Search

CN-122027770-A - Focusing type monitoring storage method and system based on video stream real-time analysis

CN122027770ACN 122027770 ACN122027770 ACN 122027770ACN-122027770-A

Abstract

The invention relates to the technical field of intelligent video monitoring, in particular to a focusing type monitoring storage method and system based on video stream real-time analysis, comprising the following steps: and (3) real-time analysis and identification of the ROI, namely, at the monitoring equipment end, carrying out real-time analysis on the collected original video stream, and identifying the ROI of the region of interest in the picture through a lightweight double-stage analysis engine. According to the invention, the lightweight dual-stage analysis engine is arranged to accurately identify the high-value monitoring target, so that interference is effectively filtered, the acquisition, analysis and control of closed loop linkage are combined, the on-demand focusing and differential coding are realized, the clear and available key targets are ensured, the resource allocation is optimized, meanwhile, the layered packaging and differential storage strategy is adopted, the occupation of storage space is obviously reduced, the storage period is prolonged, the cost is reduced, and in addition, the rapid semantic retrieval of video content is realized through the structured metadata tag, so that the post verification and analysis efficiency is greatly improved.

Inventors

  • HE ZHIWEN

Assignees

  • 何志文

Dates

Publication Date
20260512
Application Date
20260202

Claims (9)

