ScatterHough: Automatic Lane Detection from Noisy LiDAR Data.

Area of Science:

• Computer Vision
• Robotics
• Autonomous Systems

Background:

• Lane detection is critical for autonomous driving systems.
• LiDAR data offers advantages over RGB images, especially in adverse weather and lighting.
• Handling noisy, scattered LiDAR data is a key challenge in lane detection.

Purpose of the Study:

• To propose an improved algorithm for lane detection using LiDAR data.
• To address the limitations of the classic Hough Transform for scattered data.
• To enhance the accuracy and robustness of lane detection in real-time applications.

Main Methods:

• A novel Scatter Hough algorithm is introduced for lane detection.
• The method incorporates 'ρ neighbor voting' and 'ρ neighbor vote-reduction' into the Hough Transform.
• These operations enable points on curves to vote and consider neighbor information.

Main Results:

• The proposed Scatter Hough algorithm demonstrates high accuracy in fitting both straight lines and curves.
• The method effectively handles scattered and noisy LiDAR data.
• Performance is validated against benchmark and state-of-the-art lane detection techniques.

Conclusions:

• The Scatter Hough algorithm provides a robust and accurate solution for LiDAR-based lane detection.
• The approach is suitable for real-time autonomous driving applications, even in poor conditions.
• The enhanced Hough Transform effectively addresses the challenge of scattered data in lane detection.