Design And Implementation Of Intelligent Video Surveillance System Based On Abnormal Behavior Detection

With the increasing demand for video surveillance,the traditional manual video by video analysis method is no longer applicable.At the same time,people often don’t pay much attention to the daily normal events,but pay special attention to the occasional abnormal events,and people’s demand for intelligent surveillance is rising.Abnormal event detection is the main purpose of monitoring system,which has attracted more and more attention of relevant research.Based on the requirements of abnormal events and actual functions in surveillance video,this paper designs and implements a video surveillance system based on abnormal behavior detection.For complex structured video data,the algorithm model is studied from three aspects:behavior,crowd density and fireworks detection,and a complete processing framework and platform system integrating abnormal configuration,alarm query and so on is realized.Starting from the core functions,I have completed the following work:1.According to the research status and background investigation,it is determined that the subject takes the video stream protocol as the detection object and supports the detection algorithm of exception categories in three public scenes as the core to design and implement the research direction and related research content of video surveillance system.2.After learning the key technologies and theories such as convolutional neural network,pytorch,front-end framework Vue,spring MVC framework and database,the overall architecture design and module division of the system are carried out around the functional and performance requirements,the detection algorithm model is selected and designed,and the processing flow of each functional module and the specific design of front-end interface are completed.3.Research,analyze and implement the respective detection algorithm models of the three anomaly types,improve the decision processing of slowfast algorithm,improve the efficiency of action behavior detection,design the training iteration scheme of crowd and fireworks detection model based on yolov5,complete the production of specific scene type coco format data set,and finally train and simulate specific targets in the experimental environment,The detection effect and related performance of their models are tested and analyzed.4.Complete the detailed implementation of the front and rear functions of each module of the abnormal behavior detection video monitoring system.The video acquisition part generates RTMP video stream protocol in the form of nginx+ffmpeg server streaming;Around the parameter analysis method,the demand functions such as behavior customization,confidence and detection frequency are extracted and set.Through the separation of model working threads,single and multi frame separation and size processing are carried out for the unified video stream to realize the matching of detection model input.The information content of detection results is designed and labeled to realize the functions of alarm and storage.The user’s interactive query of abnormal information is realized through the back-end transfer request module;The realization of system data storage and the design of database table are completed,and RBAC scheme is used to realize user management.I finally realized the video monitoring system for abnormal behavior detection,and tested the function and performance of the system as a whole.The test results verify that the system as a whole meets the target requirements,and each module can work in coordination with each other.