Development of Robust Lane-Keeping Algorithm Using Snow Tire Track Recognition in Snowfall Situations.

Area of Science:

• Computer Vision
• Machine Learning
• Robotics

Background:

• Lane-keeping systems struggle in adverse weather like heavy snowfall.
• Existing algorithms often fail due to obscured lane markings.

Purpose of the Study:

• To develop a robust lane-keeping algorithm for snowy road conditions.
• To accurately detect snow tire tracks for reliable lane estimation.

Main Methods:

• A machine learning model (YOLOv5) was trained on custom datasets of snow tire tracks.
• The algorithm employs a two-module structure: snow tire track detection and lane center estimation.
• Techniques include binarization, bird's-eye view transformation, skeletonization, Hough transformation, and Kalman filtering.

Main Results:

• The snow tire track detection model demonstrated effectiveness in identifying tracks on snowy roads.
• The lane center estimator successfully utilized detected tracks to establish a reliable lane reference.
• The algorithm provided robust lane-keeping performance even under heavy snowfall conditions.

Conclusions:

• The proposed algorithm offers a reliable solution for lane keeping in challenging snowy environments.
• Machine learning-based tire track detection significantly enhances lane-keeping system performance in adverse weather.