Spatially divided outlier removal for LiDAR de-noising in adverse weather conditions.

Adverse weather conditions, such as rain and snow, significantly degrade the performance of Light Detection and Ranging (LiDAR) perception systems, creating a demand for real-time and high-precision denoising algorithms. This paper proposes a novel LiDAR point cloud denoising method, Spatially Divided Outlier Removal (SDOR), which spatially divides the point cloud into multiple sectors and processes them in parallel to reduce execution time. Furthermore, SDOR dynamically adjusts the search radius based on the point density within each sector, enabling more accurate noise removal than conventional distance-based filtering. Experiments on the WADS and Weather-KITTI datasets show that SDOR achieves F1-scores of 90.53% on WADS and 90.24% (Snow) / 91.20% (Fog) on Weather-KITTI, along with error rates as low as 6.17%, 0.61%, and 2.89%, respectively. In addition, SDOR demonstrates the fastest execution times on the Weather-KITTI benchmark, recording 0.219 s in Snow, 0.241 s in Rain, and 0.229 s in Fog. These results highlight that SDOR provides both high denoising accuracy and real-time performance, making it a promising solution for reliable LiDAR perception under diverse adverse weather conditions.