Particle Filter for Targets Tracking with Motion Model

Abstract

Real-time robust tracking for multiple non-rigid objects is a challenging task in computer vision research. In recent years, stochastic sampling based particle filter has been widely used to describe the complicated target features of image sequence. In this paper, non-parametric density estimation and particle filter techniques are employed to model the background and track the object. Color feature and motion model of the target are extracted and used as key features in the tracking step, in order to adapt to multiple variations in the scene, such as background clutters, object's scale change and partial overlap of different targets. The paper also presents the experimental result on the robustness and effectiveness of the proposed method in a number of outdoor and indoor visual surveillance scenes.