Object Tracking

Object tracking assigns stable IDs to detections so you know which object is which over time. On OAK, tracking runs fully on-device, pairing detector outputs with hardware-accelerated association so IDs stay steady through occlusions and motion.Detection networks combined with ObjectTracker node yield tracked objects with status and (when paired with stereo) XYZ coordinates.

Tracking resilience

Tracklets carry lifecycle flags (new/removed/persisting) and survive detector dropouts via short-term tracking. You can cap max objects, pick ID policies, and remap ROIs to depth for spatial readouts. Pair with depth or feature tracking when you need both semantics and precise motion cues.

Typical use cases

Spatial object tracking: fuse detections with stereo for XYZ-labeled tracklets to drive grasping, inspection, or collision avoidance.
RGB-only tracking: keep lightweight persistent IDs on single-camera streams for retail analytics, counting, or sports clips.
Cross-stream overlays: map RGB tracks into depth/IR space so overlays line up and which can help navigation tasks to stay accurate.

Guides and examples

People tracker example

Counts people going through FOV

Object tracker example

Spatial tracking with YOLO and stereo fusion

Object Tracker Replay

Track IDs on recorded or live video

Object Tracker Remap

Align tracked boxes across RGB and depth

Need assistance?

Head over to Discussion Forum for technical support or any other questions you might have.

ON THIS PAGE