponding target event. When multiple target events occur simultaneously, the system can still organize the event cognition completion results and risk assessment results separately according to different event identifiers, ensuring that different target events do not interfere with each other and maintaining consistency in the risk analysis process.
[0033] In actual operation, the video monitoring platform continuously receives video monitoring data uploaded by multiple camera devices. Each camera device continuously acquires video image data at 25 fps to 30 fps and organizes it uniformly with millisecond-level video time data. The system continuously analyzes target events according to continuous video frames. Throughout the processing, it maintains a consistent correspondence between event identifiers, video time data, camera device identifier data, scene area data, and early warning rule data. When multiple target events occur alternately or simultaneously within the video monitoring area, this invention can still establish event cognition coverage structures based on event identifiers and independently generate multi-algorithm collaborative orchestration strategies for their respective event cognition coverage gaps. This enables parallel processing of multiple target events without event cognition confusion or duplicate analysis caused by cross-calling of multiple early warning analysis algorithms. Throughout the processing, each early warning analysis algorithm is called on demand around the event cognition coverage gaps, avoiding duplicate analysis caused by fixed calls to all early warning analysis algorithms. At the same time, it ensures that risk analysis is based on complete event cognition, making the intelligent early warning results more complete, consistent, and effective in engineering applications. It can meet the practical application needs of intelligent video early warning scenarios such as smart parks, industrial parks, and public safety monitoring.
[0034] Table 1. Comparison of comprehensive performance of multi-algorithm collaborative intelligent early warning methods
[0035] As shown in Table 1, the fixed-process multi-algorithm early warning method, the rule-priority-based multi-algorithm early warning method, and the intelligent early warning method based on programmable multi-algorithm collaboration of this invention show significant differences in indicators such as event cognition coverage completeness, algorithm repetition rate, average analysis time per event, risk judgment consistency rate, and effective early warning generation rate. This invention achieves better overall performance in all five evaluation indicators, and the indicators show a consistent trend of change. This indicates that this invention does not rely on the performance improvement of a single early warning analysis algorithm, but rather achieves overall optimization of the entire intelligent early warning process through the combined effects of event cognition coverage structure, event cognition coverage gap, multi-algorithm collaborative programming strategy, and completeness detection.
[0036] In terms of event cognition coverage completeness, the fixed-process multi-algorithm early warning method achieves 86.7%, the rule-priority-based multi-algorithm early warning method achieves 89.4%, and this invention reaches 94.8%, which is 8.1 percentage points higher than the fixed-process multi-algorithm early warning method and 5.4 percentage points higher than the rule-priority-based multi-algorithm early warning method. Traditional methods typically output spatial location cognition results, motion trajectory cognition results, behavioral action cognition results, target relationship cognition results, and target identity cognition results separately. There is a lack of a unified data organization method among different event cognition results. When some event cognition dimensions have not been analyzed, the system has difficulty in timely detecting cognition gaps. This invention first constructs an event cognition coverage structure, uniformly organizes all event cognition results corresponding to the same event identifier, and then identifies event cognition coverage gaps based on the event cognition coverage structure. This ensures that subsequent supplementary analysis always revolves around the event cognition dimensions that have not been covered, thus obtaining more complete event cognition information and improving the event cognition coverage completeness rate.
[0037] In terms of algorithm repetition rate, the fixed-process multi-algorithm early warning method has a rate of 38.5%, the rule-priority-based multi-algorithm early warning method has a rate of 24.7%, and this invention reduces it to 11.6%, a reduction of 26.9 percentage points compared to the fixed-process multi-algorithm early warning method and a reduction of 13.1 percentage points compared to the rule-priority-based multi-algorithm early warning method. The fixed-process multi-algorithm early warning method calls all early warning analysis algorithms in a predetermined order, continuing the subsequent analysis process even if some event cognition dimensions have been analyzed. While the rule-priority-based method can reduce some calls based on preset rules, it still cannot flexibly adjust according to the current event cognition completion status. This invention generates candidate algorithm mapping results based on event cognition coverage gaps, and then combines algorithm collaboration relationships to form a multi-algorithm collaborative orchestration strategy, so that the early warning analysis algorithm is only called for the items to be supplemented in the analysis, thereby effectively reducing repetitive calls and improving the efficiency of algorithm resource utilization.
