Object Tracking

Simulated macOS track detail — Figure 1. macOS — Persistent track confidence and centroid history. Simulated research data.

Tracking assigns a fresh UUID when an object first appears. Subsequent observations match using overlap plus label consistency. Histories retain boxes, centroids implied by boxes, labels, confidences, and environment scores. Missing observations mark a track occluded; exceeding the missed-frame limit terminates it. IDs are never reused.


Frame N detection ─┐

                   ├─ overlap + label score → persistent track → searchable history

Frame N+1 detection┘                              ↓

                                      expire after missed frames

Periodic detection refresh prevents a lightweight tracker from permanently preserving an early classification. A tracking ID means “likely the same visible region,” not verified identity.