Robust Tracking in FLIR Imagery by Mean Shift Combined with Particle Filter Algorithm

Abstract

A novel target tracking algorithm for forward-looking infrared image sequences is proposed based on mean shift and particle filter algorithm. The mean shift algorithm is served as an efficient gradient estimation and mode seeking procedure in the particle filter. Particles move toward the modes of the posterior kernel density estimation. The infrared target is represented in the cascade grey space and the state transition model is established as the second-order auto-regressive model. We use the modified particle filter to track the infrared target robustly. Experiment results show that the proposed tracking algorithm is efficient and robust for the infrared targets with severe clutter background and provide better tracking performance than the conventional particle filter.