Research On Pedestrian Detection Technology In Fixed-point Video Surveillance Based On Deep Learning

The video surveillance systems have been widely used in China's social production and life,and have played enormous role in ensuring the safety of people's lives and property and maintaining the normal order of social production and life.In practical applications,the video surveillance systems are mainly based on fixed-point video surveillance mode.The data analysis in the video surveillance systems,especially the research on target detection and target recognition with the key elements of human being as the main research object,is one of the key and difficult problems faced by researchers.Exploring efficient and accurate analysis methods and effectively constructing a pedestrian detection and recognition analysis system is one of the hot research topics in the current video surveillance system research process.Research on pedestrian recognition algorithm and application technology in video surveillance system has outstanding realistic meanings.However,in the practical video surveillance and detection system,there are often problems such as pedestrian missed detection and false detection due to factors such as the prospect of pedestrians being too small.This paper studies the pedestrian detection technology in fixed-point surveillance video system,and proposes a method to improve pedestrian detection by using auxiliary camera detection.This research is mainly divided into two aspects: one is the detection of moving targets,and the other is the classification confirmation of the target,thus completing the whole process of pedestrian detection.In the aspect of moving target detection,the detection algorithms based on pedestrian motion feature,such as inter-frame difference method,optical flow method and Gaussian background modeling method,and the detection algorithms based on pedestrian features,such as color feature,Harr feature and HOG feature are analyzed.By contrast,the Gaussian background modeling method is selected for moving target detection.Furthermore,the improved methods of pedestrian head and foot fracture and the adverse effects of pedestrian shadow appear in the target detection results and carry out experimental analysis,which improves the integrity and accuracy of target detection and extraction.In view of the leading edge of convolutional neural networks in deep learning in the current research fields of image classification and recognition,the classification of targets is implemented by convolutional neural networks.Through the moving target detection module processing,the pedestrian sample data is obtained by screening and the pedestrian sample database is created,then the deep convolutional neural network is trained,and then the trained network model is deployed and implemented,and the detected targets are classified,and finally,we draw the conclusion by the practical study.Using the method of using the auxiliary zoom camera for remote pedestrian detection,the relative detection percentage of the pedestrian target is improved by 95.29% in the moving target detection module.In the pedestrian classification module,the relative percentages of the accuracy of the pedestrian classification results to the increase of the average probability are 15.14% and 3.75%,respectively,and good detection results are obtained.