  1. 1. The focusing type monitoring and storing method based on the real-time analysis of the video stream is characterized by comprising the following steps: s1, real-time analysis and identification of ROI, namely, at a monitoring equipment end, real-time analysis is carried out on an acquired original video stream, and a region of interest ROI in a picture is identified through a lightweight double-stage analysis engine; S2, dynamic focusing and differential coding, namely dynamically adjusting acquisition parameters and coding parameters of monitoring equipment according to an analysis result of the ROI, carrying out optical/electronic focusing on the ROI area, distributing high code rate and high resolution, and distributing low code rate and low resolution on a background area; S3, feature extraction and label generation, namely extracting structural feature information of the ROI area to generate a metadata label containing target types, colors and behavior features; and S4, carrying out association packaging storage, namely carrying out association storage on the video stream subjected to focusing and encoding processing and the metadata tag.
  2. 2. The method for storing and monitoring focused on the basis of real-time analysis of video streams according to claim 1, wherein the specific operation steps of real-time analysis and identification of ROI in step S1 are as follows: S11, moving target preliminary screening, namely calculating pixel motion vectors between adjacent video frames by adopting an optical flow method, wherein a basic constraint equation is expressed as follows: Wherein, the For the purpose of the gradient operator, For the intensity of the pixel(s), As a component of the movement velocity, Representing pixel intensity For time of day Is a partial derivative of (2); Meanwhile, combining a lightweight deep learning model YOLO-Nano to perform target classification and preliminary positioning on the motion salient region, and outputting a target boundary frame , wherein, And Respectively representing object bounding boxes Upper left corner Coordinates and method for producing the same The coordinates of the two points of the coordinate system, And Respectively representing object bounding boxes Lower right corner Coordinates and method for producing the same Coordinates; s12, high-value ROI determination based on the target bounding box output in the step S11 Judging the stability of the motion trail, and judging the semantic value after the stability of the motion trail meets the requirement, specifically: outputting the confidence level of the target category through the YOLO-Nano model or a lightweight classifier linked with the YOLO-Nano model When confidence is When >0.85, the region is determined as a region of interest ROI of high value.
  3. 3. The method for storing and monitoring the focus based on the real-time analysis of the video stream according to claim 2, wherein the specific operation steps for determining the stability of the motion trail in step S12 are as follows: Calculating the target is in succession Motion stability in a frame, the formula is: Wherein, the Representing a score of the stability of the motion profile, Representing the number of consecutive video frames used to calculate the stability score, Is the cross-over ratio of the two adjacent layers, , Representing the object as at the first The location of the bounding box in the frame, Representing the object as at the first A bounding box position in the frame; When (when) And when the target area is more than 0.7 and the proportion of the target area to the picture is more than 3 percent, judging that the stability of the motion track meets the requirement, and entering the next step.
  4. 4. The method for monitoring and storing video stream based on real-time analysis according to claim 2, wherein the specific operation steps of dynamic focusing and differential encoding in step S2 are as follows: S21, optical focusing control, namely calculating the focal length adjustment amount of the lens according to the center coordinates of the high-value region of interest (ROI) determined in the step S12 The control module drives the automatic focusing motor to make the focal length of the lens from a default value Adjust to target value The relationship is expressed as: Wherein, the The offset of the center position of the ROI relative to the center of the image and the object distance estimation model are determined together so as to ensure that the imaging of the target in the ROI is clear; S22, carrying out coding parameter differentiation configuration, namely carrying out dynamic configuration on parameters of a video encoder according to boundary frame coordinates of the ROI, and specifically: allocating a coding code rate not lower than 4Mbps to the ROI area, and coding by adopting equivalent resolution not lower than 3840 multiplied by 2160 pixels; the background area is assigned a coding rate of not higher than 512 Kbps and is coded with a resolution of not higher than 848×480 pixels.
  5. 5. The method for focused monitoring and storing based on real-time analysis of video streams according to claim 2, wherein the specific operation steps of feature extraction and tag generation in step S3 are as follows: S31, performing feature extraction operation on the high-value region of interest (ROI) identified in the step S12, and extracting color features and behavior features; S32, semantic mapping and label generation, wherein the features extracted in the step S31 are mapped to a predefined semantic library to generate a standardized structured metadata label, and the label at least comprises a target type, a color, a behavior, a confidence level and a time stamp field; and S33, dynamically updating, namely updating the metadata tag of the ROI at a frequency not lower than 1 time per second, triggering re-analysis and tag updating when the obvious change of the characteristics of the ROI is detected, and storing or transmitting only the tag field with the change by adopting an incremental updating mechanism.
  6. 6. The method for focused monitoring and storing based on real-time analysis of video streams according to claim 5, wherein the specific operation steps of feature extraction in step S31 are as follows: s311, identifying the target type, namely acquiring probability distribution vectors of targets belonging to predefined categories through a lightweight convolutional neural network classifier , wherein, Representing the total number of predefined target classes, taking The corresponding category is taken as the target type, Representing the index corresponding to the maximum probability value; S312, extracting color features, and calculating average value vectors of ROI areas in HSV color space , wherein, 、 And Representing the average value of hue, saturation and brightness respectively and using preset color cluster dictionary Mapping to a dominant color description; S313, extracting behavior characteristics, namely calculating a speed vector according to the coordinate change of the central point of the ROI in the continuous frames , wherein, Representing the velocity component in the horizontal direction, Represents a velocity component in the vertical direction and determines the behavior state based on the magnitude and direction of the velocity.
  7. 7. The method for storing and monitoring focused on the basis of real-time analysis of video streams according to claim 1, wherein the specific operation steps of associating and packaging storage in step S4 are as follows: s41, time stamp alignment and frame level binding, namely extracting time stamps from the differentially encoded video stream according to frames At the same time, the metadata tag generated in step S3 is time stamped Alignment is performed when meeting Binding the frame of video data with the corresponding metadata tag, wherein A preset time synchronization tolerance threshold value; S42, hierarchical encapsulation and index construction, namely dividing a video stream into an ROI data layer and a background data layer by adopting a custom encapsulation format, respectively encapsulating the ROI data layer and the background data layer, and generating an index table for each segment of video data, wherein the index table at least comprises the following fields of a start time stamp, an end time stamp, an ROI region coordinate, a target type, a behavior label and a storage offset; S43, executing a differential storage strategy, namely adopting a high-reliability storage strategy for the ROI data layer, comprising write-in verification and redundancy backup, and adopting a high-compression storage strategy for the background data layer, and allowing lossy compression or periodic coverage when the storage pressure is high; s44, metadata plug-in and quick retrieval support, namely storing an index table and a metadata tag together with a video stream in a lightweight plug-in file form, and supporting quick retrieval and target positioning by directly using metadata on the premise of not decoding the video.
  8. 8. A focused monitoring storage system based on real-time analysis of video streams, the system comprising: The acquisition module is used for acquiring an original video stream; The analysis module is used for carrying a lightweight double-stage analysis engine and identifying the ROI in real time; The control module is used for adjusting the focal length of the lens and the coding parameters according to the analysis result; the metadata generation module is used for extracting the ROI characteristics and generating a structured label; And the storage module is used for storing the processed video stream and the metadata tag in an associated mode.
  9. 9. The focused monitoring storage system based on real-time analysis of video streams according to claim 8, wherein the output end of the acquisition module is connected with the input end of the analysis module, and is used for transmitting the acquired original video streams to the analysis module in real time; The output end of the analysis module is respectively connected with the input end of the control module and the input end of the metadata generation module, and is used for transmitting the identified ROI information, the boundary frame coordinates thereof and the target class confidence coefficient to the control module and transmitting the ROI region image data and the space-time information thereof to the metadata generation module; the output end of the control module is connected with the control interface of the acquisition module and is used for sending an optical focusing adjustment instruction and a coding parameter configuration instruction to the acquisition module; The output end of the metadata generation module is connected with the input end of the storage module and is used for outputting the structured metadata tag to the storage module; The acquisition module is also connected with the storage module through an encoding output interface and is used for transmitting the video stream subjected to the differential encoding treatment to the storage module; The storage module is used for receiving and storing the video stream from the acquisition module and the metadata tag from the metadata generation module in a correlated mode, and establishing an index relation between the video stream and the metadata tag to support quick retrieval.

