Design And Implementation Of Abnormal Behavior Analysis System Based On Video Object And Deep Learning

In recent years,deep learning has developed rapidly and made breakthroughs in the fields of machine vision,natural language processing and speech recognition.At the same time,with the development of society,security equipment began to be applied to urban security and the rapid increase of urban population,which brought great challenges to urban security work.At present,there are some related researches on abnormal behavior analysis based on security surveillance video,but there is no more general method for dealing with actual surveillance video.Therefore,research on abnormal behavior analysis system based on video object has important practical significance for urban security.The rapid development of deep learning technology also makes it possible to implement such a system.Therefore,based on deep learning techniques,this paper will be developed in the following directions:Firstly,the depth detection-based target detection algorithm Faster-RCNN is studied,which is combined with the needs of real-world scenarios to modify and optimize.Based on the model pre-trained by ImageNet data,the migration learning training is based on the data from the real scene and the public dataset data,and the real-world scene data is used for the detection and optimization and correction of the network structure parameters to construct a realistic scene.In addition,the target tracking algorithm KCF algorithm is improved,and a multi-scale target frame is introduced to construct a more effective tracker.Anomalous behavior analysis is then implemented based on the improved target detection and target tracking algorithms described above.The target data detection algorithm is used to perform target detection on the video data,and the KCF tracker is initialized by using the detected target frame,and the target detection frequency is reduced by the tracker to improve the overall running speed.After the tracker obtains the target area and motion information,the abnormal behavior analysis model is designed according to the real scene requirements,and the abnormal behavior modeling of the number of people exceeding the limit,crowd gathering,rapid motion and regional invasion is completed.Finally,an abnormal behavior analysis system is constructed,which completes the automatic analysis of video data and realizes the analysis and early warning of abnormal behavior of monitoring video data.By monitoring the access and management of video data,the system can display surveillance video on the one hand,and automatically analyze and predict the abnormal behavior of surveillance video by combining target tracking algorithm and abnormal behavior analysis.The system can effectively utilize monitoring data,realize automatic analysis and early warning of monitoring data,improve the efficiency of security work and effectively prevent the occurrence of security accidents.