Video Object Counting With Scene-Aware Multi-Object Tracking

The critical challenge of video object counting is to avoid the same multiple times in different frames. By comparing appearance and motion feature information detection results, authors use multi-object tracking method assign an independent ID number each object. From time tag obtained until end video, counted only once. However, even minor amounts image noise can cause irreversible changes information, resulting severe drifts. This paper introduces concept scene awareness addresses unreasonable assignment caused by unreliable matching context region division. Through macro analysis scene, define (called transition region) where objects increase or decrease require that all assignments for new deletions existing take place region. Because actual non-transition constant, they rematch unmatched with IDs relocation) because are failure. In this paper, create algorithms dynamically generating regions, detecting increases decreases, relocating IDs. Experimental results show effectively improves accuracy counting.