Description

Focusing type monitoring storage method and system based on video stream real-time analysis Technical Field The invention relates to the technical field of intelligent video monitoring, in particular to a focusing type monitoring storage method and system based on video stream real-time analysis. Background With the rapid popularization of video monitoring technology, the storage and efficient retrieval of massive video data become key bottlenecks for restricting the development of industry. Conventional monitoring systems typically employ a fixed code rate or a simple dynamic code rate for full-picture recording, resulting in a large amount of storage space being occupied by a static background of low information density, while truly valuable objects (e.g., pedestrians, vehicles) are difficult to identify due to insufficient resolution or compression distortion. Although the prior art coding schemes such as h.265, AV1, and the like and the generated compression technique (GVC) reduce video volume to some extent, the fundamental problem of "low storage content information density, unusable key targets" is not solved. The invention patent of China with the authorized bulletin number of CN119996723B discloses a video monitoring storage system and a method based on distributed cloud storage, which are characterized in that a monitoring video stream is acquired through a camera, and is subjected to time blocking based on a video searching and inquiring state of a user to form a plurality of video data blocks, so that a video distributed cloud storage architecture is constructed, the storage position of each video data block in the distributed cloud storage is determined by utilizing a consistent hash algorithm, and the use frequency of each video data block is monitored in real time, so that the storage priority of each video data block is dynamically adjusted. Although the patent technology provides a video data blocking and dynamic priority adjusting method based on distributed cloud storage, the method has the focus of management and retrieval efficiency optimization of back-end storage resources, and content analysis and enhancement in a video acquisition stage are not involved. The scheme still depends on the original video stream, can not promote the visual usability of a long-distance small target, and can not endow the video content with a retrievable semantic tag. In addition, although AI is introduced into the existing part of intelligent monitoring system to detect targets, all moving objects are generally regarded as regions of interest (ROI), so that the misjudgment rate is high (such as leaf shaking and light and shadow change), and closed-loop linkage is not realized with the focusing control and coding strategies of the acquisition equipment, so that 'enhancement on demand and storage on demand' cannot be realized. Therefore, a technical scheme capable of performing intelligent analysis, dynamic focusing, local enhancement and fusion of structural semantic information at a video acquisition source is urgently needed, so that storage redundancy is greatly reduced and retrieval efficiency is improved while clear and usable key information is ensured. Disclosure of Invention The invention aims to provide a focusing type monitoring storage method and system based on video stream real-time analysis, which are used for solving the problems of high storage redundancy, unobtrusive key information and low post retrieval efficiency in the prior monitoring technology in the background technology. In order to achieve the above object, the present invention provides a focusing type monitoring storage method based on real-time analysis of video stream, comprising the following steps: s1, real-time analysis and identification of ROI, namely, at a monitoring equipment end, real-time analysis is carried out on an acquired original video stream, and a region of interest ROI in a picture is identified through a lightweight double-stage analysis engine; S2, dynamic focusing and differential coding, namely dynamically adjusting acquisition parameters and coding parameters of monitoring equipment according to an analysis result of the ROI, carrying out optical/electronic focusing on the ROI area, distributing high code rate and high resolution, and distributing low code rate and low resolution on a background area; S3, feature extraction and label generation, namely extracting structural feature information of the ROI area to generate a metadata label containing target types, colors and behavior features; and S4, carrying out association packaging storage, namely carrying out association storage on the video stream subjected to focusing and encoding processing and the metadata tag. As a further improvement of the present technical solution, the specific operation steps of real-time analysis and identification of ROI in step S1 are: S11, moving target preliminary screening, namely calculating pixel motion vectors betw