Real-time video analysis system and method based on edge calculation

Abstract

The invention relates to the technical field of image processing, in particular to a real-time video analysis system and method based on edge computing, and the system comprises a central management and control service module, an edge computing module, an algorithm adaptation module, a node management optimization module and a network scheduling module. According to the invention, through analysis of the video stream frame rate and the processing delay change amplitude, a high-load time period can be identified in real time, task allocation can be dynamically optimized, a frame processing strategy is ensured to be matched with a current load state, system delay is reduced to the greatest extent, and meanwhile, the accuracy of video analysis is ensured; according to the method, the network influence time period can be accurately identified, task scheduling can be optimized according to network fluctuation, inaccurate analysis results or task execution delay caused by network problems are avoided, the real-time performance and stability of a frequency analysis system are remarkably improved by comprehensively considering load changes and environmental factors, and the method is suitable for large-scale popularization and application. And the performance bottleneck caused by fixed frame interval or excessive dependence on the central server is avoided.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a real-time video analysis system and method based on edge computing. Background Art

[0002] The field of image processing technology mainly involves methods and means for collecting, analyzing, identifying and processing static images or dynamic video data. Its core issues include image acquisition and preprocessing, feature extraction and expression, target detection and recognition, scene understanding and analysis, etc. In practical applications, it widely serves security monitoring, industrial automation, medical diagnosis, traffic management and other scenarios. With the development of the Internet of Things and artificial intelligence technology, this technology field is gradually evolving towards higher real-time performance, stronger processing efficiency and deeper intelligence. Among them, the traditional real-time video analysis system refers to a system architecture that sinks some computing tasks to edge nodes close to the data source to relieve the pressure on the central server and reduce latency. The system mainly focuses on technical issues such as high-concurrency processing capabilities, flexible algorithm deployment, and real-time analysis results in video data processing. Traditional solutions complete the above tasks by centrally processing video streams on a single server, statically deploying with fixed algorithm configurations, and setting fixed frame intervals for image sampling. Some improved solutions attempt to achieve data diversion processing by introducing a distributed computing architecture, or adopt a plug-in structure to achieve algorithm replacement and loading, but most lack a fine-grained management mechanism for plug-in versions and online hot update support, and fail to dynamically adjust the image sampling frequency according to the load conditions in the frame processing strategy, resulting in dual limitations on processing performance and real-time performance.

[0003] When processing highly concurrent video streams, existing technologies rely on centralized servers for video stream processing. Because computing tasks cannot be flexibly adjusted based on actual load, processing latency and real-time performance are affected. Especially under large load fluctuations, the frame processing frequency cannot be adjusted promptly, making it difficult for the system to maintain stable processing power and low latency under high loads. Fixed algorithm configurations and a lack of online hot updates result in poor algorithm flexibility, making it difficult to cope with complex scene changes, and unable to quickly switch algorithms based on different situations, even resulting in reduced analysis accuracy. Especially in the case of unstable network bandwidth, the scheduling methods of existing technologies fail to account for the actual impact of network fluctuations on task execution, thereby affecting the overall system performance and user experience. Summary of the Invention

[0004] The purpose of the present invention is to solve the shortcomings of the existing technology and propose a real-time video analysis system and method based on edge computing.

[0005] In order to achieve the above objectives, the present invention adopts the following technical solutions: A real-time video analysis system based on edge computing includes: Based on edge node operation data, the central control service module analyzes load fluctuations by comparing the changes in video stream frame rate and processing latency, screens key high-load time periods, extracts critical conditions for task scheduling, and obtains task allocation optimization mark values. The edge computing module optimizes the marking value based on the task allocation, matches the video stream data of the corresponding time period with the frame processing strategy, compares the frame processing delay point in the marked segment with the current load status, filters the frame processing instructions corresponding to the deviation rate, and obtains the frame processing offset control instruction group; The algorithm adaptation module extracts the analysis accuracy value of the corresponding node before and after the adjustment instruction according to the frame processing offset control instruction group, identifies the target detection accuracy path under the unit frame in combination with the frame processing trend, extracts the change section to form a response interval, and obtains the node unit adaptation algorithm trend; The node management optimization module calls the node unit adaptation algorithm trend, collects the external environment data of the edge node, analyzes the overlap time between the edge computing optimization algorithm change and the network bandwidth jump period, and outputs the network impact duration.

[0006] As a further solution of the present invention, the task allocation optimization mark value includes a frame rate fluctuation threshold, a processing delay variation range, and a load rise rate characteristic; the frame processing offset control instruction group includes a frame processing deviation amount, a delay variation anomaly value, and a load state offset rate; the node unit adaptation algorithm trend includes an analysis accuracy variation interval, a target detection accuracy path type, and an algorithm response amplitude; the network impact duration includes a bandwidth fluctuation duration, a delay jump overlap period, and a network disturbance duration.

[0007] As a further solution of the present invention, the central management and control service module includes: The frame rate fluctuation extraction submodule extracts the video stream time series based on the edge node operation data, calculates the frame rate difference between adjacent sampling points, determines the fluctuation period and extracts the peak-valley interval, and selects the segments where the period difference exceeds the threshold to obtain the frame rate period fluctuation interval value; The delay variation calculation submodule calls the frame rate period fluctuation interval value, identifies the corresponding delay data, analyzes the absolute delay variation amplitude of adjacent time periods, compares it with the set delay amplitude threshold, locates the mutation node, and obtains the delay mutation amplitude interval value; The task status marking submodule identifies the corresponding load and algorithm complexity data according to the delay mutation amplitude interval value, extracts the load and algorithm data, combines the delay mutation and frame rate fluctuation, calculates the load complexity offset value, sets the offset threshold, marks the time segment exceeding the threshold, and obtains the task allocation optimization marking value.

[0008] As a further solution of the present invention, the edge computing module includes: The frame processing data matching submodule extracts the video stream frame rate and delay data of the corresponding time period based on the task allocation optimization mark value, calculates the delay fluctuation amplitude according to the sampling interval, aligns the frame rate and delay amplitude at the same time, and obtains the frame delay linkage interval group; The frame processing deviation judgment submodule calls the frame delay linkage interval group, extracts the load state sequence, compares the frame processing change with the state difference, compares it with the preset offset boundary after normalization, calculates the cumulative value of the deviation intensity of the frame processing point, extracts the frame processing position index with a deviation intensity greater than the reference judgment value, and establishes an offset intensity index group; The frame processing instruction extraction submodule filters the position of the corresponding time point in the task instruction set based on the offset strength index group, extracts the instruction value, sorts it by time series, removes duplicate instructions, and obtains the frame processing offset control instruction group.

[0009] As a further solution of the present invention, the algorithm adaptation module includes: The analysis accuracy extraction submodule extracts the frame processing accuracy data of the node in the cycle before and after the adjustment according to the frame processing offset control instruction group, identifies the analysis boundary conditions, and obtains the analysis accuracy difference value; The target detection accuracy identification submodule calls the analysis accuracy difference value, identifies the target detection accuracy change trajectory in the unit frame path, determines the target number trend corresponding to the accuracy fluctuation, compares the accuracy drop and the target number fluctuation in adjacent time periods, calculates the accuracy concentration of the segment in the unit frame path, filters the synchronous fluctuation segment, and obtains the target detection accuracy feature segment; The algorithm response interval identification submodule analyzes the time and precision displacement trends according to the target detection accuracy feature segment, determines the direction consistency and amplitude change characteristics, screens the fluctuation segments and aggregates them, and obtains the node unit adaptation algorithm trend.

