Edge computing-based methods and systems for processing video streams in community surveillance
Patent Information
- Application Number
- CN202610990792.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-03
- Publication Date
- 2026-09-11
AI Technical Summary
边缘节点分配机制不合理,负载不均衡问题突出;在任务分配时,通常仅简单考量边缘节点的当前资源负载,未综合匹配任务的资源消耗特性与节点硬件能力,也未评估摄像头与边缘节点之间的数据传输延迟,这导致部分边缘节点因承担过多高资源消耗任务而过载,而其他节点资源大量闲置;同时,将任务分配给传输延迟过高的节点会进一步加剧分析延迟,最终降低系统整体吞吐量和实时性
本发明通过构建摄像头动态历史行为画像与多因子动态优先级调度机制,显著提升小区安防响应效率与边缘资源利用率,通过量化每个摄像头的事件可信度与资源消耗特性,从源头过滤高误报、高消耗的无效任务,将边缘节点有限的CPU、GPU资源向高价值安全事件倾斜;同时融合时间重要度与视频内容紧急度因子,实现任务优先级的动态调整,有效弥补了传统先到先服务调度策略的安防漏洞;
Smart Images

Figure CN122741731A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of edge computing and video surveillance technology, and more specifically, to a method and system for processing community surveillance video streams based on edge computing. Background Technology
[0002] With the rapid advancement of smart community construction, the level of community security and intelligence is constantly improving. Modern communities typically deploy dozens to hundreds of high-definition network cameras to build a fully covered, all-weather video surveillance network.
[0003] However, current edge computing-based community surveillance video stream processing systems generally adopt a first-come, first-served task scheduling strategy, which still has the following technical shortcomings when facing peak scenarios where multiple cameras concurrently trigger a large number of analysis tasks: Existing technologies suffer from a lack of task safety priority, applying an indiscriminate processing order to all analysis tasks without distinguishing between historical risk data and real-time data from different camera locations. Most data are processed in order of trigger time, resulting in a low level of intelligent processing. The edge node allocation mechanism is unreasonable, resulting in a significant load imbalance. When allocating tasks, only the current resource load of the edge node is usually considered, without comprehensively matching the resource consumption characteristics of the task with the node's hardware capabilities, or assessing the data transmission latency between the camera and the edge node. This leads to some edge nodes becoming overloaded due to taking on too many resource-intensive tasks, while other nodes have a large amount of idle resources. At the same time, assigning tasks to nodes with excessively high transmission latency will further exacerbate the analysis latency, ultimately reducing the overall system throughput and real-time performance.
[0004] In summary, a method and system for processing community surveillance video streams based on edge computing are proposed. Summary of the Invention
[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method and system for processing cell surveillance video streams based on edge computing.
[0006] To achieve the above objectives, the present invention provides the following technical solution: A cell surveillance video stream processing method based on edge computing includes: Behavioral profile construction: Store the execution results and resource consumption data of all historical analysis tasks as historical data by camera ID and time granularity; when a camera triggers an analysis task, identify the time period in which the current task is triggered as the evaluation time period, retrieve the historical data of the x previous evaluation time periods, and extract multi-dimensional parameters from the retrieved historical data to generate a dynamically updated historical behavioral profile for each camera. Real-time parameter acquisition: Collect time information, camera information, edge node resource status and instantaneous characteristics of video stream when the current task is triggered, and determine time importance factor and content urgency factor after analysis and processing; Tiered task scheduling: Based on historical behavior profiles, combined with time importance factors and content urgency factors, a weighted fusion logic is used to calculate the priority coefficient of each analysis task. The tasks are divided into different queues for processing according to the priority coefficient. During the processing, the node status is taken into account to calculate the suitability score of each node for the current task, and the optimal allocation target is determined before allocation.
[0007] Specifically, the execution results include the task trigger time, camera ID, analysis target, task recognition result, event level, and event type; Resource consumption data includes task execution time, CPU utilization before and after task execution, and GPU utilization.
