AI Person Tracking: Single Camera to Multi-Camera ReID
Person tracking with AI goes beyond following a dot on a map — it builds a temporal biography of where a person was, what they did, and when, across every camera in your network and every session in your library.
Single-camera vs multi-camera tracking
Within a single camera, person tracking assigns persistent IDs to detected persons across frames using IoU (Intersection over Union) matching and Kalman filter prediction. This enables measuring how long each person spends in defined zones, their movement path, and activity classification.
Multi-camera (MTMC) tracking is significantly harder. When a person exits one camera's field of view and appears in another (with an unknown time gap, possibly changed appearance), the system must match them by appearance alone. VidForgeX Forensic uses Transformer-based ReID networks that are invariant to viewpoint changes and moderate appearance variation.
Tracking Profile system (sports and coaching)
In the sports and coaching context, Tracking Profiles serve a different purpose: building longitudinal performance records for athletes. Upload a reference image of an athlete, create their Tracking Profile, and every subsequent analysis automatically attributes performance metrics (distance covered, actions performed, position heatmap) to their profile.
The Person Profile page in VidForgeX aggregates across all tagged sessions: performance trends over time, comparison against team averages, AI-generated coaching insights, and a cross-session AI chat interface that answers questions like "Has this player's pressing intensity improved over the last 6 sessions?"
Technical accuracy and limitations
| Scenario | Typical accuracy | Limitation |
|---|---|---|
| Single camera, distinct individuals | 95%+ tracking consistency | Occlusion breaks tracking |
| ReID across cameras, same outfit | Top-1: 80–90% | Lighting/angle variation |
| ReID across cameras, outfit change | Top-1: 50–70% | Appearance basis compromised |
| Crowded scenes (>5 persons/m²) | Degrades significantly | Occlusion, overlap |