Data management system based on monitoring video stream

Abstract

The invention discloses a data management system based on a monitoring video stream, belongs to the field of data analysis, and aims to solve the problems of camera connection acquisition, image quality optimization, pedestrian related information analysis and data storage retrieval and backup in video monitoring. The system comprises a data acquisition module, a target detection module and a data storage module. The data acquisition module is used for identifying and connecting a video monitoring camera, performing video stream data acquisition, adjusting an acquisition frame rate by estimating a data size, and evaluating and optimizing image quality to obtain a detection image data set; the target detection module is used for carrying out feature extraction, classification and marking on pedestrians in the detection image data set, recording pedestrian trajectory information, carrying out real-time tracking and updating, identifying a behavior mode according to the trajectory information, dividing a gathering area and obtaining the pedestrian flow; and the data storage module is used for classifying and labeling the data, constructing an index structure to guarantee efficient retrieval, dividing data grades according to multi-dimensional features, and dynamically adjusting a backup time interval.

Description

Technical Field

[0001] The present invention relates to the field of data analysis, and more particularly to a data management system based on monitoring video streams. Background Art

[0002] Among all kinds of activities in human society, the flow of people is undoubtedly a crucial feature of social operation. With the continuous acceleration of the global urbanization process and the significant increase in the density of people in public places, the accurate monitoring and efficient management of human flow has become a core component of modern urban management, public safety protection and commercial operation optimization. However, traditional methods of human flow monitoring often rely on manual statistics or simple sensor equipment. Although these methods can meet basic needs in some situations, they have many limitations. For example, manual statistics are not only time-consuming and labor-intensive, but also difficult to ensure the accuracy and real-time nature of the data; and traditional sensor equipment has limited coverage and poor adaptability to complex environments, and often cannot provide comprehensive and accurate information. These problems make traditional monitoring methods face considerable challenges in practical applications.

[0003] However, with the rapid development of computer vision technology, a new and more efficient solution has been provided through in-depth processing and intelligent analysis of video data. Computer vision technology can not only capture and analyze a large amount of video data in real time and accurately, and identify the behavior patterns of individuals and groups, but also automatically realize accurate counting and dynamic monitoring of human traffic through algorithm optimization. This method not only effectively makes up for the shortcomings of traditional methods, but also has significant advantages in accuracy, real-time performance, coverage and cost-effectiveness, and has gradually become an innovative highlight in the field of modern urban management and public safety.

[0004] A Chinese patent with announcement number CN102855466B discloses a method for counting people based on video image processing, the method comprising the following steps: S1, obtaining a video stream image of a monitoring area; S2, obtaining a pedestrian target by a frame difference method, and performing mathematical morphological operations on the frame difference result to obtain a pedestrian motion area; S3, segmenting the pedestrian motion area at least once by a coordinate projection segmentation method to segment a plurality of sub-areas; S4, searching and matching sub-areas of adjacent frames in combination with the center point distance of the sub-area and the grayscale information distribution characteristics; S5, instantly updating the center point position information of the sub-area and saving it in a storage device, and obtaining a result of counting people by calculating the trajectory information of the center point of the sub-area in the monitoring area.

[0005] Although the existing technology can effectively improve the efficiency and accuracy of real-time counting of people in public areas, it still fails to solve the problems of intelligent connection and frame rate adaptation during video data acquisition, multi-dimensional precise optimization of image quality, comprehensive and in-depth analysis of pedestrian information, and efficient retrieval and dynamic backup during data storage. Therefore, in order to overcome these limitations, the present invention proposes a data management system based on surveillance video streams. Summary of the invention

[0006] In view of the shortcomings of the prior art, the purpose of the present invention is to provide a data management system based on surveillance video streams, which solves the problems of camera connection and data acquisition, optimization of acquired image quality, pedestrian feature extraction and trajectory and other related information analysis, efficient retrieval of stored data and reasonable backup according to importance, so as to achieve more effective management and utilization of surveillance video stream data.

[0007] To achieve the above object, the present invention provides the following technical solutions: A data management system based on monitoring video stream, comprising a data acquisition module, a target detection module and a data storage module; The data acquisition module is used to identify connected video surveillance cameras and collect video stream data. It estimates the amount of data to adjust the acquisition frame rate, builds an evaluation image data set, and performs multi-dimensional image quality evaluation. Based on the evaluation results, the image clarity, contrast, and noise level are adjusted and optimized to obtain a detection image data set. The target detection module is used to extract features and classify pedestrians in the detection image data set to mark pedestrians and their categories, record pedestrian trajectory information and track and update it in real time, identify pedestrian behavior patterns based on trajectory information, divide gathering areas, and calculate real-time pedestrian flow in gathering areas; The data storage module is used to classify and label the stored data, establish an index structure according to multiple dimensions to achieve efficient retrieval, collect multi-dimensional features of the data to calculate importance scores, and divide the importance levels of data types to adjust the backup time intervals for data of different levels.

[0008] Specifically, the data acquisition module includes a video source adaptation unit and an image preprocessing unit; The video source adapter unit is configured with an intelligent connection strategy, which is used to identify compatible devices by transmitting detection signals, extract and match protocol types, adapt to the optimal parameter combination, configure devices and establish stable connections, so as to realize the integration of video surveillance cameras and video stream data collection; The image preprocessing unit is equipped with an adaptive optimization strategy, which is used to adjust the acquisition frame rate by setting the acquisition time and estimating the data volume, and to perform image evaluation on the constructed evaluation image data set, mark abnormalities according to the threshold setting, and calculate the evaluation failure rate, and adjust the initial image optimization parameters.

[0009] Specifically, the steps of the adaptive optimization strategy include: Set the initial duration of the video surveillance camera in the device list to collect video stream data, and estimate the amount of image data in the video stream data based on the parameter configuration of the video surveillance camera, that is: ; in, is the amount of image data in the video stream data estimated at the initial duration, and Respectively The number of pixels in the horizontal and vertical directions of a video surveillance camera, yes The frame rate of a video surveillance camera, is the number of video surveillance cameras in the device list, is the initial duration of the video surveillance camera collecting video stream data, is the number of bytes per pixel; Configure the ratio threshold, obtain the current available memory resources, and dynamically adjust the acquisition frame rate by calculating the ratio of the image data volume to the available memory resources, that is: ; in, is the amount of image data per unit of available memory resources, is the currently available memory resource; if If the ratio is greater than the threshold, the frame rate of the video surveillance camera is adjusted, that is: ;in, is the ceiling function, is the adjusted The frame rate of a video surveillance camera, is the adjustment factor; Configure the initial image optimization parameters, including sharpening parameters, contrast stretching ratio coefficient, and filter kernel size, and collect video stream data according to the adjusted frame rate of the video surveillance camera; The collected video stream data is processed by the initial image optimization parameters to form a benchmark data set. The video stream data collected by each video surveillance camera is randomly extracted from the benchmark data set. image pairs to construct the evaluation image dataset.