[0038] In terms of average analysis time per single event, the fixed-process multi-algorithm early warning method takes 426 ms, the rule-priority-based multi-algorithm early warning method takes 351 ms, and this invention reduces it to 276 ms, which is 150 ms less than the fixed-process multi-algorithm early warning method and 75 ms less than the rule-priority-based multi-algorithm early warning method. Although this invention adds processing steps such as event recognition coverage analysis, candidate algorithm mapping, and multi-algorithm collaborative orchestration, it avoids repeated calls and invalid analysis, making the overall processing flow more compact. This allows multiple early warning analysis algorithms to complete collaborative calls according to data input dependencies, data output dependencies, and execution order dependencies, thus further improving the overall analysis efficiency.
[0039] In terms of risk assessment consistency, the fixed-process multi-algorithm early warning method achieves 84.2%, the rule-priority-based multi-algorithm early warning method achieves 87.8%, and this invention reaches 92.5%, which is 8.3 percentage points higher than the fixed-process multi-algorithm early warning method and 4.7 percentage points higher than the rule-priority-based multi-algorithm early warning method. Traditional methods conduct risk analysis before some event cognition dimensions are fully analyzed, which can easily lead to inconsistent risk assessments at different times or in different scenarios due to incomplete event cognition information. After supplementary analysis, this invention first updates the event cognition coverage structure, and then completes the integrity check based on the event cognition coverage detection results. Only when all event cognition dimensions are fully covered is the event cognition completion result generated, ensuring that risk analysis is always based on complete event cognition. Therefore, the risk assessment results are more stable and consistent.
[0040] In terms of effective early warning generation rate, the fixed-process multi-algorithm early warning method achieves 82.6%, the rule-priority-based multi-algorithm early warning method achieves 86.1%, and this invention reaches 91.3%, which is 8.7 percentage points higher than the fixed-process multi-algorithm early warning method and 5.2 percentage points higher than the rule-priority-based multi-algorithm early warning method. This invention can utilize the event cognition coverage structure to uniformly organize all event cognition results, and dynamically organize multiple early warning analysis algorithms to complete supplementary analysis based on the event cognition coverage gaps. Furthermore, it ensures that the risk analysis is based on complete event cognition through integrity detection. Therefore, under conditions such as continuous video monitoring, multi-target concurrency, and complex scene changes, the system can continuously output more complete and reliable intelligent early warning results, thereby further improving the effective early warning generation rate.
[0041] As shown in Table 1, this invention outperforms the two comparative methods in all five evaluation indicators, and the indicators are mutually reinforcing. The event cognition coverage completeness rate increased from 86.7% to 94.8%, providing a more complete data foundation for subsequent risk analysis; the algorithm repetition rate decreased from 38.5% to 11.6%, shortening the average analysis time per event from 426 ms to 276 ms; the complete event cognition results further improved the risk judgment consistency rate from 84.2% to 92.5%, and increased the effective early warning generation rate from 82.6% to 91.3%. These results indicate that the event cognition coverage structure, event cognition coverage gap, multi-algorithm collaborative orchestration strategy, and integrity detection proposed in this invention constitute a complete data processing link, enabling multiple early warning analysis algorithms to conduct collaborative analysis around the event cognition coverage. While ensuring the integrity of event cognition, it improves the efficiency of intelligent early warning processing and the reliability of risk analysis, demonstrating good engineering application value.
