Fast and Robust Ground Surface Estimation from LiDAR Measurements using Uniform B-Splines

Mapping LiDAR and Camera Measurements in a Dual Top-View Grid Representation Tailored for Automated Vehicles

“…For the condition "sup (D C ) > 0", the ground surface estimation must be very accurate to exclude all the measurements reflected on the ground. Hence, a small tolerance margin δ G > 0 is added in some publications such as [22] so that the condition becomes "sup…”

“…The calculation of surface normal vectors used in Definition 2 requires finding neighboring elements which is computational expensive. Therefore, occupancy is modeled according to Definition 1 which means that the detection point set is segmented into obstacle and ground detections based on a ground surface model such as presented in [22]. In Equation ( 12), the occupancy probability p occ is then set to one if m was classified as occupying and to zero otherwise.…”

“…This information is needed to exclude measurements outside the considered driving corridor and to clip measurement rays that intersect with the free space corridor boundaries. One option is to explicitly estimate a ground surface using a parametric model such as a plane or a 2D Bspline in the Cartesian space as in [22]. However, estimating a parametric model often comes along with solving an optimization problem with computationally time-consuming numerical solvers.…”

“…Derive evidence for occupancy according to Definition 1 as described in Section IV-A. The ground surface is represented by the uniform B-spline model proposed by Wirges et al [22] and the tolerance margin is set to δ G = 0.3m. The resulting BBA is denoted as m spline .…”

Parametric Cubic Spline Interpolating Surfaces for Arbitrary Polygonal Meshes

“…A simultaneous localization and mapping (SLAM) based technique that builds on top of a B-spline surface map was presented in [11]. A fast and robust method to estimate the ground surface from LiDAR (light detection and ranging) measurements on an automated vehicle, where the ground surface was modeled as a uniform B-spline, was proposed in [16] Neural splines for 3d surface reconstruction that is based on random feature kernels arising from ReLU (rectified linear unit) networks was presented in [17]. An adaptive state estimation method by using cubic spline interpolation for particle filter tracking was presented in [18].…”