Dynamic Road Boundary Detection Using Sparse 3D-LiDAR in Complex Road and Environmental Conditions

Abstract

The integration of autonomous vehicles into the mainstream transportation system requires advanced perception technologies to accurately detect and navigate road boundaries. This study addresses the challenge by utilizing a cost-effective 16-beam light detection and ranging (LiDAR) sensor combined with robust data processing techniques. The experiments were conducted in various environments using an instrumented vehicle. Ground and non-ground points were segregated using the random sample consensus (RANSAC) algorithm. The road boundary detection algorithm utilized the shortest-beam method and median filtering. This approach was used to enhance detection accuracy even in complex road and environmental conditions, such as different types of intersection, adverse weather, and varying terrain and lighting conditions. The RANSAC polynomial fitting method was used to fit the road boundaries. This methodology demonstrates significant improvements in identifying road boundaries, proving to be a viable alternative to more expensive, high-resolution LiDAR systems. Experimental results highlight the stability of the proposed approach, showing its potential to facilitate the broader adoption of autonomous driving technology by making it more economically accessible and scalable.