Dynamic Vertical and Low-Intensity Outlier Removal for Efficient Snow Noise Removal from LiDAR Point Clouds in Adverse Weather

As autonomous driving technology advances, the reliability of LiDAR sensors in adverse weather conditions, such as snow, has become a pressing concern. Traditional denoising algorithms have limited effectiveness in handling snow noise, making it challenging to distinguish dynamic noise points from environmental features. Researchers at Shanghai Dianji University have proposed a novel algorithm, Dynamic Vertical and Low-Intensity Outlier Removal (DVIOR), designed specifically to optimize LiDAR point cloud data under snowy conditions.

Key Takeaways:

• The DVIOR algorithm combines intensity and vertical height information to dynamically adjust filter parameters, effectively filtering out snow noise while preserving environmental features.
• Experiments on publicly available datasets, including the Winter Adverse Driving Scenarios (WADS), the Canadian Adverse Driving Conditions (CADC), and the Radar Dataset for Autonomous Driving in Adverse weather conditions (RADIATE), demonstrated notable improvements over existing methods.
• Compared to the mainstream dynamic distance-intensity hybrid algorithm (DDIOR) and the representative intensity-based filter (LIOR), DVIOR achieved a 10.2-point higher F1-score on the WADS dataset and an 11.8-point higher F1-score on the same dataset.
• On the CADC and RADIATE datasets, DVIOR achieved F1-scores of 87.35 and 86.68, representing an improvement of 19.82 and 36.9 points over DDIOR and 4.67 and 17.95 points over LIOR, respectively.
• The DVIOR algorithm outperforms existing methods in snow noise removal, particularly in complex snowy environments.

Statistics:

• 10.2-point higher F1-score on the WADS dataset compared to DDIOR
• 11.8-point higher F1-score on the WADS dataset compared to LIOR
• 79.00 F1-score on the WADS dataset with DVIOR
• 87.35 F1-score on the CADC dataset with DVIOR
• 86.68 F1-score on the RADIATE dataset with DVIOR
• 19.82-point improvement over DDIOR on the CADC dataset
• 36.9-point improvement over LIOR on the RADIATE dataset

Sources:

• Dvior: Dynamic Vertical and Low-intensity Outlier Removal for Efficient Snow Noise Removal From Lidar Point Clouds In Adverse Weather. Electronics, 2025;14(18):3662
• Electronics can be contacted at: Mdpi, St Alban-Anlage 66, Ch-4052 Basel, Switzerland
• Fanhao Kong, Shanghai Dianji University, School of Mechanical Engineering, Shanghai 201306, People's Republic of China
• Guanqiang Ruan, Kuo Yang, Tao Hu, Chenglin Ding, and Rong Yan (additional authors)