Multi-Object Tracking with Multiple Cues and Switcher-Aware Classification

• DOI: 10.48550/arxiv.1901.06129
• Publication Date: 2019-01-01
• Authors: Weitao Feng; Zhihao Hu; Wei Wu; Junjie Yan; Wanli Ouyang
• Keywords: Robustness; Sensory cue

In this paper, we propose a unified Multi-Object Tracking (MOT) framework learning to make full use of long term and short cues for handling complex cases in MOT scenes. Besides, better association, switcher-aware classification (SAC), which takes the potential identity-switch causer (switcher) into consideration. Specifically, proposed includes Single Object (SOT) sub-net capture cues, re-identification (ReID) extract classifier matching decisions using extracted features from main target switcher. Short help find false negatives, while avoid critical mistakes when occlusion happens, SAC learns combine multiple an effective way improves robustness. The method is evaluated on challenging benchmarks achieves state-of-the-art results.