Human tracking by employing the scene information in underground coal mines

In this paper, we propose a human tracking method by fusion of the scene information for underground coal mines environment, which aims to improve the tracking accuracy and robustness in the presence of complicated factors (e.g. low illumination and low intensity contrast, bright spots and shadow disturbance). For object detection, in order to decrease the influence of bright spots, we present a pixel-to-region hierarchical bright spots removing criterion by fusion of the learned background model priors. Moreover, within the particle filtering tracking framework, we propose a novel particle weight calculation method, which adopts the nonparametric kernel density estimation method to evaluate the likelihoods of each particle belonging to the object and the background, respectively. The final tracking results are refined based on a shadow removing mechanism by making use of the background priors. Experimental results on several real underground coal mine video sequences demonstrate the effectiveness and robustness of our tracking method.