Reproduce “No Blind Spots: Full-Surround Multi-Object Tracking for Autonomous Vehicles using Cameras & LiDARs” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Akshay Rangesh, Mohan M. Trivedi
The M³OT framework from UC San Diego enables full-surround multi-object tracking for autonomous vehicles by extending a 2D Markov Decision Process to 3D global coordinates, integrating multi-modal camera and LiDAR data. This method achieves robust, long-term tracking with improved accuracy through early sensor fusion and novel global position-based features.
Key insights
- Accurate 3D localization, particularly when utilizing LiDAR-based projection to estimate object positions, significantly improves track stability and duration compared to vision-only IPM-based 3D estimation.
- The integration of LiDAR sensors with cameras substantially enhances tracking performance, yielding higher MOTA, lower MOTP, and fewer ID switches, demonstrating the complementary benefits of multi-modal fusion.
- The framework's modular design ensures consistent tracking performance across varying camera configurations (e.g., 2 to 8 cameras) and exhibits robustness to the performance variations of underlying object detectors.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.