Radar-Driven Occupancy Grid Maps for Robust Perception in Adverse Fog Conditions

Adverse weather conditions significantly challenge scene understanding of autonomous vehicles by degrading perception system performance. Despite advancements, the impacts of such conditions still require further investigation. In this paper, we propose a novel approach that leverages the advantages of Radar sensors against various weather types to improve robustness under foggy weather. Our framework refines Radar measurements through the Bayesian Filtering algorithm to enhance data quality and sparsity, generating informative representations as probabilistic Occupancy Grid Maps (OGM). We address sensor synchronization challenges to ensure accurate fusion across different modalities. We evaluate our framework and the effectiveness of the OGMs fused with LiDAR data under various fog densities. The experimental results demonstrated improvements in robustness and reduced detection errors when compared with LiDAR-only object detection.