Abstract
Sparse modeling has been successfully applied in object tracking methods. When the algorithms lose track of the target, it usually keeps locating a part of the background or starts locating another different object, which has a similar appearance to the original one. In this paper, we present a novel-tracking algorithm based on sparse representation and back-projection technique for feature and region extraction. We address the issue of the tracking by modeling the target appearance using the sparse approximation, thereafter, we apply a back-projection process to identify its region. We exploit the spatial information by back-projecting the sparse coefficient of the template in each frame. Thereby, we guarantee a more robust localization of the target as we handle the foreground/background separation. Our tracker proved to be more stable and less prone to drift away.