Object matching using hybrid modified RGB color model and HRR-based background detection

This study presents a pixel-wise statistical approach to distinguish every pixel into foreground, background, shadow or highlight for background subtraction and foreground detection. The modified RGB color model is proposed to effectively reduce misclassifying the foreground pixels into highlights. The modified Highest Redundancy Ratio (HRR)-based background update method is also proposed to overcome the lighting variation and slow motion object problems in background reconstruction. In tracking procedure, a decision function with low computational complexity is proposed to sequentially evaluate the objects’ correlation between consecutive frames. The decision function consists of the objects’ centroid distances, objects’ area differences, and objects’ overlapping areas between current frame and previous frame. As documented in experimental results, the proposed method can achieves high matching rate, which is great advantageous in surveillance systems.