Color-Based Object Tracking Using a Hybrid Particle Filter

摘要

The efficient tracking of visual objects in complex environments is a challenging task for the vision community. In real-rime particle filter tracking, the particles are a constrained resource and should be well used. We propose a fast hybrid tracking method. At each time step, the procedure consists of a detection based particle filtering and a mean shift iteration followed. The detection based particle filter method efficiently explores the possible state space by evenly spreading particles and keeping a certain number of particles in each possible target region. The mean shift iteration is used to further improve the accuracy of state estimate. Color clues are used in this method.Experimental results show that when tracking rapidly moving targets, this method has better performance but need smaller particle number compared with standard particle filter method.