Surface Reconstruction from Arbitrarily Large Point Clouds
In Proceedings of the Second IEEE International Conference on Robotic Computing, (IRC-2018), 31.1.-02.2.2018, Laguna Hills, CA, IEEE Press, 2018.
Generating 3D robotic maps from point cloud data is an active field of research. To handle high resolution data from terrestrial laser scanning to generate maps for mobile robots is still challenging, especially for city scale environments. In this paper, we present an approach that allows to generate polygonal reconstructions from arbitrarily large point clouds in parallel on single PCs or computing clusters. To achieve this, we serialize the large input data into suitable chunks, that are serialized to a shared hard drive. After computation, the partial results are fused into a globally consistent reconstruction. To speed up the process, we use a GPU accelerated algorithm for surface normal estimation.