Multi-scale Correlation Filter Tracking with Adaptive Re-detection Mechanism

Abstract

Tracking is a popular research topic in artificial intelligence, but how to handle the severe occlusion and deformation remains a challenging problem. Focusing on this issue, we propose a multi-scale correlation filter tracker using a re-detection module. Specifically, we utilize a reliable confidence strategy to estimate the reliability of initial results, and introduce a novel template matching technique to solve the target relocation problem. Experiment results demonstrate that our method can outperform several classic trackers.