Residual Network Based on Multi-Features Combination for Tracking

Abstract

Correlation filter (CF) based tracking algorithms have shown favorable performance in recent years and have the impressive performance on benchmark datasets. The combination of deep learning and correlation filtering has also become a research hotspot. However, the tracking model has limited information about their context and easily drift in cases of fast motion, occlusion or background clutter, and the trackers update tracking models at each frame without considering whether the detection is accurate or not. In this paper, we present a tracking strategy based on the multi-features combination and use the residual network to enhance the learning ability that makes our trackers can take full advantage of multi-features. Experimental results on the benchmark datasets show that the performance of the model has been improved effectively.