Improved compressive tracking based on pixel-wise learner
This publication appears in: Journal of Electronic Imaging
Authors: T. Chen, H. Sahli, Y. Zhang and T. Yang
Publication Date: Jan. 2018
This work extends upon state-of-the-art multi-scale tracking based on compressive sensing (CT) by increasing the overall tracking accuracy. A pixel-wise classification stage is incorporated in CT-based tracker to obtain relatively stable appearance model, by distinguishing object pixels from the background. Additionally, we identify potential distracting regions which are used in a feedback strategy to handle occlusion and avoid drifting toward nearby regions with similar appearances. We evaluate our approach on several benchmark datasets to demonstrate its effectiveness with respect to state-of-the-art tracking algorithms.