Patch-based object tracking using corner and color with partial occlusion handling

This paper proposes a novel patch-based object tracking algorithm combining Harris-SIFT and color histogram features to handle the partial occlusion problem. The object is represented as a number of non-overlapping patches. Harris-SIFT feature is defined as Harris corner and its SIFT feature vector, which can represent steady parts of the object when it is partially transformed or occluded. The Harris-SIFT corner matching results are then used for filtering out invalid patches of the object. Patch-based color histogram takes spatial information into account and is guided by valid patch selection, thus provides a richer description of the object than the traditional color histogram. The average valid patch color histogram similarity is used in particle filter to locate the object. The experimental results show that the proposed algorithm is more accurate, robust and efficient than state-of-the-art object tracking algorithms in occlusion scenario.