Local appearance based robust tracking via sparse representation
Avinash Ruchandani, R. Venkatesh Babu · 2012
Visual tracking is an important task in many computer vision applications. In this paper, we present a novel approach for visual tracking using a local sparse appearance model that exploits spatial information. The sparse information along with the spatial information of local patches within this representation is used to determine the motion of the object. When the match between the target and candidate patches are represented in matrix form, the translation motion of the object can be obtained by analyzing the diagonals of the mapping matrix. A novel local template update strategy is proposed to update the relevant parts of the object within the candidate undergoing changes. Along with the local patch update, we use an adaptive/permanent template update strategy which gives less update priority to the transient local patches accounting for partial occlusion. This approach performs comparably against state-of-the-art tracking techniques, in various challenging videos involving changes in scale, pose, illumination and partial occlusion. The proposed approach tracks at a processing speed of over 40 frames per second and it is suitable for various real-time applications.