Traditional method can’t follow the scale and orientation change of the object, which usually causes inaccurate
localization. Therefore, this paper combines SIFT(scale invariant feature transform) feature detecting with the mean-shift
tracking method, which proposes that the scale change of the SIFT keypoint is proportional to that of the object, and the
dominant direction change of the SIFT keypoint is the same as that of the object. The algorithm for adaptively constructing
the scale and the orientation information from SIFT features of the object is presented. And a variable-bandwidth and
orientation ellipse kernel is used to improve the traditional mean-shift method. The experimental results show that the
proposed algorithm provides good tracking of the scale and orientation change of the object, and the localization is more
accurate.