A New Grid Map Construction Method for Autonomous Vehicles

Abstract

Grids map provide a simple but accurate way of understanding surroundings which means they could play a vital role in methods of environment perception. A new grid map construction approach to environment perception aimed at unmanned ground vehicles is proposed in this paper. First, the pose of the raw point cloud from the LiDAR system (Velodyne HDL-32E) is aligned by introducing data from IMU. Then, the RANSAC algorithm is utilized to remove the ground part of the point cloud and a grid map is established using octrees. The probability of occupancy grid map is updated based on data fusion with Bayesian inference and the Dezert–Smarandache theory combination rule. Finally, a cluster analysis is performed and moving objects are detected on the grid map, in order to facilitate obstacle detection and selection of the accessible road area. Experimental results show that the resulting grid map based on octrees and data fusion can be reliably applied to vehicle perception and that the approach is highly practicable.