HybridTrack is a research project focused on robust multi-object tracking (MOT) for autonomous driving, with an emphasis on stable and reliable tracking over time.

Core idea

HybridTrack combines learning-based components with a real-time, data-driven Kalman Filter to improve temporal stability and robustness. The goal is to get the best of both worlds: strong per-frame perception and consistent track continuity.

What I did

• Developed HybridTrack as a 3D vehicle tracking system on an autonomous driving dataset.
• Designed and integrated a data-driven Kalman Filter for real-time tracking.
• Focused on practical robustness: reduced identity switches and improved temporal consistency.

Publication

Published in IEEE Robotics and Automation Letters (RA-L/ICRA) (2025).

Project date

Apr 1, 2025