[0008] Specifically, the multi-dimensional parameters include the event impact performance coefficient and the resource consumption performance coefficient; The event impact performance coefficient is determined by weighted fusion processing based on true positive rate, false positive rate, effective event type value score, and user feedback impact rate. The resource consumption performance coefficient is determined by weighted fusion processing based on average processing time, average CPU usage increment, and average GPU usage increment.
[0009] Specifically, the logic for obtaining the relevant parameters of the event impact performance coefficient; The true positive rate is the ratio of the number of true positives to the sum of the number of true positives and the number of false negatives. The number of true positives is the number of tasks that camera i triggers and identifies valid targets within the historical evaluation time period T. The number of false negatives is the number of times that camera i does not trigger but actually has valid targets within the historical evaluation time period T. The false alarm rate is the ratio of the number of false positives to the sum of the number of false positives and the number of true negatives. The number of false positives is the number of tasks triggered by camera i within the historical evaluation period T but without a valid target, and the number of true negatives is the number of times camera i did not trigger and had no valid target within the historical evaluation period T. The user feedback impact rate is the ratio of the number of events of importance level to the total number of events for camera i within the historical evaluation period T. The event level is marked as important or ignored by the manager after the task. The effective event type value score is obtained by accumulating the security impact scores corresponding to different preset event types.
[0010] Specifically, the logic for obtaining the relevant parameters of the resource consumption performance coefficient; The average processing time is the ratio of the total execution time of all historical analysis tasks of camera i to the total number of historical analysis tasks; The average CPU usage increment is the ratio of the sum of the differences in CPU utilization before and after each historical analysis task to the total number of historical analysis tasks. The average GPU utilization increment is the ratio of the sum of the differences in GPU utilization before and after each historical analysis task to the total number of historical analysis tasks.
[0011] Specifically, the calculation logic for the time importance factor and the content urgency factor; The time importance factor is determined by mapping the evaluation time period through a preset time weight table; The content urgency factor is determined by a weighted fusion of the target impact ratio and the operational amplitude in the instantaneous features of the video stream.
[0012] Specifically, the logic for obtaining the target impact ratio and operational magnitude; The target impact ratio is the ratio of the highest number of targets in each frame of the video stream of camera i that is currently triggering the task to the preset saturation number of targets at the location of the camera. The maximum permissible physical speed at the location of the preset camera i ; Operating range ; The average displacement of all pixels is given by the formula. The result is obtained through calculation; where K is the number of sampling points; The x-direction optical flow component of the i-th sampling point; Let y be the optical flow component of the i-th sampling point.
[0013] Specifically, the queue partitioning logic; Using formula The priority coefficient for each analysis task is calculated. ; in , , , These represent the event impact performance coefficient, resource consumption performance coefficient, time importance factor, and content urgency factor, respectively. , , , These are preset weighting coefficients, and their sum is one.
[0014] Three sets of coefficient intervals are defined for each priority coefficient, and each set of coefficient intervals corresponds to a partition queue; the partition queues include high-priority queues, medium-priority queues, and low-priority queues. The priority coefficient of each analysis task is matched with the corresponding coefficient range to determine the partition queue to which the target analysis task belongs.
[0015] Specifically, the logic for determining the optimal allocation target; For the analysis task required for the current goal, the task characteristics and node status are considered together, and the fitness score of each node to the current task is calculated. The node with the highest fitness score is taken as the optimal target. First, identify the pre-marked queues of processable tasks for each node. For the task to be analyzed by the target, identify its queue and match it with the queues of processable tasks of the node. If the match fails, remove the node and use the remaining nodes as the assignable nodes. That is, through the formula The fit scores between the analysis task required for the current objective and each node are calculated. ; , These are preset weighting coefficients, and their sum is one; To determine resource matching accuracy, obtain the CPU and GPU utilization of edge node resources and output the resource availability factor; use the formula... Calculate the resource availability factor ;in and The preset weighting coefficients,
[0016] By obtaining the resource consumption performance coefficient of camera i And the maximum resource consumption coefficient that node j can withstand. ; ; To determine the delay matching degree; obtain the data transmission delay from camera i to node j; ;in This is the preset maximum allowable delay.
