Developing and Validating a Predictive Model of Measurement Uncertainty for Multi-Beam Lidars: Application to the Velodyne VLP-16
Abstract
A key feature for multi-sensor fusion is the ability to associate, to each measured value, an estimate of its uncertainty. We aim at developing a point-to-pixel association based on UAV-borne LiDAR point cloud and conventional camera data to build digital elevation models where each 3D point is associated to a color. In this paper, we propose a convenient uncertainty prediction model dedicated to multi-beam LiDAR systems with a new consideration on laser diode stack emitted footprints. We supplement this proposition by a novel reference-free evaluation method of this model. This evaluation method aims at validating the LiDAR uncertainty prediction model and estimating its resolving power. It is based on two criteria: one for consistency, the other for specificity. We apply this method to the multi-beam Velodyne VLP-16 LiDAR. The sensor's prediction model validates the consistency criterion but, as expected, not the specificity criterion. It returns coherently pessimistic prediction with a resolving power upper bounded by 2 cm at a distance of 5 m.