Visual Object Tracking via Cascaded RPN Fusion and Coordinate Attention

Recently, Siamese-based trackers have achieved excellent performance in object tracking. However, the high speed and deformation of objects movement process make tracking difficult. Therefore, we incorporated cascaded region-proposal-network (RPN) fusion coordinate attention into Siamese trackers. The proposed network framework consists three parts: a feature-extraction sub-network, block, RPN block.We exploit which can embed location information channel attention, to establish long-term spatial dependence while maintaining associations. Thus, features different layers are enhanced by block. We then send these separately for classification regression. According two regression results, final position target is obtained. To verify effectiveness method, conducted comprehensive experiments on OTB100, VOT2016, UAV123, GOT-10k datasets. Compared with other state-of-the-art trackers, tracker good run at real-time speed.