[0017] A cell surveillance video stream processing system based on edge computing includes: Historical data storage module: Stores the execution results and resource consumption data of all historical analysis tasks by camera ID and time granularity; execution results include task trigger time, camera ID, analysis target, task identification result, event level, and event type; resource consumption data includes task execution time, CPU utilization before and after task execution, and GPU utilization.
[0018] Behavioral profiling module: When a camera triggers an analysis task, it identifies the evaluation time period triggered by the current task, retrieves historical data from the x previous evaluation time periods of the same time period, extracts the event impact performance coefficient and resource consumption performance coefficient, and generates a dynamically updated historical behavioral profile for each camera. Real-time parameter acquisition module: Collects time information, camera information, edge node resource status and instantaneous characteristics of video stream when the current task is triggered, and determines time importance factor and content urgency factor after analysis and processing; The hierarchical task scheduling module calculates the priority coefficient of each analysis task based on historical behavior profiles and combines time importance factors and content urgency factors. Based on the priority coefficient, the tasks are divided into three priority queues: high, medium, and low. The module also calculates the suitability score of each node for the current task by taking into account the status of edge nodes. After determining the optimal allocation target, the task is assigned.
[0019] The technical effects and advantages of this invention are as follows: This invention significantly improves the security response efficiency and edge resource utilization of communities by constructing dynamic historical behavior profiles of cameras and a multi-factor dynamic priority scheduling mechanism. By quantifying the event credibility and resource consumption characteristics of each camera, it filters out invalid tasks with high false alarms and high resource consumption from the source, and tilts the limited CPU and GPU resources of edge nodes towards high-value security events. At the same time, it integrates time importance and video content urgency factors to achieve dynamic adjustment of task priority, effectively making up for the security vulnerabilities of traditional first-come-first-served scheduling strategies. This invention achieves a balance between load balancing and scheduling fairness at edge nodes through a dual-dimensional node adaptability allocation and dynamic priority enhancement mechanism. It calculates node adaptability by combining resource matching degree and latency matching degree, and prioritizes the allocation of tasks to nodes with sufficient resources and low transmission latency, avoiding the problem of some nodes being overloaded and others being idle, thereby improving the overall throughput of the system. At the same time, it introduces a priority aging mechanism based on waiting time to prevent low-priority tasks from being starved for a long time while ensuring that high-priority tasks are processed first. Attached Figure Description
[0020] Figure 1 This is a flowchart of the cell surveillance video stream processing method based on edge computing according to the present invention. Detailed Implementation
[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0022] Example 1 like Figure 1 As shown, the edge computing-based method for processing community surveillance video streams is as follows: Behavioral profile construction: Store the execution results and resource consumption data of all historical analysis tasks as historical data by camera ID and time granularity; when a camera triggers an analysis task, identify the time period in which the current task is triggered as the evaluation time period, retrieve historical data of the x previous evaluation time periods, and extract multi-dimensional parameters from the retrieved historical data to generate a dynamically updated historical behavioral profile for each camera; x > 5, specifically set by the administrator. Additional notes: Time granularity can be divided into hourly granularity (0-23 hours).
[0023] Specifically: The execution results include the task trigger time, camera ID, analysis target (e.g., face / license plate), task recognition result (true positive / false positive / false negative / true negative), event level, and event type; Resource consumption data includes task execution time, CPU utilization before and after task execution, and GPU utilization. Multi-dimensional parameters include the event impact performance coefficient and the resource consumption performance coefficient; The event impact performance coefficient is determined by weighted fusion processing based on true positive rate, false positive rate, effective event type value score, and user feedback impact rate. The specific calculation is as follows: Normalization of valid event type value scores, i.e., through valid event type value scores ; Value scoring for valid event types that have not been normalized. The maximum valid event type value score for all cameras within the same evaluation time period T; Event Impact Performance Coefficient ; In the formula , , , , respectively represent the true positive rate, user feedback impact rate, false positive rate, and valid event type value score; , , , These are preset weighting coefficients, and their sum is one.
[0024] To measure the credibility and security value of camera-triggered tasks, the higher the parameter value, the greater the impact of the event on community security and the greater the probability that the task is a valid event.
