(A) scale-invariant object tracking method using strong corner in scale-domain

Abstract

An object tracking method using scale-invariant feature shows good performance in tracking. It is efficiently applicable to the rotated or size-changed target, and also maintains good performances in an occluded and intensity-changed image. However, SIFT algorithm has high computational cost. In addition, for enough features for matching, target object should be sufficiently big. In this thesis, a scale-invariant object tracking method using strong corner in scale-domain is proposed. While reducing computational cost of SIFT tracker, the proposed method can track an object of smaller size than SIFT tracker by extracting relatively lots of features. The proposed method extracts extremum point in each scale-domain, thereby increasing the feature number. With matching features, the method finds relations between adjacent frames, and updates previous frame. In the experimental results, the proposed algorithm shows better performances than the existing SIFT tracker while reduces its computational complexity, and makes object also to be tracked in a small size.