Online fragments-based scale invariant electro-optic tracking with SIFT

A novel online fragments-based scale invariant electro-optic tracking algorithm is proposed to deal with partial occlusion and cluttered background. First, the bounding box of the object is divided into equal-size fragments. A criterion based on FBE (Forward–Backward Error) is adopted to select reliable fragments that are used to seek the location of the object. Then, the size of the bounding box is calculated with SIFT (Scale Invariant Feature Transform) keypoints modified via geometry constraints regularly. Finally, fragments are updated separately by combining SAD (Sum of Absolute Difference) with MGC (Major Gray Component) anti-projection to distinguish partial occlusion from appearance changes of the object itself every frame. Numerical experiments certificate that the proposed algorithm outperform state-of-the-art trackers on plentiful video sequences, especially, under conditions of scale changes of the object and occlusion.