[0010] As a further solution of the present invention, the node management optimization module includes: The algorithm trend extraction submodule calls the node unit adaptation algorithm trend, extracts the node task record and periodic algorithm data, identifies the fluctuation amplitude and fluctuation frequency within a single cycle, and obtains the algorithm change trend value; The network jump identification submodule selects bandwidth and delay records within the same period based on the algorithm change trend value and the external network data of the edge node, compares the data change amplitude on a daily basis, and selects time nodes with amplitudes greater than the network jump threshold to obtain the bandwidth delay jump period; The network impact quantification submodule identifies the intersection duration of the algorithm amplitude variation and the network jump according to the bandwidth delay jump period, performs weighted processing, and normalizes the usage duration under the differentiated period with reference to the amplitude variation frequency, the combined amplitude variation value and the period fluctuation, and outputs the network impact duration.

[0011] As a further solution of the present invention, the system also includes a network scheduling module: The network scheduling module filters the task scheduling instructions affected by the network interference based on the duration of the network impact, classifies the load switching time and adjusts the trigger frequency, identifies the load jump exceeding the limit period, and obtains the load interference impact frequency of the video analysis task; The frequency of load interference impact of the video analysis task includes load switching frequency, adjustment instruction triggering times, and jump period exceeding times.

[0012] As a further solution of the present invention, the network scheduling module includes: The instruction screening submodule screens matching task scheduling instructions based on the duration of the network impact, identifies the load period and load state, compares the disturbance period with the instruction period, eliminates low-matching instructions, and obtains an instruction set affected by the network interference; The load classification submodule calls the instruction set affected by network interference, extracts the load switching time and adjustment trigger frequency corresponding to the instructions, arranges them in order of switching time, classifies them by frequency interval, counts the correspondence between frequency bands and load time periods, and obtains the load adjustment frequency band distribution value; The interference identification submodule adjusts the frequency band distribution value according to the load, collects the load jump amplitude and duration within the frequency band, determines whether it exceeds the jump amplitude threshold, identifies the frequency band interference intensity, sorts the interference intensity of all frequency bands, determines the frequency band and jump period of the interference intensity, and obtains the load interference impact frequency of the video analysis task.

[0013] The edge computing-based real-time video analysis method is performed based on the above-mentioned edge computing-based real-time video analysis system, and includes the following steps: S1: Based on edge node operation data, it analyzes frame rate fluctuations and processing delay changes, extracts periods of sudden load peaks and sharp changes in algorithm complexity, identifies state switching points where delay and load changes are synchronized, marks task levels, and generates task state switching identification fragments. S2: Based on the task state switching identification segment, extract the frame processing change data and the delay change amplitude, compare the frame processing mutation with the delay fluctuation amplitude, analyze the correspondence between the frame processing change and the load response, and obtain the frame processing adjustment response trajectory segment; S3: Based on the frame processing, the response trajectory segment is adjusted, the change in analysis accuracy value and target detection accuracy gradient before and after the response is extracted, the target detection accuracy drop and the increase in the number of targets at the node are analyzed, the accuracy decay path is identified and compared with the original cycle data, the segments with accuracy decay characteristics are screened, and the accuracy adaptation decay path distribution set is generated; S4: Based on the precision adaptive attenuation path distribution set, identifying external network data within a corresponding time period, analyzing the overlapping period of bandwidth jump and precision attenuation path, extracting the associated time period, and obtaining the network parameter interference section for frame processing adjustment; S5: Based on the network parameter interference section adjusted by the frame processing, the scheduling control strategy records within the time period are extracted, the time interval of the load switching and the adjustment trigger frequency are analyzed, and the high-frequency adjustment strategy segments are screened to obtain the load interference impact frequency of the video analysis task.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, by analyzing the variation range of the video stream frame rate and processing delay, it is possible to identify high-load periods in real time and dynamically optimize task allocation, ensuring that the frame processing strategy matches the current load status, and then adjusting the frame processing frequency to minimize system delays while ensuring the accuracy of video analysis. In addition, by combining the external environmental data of the edge node and the changes in network bandwidth, it is possible to accurately identify network-affected periods and optimize task scheduling based on network fluctuations, so that the task load can be adjusted according to the actual network conditions, thereby avoiding inaccurate analysis results or task execution delays caused by network problems. By comprehensively considering load changes and environmental factors, the real-time and stability of the video analysis system are significantly improved, avoiding performance bottlenecks caused by fixed frame intervals or over-reliance on central servers. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 is a system flow chart of the present invention; Figure 2 This is a flow chart of the central control service module of the present invention; Figure 3 This is a flow chart of the edge computing module of the present invention; Figure 4 This is the flow chart of the algorithm adaptation module of the present invention; Figure 5 This is a flow chart of the node management optimization module of the present invention; Figure 6 This is a flow chart of the network scheduling module of the present invention. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0017] In the description of the present invention, it should be understood that the terms "length," "width," "up," "down," "front," "back," "left," "right," "vertical," "horizontal," "top," "bottom," "inside," "outside," and the like, indicating positions or relationships, are based on the positions or relationships shown in the accompanying drawings and are intended only to facilitate the description of the present invention and simplify the description. They do not indicate or imply that the devices or elements referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limiting the present invention. Furthermore, in the description of the present invention, "plurality" means two or more, unless otherwise expressly and specifically defined.

[0018] See also Figure 1 , the present invention provides a technical solution: a real-time video analysis system based on edge computing includes: Based on edge node operation data, the central control service module analyzes load fluctuations by comparing the changes in video stream frame rate and processing latency, screens key high-load time periods, extracts critical conditions for task scheduling, and obtains task allocation optimization mark values. The edge computing module optimizes the tag value based on task allocation, matches the video stream data of the corresponding time period with the frame processing strategy, compares the frame processing delay point in the marked segment with the current load status, and selects the frame processing instructions corresponding to the deviation rate to obtain the frame processing offset control instruction group; The algorithm adaptation module extracts the analysis accuracy value of the corresponding node before and after the adjustment instruction according to the frame processing offset control instruction group, combines the frame processing trend to identify the target detection accuracy path under the unit frame, extracts the change section to form the response interval, and obtains the node unit adaptation algorithm trend; The node management optimization module calls the node unit adaptation algorithm trend, collects the external environment data of the edge node, analyzes the overlap time between the edge computing optimization algorithm change and the network bandwidth jump period, and outputs the network impact duration; The network scheduling module filters the task scheduling instructions affected by network interference based on the duration of network impact, classifies the load switching time and adjusts the trigger frequency, identifies the load jump exceeding the limit period, and obtains the load interference impact frequency of the video analysis task.

