A New Multiple-Objects Tracking Method with Particle Filter

Abstract

The new method stated in this paper is to model the multiple objects in the visual sequence into two-dimensional multi-peak probability distribution, which raised a new multiple-objects tracking method with particle filter. The results of importance resampling by the particle filter represent the probability distributions of the objects. Firstly, it gains the probability distribution model points of each object through mean-shift algorithm, and FCM (Fuzzy C - Means) is used to get the particle subset of the respective objects. Then final state of each object can be estimated and mean-shift kernel bandwidth parameter can be updated through particle subset. Finally, the movement of the objects can be tracked through data association. Experiments prove that this algorithm can be more effectively and more stably applied onto the tracking of multiple-objects complicated movements, such as spinning, zooming, masking, etc.