[0010] Specifically, the steps of the adaptive optimization strategy also include: Perform image evaluation on the images in the evaluation image dataset that have been processed by image optimization parameters, and calculate image evaluation indicators, including: To evaluate the clarity of multi-scale images, wavelet transform is used to perform scale analysis on each image, obtain image representations at different scales, and calculate the image gradient amplitude at each scale to obtain the comprehensive image gradient amplitude, namely: ; in, is the comprehensive image gradient amplitude, yes Direction, the scale is The directional gradient when yes Direction, the scale is The directional gradient at time, scale The value range is [1, ], The scale is The weight of Perform local weighted contrast evaluation, evaluate the image contrast by comparing the standard deviation change of the pixel value sequence of each image with the pixel value sequence of its contrast-stretched image, divide each image into multiple non-overlapping sub-regions, and determine its weight according to the pixel distribution characteristics of each sub-region to obtain the pixel mean and pixel standard deviation of each image to calculate the image contrast ratio, that is: ; in, is the contrast ratio, is the standard deviation of the contrast-stretched image, is the standard deviation of the original image; The noise level of the optimized images in the evaluation image dataset is evaluated, and the noise level of the image is calculated by the signal-to-noise ratio.

[0011] Specifically, the steps of the adaptive optimization strategy also include: Configure evaluation thresholds, including clarity threshold, contrast threshold, and signal-to-noise ratio threshold; If the integrated image gradient amplitude of the image is less than the clarity threshold, it is marked as an abnormal clarity image, otherwise no operation is performed; if the contrast ratio of the image is less than the contrast threshold, it is marked as an abnormal contrast image, otherwise no operation is performed; if the noise level of the image is greater than the signal-to-noise ratio threshold, it is marked as an abnormal noise image, otherwise no operation is performed; Configure evaluation failure thresholds, including clarity failure threshold, contrast failure threshold, and noise failure threshold, and calculate evaluation failure rates of images captured by video surveillance cameras in the evaluation image dataset, including clarity failure rate, contrast failure rate, and noise failure rate; If the evaluation failure rate of the images collected by the video surveillance camera is greater than the evaluation failure threshold, the image optimization parameters are gradually adjusted; According to the adjusted image optimization parameters, the original images in the evaluation image data set are optimized, the image evaluation is re-performed, and the change rate of the image evaluation index is calculated; Configure the evaluation change thresholds, including the clarity change threshold, contrast change threshold, and signal-to-noise ratio change threshold. If the image evaluation index change rate is greater than the evaluation change threshold, retain the image optimization parameters before adjustment and issue a parameter warning. Otherwise, use the adjusted image optimization parameters. After adjusting the image optimization parameters, re-evaluate the image evaluation failure rate and determine whether to continue adjusting the image optimization parameters; Configure the adjustment times threshold to count the times of image optimization parameter adjustment. If the image optimization parameter adjustment times are greater than the adjustment times threshold, stop the image optimization parameter optimization and output the adjusted image optimization parameters.

[0012] Specifically, the target detection module includes a target recognition unit, a tracking and analysis unit, and a flow calculation unit; The target recognition unit is configured with a deep recognition strategy, which is used to extract and classify pedestrian features in the detection image data set according to the convolutional neural network model, and annotate pedestrians and pedestrian categories in the form of bounding boxes; The tracking and analysis unit is equipped with a continuous recognition strategy, which is used to establish a tracker for the pedestrians recognized for the first time, record the pedestrian trajectory information, including the initial position, movement speed, and movement position, and perform tracking updates through multi-modal information fusion; The flow calculation unit is configured with a regional division strategy, which is used to identify pedestrian behavior patterns based on pedestrian trajectory information, including stop behavior and walking behavior, perform aggregate behavior recognition based on the pedestrian behavior model, divide the aggregation area, and calculate the real-time pedestrian flow in the aggregation area.

[0013] Specifically, the steps of the continuous identification strategy include: In the starting frame or the image frame where a new pedestrian enters the monitoring area, a tracker is created for each pedestrian detected for the first time, and a unique identification number is set for each tracker; The initial position information and initial frame of the pedestrian are recorded in the matching tracker, and based on the feature point matching method, the position change of the pedestrian in the image frame is obtained to generate the motion trajectory; Configuring Interval Thresholds , used to estimate the initial speed of the pedestrian, and obtain the image frames, and estimate the initial motion speed of the pedestrian according to the position change of the pedestrian in the image frame, that is: ; in, It is The initial motion speed of pedestrians in image frames, is the impact factor, yes Direction The image frame rate, yes Direction The image frame rate, The value range is [1, }; Configuring the Initial Update Threshold , according to the initial movement speed of the pedestrian, adjust the update interval, update the pedestrian's movement speed, and record it in the matching tracker. The formula for adjusting the update interval is: in, is the adjusted update interval, is the speed adjustment factor.

[0014] Specifically, the steps of the regional division strategy include: Obtain pedestrian trajectory information, including the location and speed of pedestrians at different times, configure the time window size, and calculate the average speed of pedestrians within the time window; Configure the speed threshold. If the average speed of the pedestrian in the time window is less than the speed threshold, the pedestrian's behavior mode is judged as staying behavior; otherwise, the pedestrian's behavior mode is judged as walking behavior. Discretize the monitoring scene and divide the monitoring area into multiple discrete areas. For each discrete area, count the number of pedestrians whose behavior pattern is the stop behavior. The number of pedestrians whose behavior pattern is stop behavior and the spatial distance between pedestrians in the discrete area are integrated to calculate the density score, that is: ; in, It is The dense score of discrete regions, It is The number of pedestrians whose behavior pattern is stop behavior in discrete areas, It is The comprehensive spatial distance between pedestrians in discrete areas, It is The area of ​​a discrete region, and is the weight coefficient; Configure a dense threshold. When the dense score of a discrete area is greater than the dense threshold, it is marked as a clustered area. Otherwise, no processing is performed. Configure the statistical interval to count the flow of people in the gathering area according to the statistical interval, and count the number of pedestrians entering, leaving, and staying in the gathering area.

