LiDAR-Based Long-Term Mapping in Snow-Covered Environments.

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Autonomous driving systems struggle with environmental uncertainties, particularly dynamic elements like snow.
• Accumulated snow degrades map accuracy and localization performance due to temporal changes and occlusion.

Purpose of the Study:

• To develop a robust mapping strategy for snow-covered environments.
• To mitigate localization errors and improve map update accuracy in the presence of snow.

Main Methods:

• A deep learning approach for detecting and removing accumulated snow from sensor data.
• Map updating using snow-free data and restoration of occluded ground information.

Main Results:

• Achieved 78.6% Intersection over Union (IoU) for snow detection.
• Reduced map alignment errors by 12.5% (RMSE) and 15.6% (Chamfer Distance).
• Validated effectiveness in real-world snow-covered environments.

Conclusions:

• The proposed method effectively handles snow-related mapping challenges.
• Enables reliable long-term autonomous driving in snow-covered regions by maintaining map quality.