[0019] The task allocation optimization mark values ​​include the frame rate fluctuation threshold, the processing delay variation range, and the load rise rate characteristics. The frame processing offset control instruction group includes the frame processing deviation, the delay variation anomaly value, and the load state offset rate. The node unit adaptation algorithm trend includes the analysis accuracy variation range, the target detection accuracy path type, and the algorithm response amplitude. The network impact duration includes the bandwidth fluctuation duration, the delay jump overlap period, and the network disturbance duration. The video analysis task load interference impact frequency includes the load switching frequency, the number of adjustment instruction triggers, and the number of jump period exceedances.

[0020] See also Figure 2 , the central management and control service modules include: The frame rate fluctuation extraction submodule extracts the video stream time series based on the edge node operation data, calculates the frame rate difference between adjacent sampling points, determines the fluctuation period and extracts the peak-valley interval, and selects the segments where the period difference exceeds the threshold to obtain the frame rate period fluctuation interval value; Based on edge node operating data, the frame rate performance of the processed video stream is monitored in real time. A frame rate sequence at different timestamps is obtained, such as FPS = [25, 26, 24, 23, 27, 28, 26, 25, 24, 22]. The difference between adjacent frame rates is calculated to obtain [-1, 2, -1, 4, 1, -2, -1, -1, -2]. FFT is then used to analyze its periodicity, identifying a frame rate fluctuation period of approximately 2 seconds. Peaks and troughs within this period are extracted and their difference is calculated, for example, 28 − 22 = 6 frames per second. The frame rate fluctuation threshold is set to 3 frames per second (based on historical data statistics and test verification). Segments exceeding the threshold are selected as sections with severe fluctuations. The resulting frame rate period fluctuation intervals are [1s, 3s] and [5s, 7s], which are used to identify performance anomalies.

[0021] The delay variation calculation submodule calls the frame rate period fluctuation interval value, identifies the corresponding delay data, analyzes the absolute delay variation amplitude of adjacent time periods, compares it with the set delay amplitude threshold, locates the mutation node, and obtains the delay mutation amplitude interval value; Call the frame rate period fluctuation interval value, such as the frame rate period fluctuation interval value [1s, 3s], [5s, 7s], identify the delay data corresponding to the interval, for example, in the time period from 1s to 3s, the delay data sequence obtained is [50ms, 55ms, 60ms, 65ms, 70ms], and in the time period from 5s to 7s, the delay data sequence obtained is [40ms, 42ms, 45ms, 60ms, 80ms]. Analyze the absolute change amplitude of the delay in adjacent time periods. For example, for the time period from 1s to 3s For the delay data within the time range of 5s to 7s, the absolute variation of the delay between adjacent sampling points is |55-50|=5ms, |60-55|=5ms, |65-60|=5ms, |70-65|=5ms. For the delay data within the time range of 5s to 7s, the absolute variation of the delay between adjacent sampling points is |42-40|=2ms, |45-42|=3ms, |60-45|=15ms, |80-60|=20ms. The absolute variation is compared with the set delay amplitude threshold, which is used to identify sudden changes in delay. For example, The delay amplitude threshold is set to 10ms. This threshold is based on statistics on the delay fluctuation range under normal operating conditions. By collecting delay data from edge nodes under stable load for at least 24 hours, the mean absolute deviation of the delay fluctuation is calculated as the threshold. Appropriate adjustments are made based on this threshold to balance sensitivity and false alarm rate. In actual applications, after months of operational data accumulation, it has been found that setting the delay amplitude threshold to 10ms can effectively identify over 90% of delay mutation events while maintaining a low false alarm rate. The above delay absolute change amplitude series is compared with the 10ms threshold. In the time period from 5s to 7s, the delay changes of 15ms and 20ms exceed the threshold. The mutation nodes exceeding the threshold are located, that is, the specific time points when the delay mutation occurs. For example, in the time period from 5s to 7s, the time points when the delay mutation changes from 45ms to 60ms and from 60ms to 80ms are located as mutation nodes. For example, [5.5s, 6.5s] indicates that the delay mutation occurred in this time period, resulting in the delay mutation amplitude interval value [5.5s, 6.5s].

[0022] The task status marking submodule identifies the corresponding load and algorithm complexity data based on the delay mutation amplitude interval value, extracts the load and algorithm data, combines the delay mutation and frame rate fluctuation, and uses the formula: ; Calculate the load complexity offset value, set the offset threshold, mark the time segment exceeding the threshold, and obtain the task allocation optimization mark value; in, Indicates the load complexity offset value, For the The load increase rate at the moment, For the The algorithm complexity value at the moment, For the The frame rate fluctuation difference at each moment, is the delay correction factor, is the load complexity offset parameter, Indicates the total number of moments; According to the delay mutation amplitude interval value, for example, the delay mutation amplitude interval value obtained above , the load data and algorithm complexity data corresponding to the identification interval, for example, arrive During the time period, the load increase rate of the edge node is recorded for , the corresponding algorithm complexity value for , and frame rate fluctuation difference for Frames / second, payload complexity offset parameter for , where the load rise rate It is calculated by monitoring the change in the CPU utilization, memory usage and other parameters of the edge node per unit time. For example, if the CPU utilization increases from 50% to 60% in 1 second, then 10% / second, algorithm complexity value The complexity of the currently running video analysis algorithm (such as the target detection algorithm YOLOv5 or the image recognition algorithm ResNet) is quantified by the amount of computation, memory usage and other indicators. For example, for the YOLOv5 model, its complexity can be quantified as , after normalization The value is 0.8, the frame rate fluctuation difference is the output frame rate fluctuation amplitude, and the load complexity offset parameter This value reflects the experimental value of load or algorithm adjustment to maintain service quality under specific delay mutations. The value is initially set to 0.1 and is dynamically adjusted based on the adjustment effect in historical delay mutation events. For example, if the historical adjustment effect is not good, the value will be appropriately increased to increase the weight of delay. is the delay correction factor, and its value is set to 0.6. The setting is to simulate the real network environment in the experiment, The calculated results of the load complexity offset value under the value of 1 are compared with the actual performance degradation, and the value that makes the calculated results reflect the actual performance degradation most accurately is selected. The value, experimentally proved, When set to 0.6, the calculated load complexity offset value is highly consistent with the actual observed performance degradation trend, ensuring that the formula accurately quantifies the impact of latency. Indicates the number of sampling points in the statistical time period, for example, here , using the formula Calculate the load complexity offset value , Indicates the load complexity offset value, which quantifies the comprehensive pressure caused by the current task on the system by comprehensively considering the load increase rate, algorithm complexity, frame rate fluctuation and delay correction parameters. Item indicates the The degree of influence of momentary load and algorithm complexity on frame rate fluctuation, among which and The product of represents the comprehensive computational burden of the current task, As a divisor, it represents the diluting effect of frame rate fluctuation on the burden. The term represents the correction effect of delay on the load complexity offset, where Is the delay correction factor, used to adjust the delay offset parameter The weight in the total offset value enables the formula to more accurately reflect the impact of latency fluctuations on system load. The overall operation logic of this formula is to combine the computational load of the task itself with the system's sensitivity to latency fluctuations to quantify the resource demand offset. The larger the value, the greater the system load pressure. The specific calculation process is as follows: ; Calculated load complexity offset value , set the offset threshold, which is used to judge whether the load complexity offset reaches the level that requires task allocation optimization. The offset threshold is set to 0.3. The setting of this threshold is based on the analysis of stable operation historical data. The 90% percentile is used as the threshold for statistics of the load complexity, and fine-tuned according to the false alarm rate and missed alarm rate in actual operation. Experimental verification shows that when the threshold is set to 0.3, 85% of overload situations can be effectively identified, while avoiding unnecessary task scheduling overhead. The calculated load complexity offset value Comparing with the offset threshold of 0.3, , indicating that the current load complexity offset value exceeds the threshold, marking the time segment exceeding the threshold. The benefit of the formula is that by combining the load increase rate, algorithm complexity, frame rate fluctuation and delay correction factor, it can comprehensively and accurately quantify the comprehensive load pressure of edge nodes in video analysis tasks, especially by introducing the delay correction factor and load complexity offset parameter , which enables the model to more sensitively capture the system performance degradation caused by delay fluctuations, thereby considering network instability factors in task allocation decisions, improving the accuracy of task allocation and the utilization efficiency of system resources.

