A Novel Strategy for Kernel-Based Small Target Tracking against Varying Illumination with Multiple Features Fusion

This paper presents a novel kernel-based method for small Target Tracking with multi-feature fusion against Varying Illumination. Firstly, the conventional tracker based on color histogram is unreliable or even failed under varying illumination. Therefore, a new fuzzy color histogram creation is proposed based on the HSV color space and utilizes the local background information around tracking target to dynamically correct its fuzzy color histogram model and eliminates the sensitive of conventional color histogram to illumination change and noise. Secondly, there is still not an effective method to cope with object occlusion, angle variation, scale change etc. The tracking algorithm utilizes feature points extracted by improved SIFT as the reference points of Mean-Shift and calculates the target area center, which combines the two methods together seamlessly. Lastly, the whole tracking algorithm utilizes fuzzy color histogram model and combination of improved SIFT as the reference points of Mean-Shift for small target tracking. Experiment results show that the proposed algorithm can keep tracking object of varying scales and various illumination even when the surrounding background being similar to the object's appearance.