Tracking-by-Counting: Using Network Flows on Crowd Density Maps for Tracking Multiple Targets

State-of-the-art multi-object tracking~(MOT) methods follow the tracking-by-detection paradigm, where object trajectories are obtained by associating per-frame outputs of detectors. In crowded scenes, however, detectors often fail to obtain accurate detections due heavy occlusions and high crowd density. this paper, we propose a new MOT tracking-by-counting, tailored for scenes. Using density maps, jointly model detection, counting, tracking multiple targets as network flow program, which simultaneously finds global optimal over whole video. This is in contrast prior that either ignore thus prone errors or rely on suboptimal two-step process using heuristic density-aware point-tracks matching targets.Our approach yields promising results public benchmarks various domains including people tracking, cell fish tracking.