Research And Implementation Of Vehicle Tracking Algorithm

In the intelligent transportation,the accurate position of vehicle could be acquired by use of computer vision method to detect and track vehicle in the video.While,due to the complexity of application situation,such as the apperance of moving object varies,the environment change over time and the occlusion between objects,it is very diffcult to acquire a robust,accurate tracking method fitting in every situation.As moving object detection is necessary before tracking,this paper also studied the moving object detection algorithm.For tracking,the Mean Shift tracking algorithm was choosed for study and improvement.Fistly,three commonly used moving object detection algorithms were studied and analyzed.And The image post-processing algorithms for enhancing the output of detection algorithms were study.A new binary image processing algorithm was developed for the case that the current binary image processing algorithm can not get the ideal foreground mask image well.For each pixel in the image,a new feature is defined according to its surrounding environment and the position in the environment.After getting the feature value of every pixel,the image was adjusted according to a given threshold.After the operation,the noise in the image was smaller.Then set the adjusted image as input,repeat the above procedure until the change of result image tends to be stable and get the final result.This algorithm can effectively remove the noise while maintaining the edge shape of moving object.Finally,the experimental result show that the algorithm has high practicability.The Mean Shift algorithm and its application in moving object tracking are deduced in detail.Then the Mean Shift tracking algorithm is implemented,and the algorithm is tested by a traffic monitoring video.The validity of the algorithm is proved.The limitation of the Mean Shift tracking algorithm is analyzed by combining theory with experiment.In order to solving the dimension adaptation problem of Mean Shift tracking algorithm,the traditional Mean Shift algorithm was improved taking advantage of the fact that the distance ofpoints in the object is changing with the object.The algorithm fist produces some candinate points in the object window and uses the L-K optical flow method to calculate the position of candinate points in the previous frame.Then the local consistency constraint and the time-flow direction consistency constraint are used to remove the wrong candidate points which can not get correct position in the previous frame.After the correct points and their corresponding points in the previous frame have gotten,the change of distance of points between the neighbour frames was used to estimate the change of the object size.Finally,the validity of the algorithm is proved by experiments.