Research On Anomaly Detection Technology In Video Surveillance Based On Deep Learning

With the continuous development of artificial intelligence technology,the application scope of video surveillance is constantly expanding.People can not only view the monitoring picture in video surveillance,but also find abnormal phenomena.Anomaly detection technology is a key technology in intelligent video surveillance,which is conducive to the construction and development of smart cities and safe cities,and also conducive to maintaining social public safety and stability,and protecting citizens’ life and property safety.For anomaly detection,a very key link is to design an effective,stable and intelligent algorithm to deal with the abnormal behavior in the video.In practical application,it can effectively realize a variety of real situations of high precision detection.This topic is a research on anomaly detection technology in video surveillance based on deep learning,focusing on feature extraction,human posture estimation and abnormal behavior detection in monitoring scenarios.The main research contents and relevant achievements of this paper are as follows:(1)In view of the defects of traditional methods in feature extraction,a feature extraction method based on improved attention mechanism and residual neural network is studied.This method uses attention mechanism to weight the features of different channels and suppress useless information,thus greatly improving the ability of feature recognition.(2)A human posture estimation model is designed based on the fusion of the improved attention mechanism SE-Res Net feature extraction network and convolutional neural network CNN.This model can effectively improve the generalization performance by combining residual neural network and convolutional neural network.The evaluation indexes PCKh and m AP of human pose estimation on MPII data set have achieved excellent results.Quantitative analysis,comparative analysis and visualization analysis of experimental results have verified that the human pose estimation model proposed in this paper can realize single and multi-person pose estimation in complex situations.(3)An anomaly detection algorithm based on human posture estimation in video surveillance is proposed.Based on the human pose estimation model proposed in this paper and the multi-classification method based on directed acyclic graph DAG-SVM,the algorithm realizes abnormal behavior detection in monitoring scenarios.In order to verify this algorithm,experimental analysis is carried out on the data set.Compared with other algorithms,the detection accuracy and Kappa coefficient have achieved strong competitiveness,and the expected effect has been achieved.It is proved that the anomaly detection algorithm in this paper can adapt to different visual angles and different scales of video data.(4)An anomaly detection system in video surveillance based on the human posture estimation network model and anomaly detection algorithm proposed in this paper is developed,which can accurately detect human behaviors in video surveillance,further proving the effectiveness and feasibility of the model and algorithm in this paper.