A novel affine invariant feature set and its application in motion estimation

This paper proposed a novel semi-local feature set, which is proved to be affine invariant. The proposed feature set can be local or global based on the mask parameter selected. Small mask indicates local features, which are more precise and not sensitive to occlusion problems. Large mask indicates global features, which are more robust and less sensitive to noise. Using this feature set a correspondence based velocity estimation algorithm is developed. This motion estimation method can be used in motion detection, motion segmentation, and motion based recognition