[0025] The number of tasks triggered and identified by camera i within the historical evaluation time period T is used as the true positive count. The number of times camera i did not trigger a valid target within the historical evaluation period T is counted as the false negative count. The true positive rate is obtained by summing the number of true positives and false negatives, and then dividing the number of true positives by the sum. The number of tasks triggered by camera i within the historical evaluation time period T but without a valid target is counted as the false positive count. The number of times camera i did not trigger and had no valid target within the historical evaluation time period T is counted as the true negative count; The false positive rate is obtained by summing the number of false positives and the number of true negatives, and then dividing the number of false positives by the sum. The management personnel make adjustments by determining the effectiveness of the objectives and whether effective objectives exist after each task.
[0026] Obtain the event level of each event of camera i within the historical evaluation time period T, including important and ignored; Event levels are marked by administrators after the task.
[0027] The number of events of importance level for camera i within the historical evaluation period T is counted, and then divided by the total number of events to obtain the user feedback impact rate. Preset security impact scores corresponding to different event types; identify the event type of events that occur in camera i within the historical evaluation time period T, convert them into security impact scores, and accumulate them to obtain the effective event type value score of camera i within the historical time period T. The safety impact score is limited to 1-10, and the higher the score, the greater the degree of safety hazard caused by the event. For example, stranger intrusion = 10, non-resident vehicle = 5, gathering of people = 3, other anomalies = 1.
[0028] The resource consumption performance coefficient is determined by weighted fusion processing based on average processing time, average CPU usage increment, and average GPU usage increment. The specific calculation process is as follows: After normalizing the average processing time, average CPU usage increment, and average GPU usage increment, they are respectively labeled as follows: , , ; Resource consumption performance coefficient ; , , These are preset weighting coefficients, and their sum is one; This parameter measures the resource consumption of edge nodes by camera analysis tasks. The higher the parameter value, the greater the pressure on system resources.
[0029] Retrieve all historical analysis tasks of camera i, sum up the execution time of each historical analysis task, and divide by the total number of historical analysis tasks to obtain the average processing time of the analysis tasks associated with camera i. For each historical analysis task, obtain the CPU utilization rate after the k-th task runs and before the task runs, calculate the difference, and output the CPU increment of the k-th task. Sum the CPU increments of each group and divide by the total number of historical analysis tasks to obtain the average CPU usage increment of camera i for edge node resources. Get the GPU utilization after the k-th task runs and before the task runs, calculate the difference and output the GPU increment of the k-th task. Sum the GPU increments of each group and divide by the total number of historical analysis tasks to get the average GPU usage increment of camera i on edge node resources. Real-time parameter acquisition: Collect time information, camera information, edge node resource status and instantaneous characteristics of video stream when the current task is triggered, and determine time importance factor and content urgency factor after analysis and processing; Specifically: The time period in which the current task is triggered is identified as the evaluation time period. A preset time weight table is used to map the evaluation time period and determine the time importance factor of camera i corresponding to the current task trigger time. That is, by setting time importance factors corresponding to different time periods, the mapping value range is mapped to 0-1. For example, if the evaluation time period is at night, the mapping result will be closer to 1.
[0030] For the analysis task currently triggered by camera i, count the number of valid targets in each frame, extract the highest number of targets, identify the location of camera i, and predefine the saturation number of targets for different camera locations; through pre-setting by the administrator, it represents "the maximum number of salient targets that the scene can tolerate"; The number of targets in each frame of a video stream can be obtained through target detection models (such as YOLO and SSD).
[0031] Divide the highest number of targets in camera i by the corresponding number of saturated targets to output the target influence ratio; Calculate the average displacement of all pixels That is, through the formula The result is obtained through calculation; where K is the number of sampling points; The x-direction optical flow component of the i-th sampling point; Let y be the optical flow component of the i-th sampling point; The maximum permissible physical speed at the location of the preset camera i ; Additional note: The maximum permissible physical speed must be considered during the calculation. Perform the conversion between physical speed and pixel speed.
