We propose a road-surface abstraction algorithm which is suitable for structured or semi-structured road environment and extracts effectively a road-surface for ground robot navigation. A fuzzy cluster method based on maximum entropy theories was employed to cluster Lidar points. After fitting clustered data linearly

the algorithm extracts seams that belong to road-surface by their location and angle. Current referenced horizontal can be acquired by comparing several continuous Lidar frames and then the algorithm extracts the obstacles in road area. Experiments show the algorithm works well in spite of road-boundary has regular shape or not

and is free from the impact of complex texture or irregular illumination of the road.