Adaptive particle swarm optimization-based particle filter for tracking maneuvering object

Particle filter is a powerful tool for visual tracking based on Sequential Monte Carlo framework. However,it is faced with a fatal problem due to its suboptimal sampling mechanism in the importance sampling process and thus leads to the well-know sample impoverishment problem. Although last decade people have developed a lot resampling schemes to solve this problem, no one can be suitable for all situation. The standard PSO-PF algorithm, which applying particle swarm optimization algorithm into particle filter framework, is an excellent one of them. However, we found that the standard PSO-PF not suitable for tracking object with complicated motion model such as maneuvering object, which is moving with various motion state such as abrupt motion or mildly motion or even stand still. In this paper, we propose an algorithm, which we call it Adaptive Particle Swarm Optimization-Based Particle Filter(APSO-PF), to track the maneuvering object. Instead of constant parameters used in the standard PSO-PF, APSO-PF can adaptively change its parameters according to the motion state of the object and thus improve the tracking performance significantly. Experimental results show that our algorithm surpasses the standard PSO-PF on both robustness and accuracy.