[0032] Operating range ; After weighted fusion processing of the target impact ratio and running amplitude in the instantaneous features of the video stream, the content urgency factor is obtained. The specific calculation process is as follows: Content urgency factor In the formula Indicates the target impact ratio. and These are preset weighting coefficients, and their sum is one.
[0033] Tiered task scheduling: Based on historical behavior profiles, combined with time importance factors and content urgency factors, a weighted fusion logic is used to calculate the priority coefficient of each analysis task. The tasks are divided into different queues for processing according to the priority coefficient. During the processing, the node status is considered to calculate the suitability score of each node for the current task, and the optimal allocation target is determined before allocation. Specifically: That is, using formulas The priority coefficient for each analysis task is calculated. ; in , , , These represent the event impact performance coefficient, resource consumption performance coefficient, time importance factor, and content urgency factor, respectively. , , , These are preset weighting coefficients, and their sum is one.
[0034] Three sets of coefficient intervals are defined for each priority coefficient, and each set of coefficient intervals corresponds to a partition queue; the partition queues include high-priority queues, medium-priority queues, and low-priority queues. Match the priority coefficient of each analysis task with the corresponding coefficient range to determine the partition queue to which the target analysis task belongs; As a supplementary note, the higher the priority coefficient, the greater the likelihood of matching a high-priority queue.
[0035] For the analysis task required for the current goal, the task characteristics and node status are considered together, and the fitness score of each node to the current task is calculated. The node with the highest fitness score is taken as the optimal target. First, identify the pre-marked queues of processable tasks for each node. For the task to be analyzed by the target, identify its queue and match it with the queues of processable tasks of the node. If the match fails, remove the node and use the remaining nodes as the assignable nodes. That is, through the formula The fit scores between the analysis task required for the current objective and each node are calculated. ; , These are preset weighting coefficients, and their sum is one; Resource matching degree measures whether node resources meet task requirements; Obtain the CPU and GPU utilization of edge node resources and output the resource availability factor; use the formula... Calculate the resource availability factor ;in and The weighting factor is a preset value, and the sum of the weights is one. The more idle the resources, the larger this factor becomes, indicating a higher margin for tasks to be processed in a timely manner.
[0036] By obtaining the resource consumption performance coefficient of camera i And the maximum resource consumption coefficient that node j can withstand. Pre-configured according to hardware settings; ; Additional explanation, when > If so, then sort the node directly.
[0037] For latency matching; measures the impact of data transmission latency on task real-time performance; Obtain the data transmission delay from camera i to node j; Data transmission latency is determined by the time it takes for the video frame encoding to be completed at the camera end and the time it takes for the edge node to receive the complete data of that video frame.
[0038] ;in The preset maximum allowable delay; Additional explanation, when > If so, exclude that node directly.
[0039] Additional explanation: Dynamic priority boosting based on different queue waiting times; That is, using formulas Calculate the updated priority coefficients ;in The preset aging coefficient is limited to a range of 0.3-0.5; To analyze the waiting time of the task, The maximum allowed waiting time is preset for the corresponding queue; For medium-priority tasks that wait for more than 2.5 seconds, the priority will be increased every 500ms. If a low-priority task waits for more than 15 seconds, its priority will be increased every second. Maximum upgrade limit: Low → Medium, Medium → High.
[0040] Example 2 Based on the edge computing-based cell surveillance video stream processing method provided in Embodiment 1 of this application, Embodiment 2 of this application proposes an edge computing-based cell surveillance video stream processing system. Embodiment 2 is merely a preferred embodiment of Embodiment 1, and its implementation will not affect the individual implementation of Embodiment 1.
[0041] Specifically, the edge computing-based cell surveillance video stream processing system provided in Embodiment 2 of this application includes: Historical data storage module: Stores the execution results and resource consumption data of all historical analysis tasks by camera ID and time granularity; execution results include task trigger time, camera ID, analysis target, task identification result, event level, and event type; resource consumption data includes task execution time, CPU utilization before and after task execution, and GPU utilization.
