3D MOT

3D multi-object tracking follows objects through time, estimating their positions and preserving identities as they move, disappear behind obstacles, or enter a scene. Many systems detect objects in each frame and associate them using motion models such as Kalman filters, while learned methods use recurrent queries, transformers, appearance cues, or geometric features to make associations. Research spans autonomous-driving benchmarks such as KITTI, nuScenes, and Waymo, as well as multi-camera, cooperative-vehicle, robotics, and people-tracking settings. A central challenge is improving identity continuity and position accuracy under occlusion, sensor gaps, and changing environments while keeping systems fast, well-calibrated, and practical to deploy.