The Study Of Vehicle Detection And Tracking Methods Based On Machine Vision

With China’s urbanization and high speed economic development, the number ofvehicles in our country is rising significantly year after year. These tools give us muchconvenience,but at the same time it also brings a series of problems such asenvironmental pollution, traffic congestion and traffic accidents. In this context,intelligent traffic control system based on machine vision came into being. In this system,the vehicle target detection, tracking and behavior analysis are the most basic parts. Fromthe most commonly used detection and tracking algorithms to start,this article mainlymakes the following aspects of work:(1) The traditional Gaussian Mixture Model algorithm only sets a fixed backgroundlearning rate, when the environment is mutated, the traditional algorithm is difficult toreturn to normal in a short time. Aiming at this problem, this article first designs aenvironmental change judgment factor, its value is the number of foreground pixelsdivided by the total number of pixels, then according to the factor value to adjust thebackground learning rate. Experimental results show that this method can help thedetection system quickly overcome the effects caused by the environment mutation.(2) The CamShift algorithm is easy to generate the problem of tracking failure whenthe target motion state changed, aiming at this problem,this article introduces the KalmanFilter based on CamShift algorithm. This article first uses Kalman Filter to estimate themotion states and then obtains the target’s pre-estimated location in the current frame,then runs CamShift algorithm at the estimated position to arrive the true position of thetarget. Experimental results show that this method can effectively help CamShiftalgorithm eliminate the effects brought by the changes of target motion state.(3) The traditional CamShift algorithm requires people use hands to set which targetto be tracked when tracking targets,and it can not realize multi target tracking. Aiming atthe problem of CamShift algorithm can not track automatically,first uses improvedGaussian Mixture Model to detect targets, and then adds these targets to the tracking list,at last run CamShift algorithm for the tracked targets. At the same time,this article alsodesigns a multiple target tracking matrix to track single target and multi targets.Experiments show that the algorithm can track multi targets automatically.(4) At last,this article described the methods of vehicle counting, speed calculation and retrograde judgment, but for the relevant experiments and simulation. The resultsshow that the algorithm can effectively be applied to the vehicle behavior analysis work.