[0015] Specifically, the data storage module includes a storage retrieval unit and a data backup unit; The storage and retrieval unit is equipped with an intelligent indexing strategy, which is used to classify and label the stored data, establish a multi-dimensional indexing system based on the data type, data source and data association information, and classify and index the data in combination with the data timestamp and the module to which the data belongs, to build a data index structure; The data backup unit is configured with a dynamic backup strategy. The dynamic backup strategy collects multi-dimensional data features by configuring evaluation intervals to calculate importance scores, divides data types into three importance levels, and adjusts the backup time intervals for different levels of data based on system data traffic, load, and business period.

[0016] Specifically, the steps of the dynamic backup strategy include: Configuring the Evaluation Interval , for each type of stored data, collect its multi-dimensional data feature information, including the usage frequency of each type of data within the evaluation interval, the degree of dependency of related businesses, and the stability impact of data changes, so as to calculate the importance score of each type of data, namely: in, It is Class Data The importance score of is the evaluation interval Middle Day data The number of times it is used, Is the business process to data The number of calls, is the number of calls to the business process for all data types, It’s data The stability effect of changes, , and is the weight coefficient; Configure the evaluation threshold, including the upper evaluation threshold and the lower evaluation threshold. According to the importance evaluation results, the stored data is divided into three importance levels. If the importance score of the data is greater than the upper evaluation threshold, the data is marked as high importance level. If the importance score is less than the lower evaluation threshold, the data is marked as low importance level. Otherwise, the data is marked as medium importance level. According to the importance level of the data type, and in combination with the system data flow, system load, and business period, the time interval for backing up different data types is dynamically adjusted, namely: in, The importance level is The time interval for data backup is The importance level is The adjustment constant when is the system data flow, is the system load, is the business period discriminant function. During the business peak period, , is a constant greater than 1. During non-peak business hours, .

[0017] Beneficial effects of the present invention: 1. Through the intelligent connection strategy, it can accurately identify and adapt video surveillance cameras to achieve stable connection and video stream data collection; the adaptive optimization strategy can dynamically adjust the acquisition frame rate according to the estimated data volume, and can also optimize and adjust parameters such as image clarity, contrast, and noise level based on multi-dimensional image quality evaluation to ensure the quality of the collected image data and provide high-quality basic data for subsequent target detection and other links, which improves the accuracy, flexibility and data quality of data collection as a whole. 2. The deep recognition strategy uses a diverse and finely labeled training data set to train the convolutional neural network model to accurately extract and classify pedestrian features; the continuous recognition strategy can effectively track pedestrian trajectories and update relevant information in real time; the area division strategy can accurately judge behavior patterns, divide gathering areas, and calculate real-time pedestrian flow based on pedestrian trajectories. This series of strategies enables the system to more accurately grasp pedestrian-related situations in target detection, providing strong support for analyzing various conditions in monitoring scenarios.

[0018] 3. The intelligent indexing strategy supports efficient retrieval in a large-scale data storage environment through multi-dimensional classification and labeling processing and the construction of a reasonable index structure; the dynamic backup strategy can dynamically adjust the backup time interval based on the data importance level and the actual operation of the system. At the same time, with the help of log records, it is convenient for operation and maintenance personnel to control the backup situation, effectively ensuring the retrieval efficiency of data storage, data security, and the rationality and flexibility of backup management. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A structural diagram of a data management system based on monitoring video streams according to the present invention; Figure 2 A flowchart of the steps of adjusting the acquisition frame rate of the present invention; Figure 3 A flowchart of the steps of image evaluation of the present invention; Figure 4 A flowchart of the steps of adjusting the image optimization parameters of the present invention; Figure 5 A flowchart of the specific steps of the continuous identification strategy of the present invention; Figure 6 The figure is a flow chart of the specific steps of the area division strategy of the present invention. DETAILED DESCRIPTION

[0020] See also Figure 1 ,This embodiment introduces a data management system based on surveillance video stream, including a data acquisition module, a target detection module and a data storage module; The data acquisition module is used to identify connected video surveillance cameras and collect video stream data. It adjusts the acquisition frame rate by estimating the data volume, builds an evaluation image dataset and performs multi-dimensional image quality evaluation. Based on the evaluation results, the image clarity, contrast, and noise level are adjusted and optimized to obtain a detection image dataset.

[0021] Preferably, the data acquisition module includes a video source adaptation unit and an image preprocessing unit. The video source adaptation unit is configured with an intelligent connection strategy, which is used to identify compatible devices by transmitting detection signals, extract and match protocol types, adapt optimal parameter combinations, configure devices and establish stable connections, so as to realize the integration of video surveillance cameras and video stream data acquisition; An adaptive optimization strategy is configured in the image preprocessing unit. The adaptive optimization strategy is used to adjust the acquisition frame rate by setting the acquisition time and estimating the data volume, and to perform image evaluation on the constructed evaluation image data set, mark abnormalities according to the threshold setting, and calculate the evaluation failure rate to adjust the initial image optimization parameters.

[0022] Preferably, the specific steps of the intelligent connection strategy include: Start the network scanning process, and transmit a detection signal at a fixed time interval within the preset network scanning frequency band. The detection signal contains the identification information of the discovery protocol. If the device receiving the detection signal is a video surveillance camera and supports the discovery protocol, a response message is sent to return its device information, which includes a unique identification name and MAC address. Otherwise, no operation is performed, and all response information is received and summarized into a device list. Extract the device identification field and protocol description field of the collected device response information to match the protocol type of the device; According to the identified device protocol type, construct parameter acquisition request information to obtain parameter information of the video surveillance camera, including resolution, frame rate, and encoding format; According to the parameter information of the video surveillance camera, the parameters are adapted to calculate the bandwidth requirements corresponding to the resolution supported by the video surveillance camera and the processing energy requirements corresponding to the frame rate of the video surveillance camera. According to the preset parameter threshold, a video surveillance camera parameter combination is selected. If there are multiple combinations that meet the conditions, the combination with the highest resolution and the highest frame rate is selected. According to the selected parameter combination, parameter configuration instructions are sent to the video surveillance camera, a connection is established, and a completeness and accuracy check is performed. If the connection check test passes, the video surveillance camera is marked as successfully connected and can collect data normally. If the connection check test fails, troubleshooting and repair operations are performed according to the error type, including rechecking whether the protocol parsing and parameter adaptation are correct, and checking whether the network connection is stable, until a connection is successfully established.

