Research On Moving Targets Tracking Algorithms Based On Particle Filter

Tracking moving object is one of the key subjects in the field of computer vision, whichhas a very wide range of applications in many fields such as civil and military.In recent years,with the rapid development of computer science and the pressing requirements of life andwork, object tracking has become a popular issue on computer vision technology.The thesis work is summarized as follows:(1)This thesis analyzes several common approaches in this field of motion object detectionincluding background subtraction as well as frame difference. An improved three framedifferences which combined with morphological processing is proposed to solve the "hole"phenomenon which caused by the frame difference method. The algorithm can improve theaccuracy of the moving targets detection result.(2)A multi-feature fusion tracking algorithm based on particle filter is put forward toovercome the disadvantages of single color feature object tracking algorithm which indicatethe target in the complex environment insufficiently so well as track the target unstableenough. Two complementary features: color and texture features are combined to build theobject’s reference model based on weighted color histogram means. Experimental resultsshow that this algorithm is more accurate than the single color feature algorithm in tracking.(3)For sub-feature fusion coefficient of multi-feature fusion tracking algorithm is fixed, anadaptive tracking algorithm combining multi-cue based on particle filter is proposed. In orderto increase the stability of the multi-feature fusion tracking algorithm, Sigmoid kernelfunction is introduced to adjust the feature weight dynamically. Groups of experimentsdemonstrate that, the adaptive tracker based on multi-cue particle works can improve theaccuracy of object tracking as well as has good robustness.(4)In order to realize the tracking of multiple moving targets, multiple particle filtertrackers are combined based on single target tracking. Simulation experiments further verifythe reliability of the algorithm.(5)Tracking demonstration platform based on Matlab GUI which combined with trackingalgorithms in this paper is designed. Experiments of different algorithms are shown in thisdemonstration platform to further illustrate the reliability of the algorithms in this paper.