Robust object tracking algorithm based on sparse eigenbasis

To reduce the computation and to improve the performance of object detection and tracking algorithm with object appearance variation, a tracker based on sparse eigenbasis is proposed. According to the compressive sensing theory, the objects are described in a low-dimensional sub-space representation based on Karhunen-Loeve transform learned online. Meanwhile, combining the Bayesian inference, an adaptive object tracker is presented. First, the authors represent the appearance of the object in a low-dimensional sub-space, then the authors obtain the optimal estimation of the state parameters by Bayesian inference. Finally, the authors update the eigenbasis space using the optimal observations. Experimental results show that the proposed method is able to track the objects effectively and robustly under temporary occlusion and large illumination changes.