[0023] Preferably, the specific steps of the adaptive optimization strategy include: See also Figure 2 , set the initial duration of the video surveillance camera in the device list to collect video stream data, and estimate the amount of image data in the video stream data based on the parameter configuration of the video surveillance camera, that is: ; in, is the amount of image data in the video stream data estimated at the initial duration, and Respectively The number of pixels in the horizontal and vertical directions of a video surveillance camera, yes The frame rate of a video surveillance camera, is the number of video surveillance cameras in the device list, is the initial duration of the video surveillance camera collecting video stream data, It is the number of bytes per pixel, which depends on the color encoding method of the image; Configure the ratio threshold, obtain the current available memory resources, and dynamically adjust the acquisition frame rate by calculating the ratio of the image data volume to the available memory resources, that is: ; in, is the amount of image data per unit of available memory resources, is the currently available memory resource, if If the ratio is greater than the threshold, the frame rate of the video surveillance camera is adjusted, that is: ;in, is the ceiling function, is the adjusted The frame rate of a video surveillance camera, It is the adjustment coefficient, the value range is (0,1), which is set according to system performance and actual needs, otherwise no operation is performed; See also Figure 3 , configure the initial image optimization parameters, which include sharpening parameters, contrast stretching ratio coefficients, and filter kernel sizes, and collect video stream data according to the adjusted frame rate of the video surveillance camera; After the collected video stream data is processed by the initial image optimization parameters, a benchmark data set is constructed. The benchmark data set contains images that have not been processed by the initial image optimization parameters and images after processing. The images collected by each video surveillance camera are randomly selected from the benchmark data set. image pairs to construct the evaluation image dataset; See also Figure 3 , perform image evaluation on the images in the evaluation image dataset that have been processed by image optimization parameters, and calculate image evaluation indicators, including: To evaluate the clarity of multi-scale images, a multi-scale analysis mechanism is introduced. Wavelet transform is used to perform scale analysis on each image to obtain image representations at different scales, and the image gradient amplitude at each scale is calculated to obtain the comprehensive image gradient amplitude, namely: ; in, is the comprehensive image gradient amplitude, yes Direction, the scale is The directional gradient when yes Direction, the scale is The directional gradient at time, scale The value range is [1, ], The scale is The weight of is determined by experimental analysis; Perform local weighted contrast evaluation, evaluate the image contrast by comparing the standard deviation change of the pixel value sequence of each image with the pixel value sequence of its contrast-stretched image, divide each image into multiple non-overlapping sub-regions, and determine its weight according to the pixel distribution characteristics of each sub-region to obtain the pixel mean and pixel standard deviation of each image to calculate the image contrast ratio, that is: ; ; ; ; ; in, It is the contrast ratio, which quantifies the change in image contrast by comparing the standard deviation changes of the original image and the stretched image. It is used to carefully depict the contrast changes in different areas of the image, so as to more accurately evaluate the contrast adjustment effect and the overall contrast level of the image. is the standard deviation of the contrast-stretched image, is the standard deviation of the original image, , They are the contrast stretched image and the original image in the sub-regions The mean value of pixels within is the index of the sub-region in the horizontal and vertical directions, and its value range is [1, ] and [1, ]; , Respectively represent the number of sub-regions in the horizontal and vertical directions when the image is divided into non-overlapping sub-regions, through and Uniquely identify a sub-region in an image , Assign sub-area The weight of the sub-region Different values ​​are assigned to the internal pixel distribution characteristics. For example, the variance of the grayscale values ​​of pixels in the sub-region is calculated. The larger the variance, the more drastic the pixel changes in the sub-region, and the greater the impact on the overall contrast. The higher it is, and They are the contrast of the stretched image and the original image. The pixel value of each pixel is the basic element that constitutes the original image information and is used to subsequently calculate indicators such as the pixel mean and standard deviation of the original image in each sub-region.

[0024] Evaluate the noise level of the optimized image in the evaluation image data set, and calculate the noise level of the image by using the signal-to-noise ratio; Configure the evaluation thresholds, including the clarity threshold, contrast threshold, and signal-to-noise ratio threshold. If the comprehensive image gradient amplitude of the image is less than the clarity threshold, it will be marked as an abnormal clarity image, otherwise no operation will be performed; if the contrast ratio of the image is less than the contrast threshold, it will be marked as an abnormal contrast image, otherwise no operation will be performed; if the noise level of the image is greater than the signal-to-noise ratio threshold, it will be marked as an abnormal noise image, otherwise no operation will be performed; Configure the definition failure threshold and statistically evaluate the definition failure rate of images collected by video surveillance cameras in the image dataset, that is: in, is the clarity failure rate, It is The number of images marked as having abnormal clarity by a video surveillance camera; See also Figure 4 , if the clarity failure rate of the image collected by the video surveillance camera is greater than the clarity failure threshold, the clarity is adjusted, the current sharpening parameters are obtained, and the parameters are adjusted step by step, that is: in, is the sharpening parameter after the first adjustment. is the initial sharpening parameter, is the adjustment step size. According to the adjusted sharpening parameters, the unoptimized images in the evaluation image dataset are optimized, and their comprehensive image gradient amplitudes and the rate of change of the comprehensive image gradient amplitudes are calculated, namely: in, is the rate of change of the comprehensive image gradient amplitude, is the comprehensive image gradient amplitude of the image after the sharpening parameters are optimized after the first adjustment, is the comprehensive image gradient amplitude of the image after the initial sharpening parameters are optimized; Configure the amplitude change threshold. If the comprehensive image gradient amplitude change rate is greater than the amplitude change threshold, retain the sharpening parameters before adjustment and issue a parameter warning. Otherwise, use the adjusted sharpening parameters. After adjusting the sharpening parameters, re-evaluate the clarity failure rate and determine whether to continue adjusting the sharpening parameters based on the clarity failure rate; Configure the adjustment times threshold, count the number of sharpening parameter adjustments, and if the number of sharpening parameter adjustments is greater than the adjustment times threshold, stop the sharpening parameter optimization and output the adjusted sharpening parameters; Configure the contrast failure threshold, statistically evaluate the contrast failure rate of the images collected by the video surveillance camera in the image data set, and if the contrast failure rate of the images collected by the video surveillance camera is greater than the contrast failure threshold, perform contrast adjustment, obtain the current contrast stretching ratio coefficient, and perform step-by-step parameter adjustment; According to the adjusted contrast stretching ratio coefficient, the unoptimized images in the evaluation image data set are optimized, the image contrast ratio thereof is calculated, and the rate of change of the image contrast ratio is calculated; Configure the contrast change threshold. If the image contrast ratio change rate is greater than the contrast change threshold, the contrast stretching ratio coefficient before adjustment is retained and a parameter warning is issued. Otherwise, the adjusted contrast stretching ratio coefficient is used. After adjusting the contrast stretching ratio coefficient, evaluate the contrast failure rate of the images captured by the video surveillance camera in the image data set, and determine whether to continue adjusting the contrast stretching ratio coefficient according to the clarity failure rate; Count the number of sharpening parameter adjustments. If the number of sharpening parameter adjustments is greater than the adjustment number threshold, stop optimizing the contrast stretching ratio coefficient and output the adjusted contrast stretching ratio coefficient. Configure the noise failure threshold, statistically evaluate the noise failure rate of the images collected by the video surveillance camera in the image dataset, and if the noise failure rate of the images collected by the video surveillance camera is greater than the noise failure threshold, perform noise adjustment, obtain the current filter kernel size, and perform step-by-step parameter adjustment. According to the adjusted filter kernel size, the unoptimized images in the evaluation image dataset are optimized, the image signal-to-noise ratio is calculated, and the rate of change of the image signal-to-noise ratio is calculated; Configure the signal-to-noise ratio change threshold. If the image signal-to-noise ratio change rate is greater than the signal-to-noise ratio change threshold, retain the filter kernel size before adjustment and issue a parameter warning. Otherwise, use the adjusted filter kernel size. After adjusting the filter kernel size, the noise failure rate of the images collected by the video surveillance camera in the image dataset is evaluated, and whether to continue adjusting the filter kernel size is determined based on the noise failure rate; The number of times the filter kernel size is adjusted is counted. If the number of times the filter kernel size is adjusted is greater than the adjustment number threshold, the filter kernel size optimization is stopped and the adjusted filter kernel size is output.