[0023] See also Figure 3 , the edge computing module includes: The frame processing data matching submodule extracts the video stream frame rate and delay data for the corresponding period based on the task allocation optimization tag value, calculates the delay fluctuation amplitude according to the sampling interval, and aligns the frame rate and delay amplitude at the same time to obtain the frame delay linkage interval group; Based on the task allocation optimization tag value, such as the aforementioned task allocation optimization tag value [5.5s, 6.5s], the video stream frame rate and delay data of the corresponding time period are extracted. For example, in the time period from 5.5s to 6.5s, the system extracts video stream frame rate data of [25, 23, 20, 18] frames / second, and the corresponding delay data is [70, 85, 100, 120] milliseconds. The delay fluctuation amplitude is calculated according to the preset sampling interval. For example, the sampling interval is set to 0.1 seconds, and the delay fluctuation amplitude is calculated as the difference between the delays of adjacent sampling points, such as |85-70|=15ms, |100-85|=15ms, |120-100|=20ms s, align the frame rate and delay amplitude at the same time. For example, align the above frame rate and delay amplitude data with a sampling interval of 0.1 seconds to obtain the frame rate and corresponding delay fluctuation amplitude at each time point. For example, the time point 5.6s corresponds to a frame rate of 23 frames / second and a delay fluctuation amplitude of 15ms, the time point 5.7s corresponds to a frame rate of 20 frames / second and a delay fluctuation amplitude of 15ms, and the time point 5.8s corresponds to a frame rate of 18 frames / second and a delay fluctuation amplitude of 20ms. The frame delay linkage interval groups are [5.6s, 23fps, 15ms], [5.7s, 20fps, 15ms], and [5.8s, 18fps, 20ms].

[0024] The frame processing deviation judgment submodule calls the frame delay linkage interval group, extracts the load state sequence, compares the frame processing changes with the state differences, and compares them with the preset offset boundaries after normalization. The formula is used: ; Calculate the cumulative value of the deviation intensity of the frame processing point, extract the frame processing position index whose deviation intensity is greater than the reference judgment value, and establish the deviation intensity index group; in, Indicates the cumulative value of the deviation intensity of the frame processing point, represents the normalized frame number, Representative The current state value of the frame, Representative The preset boundary value of the frame, Representative Frame processing weight coefficient, Representative The state change rate of the frame, Represents the average value of the frame state change rate; Call the frame delay linkage interval group, such as the aforementioned frame delay linkage interval group ,Extract the load state sequence, the load state sequence reflects the resource usage of the edge node when processing the video frame. For example, based on the real-time monitoring data such as CPU utilization and memory usage, the load state sequence is 、 、 They are (normalized value between 0-1, 1 represents full load), compare the frame processing change with the load state difference, frame processing change refers to the correlation between frame rate and delay fluctuation amplitude, for example, when the frame rate drops and the delay fluctuation amplitude increases, it indicates that the frame processing performance has dropped, and the frame processing change is normalized with the load state difference, for example, the frame rate and delay fluctuation amplitude are normalized so that they are compared with the load state sequence in the same dimension, and compared with the preset offset boundary after normalization. The preset offset boundary is used to define the normal range of frame processing deviation, for example, the offset boundary is set to 0.15. The setting of the offset boundary is based on the relationship between the frame processing performance and the load state of the edge node under ideal operating conditions. By conducting a large number of tests under different load intensities, the difference between frame processing and load state is recorded, and the 90% percentile of the difference is used as the offset boundary. In actual applications, through long-term operation and data analysis, we found that setting the offset boundary to 0.15 can effectively identify 92% of frame processing deviations while avoiding misjudgment of normal fluctuations. The formula is used. Calculate the cumulative value of the deviation intensity of the frame processing point , Indicates the cumulative value of the deviation intensity of the frame processing point, quantifying the degree of deviation of each frame processing point from the preset normal behavior. The larger the value, the more serious the deviation. Represents the number of normalized frames, indicating that The number of frames considered when , in this case, , Representative The current state value of the frame refers to the current moment when the system processes the Normalized load state value at frame time, such as the load state sequence above , Representative The preset boundary value of the frame indicates that the system is processing the The upper limit of the load state expected at the frame time, for example, the default boundary value is set to 0.7, Representative Frame processing weight coefficient is used to adjust the importance of different frames in the deviation strength calculation. For example, when frame processing is stuck or lost, its weight coefficient will be increased accordingly. The value is , the weight coefficient The setting is based on the priority evaluation of different frame processing situations. For key video streams (such as important areas in security monitoring), the frame processing weight coefficient will be set to a higher value to ensure that its deviation can be identified in time. Through empirical adjustment and actual testing, it is found that Setting it between 0.05 and 0.2 can effectively reflect the importance of different frames and make it play a reasonable trade-off role in deviation calculation. Representative The state change rate of the frame refers to the The rate at which the load state of a frame changes relative to the previous frame, for example, calculated from the load state sequence , Represents the average value of the frame state change rate, calculated as all The average value, for example ,molecular Indicates the difference between the current state and the preset boundary, the denominator The deviation intensity is weighted and normalized, where and Ensure that the denominator is not zero and assign higher deviation sensitivity to high-weight frames and frames with drastic state changes. The overall operation logic of this formula is to accurately quantify the degree of deviation of frame processing by comparing the current state of the frame with the preset boundary and weighting it with the frame processing weight and state change rate. The larger the value, the more severe the deviation. The specific calculation process is as follows: for (Time point ): , , , , ; ; for (Time point ): , , , , ; ; for (Time point ): , , , , ; ; ; Calculate the cumulative value of the deviation intensity of the frame processing point , extract the frame processing position index with a deviation intensity greater than the benchmark judgment value. The benchmark judgment value is used to determine which frame processing points have a significant enough deviation and need intervention. For example, the benchmark judgment value is set to 0.08. The setting of this benchmark judgment value is based on the statistical analysis of the deviation intensity of frame processing under normal operating conditions. By collecting a large amount of frame processing data in typical scenarios, the The average value and standard deviation of the frame processing are calculated, and the average value plus 1.5 times the standard deviation is used as the benchmark judgment value to ensure that most abnormal situations can be identified. Experimental data show that when the benchmark judgment value is set to 0.08, the recognition accuracy of frame processing deviation reaches more than 90%. The calculated deviation intensity cumulative value Compared with the benchmark value of 0.08, , indicating that the deviation intensity exceeds the benchmark judgment value, so an offset intensity index group is established. The index group contains all frame processing positions with deviation intensity greater than the benchmark judgment value and their corresponding deviation intensity. For example, the index group contains Frame processing at this time point, the deviation intensity is 0.1399, and the offset intensity index group is obtained. By quantifying the difference between the current state of the frame and the preset boundary, and introducing the frame processing weight coefficient and the frame state change rate as weighting factors, the judgment of frame processing deviation is made more precise and accurate. In particular, the design of the denominator effectively avoids the error of dividing by zero and ensures greater sensitivity to the deviation of key frames and frames with drastic state changes, thereby improving the ability to identify abnormal frame processing behavior.

