What is object tracking?
Object tracking is the task of following each detected object across the frames of a video and keeping its identity consistent. Object detection finds the people and vehicles in one frame; tracking links those detections over time so the system knows that the person in frame 300 is the same person it saw in frame 1. Each tracked object gets an ID, a trajectory, and a duration.
Most trackers work in two steps. A detector finds objects in each frame, and a matching step associates the new detections with existing tracks using position, motion, and appearance. When a worker walks behind a rack and reappears, a good tracker re-identifies them rather than starting a new track.
Object tracking in practice
A grocery chain wants to know how long customers wait at the deli counter. A detector alone can only count heads per frame. With tracking, each shopper who enters the queue zone is timed until they leave it, giving a wait-time distribution by hour that the store uses to schedule staff. In a distribution center, tracking forklifts through intersections produces near-miss counts for the safety committee.
What to watch for
- Crowded scenes cause identity switches, where two tracks swap, and these errors inflate or deflate counts.
- Tracking people, even without identifying them, may count as processing personal data under privacy law.
- Camera placement matters more than the algorithm, and an overhead view avoids most occlusion problems.