I. Introduction
Motion segmentation is a fundamental problem in computer vision, and a large amount of work has addressed this issue. Motion segmentation is aimed at segmenting objects with different motions. The algorithms for motion segmentation [1] can be divided into six main categories: image differencing [2], [3], statistical-based methods [4], [5] , optical flow based methods [6], [7] , wavelets [7], [8], layer-based methods [9], and factorization-based methods [10]–[12]. Most of the time, the algorithms for motion segmentation are different for different data. For example, the optical flow based methods are used to conduct motion segmentation in optical flow fields, and the image differencing and layer-based methods are mainly used to segment the different motions in dense image pixels. Meanwhile, the factorization-based methods are mainly used to solve motion segmentation for tracking features. In this letter, the proposed method is used to segment the tracking feature points belonging to different motions.