Reproduce “No More Potentially Dynamic Objects: Static Point Cloud Map Generation based on 3D Object Detection and Ground Projection” 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
Soojin Woo, Donghwi Jung, Seong-Woo Kim
A method creates static point cloud maps by removing both dynamic and potentially dynamic objects from LiDAR data through 3D object detection and ground projection. This approach demonstrated improved localization accuracy, reducing x-y error by 5.3% and yaw error by 5.7% on a proprietary dataset compared to maps without removal, while also preserving map density.
Key insights
- Identifying dynamic objects based on their inherent characteristics via 3D object detection, rather than observed motion, enables the removal of "potentially dynamic objects" such as parked cars.
- Projecting dynamic object points onto the ground plane is crucial for maintaining map density, preventing sparsity that could negatively impact SLAM performance.
- The integration of a real-time 3D object detector with ground projection and a SLAM framework provides a robust and efficient approach for generating long-term static maps.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.