[0025] The frame processing instruction extraction submodule filters the position of the corresponding time point in the task instruction set based on the offset strength index group, extracts the instruction value, sorts it by time series, removes duplicate instructions, and obtains the frame processing offset control instruction group; Based on the offset strength index group, such as the aforementioned offset strength index group [5.8s, 0.1399], the position of the corresponding time point in the task instruction set is filtered. The task instruction set contains a series of predefined frame processing optimization instructions. For example, the task instruction set contains instruction A (corresponding to the time point 5.8s, the instruction content is "reduce video encoding bit rate") and instruction B (corresponding to the time point 5.9s, the instruction content is "switch to low complexity algorithm"). The instruction values ​​are extracted. For example, for the time point 5.8s, the value content of instruction A is extracted, "reduce video encoding bit rate". The extracted instructions are sorted in time series. For example, if there are multiple time points corresponding to instructions, they are arranged in chronological order. In this example, there is only one time point 5.8s. Duplicate instructions are eliminated. For example, if there are multiple identical instructions at the same time point, only one is retained. Finally, a frame processing offset control instruction group is obtained, which is [5.8s, "reduce video encoding bit rate"].

[0026] See also Figure 4 , the algorithm adaptation module includes: The analysis accuracy extraction submodule extracts the frame processing accuracy data of the node in the cycle before and after adjustment based on the frame processing offset control instruction group, identifies the analysis boundary conditions, and obtains the analysis accuracy difference value; According to the frame processing offset control instruction group, such as the aforementioned frame processing offset control instruction group [5.8s, "Reduce video encoding bit rate"], the frame processing accuracy data of the node in the loop before and after the adjustment is extracted. The frame processing accuracy data reflects the accuracy of the video analysis task for target recognition, classification, and other tasks before and after the adjustment. For example, before the 5.8s instruction is issued, the frame processing accuracy of the node in the past 5 seconds is recorded as 92%. After the instruction is issued, the frame processing accuracy in the next 5 seconds is recorded as 88%. The analysis boundary conditions are identified and used to define the validity and comparability of the accuracy data. For example, 100 frames of video data are collected before and after the adjustment as analysis samples, and the video content type is required to remain consistent to ensure fairness of the comparison. The final analysis accuracy difference value is obtained. For example, if the accuracy drops from 92% to 88%, the analysis accuracy difference value is 92% - 88% = 4%.

[0027] The target detection accuracy recognition submodule calls and analyzes the accuracy difference value, identifies the target detection accuracy change trajectory in the unit frame path, determines the target number trend corresponding to the accuracy fluctuation, compares the accuracy drop and target number fluctuation in adjacent time periods, calculates the accuracy concentration of the segment in the unit frame path, filters the synchronous fluctuation segments, and obtains the target detection accuracy feature segment; By analyzing the accuracy difference (e.g., 4%), we identified the trajectory of target detection accuracy changes within the unit frame path, tracked the fluctuations in detection accuracy for each frame, and the corresponding target number trend. For example, in the 5.8s to 6.0s segment, the accuracy dropped from 90% to 85%, and the number of targets decreased from 10 to 8, showing a consistent decline. We compared the accuracy drop and target number fluctuations in adjacent time periods and calculated the accuracy concentration within that segment. If the concentration was 80%, it reflected the stability of detection accuracy. We then screened out segments where the accuracy and target number fluctuated synchronously, ultimately obtaining the target detection accuracy characteristic segment [5.8s, 6.0s], indicating a significant decline in algorithm performance within that period.

[0028] The algorithm response interval identification submodule analyzes the time and precision displacement trends based on the target detection accuracy feature segments, determines the direction consistency and amplitude change characteristics, screens the fluctuation segments and aggregates them to obtain the node unit adaptation algorithm trend; Analyze the time and accuracy displacement trends based on target detection accuracy characteristic segments. The time displacement trend reflects the timing and duration of accuracy fluctuations, while the accuracy displacement trend shows the trajectory of accuracy changes from normal to abnormal. For example, from 5.8s to 6.0s, accuracy drops from 90% to 85%. Determine the consistency between the direction of accuracy change and the direction of target number change. Accuracy decreases and the number of targets decreases by 5%. Fluctuation segments are screened out, which are time periods where both accuracy and target number experience abnormal changes. Adjacent segments are aggregated into larger response intervals to obtain the node unit adaptation algorithm trend. For example, [5.8s, 6.0s, "Accuracy decreases, algorithm adjustment required"] indicates that algorithm performance degraded during this time period and requires adaptation.