[0042] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. An intelligent early warning method based on programmable multi-algorithm collaboration, characterized in that, Includes the following steps: Collect video monitoring data and business-related data, and preprocess them to generate standardized event datasets; Multiple early warning analysis algorithms are invoked to analyze the standardized event dataset, generate event cognition results, and construct an event cognition coverage structure according to the cognition dimensions of each event cognition result; Coverage analysis is performed on the event cognition dimensions in the event cognition coverage structure to identify the event cognition dimensions that are not fully covered and generate event cognition coverage gaps; Based on the event cognition coverage gap matching early warning analysis algorithm, a multi-algorithm collaborative orchestration strategy is generated, and the early warning analysis algorithm is called according to the multi-algorithm collaborative orchestration strategy to supplement the event cognition coverage gap analysis; The event cognition coverage structure is updated using the newly added event cognition results obtained from the supplementary analysis, and the integrity of the updated event cognition coverage structure is checked to generate the event cognition completion result. Based on the results of event cognition, risk analysis is performed on the target event, and intelligent early warning results are generated.
2. The intelligent early warning method based on programmable multi-algorithm collaboration according to claim 1, characterized in that, The video monitoring data specifically includes video image data, video time data, and camera device identification data; the business-related data specifically includes early warning rule data and scene area data; the preprocessing specifically includes time alignment, abnormal data removal, missing data completion, and data format unification processing.
3. The intelligent early warning method based on programmable multi-algorithm collaboration according to claim 1, characterized in that, The construction of the event cognition coverage structure specifically includes: Read video image data from the standardized event dataset, divide the video time data into continuous event segments, and treat each continuous event segment as an independent event analysis object; The target detection algorithm, target tracking algorithm, behavior recognition algorithm, relationship analysis algorithm, and identity recognition algorithm are called respectively to analyze each independent event analysis object and generate spatial location recognition results, motion trajectory recognition results, behavior action recognition results, target relationship recognition results, and target identity recognition results. Extract the event identifier and cognitive dimension of each event cognitive result, merge them according to the same event identifier, and organize multiple event cognitive results belonging to the same event into an event cognitive result set; Traverse each event cognitive result set, count the cognitive dimensions contained in the event cognitive result set, construct event cognitive coverage units according to the correspondence between cognitive dimensions and event cognitive results, associate and organize all event cognitive coverage units of the same event to generate event cognitive coverage structure; The cognitive coverage structure of each event is saved based on the event identifier, forming a set of event cognitive coverage structures.
4. The intelligent early warning method based on programmable multi-algorithm collaboration according to claim 1, characterized in that, The generation of the event cognition coverage gap specifically includes: Read the event cognition coverage structure of each event, traverse all event cognition dimensions in the event cognition coverage structure, and extract the event cognition results of each event cognition dimension; Based on the event cognition results of each event cognition dimension, the coverage status of each event cognition dimension is determined. Event cognition dimensions with event cognition results are determined as event cognition dimensions that have been fully covered, and event cognition dimensions without event cognition results are determined as event cognition dimensions that have not been fully covered. Based on all event cognitive dimensions for each event, the event cognitive dimensions that have been covered and those that have not been covered are linked and organized to form an event cognitive dimension coverage distribution; Traverse the coverage distribution of each event cognition dimension, extract all event cognition dimensions that are not fully covered, and merge them according to the event identifier to generate event cognition coverage gaps; Based on the event identifiers corresponding to the event cognition coverage gaps, the event cognition coverage gaps are classified and organized to form an event cognition coverage gap set.
