Adaptive object tracking algorithm based on eigenbasis space and compressive sampling

To improve the speed of image storage, processing and transmission in visual systems, to reduce the computation of target-tracking algorithm and to improve the performance of the visual tracking methods with object appearance variation, according to the signal description and processing theory of compressive sensing, the tracking based on eigenbasis and compressive sampling is proposed, and the objects in the visual system are described in low-dimensional subspace representation learned online. Meanwhile, combining the representation with Bayesian inference, an adaptive object tracking method is presented. First, the authors represent the appearance of the object in the low-dimensional subspace, then they obtain the optimal estimation of the state parameters by Bayesian inference. Finally, the authors update the eigenbasis space using the optimal observation. Experimental results show that the proposed method is able to track objects effectively and robustly under pose variation, temporary occlusion and large illumination changes.