BSCF: Learning background suppressed correlation filter tracker for wireless multimedia sensor networks

Recently, the Internet of Things (IoT) devices have been widely used in all sectors of society, especially in the field of wireless multimedia sensor networks (WMSNs). However, massive IoT video data requires efficient back-end processing to detect and track prominent objects for performing interconnection feedback. Correlation filter (CF)-based trackers present favorable properties for our back-end processing needs, such as fast computation speed, robustness to photometric and geometric variations. In this paper, we propose a novel background suppressed correlation filter (BSCF)-based target tracking method for wireless multimedia sensor networks, this method can incorporate all global background patches to enhance the tracking performance. Firstly, we reformulate the original ridge regression objective by introducing a convolution suppression term so that all real background patches will limit the generation of the filter through the circular shift operator and cropping operator. We then provide the closed form solutions of BSCF for multi-channel features via the alternating direction method of multipliers (ADMM). Further, we suggest a decomposition strategy to apply the background suppression framework as a general module to other CF-based trackers. The extensive experiments demonstrate that the proposed method performs favorably on multiple datasets against recent state-of-the-art methods. In particular, our BSCF ranks the first place on OTB-2013 with an AUC of 0.688 and an average precision of 0.904, exceeding the advanced CF tracker STRCF.