5. The intelligent early warning method based on programmable multi-algorithm collaboration according to claim 1, characterized in that, The supplementary analysis of the gaps in event awareness coverage specifically includes: Read the incomplete event cognition dimensions contained in the event cognition coverage gap, extract the items to be supplemented for analysis based on the cognition content corresponding to each event cognition dimension, and generate gap analysis results; Based on the gap analysis results, the set of early warning analysis algorithms is traversed, and early warning analysis algorithms that can complete the analysis items to be supplemented are selected. A mapping relationship between each analysis item to be supplemented and the corresponding early warning analysis algorithm is established, and candidate algorithm mapping results are generated. Based on the candidate algorithm mapping results, the data input dependency, data output dependency, and execution order dependency among the early warning analysis algorithms are analyzed, and the collaborative calling relationship among the early warning analysis algorithms is established according to the dependency relationship to generate the algorithm collaboration relationship; Based on the algorithm collaboration relationship, the calling order of the early warning analysis algorithms in the candidate algorithm mapping results is arranged, and a multi-algorithm collaborative arrangement strategy is generated by combining the early warning analysis algorithms corresponding to each supplementary analysis item. A multi-algorithm collaborative orchestration strategy is implemented, and the early warning analysis algorithm is called sequentially according to the calling order to perform supplementary analysis on each item to be supplemented, obtain the cognitive results of the new events, and associate and update the cognitive results of the new events with the corresponding cognitive dimensions in the event cognitive coverage gap to generate gap supplementation results.
6. The intelligent early warning method based on programmable multi-algorithm collaboration according to claim 1, characterized in that, The generation of the event cognition completion result specifically includes: Read the newly added event perception results, extract the event identifier, event perception dimension, and event perception result of the newly added event perception results, match the corresponding event perception coverage structure according to the event identifier, and locate the event perception dimension of the newly added event perception results; Write the newly added event cognition results into the event cognition dimension of the corresponding event cognition coverage structure to generate the updated event cognition coverage structure; Traverse all event cognition dimensions in the updated event cognition coverage structure, check whether each event cognition dimension corresponds to an event cognition result, and record the event cognition dimensions with event cognition results and the event cognition dimensions without event cognition results to form the event cognition coverage detection result; Based on the event cognition coverage detection results, the updated event cognition coverage structure is subjected to integrity detection. When all event cognition dimensions correspond to event cognition results, the updated event cognition coverage structure is determined to meet the integrity requirements; when there are event cognition dimensions that do not correspond to event cognition results, the updated event cognition coverage structure is determined to not meet the integrity requirements. For an event cognition coverage structure that meets the integrity requirements, all event cognition results are extracted and organized according to event identifiers to generate event cognition completion results.
7. The intelligent early warning method based on programmable multi-algorithm collaboration according to claim 1, characterized in that, The generation of the intelligent early warning results specifically includes: Read the event cognition completion result, extract the event identifier and event cognition result of the event cognition completion result, and read the warning rule data and scene area data according to the event identifier; The target identity, behavior, movement trajectory, target relationship and time sequence changes in the event perception results are matched with the early warning rule data to determine the risk assessment results of each event perception result; The risk assessment results corresponding to the same event identifier are summarized and processed, and the risk type and risk level of the target event are determined by combining the scene area data; Based on event identifiers, scene area data, risk type, risk level, and event perception results, intelligent early warning results for target events are generated.
8. An intelligent early warning system based on programmable multi-algorithm collaboration, executing the intelligent early warning method based on programmable multi-algorithm collaboration as described in any one of claims 1 to 7, characterized in that, include: The data acquisition and preprocessing module is used to collect video monitoring data and business-related data, preprocess the video monitoring data and business-related data, and generate standardized event datasets. The event cognition module is used to call multiple early warning analysis algorithms to analyze standardized event datasets, generate event cognition results, and construct an event cognition coverage structure according to the cognition dimensions of each event cognition result. The event cognition coverage analysis module is used to perform coverage analysis on the event cognition dimensions in the event cognition coverage structure, identify the event cognition dimensions that are not fully covered, and generate event cognition coverage gaps. The multi-algorithm collaborative orchestration module is used to match early warning analysis algorithms based on the event cognition coverage gap, generate a multi-algorithm collaborative orchestration strategy, and call early warning analysis algorithms to supplement the event cognition coverage gap analysis according to the multi-algorithm collaborative orchestration strategy; The event cognition completion module is used to update the event cognition coverage structure using the new event cognition results obtained from the supplementary analysis, and to perform an integrity check on the updated event cognition coverage structure to generate the event cognition completion result. The intelligent early warning module is used to perform risk analysis on target events based on the results of event cognition and generate intelligent early warning results.
