Multi-target Tracking Based on Motion Estimation and RJ-MCMC Particle Filter

Aiming at the problems of multi-target tracking, such as occlusion between target and background, similarity of features and random change of the target number, this paper designs a multi-target tracking system which can be used in complex environment. In this thesis, the constraints of scene is added to the target detection algorithm based on the hybrid multiscale deformable model to improve the efficiency. At the same time, the global motion parameter is obtained by using the dual absolute conic constraint. The obtained parameters are used to modify the motion model of target tracking. Then based on the basic particle filter, MCMC (Markov Chain Monte Carlo) improvement strategy and reversibility-jump mechanism (RJ) are added to reduce the computational complexity and solve the stochastic change of target number. Finally, verify the validity of the tracking algorithm in the thesis.