Sparse Gaussian process regression based ground segmentation for autonomous land vehicles

Abstract

Ground segmentation, which is the base of the successive object detection and recognition, is a crucial component for Autonomous Land Vehicles (ALV) equipped with Velodyne LIDAR navigation in outdoor environments. This paper presents a novel algorithm based on sparse Gaussian Process Regression (GPR) for segmenting three-dimensional scans of various terrains. The 3D points of a scan are firstly mapped into 3D grid map, and then iterative two-dimensional GPR with sparse covariance function is exploited to model the ground surface directly. In order to verify the performance of our approach, It has been compared with two previous ground segmentation techniques on the data collected by our own ALV in different outdoor scenes. The results show that our approach can obtain promising performance.