Parametric Surface Fitting on Airborne Lidar Point Clouds for Building Reconstruction

Surface reconstruction is an essential step in most processing pipelines involving point clouds. By constructing a surfacic or volumetric model of the cloud, it possible to infer large-scale semantic and geometric information required for applications computer graphics, simulation virtual reality. Among different types cloud variety problems, we are interested airborne Lidar data problem building urban planning ranging from flood light exposure touristic visits. While existing methods based on characteristic features extraction clouds such as planes, ridges, contours their combination into more complex model, instead adopt template-based approach relying library primitives developed industrial context. This formulated global fitting between constrained triangulated mesh (our template) cloud. More precisely, design energy function that takes account distance both objects while integrating outliers rejection directly our numerical optimization through use M-estimator. being smooth everywhere, can be efficiently minimized by quasi-Newtonian like L-BFGS algorithm. We demonstrate reliability collection diverse roof models several publicly available datasets, well its robustness limits initialization, quality presence outliers. only onto relevant points, this method allows precise correct outlier segmentation unique step, providing reasonable initialization close barycenter