Robust Ground Segmentation for Autonomous Driving Using Sparse Gaussian Process Regression

Abstract

Abstract In this paper, a robust ground segmentation algorithm based on Gaussian Process regression is proposed for autonomous driving in urban condition. Specifically speaking, the algorithm we proposed committing to improving the quality of ground segmentation under the urban conditions which including slope road, intersection and so on. There are two mainly steps in our algorithm, in the first step, the coordinates of the points which collected by the LiDAR are mapped from Cartesian coordinate system to polar coordinate system; in the second step, we introduced Gaussian Process Regression to complete high precision ground curve fitting. Among them, we added slope judgment to improve the quality of the fitting, and introduced a sparse covariance function to speed up the iteration. Experiments conducted on the KITTI dataset proved that our algorithm achieves the desired effect and provides a good foundation for further target recognition and tracking.