Intelligent Violence Detection Based On Deep Learning Method

With the concept of the "safe city" appearing before the public,security construction has been paid attention to cities,and video has been continuously developed and applied.However,with the rapid increase of the number of cameras and more and more diversified functions of the surveillance system,intelligent video surveillance merges.The intelligent video surveillance mainly faces the following key chanlleges:(1)how to solve the problem of the efficient storage and transmission of video signal;(2)how to make the Internet share the video streams;(3)how to ensure the accuracy and efficiency of intelligent video analysis algorithms.The first two questions involve hardware and coding technology,so this paper focuses on the intelligent video analysis methods.The application of the intelligent video analysis methods can transform the post-mortem analysis into a matter analysis,and alarm the abnormal behavior in time.On the other hand,it is possible for the system to quickly retrieve and locate the target fragments in a large number of video data.In recent years,deep learning has achieved very good results in many fields,such as computer vision,natural language processing and speech recognition.It also provides new solutions for intelligent video analysis technology.Therefore,for more complex violence behavior in abnormal behavior,the following research is carried out based on the method of deep learning.Firstly,to solve the complexity of quick and accurate anomaly detection in video,especially violence detection,we proposed a novel method for detecting human violent behaviour in videos by integrating trajectory and deep convolutional neural networks,which takes advantage of hand-crafted features and deep-learned features.This method improves the accuracy of the detection of violent behavior.The processing speed of the whole detection process is 21 frames per second,which basically solves the problem of low detection efficiency,and the processing speed can be improved by improving the performance of GPU equipment.Secondly,for the problem of abnormal behavior in untrimmed long videos which is difficult to be accurately retrieved,this paper proposes a two-stage method of timing localization of violent behavior based on DEC3 D network model.Comprehensive consideration of the low detection rate and inaccurate positioning problem in long videos,we established the first stage on the basis of C3 D candidate videos generation model,and the second stage of DEC3 D network positioning model.This method realizes the time accurately for violence to frame level positioning,and improves the retrieval efficiency of the target behavior in untrimmed long videos.