[0029] See also Figure 5 , the node management optimization module includes: The algorithm trend extraction submodule calls the node unit adaptation algorithm trend, extracts the node task record and periodic algorithm data, and identifies the fluctuation amplitude and frequency within a single cycle using the formula: ; Get the algorithm change trend value; in, Represents the algorithm change trend value, Representative The fluctuation coefficient corresponding to the node task record within a period, Representative The original algorithm output value of the node in the cycle, Represents the arithmetic mean of the original algorithm output values ​​of the node in this cycle, Representative The maximum change difference of the node algorithm data sequence within a period, Represents the total number of nodes in the cycle; Call the node unit adaptation algorithm trend, such as the aforementioned node unit adaptation algorithm trend [5.8s, 6.0s, "accuracy decreased, algorithm adjustment required"], extract the node task record and periodic algorithm data. The node task record contains the performance data of the edge node when performing a specific task, such as CPU usage, memory usage, number of processed frames, etc. The periodic algorithm data contains the performance indicators of the algorithm in different cycles. For example, in the task record, it is identified that the CPU usage of the node from 5.8s to 6.0s is 85%, the memory usage is 90%, and the number of processed frames is 18 frames / second. In the periodic algorithm data, it is identified that the fluctuation range of the algorithm in the normal cycle is 2% (referring to the maximum deviation of the algorithm's running results and its average value each time), and the fluctuation frequency is 1 time / minute. Identify the fluctuation range and fluctuation frequency within a single cycle. For example, in arrive In this single cycle, the actual fluctuation range of the algorithm is identified to be 5%, and the fluctuation frequency is 2 times / minute. The system uses the formula Get algorithm change trend value ,in, Represents the trend value of algorithm change, quantifying the overall fluctuation of algorithm performance within a specific period. The larger the value, the greater the algorithm fluctuation and the more unstable the performance. Represents the total number of nodes in the cycle, indicating that The number of nodes considered when , in this case, (single node), Representative The fluctuation coefficient corresponding to the node task record in a period is used to measure the contribution of the node task record to the algorithm fluctuation. For example, The value is 0.5, the fluctuation coefficient The setting is based on the correlation analysis between the node task record data (such as CPU usage, memory usage, etc.) and the algorithm performance fluctuation. The higher the correlation, the larger the coefficient. Through regression analysis and correlation evaluation of a large amount of historical data, it is found that Setting it to 0.5 can effectively reflect the aggravating effect of node resource consumption on algorithm fluctuations. Representative The original algorithm output value of the node in the cycle, for example, In the cycle, the node original algorithm output value is (Right now arrive Frame processing accuracy during the period), Represents the arithmetic mean of the original algorithm output values ​​of the node in this cycle, for example, (Right now arrive The average expected accuracy during the period), Representative The maximum change difference of the node algorithm data sequence within a period refers to the difference between the maximum and minimum values ​​of the algorithm performance within a specific period. For example, arrive During this period, the maximum value of the algorithm performance is 90% and the minimum value is 85%. , in brackets The term measures the contribution of the node task record to the deviation of the algorithm output value from the average value, and the square term This makes both positive and negative deviations affect the results, and increases the impact of large deviations on the results. The term is simplified to , which directly reflects the maximum fluctuation of the algorithm's performance. The overall operational logic of this formula is to comprehensively quantify the overall stability of the algorithm within a specific period by weightedly measuring the degree of deviation of the algorithm's output from its average value and combining it with the maximum fluctuation of the algorithm itself. The larger the value, the greater the algorithm's fluctuation and the more unstable the performance. The specific calculation process is as follows: for (Single Node): , , , , ; ; Calculated algorithm change trend value , get the algorithm change trend value The benefit of this formula is that it not only considers the degree of deviation of the algorithm output value from the average value, but also introduces the fluctuation coefficient of the node task record to weigh the importance of this deviation. At the same time, by incorporating the maximum change difference of the algorithm itself, the trend value can more comprehensively and sensitively reflect the actual operation stability of the algorithm and the coupling effect of node resources, thereby providing a more accurate decision-making basis for subsequent network scheduling and task allocation.

[0030] The network jump identification submodule selects bandwidth and latency records within the same period based on the algorithm change trend value and the external network data of the edge node. It compares the data change amplitude on a daily basis and selects time nodes with amplitudes greater than the network jump threshold to obtain the bandwidth delay jump period. Based on the algorithm's trend values ​​and external network data from edge nodes, bandwidth and latency records within the same period are selected. The data change amplitude is compared daily, and the daily averages of bandwidth and latency are compared with the previous day to calculate the change amplitude. Time nodes with amplitudes greater than the preset network jump thresholds are selected. The bandwidth jump threshold is set to 20 Mbps, and the latency jump threshold is set to 30 ms. The threshold settings are based on historical data statistics to ensure that significant bandwidth and latency jumps are captured. Experimental tests show that when the bandwidth jump threshold is 20 Mbps and the latency jump threshold is 30 ms, 95% of abnormal network jump events can be accurately identified. By comparing the bandwidth and latency change amplitudes with the thresholds, the latency jump event at time 6.0s is selected, and the bandwidth and latency jump period is finally determined to be [6.0s, 6.0s].

[0031] The network impact quantification submodule identifies the intersection of algorithm fluctuations and network fluctuations based on the bandwidth delay jump period, performs weighted processing, and normalizes the usage duration under differentiated cycles by referring to the fluctuation frequency, combined fluctuation value, and period fluctuation to output the network impact duration. Based on the bandwidth delay jump period, such as the aforementioned bandwidth delay jump period [6.0s, 6.0s], the intersection duration of the algorithm amplitude and the network jump is identified. Algorithm amplitude refers to the period of time when the algorithm performance fluctuates significantly, and network jump refers to the period of time when the network bandwidth or delay changes significantly. The algorithm amplitude period is identified as 5.8s to 6.0s, and the network jump period is identified as 6.0s. The intersection duration of the two is 0.0s (that is, the intersection is only the moment of 6.0s). Weighted processing is performed. The weighted processing is to comprehensively consider the contribution of different factors to the impact of the network. For example, different weights can be set for the algorithm amplitude frequency, the combined amplitude value, and the periodic fluctuation. For example, the amplitude frequency weight is set to 0.3, the combined amplitude weight is set to 0.4, and the periodic fluctuation weight is set to 0.3. The weight setting is based on the analysis of different network and algorithm anomalies in historical data. Through expert experience and regression analysis, the impact of each factor on the final network is determined. To ensure the accuracy of the actual network impact, the usage duration under different cycles is normalized by referring to the amplitude fluctuation frequency, combined amplitude fluctuation value, and periodic fluctuation. The amplitude fluctuation frequency refers to the number of times the algorithm performance fluctuates within a specific cycle. For example, the algorithm performance fluctuates twice between 5.8s and 6.0s. The combined amplitude fluctuation value refers to the maximum amplitude of the algorithm performance fluctuation. For example, if the algorithm accuracy drops from 90% to 85%, the combined amplitude fluctuation value is 5%. Periodic fluctuation refers to the fluctuation range of the algorithm performance over the entire cycle. For example, the periodic fluctuation is 0.0502 (the aforementioned algorithm change trend value). Normalization is to eliminate the impact of different cycle lengths on the results. The usage duration under different cycles is standardized to a unified scale, and the final output is the network impact duration. For example, by comprehensively considering the above factors and performing weighted normalization, the network impact duration is calculated to be 2.5 seconds, resulting in a network impact duration of 2.5s.

[0032] See also Figure 6 , the network scheduling module includes: The instruction screening submodule screens matching task scheduling instructions based on the duration of network interference, identifies the load period and load status, compares the disturbance period with the instruction period, eliminates low-matching instructions, and obtains the instruction set affected by network interference; Based on the duration of the network impact, such as the aforementioned network impact duration of 2.5s, the task scheduling instructions are screened and matched. The task scheduling instructions are pre-set instructions for adjusting the task allocation of edge nodes. For example, the task scheduling instruction set includes instructions such as "reduce video resolution", "suspend non-core analysis tasks", and "switch to backup servers". The load period and load status are identified. The load period refers to the time period when the edge node is in a high-load state. The load status refers to the resource utilization of the current edge node. For example, it is identified that the edge node is in a high-load state (CPU utilization exceeds 80%) from 5.8s to 6.0s. The system compares the disturbance period with the instruction period. The action cycle refers to the duration of the network impact, and the instruction cycle refers to the execution cycle of the task scheduling instruction. For example, if the network impact duration is 2.5 seconds, matching instructions are selected. If the instruction cycle is close to the network impact duration, it is considered to be highly matched, and low-matching instructions are eliminated. For example, if the execution cycle of an instruction is 10 seconds, which does not match the 2.5-second network impact duration, the instruction is eliminated, and the final set of instructions affected by network interference is obtained. For example, the "reduce video resolution" instruction is filtered out. The execution cycle of this instruction is 2 seconds, which has a high match with the 2.5-second network impact duration, and the instruction set affected by network interference ["reduce video resolution"] is obtained.