[0042] Behavioral profiling module: When a camera triggers an analysis task, it identifies the evaluation time period triggered by the current task, retrieves historical data from the x previous evaluation time periods of the same time period, extracts the event impact performance coefficient and resource consumption performance coefficient, and generates a dynamically updated historical behavioral profile for each camera. Real-time parameter acquisition module: Collects time information, camera information, edge node resource status and instantaneous characteristics of video stream when the current task is triggered, and determines time importance factor and content urgency factor after analysis and processing; The hierarchical task scheduling module calculates the priority coefficient of each analysis task based on historical behavior profiles and combines time importance factors and content urgency factors. Based on the priority coefficient, the tasks are divided into three priority queues: high, medium, and low. The module also calculates the suitability score of each node for the current task by taking into account the status of edge nodes. After determining the optimal allocation target, the task is assigned.
[0043] The above formulas are all dimensionless calculations. Dimensionless calculations can be performed using various methods such as standardization, which will not be elaborated here. The formulas are derived from software simulations based on a large amount of collected data, and the preset parameters in the formulas can be set by those skilled in the art according to the actual situation.
[0044] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, ATA hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state ATA hard disk.
[0045] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.
[0046] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0047] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.
[0048] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.
[0049] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.
[0050] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable ATA hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0051] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A method for processing community surveillance video streams based on edge computing, characterized in that, include: Behavioral profile construction: Store the execution results and resource consumption data of all historical analysis tasks as historical data by camera ID and time granularity; when a camera triggers an analysis task, identify the time period in which the current task is triggered as the evaluation time period, retrieve the historical data of the x previous evaluation time periods, and extract multi-dimensional parameters from the retrieved historical data to generate a dynamically updated historical behavioral profile for each camera. Real-time parameter acquisition: Collect time information, camera information, edge node resource status and instantaneous characteristics of video stream when the current task is triggered, and determine time importance factor and content urgency factor after analysis and processing; Tiered task scheduling: Based on historical behavior profiles, combined with time importance factors and content urgency factors, a weighted fusion logic is used to calculate the priority coefficient of each analysis task. The tasks are divided into different queues for processing according to the priority coefficient. During the processing, the node status is taken into account to calculate the suitability score of each node for the current task, and the optimal allocation target is determined before allocation.
2. The cell surveillance video stream processing method based on edge computing according to claim 1, characterized in that: The execution results include the task trigger time, camera ID, analysis target, task recognition result, event level, and event type. Resource consumption data includes task execution time, CPU utilization before and after task execution, and GPU utilization.
3. The cell surveillance video stream processing method based on edge computing according to claim 2, characterized in that: Multi-dimensional parameters include the event impact performance coefficient and the resource consumption performance coefficient; The event impact performance coefficient is determined by weighted fusion processing based on true positive rate, false positive rate, effective event type value score, and user feedback impact rate. The resource consumption performance coefficient is determined by weighted fusion processing based on average processing time, average CPU usage increment, and average GPU usage increment.
4. The cell surveillance video stream processing method based on edge computing according to claim 2, characterized in that: Logic for obtaining the relevant parameters of the event impact performance coefficient; The true positive rate is the ratio of the number of true positives to the sum of the number of true positives and the number of false negatives. The number of true positives is the number of tasks that camera i triggers and identifies valid targets within the historical evaluation time period T. The number of false negatives is the number of times that camera i does not trigger but actually has valid targets within the historical evaluation time period T. The false alarm rate is the ratio of the number of false positives to the sum of the number of false positives and the number of true negatives. The number of false positives is the number of tasks triggered by camera i within the historical evaluation period T but without a valid target, and the number of true negatives is the number of times camera i did not trigger and had no valid target within the historical evaluation period T. The user feedback impact rate is the ratio of the number of events of importance level to the total number of events for camera i within the historical evaluation period T. The event level is marked as important or ignored by the manager after the task. The effective event type value score is obtained by accumulating the security impact scores corresponding to different preset event types.
5. The cell surveillance video stream processing method based on edge computing according to claim 3, characterized in that: Logic for obtaining the relevant parameters of the resource consumption performance coefficient; The average processing time is the ratio of the total execution time of all historical analysis tasks of camera i to the total number of historical analysis tasks; The average CPU usage increment is the ratio of the sum of the differences in CPU utilization before and after each historical analysis task to the total number of historical analysis tasks. The average GPU utilization increment is the ratio of the sum of the differences in GPU utilization before and after each historical analysis task to the total number of historical analysis tasks.