[0025] The video stream data is collected according to the adjusted frame rate, and the collected video stream data is processed frame by frame according to the image optimization parameters, and stored in the detection image data set in sequence; The target detection module is used to extract features and classify pedestrians in the detection image data set to mark pedestrians and their categories, record pedestrian trajectory information and track and update it in real time, identify pedestrian behavior patterns based on trajectory information, divide gathering areas, and calculate real-time pedestrian flow in gathering areas.

[0026] Preferably, the target detection module includes a target recognition unit, a tracking and analysis unit, and a flow calculation unit; The target recognition unit is configured with a deep recognition strategy, which is used to extract and classify pedestrian features in the detection image data set according to the convolutional neural network model, and annotate pedestrians and pedestrian categories in the form of bounding boxes; The tracking and analysis unit is configured with a continuous identification strategy, which is used to establish a tracker for the pedestrian identified for the first time, record the pedestrian's trajectory information, including the initial position, movement speed, and movement position, and track and update the pedestrian's trajectory information; The flow calculation unit is configured with a regional division strategy, which is used to identify pedestrian behavior patterns based on pedestrian trajectory information, including stop behavior and walking behavior, perform aggregate behavior recognition based on the pedestrian behavior model, divide the aggregate area, and calculate the real-time pedestrian flow in the aggregate area; Preferably, the specific steps of the deep identification strategy include: Select convolutional neural network architectures, including VGG, ResNet, and YOLO, and determine initial parameters, including the initial values ​​of weight parameters for each layer, and training-related hyperparameters, including learning rate, batch size, and number of training rounds; Image data containing rich pedestrian scenes are collected as training data sets, covering different environments, including indoors, outdoors, during the day, and at night; different pedestrian postures, including standing, walking, running, and bending; different clothing styles, and different shooting angles, to ensure data diversity so that the model can learn comprehensive pedestrian characteristics.

[0027] The collected image data is annotated to accurately frame the location of each pedestrian in the image in the form of a bounding box, and the corresponding pedestrian category is annotated, including: adults, children, men, and women; The labeled training data set is divided into training set, validation set and test set. The training set is used to learn and update the model parameters. The validation set is used to evaluate the model performance during the training process and adjust the hyperparameters accordingly. The test set is used to finally evaluate the generalization ability of the model after the training is completed. The image data in the training set are input into the selected convolutional neural network model batch by batch for model training. The trained convolutional neural network model is applied to the detection image data set to extract pedestrian features and classify the image data. The position of each pedestrian is marked with a bounding box, and the pedestrian category is identified, providing an accurate individual data basis for subsequent multi-target tracking, behavior pattern analysis, etc.

[0028] See also Figure 5 , preferably, the specific steps of the continuous identification strategy include: In the starting frame or the image frame where a new pedestrian enters the monitoring area, a tracker is created for each pedestrian detected for the first time, and a unique identification number is set for each tracker; The initial position information and initial frame of the pedestrian are recorded in the matching tracker, and based on the feature point matching method, the position change of the pedestrian in the image frame is obtained to generate the motion trajectory; Configuring Interval Thresholds , used to estimate the initial speed of the pedestrian, and obtain the image frames, and estimate the initial motion speed of the pedestrian according to the position change of the pedestrian in the image frame, that is: ; ; ; in, It is The initial motion speed of pedestrians in image frames, is the impact factor, with a value range of (0,1), which is related to the distance from the current frame. The larger the distance, The smaller, yes Direction The image frame rate, yes Direction The image frame rate, The value range is [1, }, yes Direction The pedestrian position of each image frame, yes Direction The pedestrian position of each image frame, yes Direction The pedestrian position of each image frame, yes Direction The pedestrian position of each image frame, is the frame rate of the acquisition; Configuring the Initial Update Threshold , according to the initial movement speed of the pedestrian, adjust the update interval, update the pedestrian's movement speed, and record it in the matching tracker. The formula for adjusting the update interval is: in, is the adjusted update interval, is the speed adjustment coefficient. The value range is determined according to the influence of speed on the update frequency in the actual scene. For example, in the scene where the pedestrian movement speed changes rapidly, the value is relatively large, generally in the range of (0,10).

[0029] See also Figure 6 Preferably, the specific steps of the regional division strategy include: Obtain pedestrian trajectory information, including the location and speed of pedestrians at different times, configure the time window size, and calculate the average speed of pedestrians within the time window; Configure the speed threshold. If the average speed of the pedestrian in the time window is less than the speed threshold, the pedestrian's behavior mode is judged as staying behavior; otherwise, the pedestrian's behavior mode is judged as walking behavior. Discretize the monitoring scene and divide the monitoring area into multiple discrete areas. For each discrete area, count the number of pedestrians whose behavior pattern is the stop behavior. The number of pedestrians whose behavior pattern is stop behavior and the spatial distance between pedestrians in the discrete area are integrated to calculate the density score, that is: ; ; in, It is The dense score of discrete regions, It is The number of pedestrians whose behavior pattern is stop behavior in discrete areas, It is The comprehensive spatial distance between pedestrians in discrete areas, It is The area of ​​a discrete region, and is the weight coefficient, which is determined according to the relative importance of the number of people and distance to the degree of aggregation in the actual scenario. It is The pedestrian whose behavior pattern is stop behavior and the The distance of pedestrians whose behavior pattern is stop behavior, The value range is [ , ], The value range is [ , ]; Configure a dense threshold. When the dense score of a discrete area is greater than the dense threshold, it is marked as a clustered area. Otherwise, no processing is performed. Configure the statistical interval to count the flow of people in the gathering area according to the statistical interval, and count the number of pedestrians entering, leaving, and staying in the gathering area.

