Free road space estimation based on surface normal analysis in organized point cloud

This paper describes a method to estimate free space area in front of a vehicle. The free space estimation is based on surface normal analysis. Surface normal represents the orientation of a region. Naturally, a traversable area is the surface which facing upward in 3D environment or forming a slope with limited angle. Several neighboring area which has similar surface normal orientation are merged together by connected component region labeling. Finally, the regions that considered as free space area are decided by examining its size and orientation in 3D environment. The method in this paper is evaluated using KITTI stereo vision dataset. A robust global solution for free space in road surface is obtained. In a close range examination, this method can distinguish the road surface and pedestrian area which naturally has similar characteristic as free space area.