Point cloud processing strategies for noise filtering, structural segmentation, and meshing of ground-based 3D Flash LIDAR images

It is now the case that well-performing flash LIDAR focal plane array devices are commercially available. Such devices give us the ability to measure and record frame-registered 3D point cloud sequences at video frame rates. For many 3D computer vision applications this allows the processes of structure from motion or multi-view stereo reconstruction to be circumvented. This allows us to construct simpler, more efficient, and more robust 3D computer vision systems. This is a particular advantage for ground-based vision tasks which necessitate real-time or near real-time operation. The goal of this work is introduce several important considerations for dealing with commercial 3D Flash LIDAR data and to describe useful strategies for noise filtering, structural segmentation, and meshing of ground-based data. With marginal refinement efforts the results of this work are directly applicable to many ground-based computer vision tasks.