[0030] The data storage module is used to classify and label the stored data, establish an index structure according to multiple dimensions to achieve efficient retrieval, collect multi-dimensional features of the data to calculate importance scores, and divide the importance levels of data types to adjust the backup time intervals for data of different levels.

[0031] Preferably, the data storage module includes a storage retrieval unit and a data backup unit; The storage and retrieval unit is equipped with an intelligent indexing strategy, which is used to classify and label the stored data, establish a multi-dimensional indexing system based on the data type, data source and data association information, and classify and index the data in combination with the data timestamp and the module to which the data belongs, to build a data index structure; The data backup unit is configured with a dynamic backup strategy. The dynamic backup strategy collects multi-dimensional data features by configuring evaluation intervals to calculate importance scores, divides data types into three importance levels, and adjusts the backup time intervals for different levels of data based on system data traffic, load, and business period.

[0032] Preferably, the specific steps of the intelligent indexing strategy include: The data is divided into video stream data, image data, pedestrian trajectory data, model training data, model parameter data, behavior pattern recognition data, area division data and pedestrian flow statistics data; For each type of data, a label is assigned according to the data source, which includes the video camera number, system module, system unit, and data association information; the data association information includes the association between pedestrian trajectory and corresponding image frame, and the association between regional division data and related pedestrian behavior data. For example, for video stream data, the label includes information such as camera location, acquisition time range, and data format; for pedestrian trajectory data, the label includes pedestrian identification, start time, and involved area.

[0033] Based on the classification and label information, a multi-dimensional index is established. For example, for video stream data, the timestamp is used as the primary index dimension to quickly locate video data within a specific time range; at the same time, the camera number is used as an auxiliary index to facilitate the search of the video stream of a specific camera. For example, for image data, in addition to the timestamp and the camera number index, an index is also constructed based on the pedestrian features contained. Pedestrian trajectory data uses pedestrian identification, trajectory start time, trajectory area range, etc. as key index dimensions, and associates the corresponding image frame index to quickly obtain the image information corresponding to the trajectory.

[0034] Integrate and optimize indexes of different dimensions to build a hierarchical and distributed index structure to ensure efficient data retrieval in a large-scale data storage environment. For example, an index storage structure based on a combination of hash tables and B-trees is used. For data dimensions that are often used for precise queries, including timestamps, camera numbers, etc., hash tables are used for fast positioning. For data dimensions that are used for range queries or fuzzy queries, B-tree structures are used for range searches and similarity searches.

[0035] When new data is stored, the index update mechanism is automatically triggered. According to the classification and label information of the new data, its index information is inserted into the corresponding index structure, and the index structure is optimized and maintained regularly. For example, the index data is defragmented and invalid or expired index items are deleted to improve the query efficiency of the index. When the association relationship of the data changes, the association information in the index is updated in time to ensure the consistency and accuracy of the index and data.

[0036] Preferably, the specific steps of the dynamic backup strategy include: Configuring the Evaluation Interval , for each type of stored data, collect its multi-dimensional data feature information, including the usage frequency of each type of data within the evaluation interval, the degree of dependency of related businesses, and the stability impact of data changes, so as to calculate the importance score of each type of data, namely: in, It is Class Data The importance score of The value range is [1, ], is the number of all types of data stored, is the evaluation interval Middle Day data The number of times it is used, The value range is [1, ], Is the business process to data The number of calls, is the number of calls to the business process for all data types, It’s data The stability effect of the change is, for example, calculated by processing the delay change rate, , and is the weight coefficient, with a value range of (0, 1), which is determined according to system characteristics and business requirements. For example, if more emphasis is placed on the real-time nature and business relevance of data, then and The value is relatively large; if you pay more attention to the frequency of data use, then Larger; Configure the evaluation threshold, including the upper evaluation threshold and the lower evaluation threshold. According to the importance evaluation results, the stored data is divided into three importance levels. If the importance score of the data is greater than the upper evaluation threshold, the data is marked as high importance level. If the importance score is less than the lower evaluation threshold, the data is marked as low importance level. Otherwise, the data is marked as medium importance level. According to the importance level of the data type, and in combination with the system data flow, system load, and business period, the time interval for backing up different data types is dynamically adjusted, namely: in, The importance level is The time interval for data backup is The value range is [1,2,3]. When 1 is taken, it indicates a high importance level. When it is 2, it means the importance level is medium. When it is 3, it means low importance level. The importance level is The adjustment constant is greater than 0. The adjustment constant is set according to the system performance and the initial data volume. For example, for high-importance data, due to its extremely high requirements for data security and timeliness, Relatively small to ensure that the backup interval is as short as possible if system resources allow; for data with low importance level, Relatively large, appropriately extend the backup time interval to save backup resources and system overhead. is the system data flow, It is the system load, which is obtained by combining CPU usage and memory usage. is the business period discriminant function. During the business peak period, , is a constant greater than 1. During non-peak business hours, ; The backup process is logged, and the start time, end time, amount of data backed up, backup version number, and backup target location of the backup task are collected in real time, and displayed visually to help operation and maintenance personnel understand the execution of the backup task from a macro perspective, and analyze the details of each backup operation from a micro perspective to identify potential problems in a timely manner.

[0037] Working principle and its effect: A data management system based on surveillance video streams realizes effective management of surveillance video stream data by collecting and optimizing video stream data through a data acquisition module, analyzing and detecting pedestrian related information in image data through a target detection module, and classifying and storing data and backing up data according to importance through a data storage module.

[0038] In the data acquisition module, under the intelligent connection strategy, the detection signal is transmitted in the preset frequency band to collect response information, match the protocol, obtain parameters, and then adapt the combination to establish a connection. If there is a failure, it will be repaired to complete the camera integration and data acquisition. The adaptive optimization strategy first estimates the amount of image data, adjusts the frame rate based on memory resources, configures parameters to collect data to build a data set, and then evaluates the multi-dimensional indicators of the optimized image, adjusts parameters according to thresholds and failure rates, and finally stores the detection image data set according to the adjusted frame rate, so that it can intelligently and stably connect to a variety of video surveillance cameras, collect video stream data with frame rate adaptation and multi-dimensional quality optimization, improve the quality of collected data, provide a reliable image data foundation for subsequent processing, and reduce the impact of poor image quality on subsequent analysis.

