3D Ground Point Classification for Automotive Scenarios

Autonomous driving applications must be provided with information about other road users and road side infrastructure by object detection modules. These modules often process point clouds sensed by light detection and ranging (LiDAR) sensors. Within the captured point cloud a large amount of points correspond to physical locations on the ground. These points do not hold information about road users, obstacles or road side infrastructure. Thus an important preprocessing step is identifying ground points to allow the object detection focusing on relevant measurements only. Within this paper we propose a ground point classification which relies on simple but effective geometric features. We evaluate the accuracy of the proposed algorithm on simulated data of different traffic scenarios. In addition, we evaluate the effectiveness of this preprocessing step based on the achieved speed up of an object detection algorithm on real world data.