Reproduce “Probabilistic Surfel Fusion for Dense LiDAR Mapping” on OpenResearch

About this paper

Chanoh Park, Soohwan Kim, Peyman Moghadam, Clinton Fookes, Sridha Sridharan

Researchers at CSIRO Data61 and Queensland University of Technology developed a probabilistic surfel fusion framework for dense LiDAR mapping, creating high-quality, dense surface element maps on-the-fly. The system achieved a threefold reduction in positional and normal errors and significantly lower noise compared to a state-of-the-art method in both simulated and real-world scenarios.

Key insights

• Explicitly modeling and fusing uncertainties for both surfel positions and normal vectors, particularly accounting for LiDAR sensor noise characteristics, enables more accurate and robust map updates.
• A dual map representation (sparse ESM for localization and dense DSM for reconstruction) provides a balanced approach, allowing for robust pose estimation while simultaneously building a detailed and accurate environmental model.
• Geometry-considered data association, which distinguishes between distances projected onto the surface and those along the normal, allows for adaptive resolution control and robust matching in dense LiDAR environments without relying on fixed spatial discretizations.

Summary by alphaXiv.