Human Tracking by Importance Sampling Particle Filtering on Omnidirectional Camera Platform

Abstract

This paper proposes a sequential importance sampling (SIS) particle filtering framework to track the human with overcoming the warping and low resolution. We utilize a foreground-based importance sampling mechanism for efficiently converge to the target distribution. We construct a tracking system with the fusion image likelihood, even the human raises the head. Furthermore, the two-space integration to evaluate the likelihood measurement is proposed to robustly track human for overcome the warping effect. The overall performance has been validated in the experiments.