6. The cell surveillance video stream processing method based on edge computing according to claim 3, characterized in that: Calculation logic for time importance factor and content urgency factor; The time importance factor is determined by mapping the evaluation time period through a preset time weight table; The content urgency factor is determined by a weighted fusion of the target impact ratio and the operational amplitude in the instantaneous features of the video stream.
7. The cell surveillance video stream processing method based on edge computing according to claim 6, characterized in that: Logic for obtaining target impact ratio and operational magnitude; The target impact ratio is the ratio of the highest number of targets in each frame of the video stream of camera i that is currently triggering the task to the preset saturation number of targets at the location of the camera. The maximum permissible physical speed at the location of the preset camera i ; Operating range ; The average displacement of all pixels is given by the formula. Calculated; Where K is the number of sampling points; The x-direction optical flow component of the i-th sampling point; Let y be the optical flow component of the i-th sampling point.
8. The cell surveillance video stream processing method based on edge computing according to claim 6, characterized in that: Queue partitioning logic; Using formula The priority coefficient for each analysis task is calculated. ; in , , , These represent the event impact performance coefficient, resource consumption performance coefficient, time importance factor, and content urgency factor, respectively. , , , These are preset weighting coefficients, and their sum is one; Define three sets of coefficient intervals corresponding to the priority coefficients, and each set of coefficient intervals corresponds to a partition queue; The assigned queues include high-priority queues, medium-priority queues, and low-priority queues. The priority coefficient of each analysis task is matched with the corresponding coefficient range to determine the partition queue to which the target analysis task belongs.
9. The cell surveillance video stream processing method based on edge computing according to claim 8, characterized in that: Logic for determining the optimal allocation objective; For the analysis task required for the current goal, the task characteristics and node status are considered together, and the fitness score of each node to the current task is calculated. The node with the highest fitness score is taken as the optimal target. First, identify the pre-marked queues of processable tasks for each node. For the task to be analyzed by the target, identify its queue and match it with the queues of processable tasks of the node. If the match fails, remove the node and use the remaining nodes as the assignable nodes. That is, through the formula The fit scores between the analysis task required for the current objective and each node are calculated. ; , These are preset weighting coefficients, and their sum is one; To determine resource matching accuracy, obtain the CPU and GPU utilization of edge node resources and output the resource availability factor; use the formula... Calculate the resource availability factor ;in and The preset weighting coefficients, By obtaining the resource consumption performance coefficient of camera i And the maximum resource consumption coefficient that node j can withstand. ; ; To determine the delay matching degree; obtain the data transmission delay from camera i to node j; ;in This is the preset maximum allowable delay.
10. A cell surveillance video stream processing system based on edge computing, applied to any one of the cell surveillance video stream processing methods based on edge computing as described in claims 1-9, characterized in that: Historical data storage module: Stores the execution results and resource consumption data of all historical analysis tasks by camera ID and time granularity; execution results include task trigger time, camera ID, analysis target, task identification result, event level, and event type; resource consumption data includes task execution time, CPU utilization before and after task execution, and GPU utilization. Behavioral profiling module: When a camera triggers an analysis task, it identifies the evaluation time period triggered by the current task, retrieves historical data from the x previous evaluation time periods of the same time period, extracts the event impact performance coefficient and resource consumption performance coefficient, and generates a dynamically updated historical behavioral profile for each camera. Real-time parameter acquisition module: Collects time information, camera information, edge node resource status and instantaneous characteristics of video stream when the current task is triggered, and determines time importance factor and content urgency factor after analysis and processing; The hierarchical task scheduling module calculates the priority coefficient of each analysis task based on historical behavior profiles and combines time importance factors and content urgency factors. Based on the priority coefficient, the tasks are divided into three priority queues: high, medium, and low. The module also calculates the suitability score of each node for the current task by taking into account the status of edge nodes. After determining the optimal allocation target, the task is assigned.
