Research On Object Tracking Based On CRF Model

Moving object detection and tracking in hybrid background are the keytechnologies in computer vision field. Substantial applications such as robotnavigation,car video, are closely related with moving object detection in dynamicbackground. This paper presents an improved CRF model and an improved objectmatch algorithm and summarizes a better implementation method of moving objectdetection and tracking in hybrid background based on optical flow algorithm.(1) This section describes several widely-used statistic model and emphasizemainly on the concept and principle of CRF model then demonstrates the goodimplementation of CRF model in many fields.(2) This paper presents an improved CRF model based on PF algorithm whichchanges the unreasonable time intensity β of posteriori probability functiondetermined by many trials to the scientific time intensity determined by the distancebetween the estimated pixel and the center. This improved module has better accuracybased on the simulation results.(3) This section introduces a widely-used approach of block matching anddemonstrates the possible errors and wasted calculation caused by the thresholdchoosing procedure in the traditional block matching. Hence, this paper presents ablock matching method based on distributed block Particle Filter Estimation in whichmove the former fame in negative direction acquired by the PF estimation in order toget the search center of the matching block as well as the search scope determined bythe PF estimation error in the latter fame with a improvement of accuracy and rapidityin block matching. Besides, this section presents an improved match similarityfunction which combines two-dimensional information of color and space-time whichcan solve some complex situation.(4) Feature optical flow method widely-used in moving object detection in staticbackground is introduced in this section. Combined with the improved matchingalgorithm, PF algorithm, a novel object tracking algorithm is presented for dynamic background, color confusion, object block situation respectively. Through theexperiment in Matlab and Visual Studio, this method proves to be effective to thosecomplex situations.