Research On Moving Object Recognition And Tracking Technology In Video Surveillance

With China's economic growth,social progress and the enhancement of comprehensive national strength,in order to implement the concept of national security development,in the fields of banking,transportation,electric power,military and government infrastructure construction,social citizens' security awareness is getting higher and higher.Although monitoring equipment has been installed in public places,most of the existing video monitoring systems are simple object recognition and video recording,and do not give full play to the initiative and real-time of video monitoring.Therefore,in order to maintain social order and build a harmonious society,higher requirements are put forward for the design of intelligent video surveillance system,which is also a hot research direction in the field of artificial intelligence and computer vision.At present,although China's intelligent video surveillance system has been applied to the construction of smart city,its core technology in the state of moving target recognition and tracking still has the problems of low detection accuracy and slow running speed.In addition,for the densely populated areas and dim areas,there will be the possibility of occlusion,missed detection and false detection.In a word,it is necessary to further study the recognition and tracking technology of moving objects.Traditional methods of moving target recognition and detection,such as inter frame difference,background subtraction and optical flow,have the problems of low detection accuracy and missing detection;In this paper,a network model based on deep learning is introduced,and a pyramid structure is added on the basis of the yolv3 network framework to build a yolov3 for feature selection;Because the simple network model will degenerate in the deepening process,the depth residual network is introduced to improve the detection accuracy;Update the anchor mechanism to realize the prediction of the relative coordinates of the moving target area and increase the detection range;The multi-scale fusion convolution mechanism is used to detect small objects in a large range to prevent missed detection.Aiming at the problems of background and obstacle occlusion in the process of moving target tracking,based on the extended Kalman filter,the observation equation and state estimation of sliding backward recursive extended Kalman filter(SBR-EKF)are established to realize the nonlinear solution of correlation filter;The improved Hungarian algorithm is used to match the data between adjacent video frames;The region based quality evaluation network is used to evaluate the false detected moving target and improve the reliability of tracking.Finally,the recognition model of moving target and target tracking algorithm are combined to carry out the comprehensive experiment.The experimental results show that the target detection model proposed in this paper is better than the traditional recognition method in recognition accuracy and accuracy;In the process of target tracking,when there are obstacles blocking and other external factors causing the change of the target,the method proposed in this paper is improved than the traditional target tracking algorithm.