SnowSTNet: A Spatial-Temporal LiDAR Point Cloud Denoising Network for Autonomous Driving in Snowy Weather

Abstract. Autonomous vehicles perceive their surroundings through sensors such as LiDAR. However, snowflakes are distributed within the detection range of LiDAR sensors in snowy weather, generating noise points that compromise the sensor's detection performance. To mitigate this issue, we propose SnowSTNet, a point cloud denoising network that removes snowflake noise from LiDAR point clouds. In SnowSTNet, we adopt a two-branch network structure that encodes information in both spatial and temporal dimensions, and inputs the features obtained from the spatial branch into the temporal branch as guidance. We conducted comparative experiments on the SnowyKITTI dataset, and the results show that our method significantly outperforms others, achieving an MIoU of 97.19%. The proposed SnowSTNet ensures the reliable operation of self-driving vehicles in snowy weather and promotes the widespread application of autonomous driving technology in complex environments.