Scalable compressive tracking based on motion

Visual tracking is challenging due to appearance changes caused by motion, illumination, occlusion and pose, among others. For these local changes, appearance model based tracking algorithms, such as MILtracker [8], have adopted local features and most recently extended to compressive domain, namely Compressive Tracking [13], for the real-time performance. However, the motion information is missed out from these trackers and assumptions on target motion have been made by predefined search radii and the constant scale. In this paper, the motion information has been integrated into appearance model based tracking by a two-stage strategy. With this strategy, the Compressive tracking can then adapt to target scale change. The experiments show that motion estimation could improve the performance of appearance based trackers especially when the target is with motion variety.