SiamVGG-LLC: Visual Tracking Using LLC and Deeper Siamese Networks

Recently, visual tracking has been widely concerned in computer vision applications such as behavior analysis, autonomous driving with advanced trackers. Some trackers based on Discriminative Correlation Filters (DCF) have acquired great performance on benchmark, but it is not satisfied with the real-time visual tracking; Later, some people combined Deep Neural Networks (DNNs) and DCF to form some trackers, which has significantly improved the tracking accuracy and speed. Unfortunately, it cannot be applied to long-term visual tracking in many public places. To balance accuracy and speed of tracking, combining the siamese network and VGG is advanced tracker in long-term visual tracking. However, SiamVGG get an unsatisfactory result when the object is in background clutters and motion blur. In this paper, we propose a method called SiamVGG-LLC that combining the LLC (Locality-constrained Linear Coding) with SiamVGG to achieve visual tracking. Experimental results demonstrate that our method can obtain better accuracy on OTB-100 datasets when the object is in background clutters and motion blur.