Design And Implementation Of A Multi-channel Real-time Monitoring Video Data Processing And Analysis System

Nowadays,surveillance equipment is widely used in the fields of smart security,safe city,and criminal investigation,a large number of surveillance points are constantly working,the quality of surveillance video is constantly improving,which brings about an exponential increase in surveillance video data,and it has the characteristics of unstructured,borderless and spatio-temporal continuity.Traditional video surveillance systems have a lot of research on video stream transmission,image processing and offline processing,but less research on distributed processing and real-time analysis,the technical problems of designing and implementing a multi-channel real-time surveillance video data processing and analysis system needs to be solved urgently.Based on the traditional technology of surveillance video data processing and analysis technology and mainstream stream computing frameworks,this thesis designs and implements a video data processing and analysis system with high performance,low latency and high scalability.The main innovations and work are as follows:1)This thesis designs and implements a multi-channel real-time surveillance video data processing and analysis system,using a self-designed lightweight stream computing framework to make up for the shortcomings of the current mainstream streaming computing framework,such as the bloated structure,inefficient calculation of processing unstructured video data,and the scheduling algorithm does not consider the change of cluster resources.The master node is responsible for video processing task scheduling and system resource management,and the slave node is responsible for performing specific video processing and analysis tasks.The master-slave replication strategy and the heartbeat mechanism are adopted to ensure fault tolerance,the self-designed DRSA scheduling algorithm is used to ensure system resource utilization,the video frame selection strategy based on the interval of Predictive-coded Picture is adopted to select key frames.2)In terms of video data processing,a heterogeneous collaborative work mechanism based on the combination of CPU soft decoding technology and GPU hard decoding technology is used to process massive real-time surveillance video data,and the computing resources of CPU and GPU are used to process real-time data in parallel,which improves system throughput rate.3)In terms of video data analysis,based on the spatio-temporal continuity of surveillance video data,a face tracking deduplication algorithm is designed to reduce a large amount of redundant and repetitive face information,save computing resources and ensure the true validity of the statistics of non-cooperative target personnel under complex conditions.4)This thesis conducts detailed functional and performance testing on the system,and analyzes the test results in detail.The test results show that in the scene of multichannel real-time surveillance video data,the system can meet the needs of real-time streaming data processing and analysis.