[0039] In the target detection module, the deep recognition strategy first selects the convolutional neural network architecture, determines the hyperparameters, collects and divides the labeled diverse training data sets, and uses them to detect image data sets after training to extract and identify pedestrian features and categories. The continuous recognition strategy creates a tracker when a pedestrian is first detected, records information, generates trajectories based on feature point matching, and configures thresholds to adjust the update interval to update the trajectory in real time. The area division strategy uses pedestrian trajectory information to determine behavior patterns, discretizes the scene, calculates dense scores to mark clustered areas, and counts pedestrian flow. Accurately extract features, classify and identify pedestrians, track trajectories, analyze behavior patterns, and count pedestrian flow in clustered areas. Comprehensively and meticulously grasp the various dynamic situations of pedestrians in the monitoring scene, which can provide accurate data support for application scenarios such as security monitoring and crowd control, and assist in making reasonable decisions.

[0040] The data storage module uses intelligent indexing strategies to classify data, build multi-dimensional indexes and optimize them, and has an update and maintenance mechanism to ensure retrieval efficiency. The dynamic backup strategy configures interval collection features to calculate scores, adjusts the backup time interval based on system conditions after classification, and records backup information to facilitate operation and maintenance; on the one hand, it can quickly and accurately retrieve the required data in a large-scale data storage environment, improving data query efficiency; on the other hand, it can reasonably arrange backup work based on factors such as data importance, ensure data security and recoverability, and facilitate operation and maintenance personnel to effectively manage and maintain data assets and reduce the risk of data loss.

[0041] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A data management system based on surveillance video stream, characterized in that: It includes a data acquisition module, a target detection module and a data storage module; The data acquisition module is used to identify the connected video surveillance camera and collect video stream data, adjust the acquisition frame rate by estimating the data volume, construct an evaluation image data set and perform multi-dimensional image quality evaluation, and adjust and optimize the parameters of image clarity, contrast and noise level according to the evaluation results to obtain a detection image data set; The target detection module is used to extract features and classify pedestrians in the detection image data set to mark pedestrians and their categories, record pedestrian trajectory information and track and update it in real time, identify pedestrian behavior patterns based on trajectory information, divide gathering areas, and calculate real-time pedestrian flow in the gathering areas; The data storage module is used to classify and label the stored data, establish an index structure according to multiple dimensions to achieve efficient retrieval, collect multi-dimensional features of the data to calculate importance scores, and divide the importance levels of data types to adjust the backup time intervals of data of different levels.

2. A data management system based on surveillance video stream as claimed in claim 1, characterized in that: The data acquisition module includes a video source adaptation unit and an image preprocessing unit; The video source adaptation unit is configured with an intelligent connection strategy, which is used to identify compatible devices by transmitting detection signals, extract and match protocol types, adapt to optimal parameter combinations, configure devices and establish stable connections, and realize the integration of video surveillance cameras and video stream data collection; The image preprocessing unit is configured with an adaptive optimization strategy, which is used to adjust the acquisition frame rate by setting the acquisition time and estimating the data volume, and perform image evaluation on the constructed evaluation image data set, mark abnormalities according to threshold settings and calculate the evaluation failure rate, and adjust the initial image optimization parameters.

3. A data management system based on surveillance video stream as claimed in claim 2, characterized in that: The steps of the adaptive optimization strategy include: Set the initial duration of the video surveillance camera in the device list to collect video stream data, and estimate the amount of image data in the video stream data based on the parameter configuration of the video surveillance camera, that is: ； in, is the amount of image data in the video stream data estimated at the initial duration, and Respectively The number of pixels in the horizontal and vertical directions of a video surveillance camera, yes The frame rate of a video surveillance camera, is the number of video surveillance cameras in the device list, is the initial duration of the video surveillance camera collecting video stream data, is the number of bytes per pixel; Configure the ratio threshold, obtain the current available memory resources, and dynamically adjust the acquisition frame rate by calculating the ratio of the image data volume to the available memory resources, that is: ； in, is the amount of image data per unit of available memory resources, is the currently available memory resource; if If the ratio is greater than the threshold, the frame rate of the video surveillance camera is adjusted, that is: ;in, is the ceiling function, is the adjusted The frame rate of a video surveillance camera, is the adjustment factor; Configure the initial image optimization parameters, including sharpening parameters, contrast stretching ratio coefficient, and filter kernel size, and collect video stream data according to the adjusted frame rate of the video surveillance camera; The collected video stream data is processed by the initial image optimization parameters to form a benchmark data set. The video stream data collected by each video surveillance camera is randomly extracted from the benchmark data set. image pairs to construct the evaluation image dataset.

4. A data management system based on surveillance video stream as claimed in claim 3, characterized in that: The steps of the adaptive optimization strategy also include: Perform image evaluation on the images in the evaluation image dataset that have been processed by image optimization parameters, and calculate image evaluation indicators, including: To evaluate the clarity of multi-scale images, wavelet transform is used to perform scale analysis on each image, obtain image representations at different scales, and calculate the image gradient amplitude at each scale to obtain the comprehensive image gradient amplitude, namely: ； in, is the comprehensive image gradient amplitude, yes Direction, the scale is The directional gradient when yes Direction, the scale is The directional gradient at time, scale The value range is [1, ], The scale is The weight of Perform local weighted contrast evaluation, evaluate the image contrast by comparing the standard deviation change of the pixel value sequence of each image with the pixel value sequence of its contrast-stretched image, divide each image into multiple non-overlapping sub-regions, and determine its weight according to the pixel distribution characteristics of each sub-region to obtain the pixel mean and pixel standard deviation of each image to calculate the image contrast ratio, that is: ； in, is the contrast ratio, is the standard deviation of the contrast-stretched image, is the standard deviation of the original image; The noise level of the optimized images in the evaluation image dataset is evaluated, and the noise level of the image is calculated by the signal-to-noise ratio.