[0033] The load classification submodule calls the set of instructions affected by network interference, extracts the load switching time and adjustment trigger frequency corresponding to the instructions, arranges them in order of switching time, classifies them by frequency interval, calculates the correspondence between frequency bands and load time periods, and obtains the load adjustment frequency band distribution value; Call the set of instructions affected by network interference, such as the aforementioned set of instructions affected by network interference "reduce video resolution", extract the load switching time and adjustment trigger frequency corresponding to the instructions, the load switching time refers to the time required for the load state to change after the instruction is executed, and the adjustment trigger frequency refers to the frequency at which the instruction is triggered to execute, for example, the load switching time of the "reduce video resolution" instruction is 1 second, and the adjustment trigger frequency is once every 5 seconds. Arrange them in order of switching time. For example, if there are multiple instructions, sort them in ascending order according to their load switching time, and classify them by frequency interval. Frequency interval classification is to classify instructions with similar adjustment Commands that adjust trigger frequencies are grouped together. For example, the frequency ranges can be set to 0-0.2 Hz (low frequency), 0.2-0.5 Hz (medium frequency), and 0.5-1.0 Hz (high frequency). Statistics are then compiled regarding the correspondence between frequency bands and load time periods. For example, the adjustment trigger frequency (0.2 Hz) of the "Reduce video resolution" command falls within the medium frequency band, and this command is executed during periods of high load (for example, CPU utilization exceeding 80%). Finally, the load adjustment frequency band distribution value is obtained. For example, the load adjustment frequency band distribution value is ["medium frequency", "high load period"], indicating that the medium frequency command is primarily effective during high load periods.

[0034] The interference identification submodule adjusts the frequency band distribution value according to the load, collects the load jump amplitude and duration within the frequency band, determines whether it exceeds the jump amplitude threshold, identifies the frequency band interference intensity, sorts the interference intensity of all frequency bands, determines the frequency band and jump period of the interference intensity, and obtains the load interference impact frequency of the video analysis task; According to the load adjustment frequency band distribution value, such as the aforementioned load adjustment frequency band distribution value ["medium frequency", "high load period"], the load jump amplitude and duration within the frequency band are collected. The load jump amplitude refers to the instantaneous change in the edge node load level from normal to abnormal within a specific frequency band, and the duration refers to the duration of this abnormal load state. For example, in the high load period corresponding to the medium frequency band, the collected load jump amplitude is 20% (CPU utilization jumps from 60% to 80%), and the duration is 10 seconds. It is judged whether it exceeds the jump amplitude threshold. The jump amplitude threshold is used to identify the severity of the load jump. For example, the jump amplitude threshold is set to 15%. The setting of this jump amplitude threshold is based on the statistics of load fluctuations under normal system operation. By collecting a large amount of data from edge nodes under stable load, the average absolute deviation of the load fluctuations is calculated, and adjustments are made on this basis to take into account both sensitivity and false alarm rate. In actual deployment, after several months, Based on the accumulation of operational data, it was found that setting the jump amplitude threshold to 15% can effectively identify 88% of load mutation events while maintaining a low false alarm rate. The collected load jump amplitude of 20% is compared with the threshold of 15%. 20%>15% indicates that the jump amplitude threshold is exceeded. The system identifies the frequency band interference intensity. The frequency band interference intensity is comprehensively evaluated based on the load jump amplitude and duration. For example, a large load jump amplitude and a long duration indicate high interference intensity. The system sorts the interference intensity of all frequency bands. For example, if there are multiple frequency bands, they are sorted from high to low according to interference intensity. Finally, the frequency band and jump period with the highest interference intensity are determined. For example, the mid-frequency band is determined to be the frequency band with the highest interference intensity, and the corresponding jump period is [5.8s, 6.0s]. The frequency of load interference impact on the video analysis task is obtained. For example, the load interference impact frequency of the video analysis task is 1, which occurs during the high load period in the mid-frequency band of [5.8s, 6.0s].

[0035] The real-time video analysis method based on edge computing is performed based on the above-mentioned real-time video analysis system based on edge computing, and includes the following steps: S1: Based on edge node operation data, it analyzes frame rate fluctuations and processing delay changes, extracts periods of sudden load peaks and sharp changes in algorithm complexity, identifies state switching points where delay and load changes are synchronized, marks task levels, and generates task state switching identification fragments. S2: Based on the task state switching identification segment, the frame processing change data and delay change amplitude are extracted, the frame processing mutation is compared with the delay fluctuation amplitude, the correspondence between the frame processing change and the load response is analyzed, and the frame processing adjustment response trajectory segment is obtained; S3: Based on frame processing, it adjusts the response trajectory segments, extracts the change in accuracy value and target detection accuracy gradient before and after the response, analyzes the target detection accuracy drop and the increase in the number of targets at the node, identifies the precision decay path and compares it with the original cycle data, selects the segments with precision decay characteristics, and generates the precision adaptation decay path distribution set; S4: Based on the precision adaptive attenuation path distribution set, identify external network data within the corresponding time period, analyze the overlap period between the bandwidth jump and the precision attenuation path, extract the associated time period, and obtain the network parameter interference section for frame processing adjustment; S5: Based on the network parameter interference segment adjusted by frame processing, the scheduling control strategy records within the period are extracted, the time interval of load switching and the adjustment trigger frequency are analyzed, and the high-frequency adjustment strategy segments are screened to obtain the load interference impact frequency of the video analysis task.

[0036] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A real-time video analysis system based on edge computing, characterized in that: The system comprises: Based on edge node operation data, the central control service module analyzes load fluctuations by comparing the changes in video stream frame rate and processing latency, screens key high-load time periods, extracts critical conditions for task scheduling, and obtains task allocation optimization mark values. The edge computing module optimizes the marking value based on the task allocation, matches the video stream data of the corresponding time period with the frame processing strategy, compares the frame processing delay point in the marked segment with the current load status, filters the frame processing instructions corresponding to the deviation rate, and obtains the frame processing offset control instruction group; The algorithm adaptation module extracts the analysis accuracy value of the corresponding node before and after the adjustment instruction according to the frame processing offset control instruction group, identifies the target detection accuracy path under the unit frame in combination with the frame processing trend, extracts the change section to form a response interval, and obtains the node unit adaptation algorithm trend; The node management optimization module calls the node unit adaptation algorithm trend, collects the external environment data of the edge node, analyzes the overlap time between the edge computing optimization algorithm change and the network bandwidth jump period, and outputs the network impact duration.

