Research On Vehicle Moving Target Detection And Tracking Technology Based On Video

With the development of computer hardware technology and computer vision technology,traffic monitoring system based on computer vision becomes possible.Real-time detection and tracking of video vehicles is the core part of the intelligent traffic monitoring system,as well as the foundation and key technologies in the fields of computer vision,pattern recognition,video image processing and artificial intelligence.There are still many difficulties in the application of moving target detection and tracking algorithms in intelligent traffic monitoring,such as severe weather conditions and targets being blocked for a long time.This paper mainly studied the following aspects:(1)Aiming at the problem that the existing background subtraction method is sensitive to the initial moving target and is vulnerable to dynamic background,light change,camera jitter and other factors,In this paper,a method based on markov random field(MRF)and two-mode fuzzy gaussian mixture model(T2-FGMM)is adopted to detect moving targets,which solves the problems of many discrete vulnerabilities and imprecise target contour detected by T2-FGMM in dynamic scenarios.(2)Aiming at the problem that conditional iteration mode(ICM)algorithm is prone to fall into local optimum when it uses "greedy" strategy to solve energy optimization,this paper improves ICM algorithm and proposes a "smooth" descent method,which weights the smoothing term in the minimization energy function,and solves the problem that ICM algorithm is very sensitive to initial estimation and easily leads to poor local minimum.Value problem.The improved ICM algorithm is applied to vehicle target detection in video stream,and a good result of target detection is obtained.(3)In-depth study of multi-target tracking technology based on hierarchical data association of appearance features and spatio-temporal feature model,mainly study the spatio-temporal feature model.For the space-time feature model only considersthe appearance features and optical flow histogram features,when the multi-target has a long time occlusion,it can not meet the tracking accuracy problem.In this paper,the spatio-temporal feature model is improved,and a spatio-temporal feature model based on kalman filter is proposed.The centroid of rectangle is added to the target detection as the tracking feature.The improved model is validated on the public data set,and a good target tracking effect is obtained.In this paper,vehicle moving target detection and tracking in video stream are studied.The innovation of this paper is that the improved ICM algorithm is applied to moving object detection in video stream.Secondly,the spatio-temporal feature model of multi-target tracking algorithm with hierarchical correlation is improved.This method is suitable for large area and multi-target complex scenes with strong anti-jamming ability and high detection and tracking accuracy.It can be used in vehicle monitoring and tracking system of Expressway and urban traffic.