Graph-based ground segmentation of 3D LIDAR in rough area

This paper describes a new approach for 3D LIDAR data segmentation in rough area. As 3D LIDARs become popular equipments in robotics, processing data in real time and safely driving on challenge environments are two important problems of intelligent vehicles. For overcoming roughness and unpredictable inclination in rough area, we design a graph-based segmentation framework. Each LIDAR scan line is divided into line segments by least square linear regression. Then a Markov Random Field (MRF) energy function is built on line segment nodes. It is solved by graph-cut to classify the line segments into two categories: ground, non-ground. To validate our algorithm, experiments in typical rough environments are taken by our intelligent vehicle which, provide quantitative results. Meanwhile, we also compare it to two state-of-art segmentation methods. Experimental results show that our method performs better than the existing methods in terms of both visual and metric qualities.