2. The real-time video analysis system based on edge computing according to claim 1 is characterized in that: The task allocation optimization mark value includes the frame rate fluctuation threshold, the processing delay change amplitude range, and the load increase rate characteristics; the frame processing offset control instruction group includes the frame processing deviation amount, the delay change anomaly value, and the load state offset rate; the node unit adaptation algorithm trend includes the analysis accuracy change interval, the target detection accuracy path type, and the algorithm response amplitude; the network impact duration includes the bandwidth fluctuation duration, the delay jump overlap period, and the network disturbance duration.

3. The real-time video analysis system based on edge computing according to claim 1 is characterized in that: The central management and control service module includes: The frame rate fluctuation extraction submodule extracts the video stream time series based on the edge node operation data, calculates the frame rate difference between adjacent sampling points, determines the fluctuation period and extracts the peak-valley interval, and selects the segments where the period difference exceeds the threshold to obtain the frame rate period fluctuation interval value; The delay variation calculation submodule calls the frame rate period fluctuation interval value, identifies the corresponding delay data, analyzes the absolute delay variation amplitude of adjacent time periods, compares it with the set delay amplitude threshold, locates the mutation node, and obtains the delay mutation amplitude interval value; The task status marking submodule identifies the corresponding load and algorithm complexity data according to the delay mutation amplitude interval value, extracts the load and algorithm data, combines the delay mutation and frame rate fluctuation, calculates the load complexity offset value, sets the offset threshold, marks the time segment exceeding the threshold, and obtains the task allocation optimization marking value.

4. The real-time video analysis system based on edge computing according to claim 3 is characterized in that: The edge computing module includes: The frame processing data matching submodule extracts the video stream frame rate and delay data of the corresponding time period based on the task allocation optimization mark value, calculates the delay fluctuation amplitude according to the sampling interval, aligns the frame rate and delay amplitude at the same time, and obtains the frame delay linkage interval group; The frame processing deviation judgment submodule calls the frame delay linkage interval group, extracts the load state sequence, compares the frame processing change with the state difference, compares it with the preset offset boundary after normalization, calculates the cumulative value of the deviation intensity of the frame processing point, extracts the frame processing position index with a deviation intensity greater than the reference judgment value, and establishes an offset intensity index group; The frame processing instruction extraction submodule filters the position of the corresponding time point in the task instruction set based on the offset strength index group, extracts the instruction value, sorts it by time series, removes duplicate instructions, and obtains the frame processing offset control instruction group.

5. The real-time video analysis system based on edge computing according to claim 4 is characterized in that: The algorithm adaptation module includes: The analysis accuracy extraction submodule extracts the frame processing accuracy data of the node in the cycle before and after the adjustment according to the frame processing offset control instruction group, identifies the analysis boundary conditions, and obtains the analysis accuracy difference value; The target detection accuracy identification submodule calls the analysis accuracy difference value, identifies the target detection accuracy change trajectory in the unit frame path, determines the target number trend corresponding to the accuracy fluctuation, compares the accuracy drop and the target number fluctuation in adjacent time periods, calculates the accuracy concentration of the segment in the unit frame path, filters the synchronous fluctuation segment, and obtains the target detection accuracy feature segment; The algorithm response interval identification submodule analyzes the time and precision displacement trends according to the target detection accuracy feature segment, determines the direction consistency and amplitude change characteristics, screens the fluctuation segments and aggregates them, and obtains the node unit adaptation algorithm trend.

6. The real-time video analysis system based on edge computing according to claim 5 is characterized in that: The node management optimization module includes: The algorithm trend extraction submodule calls the node unit adaptation algorithm trend, extracts the node task record and periodic algorithm data, identifies the fluctuation amplitude and fluctuation frequency within a single cycle, and obtains the algorithm change trend value; The network jump identification submodule selects bandwidth and delay records within the same period based on the algorithm change trend value and the external network data of the edge node, compares the data change amplitude on a daily basis, and selects time nodes with amplitudes greater than the network jump threshold to obtain the bandwidth delay jump period; The network impact quantification submodule identifies the intersection duration of the algorithm amplitude variation and the network jump according to the bandwidth delay jump period, performs weighted processing, and normalizes the usage duration under the differentiated period with reference to the amplitude variation frequency, the combined amplitude variation value and the period fluctuation, and outputs the network impact duration.

7. The real-time video analysis system based on edge computing according to claim 1, characterized in that: The system also includes a network scheduling module: The network scheduling module filters the task scheduling instructions affected by the network interference based on the duration of the network impact, classifies the load switching time and adjusts the trigger frequency, identifies the load jump exceeding the limit period, and obtains the load interference impact frequency of the video analysis task; The frequency of load interference impact of the video analysis task includes load switching frequency, adjustment instruction triggering times, and jump period exceeding times.

8. The real-time video analysis system based on edge computing according to claim 7, characterized in that: The network scheduling module includes: The instruction screening submodule screens matching task scheduling instructions based on the duration of the network impact, identifies the load period and load state, compares the disturbance period with the instruction period, eliminates low-matching instructions, and obtains an instruction set affected by the network interference; The load classification submodule calls the instruction set affected by network interference, extracts the load switching time and adjustment trigger frequency corresponding to the instructions, arranges them in order of switching time, classifies them by frequency interval, counts the correspondence between frequency bands and load time periods, and obtains the load adjustment frequency band distribution value; The interference identification submodule adjusts the frequency band distribution value according to the load, collects the load jump amplitude and duration within the frequency band, determines whether it exceeds the jump amplitude threshold, identifies the frequency band interference intensity, sorts the interference intensity of all frequency bands, determines the frequency band and jump period of the interference intensity, and obtains the load interference impact frequency of the video analysis task.

9. A real-time video analysis method based on edge computing, characterized in that: The method is used to implement the real-time video analysis system based on edge computing according to any one of claims 1 to 8, comprising the following steps: S1: Based on edge node operation data, it analyzes frame rate fluctuations and processing delay changes, extracts periods of sudden load peaks and sharp changes in algorithm complexity, identifies state switching points where delay and load changes are synchronized, marks task levels, and generates task state switching identification fragments. S2: Based on the task state switching identification segment, extract the frame processing change data and the delay change amplitude, compare the frame processing mutation with the delay fluctuation amplitude, analyze the correspondence between the frame processing change and the load response, and obtain the frame processing adjustment response trajectory segment; S3: Based on the frame processing, the response trajectory segment is adjusted, the change in analysis accuracy value and target detection accuracy gradient before and after the response is extracted, the target detection accuracy drop and the increase in the number of targets at the node are analyzed, the accuracy decay path is identified and compared with the original cycle data, the segments with accuracy decay characteristics are screened, and the accuracy adaptation decay path distribution set is generated; S4: Based on the precision adaptive attenuation path distribution set, identifying external network data within a corresponding time period, analyzing the overlapping period of bandwidth jump and precision attenuation path, extracting the associated time period, and obtaining the network parameter interference section for frame processing adjustment; S5: Based on the network parameter interference section adjusted by the frame processing, the scheduling control strategy records within the time period are extracted, the time interval of the load switching and the adjustment trigger frequency are analyzed, and the high-frequency adjustment strategy segments are screened to obtain the load interference impact frequency of the video analysis task.

Images