02 — Trajectory
Object Tracking
Multi-camera, multi-object persistence across complex scenes. Re-ID, trajectory inference, and behaviour classification — running at 60fps on commodity GPUs.
Multi-object, multi-camera persistence across complex scenes. Re-identification through occlusion, trajectory inference at 60fps, and behaviour classification that holds up under crowd density.
How it ships
Operational discipline, not slide-ware.
01Re-ID across camera handoffs without identity-leakage.
02Trajectory prediction under partial occlusion.
03Behaviour classifiers: dwell, loiter, deviation, anomaly.
04GPU footprint: a single T4 supports 24 concurrent streams.
Reference Specs
24 / T4
Concurrent streams
92.1%
Re-ID confidence (occluded)
240 frames
Trajectory horizon
60fps continuous
Output cadence