A method for estimating object detection probability, lateral resolution, and errors in 3D-LiDARs

3D-LiDARs are nowadays used for many applications, the success of which certainly depends on the processing of the LiDAR output—the point cloud, PC,—but it also inexorably depends on the quality of the PC data. In this study, we propose an experimental method aimed at allowing estimating the errors and deformations that will statistically affect the LiDAR output — the PC. Taking advantage of the fact that LiDARs sample the surrounding space by observing it along divergent lines, hereinafter referred to as rays, this study proposes a simple method based on the experimental determination of the ray detection probability — the probability that a single ray detects the hit object, or a fraction of it, by adding a point in the point cloud. All other probabilities of interest are derived from such a probability. The proposed method also allows highlighting unexpected errors such as cross-talk. As will be shown by the examples given, due to cross-talk, small objects may be deformed and enlarged on a significantly greater number of points in the PC. Likewise, objects angularly separated by an angle greater than the angular resolution declared by the manufacturer may unexpectedly result in a continuum of points. Such errors may compromise the ability to perform very important tasks such as detection, classification, and tracking of dynamic and static objects, as well as the partition of the scene into drivable and non-drivable regions and the path planning around generic obstacles in 3D space.