5. A data management system based on surveillance video stream as claimed in claim 4, characterized in that: The steps of the adaptive optimization strategy also include: Configure evaluation thresholds, including clarity threshold, contrast threshold, and signal-to-noise ratio threshold; If the integrated image gradient amplitude of the image is less than the clarity threshold, it is marked as an abnormal clarity image, otherwise no operation is performed; if the contrast ratio of the image is less than the contrast threshold, it is marked as an abnormal contrast image, otherwise no operation is performed; if the noise level of the image is greater than the signal-to-noise ratio threshold, it is marked as an abnormal noise image, otherwise no operation is performed; Configure evaluation failure thresholds, including clarity failure threshold, contrast failure threshold, and noise failure threshold, and calculate evaluation failure rates of images captured by video surveillance cameras in the evaluation image dataset, including clarity failure rate, contrast failure rate, and noise failure rate; If the evaluation failure rate of the images collected by the video surveillance camera is greater than the evaluation failure threshold, the image optimization parameters are gradually adjusted; According to the adjusted image optimization parameters, the original images in the evaluation image data set are optimized, the image evaluation is re-performed, and the change rate of the image evaluation index is calculated; Configure the evaluation change thresholds, including the clarity change threshold, contrast change threshold, and signal-to-noise ratio change threshold. If the image evaluation index change rate is greater than the evaluation change threshold, retain the image optimization parameters before adjustment and issue a parameter warning. Otherwise, use the adjusted image optimization parameters. After adjusting the image optimization parameters, re-evaluate the image evaluation failure rate and determine whether to continue adjusting the image optimization parameters; Configure the adjustment times threshold to count the times of image optimization parameter adjustment. If the image optimization parameter adjustment times are greater than the adjustment times threshold, stop the image optimization parameter optimization and output the adjusted image optimization parameters.

6. A data management system based on surveillance video stream as claimed in claim 1, characterized in that: The target detection module includes a target recognition unit, a tracking and analysis unit, and a flow calculation unit; The target recognition unit is configured with a deep recognition strategy, which is used to extract and classify pedestrian features in the detection image data set according to a convolutional neural network model, and mark pedestrians and pedestrian categories in the form of bounding boxes; The tracking and analysis unit is configured with a continuous identification strategy, which is used to establish a tracker for the pedestrian identified for the first time, record the pedestrian's trajectory information, including initial position, movement speed, and movement position, and perform tracking updates through multi-modal information fusion; The flow calculation unit is configured with a regional division strategy, which is used to identify pedestrian behavior patterns based on pedestrian trajectory information, including stop behavior and walking behavior, perform aggregate behavior recognition based on pedestrian behavior models, divide aggregation areas, and calculate real-time pedestrian flow in the aggregation areas.

7. A data management system based on surveillance video stream as claimed in claim 6, characterized in that: The steps of the continuous identification strategy include: In the starting frame or the image frame where a new pedestrian enters the monitoring area, a tracker is created for each pedestrian detected for the first time, and a unique identification number is set for each tracker; The initial position information and initial frame of the pedestrian are recorded in the matching tracker, and based on the feature point matching method, the position change of the pedestrian in the image frame is obtained to generate the motion trajectory; Configuring Interval Thresholds , used to estimate the initial speed of the pedestrian, and obtain the image frames, and estimate the initial motion speed of the pedestrian according to the position change of the pedestrian in the image frame, that is: ； in, It is The initial motion speed of pedestrians in image frames, is the impact factor, yes Direction The image frame rate, yes Direction The image frame rate, The value range is [1, }; Configuring the Initial Update Threshold , according to the initial movement speed of the pedestrian, adjust the update interval, update the pedestrian's movement speed, and record it in the matching tracker. The formula for adjusting the update interval is: ; in, is the adjusted update interval, is the speed adjustment factor.

8. A data management system based on surveillance video stream as claimed in claim 6, characterized in that: The steps of the area division strategy include: Obtain pedestrian trajectory information, including the location and speed of pedestrians at different times, configure the time window size, and calculate the average speed of pedestrians within the time window; Configure the speed threshold. If the average speed of the pedestrian in the time window is less than the speed threshold, the pedestrian's behavior mode is judged as staying behavior; otherwise, the pedestrian's behavior mode is judged as walking behavior. Discretize the monitoring scene and divide the monitoring area into multiple discrete areas. For each discrete area, count the number of pedestrians whose behavior pattern is the stop behavior. The number of pedestrians whose behavior pattern is stop behavior and the spatial distance between pedestrians in the discrete area are integrated to calculate the density score, that is: ； in, It is The dense score of discrete regions, It is The number of pedestrians whose behavior pattern is stop behavior in discrete areas, It is The comprehensive spatial distance between pedestrians in discrete areas, It is The area of ​​a discrete region, and is the weight coefficient; Configure a dense threshold. When the dense score of a discrete area is greater than the dense threshold, it is marked as a clustered area. Otherwise, no processing is performed. Configure the statistical interval to count the flow of people in the gathering area according to the statistical interval, and count the number of pedestrians entering, leaving, and staying in the gathering area.

9. A data management system based on surveillance video stream as claimed in claim 1, characterized in that: The data storage module includes a storage retrieval unit and a data backup unit; The storage and retrieval unit is configured with an intelligent indexing strategy, which is used to classify and label the stored data, establish a multi-dimensional indexing system according to the data type, data source and data association information, and classify and label and index the data in combination with the data timestamp and the module to which the data belongs, so as to construct a data index structure; The data backup unit is configured with a dynamic backup strategy, which collects multi-dimensional data features by configuring evaluation intervals to calculate importance scores, divides data types into three importance levels, and adjusts the backup time intervals of different levels of data in combination with system data traffic, load, and business period.

10. A data management system based on surveillance video stream as claimed in claim 9, characterized in that: The steps of the dynamic backup strategy include: Configuring the Evaluation Interval , for each type of stored data, collect its multi-dimensional data feature information, including the usage frequency of each type of data within the evaluation interval, the degree of dependency of related businesses, and the stability impact of data changes, so as to calculate the importance score of each type of data, namely: ; in, It is Class Data The importance score of is the evaluation interval Middle Day data The number of times it is used, Is the business process to data The number of calls, is the number of calls to the business process for all data types, It’s data The stability effect of changes, , and is the weight coefficient; Configure the evaluation threshold, including the upper evaluation threshold and the lower evaluation threshold. According to the importance evaluation results, the stored data is divided into three importance levels. If the importance score of the data is greater than the upper evaluation threshold, the data is marked as high importance level. If the importance score is less than the lower evaluation threshold, the data is marked as low importance level. Otherwise, the data is marked as medium importance level. According to the importance level of the data type, and in combination with the system data flow, system load, and business period, the time interval for backing up different data types is dynamically adjusted, namely: ; in, The importance level is The time interval for data backup is The importance level is The adjustment constant when is the system data flow, is the system load, is the business period discriminant function. During the business peak period, , is a constant greater than 1. During non-business peak hours, .

Images