Robust Object Tracking Using Regional Mutual Information and Normalized Cross Correlation

In this paper, a novel feature point-based background detection algorithm is proposed to distinguish crowded and un-crowded background. This algorithm uses regional mutual information (RMI) and normalized cross correlation (NCC) as similarity measure based on background type criterion for template matching. RMI is suitable as similarity measure for object tracking in order to reduce sensitivity to noise, partial occlusion and illumination variation. Experimental results demonstrate that our proposed algorithm has high ability to tracking object when the background changes from un-crowded background to crowded background or vice versa.