Automatic merging of LIDAR three-dimensional point clouds using inertial navigation systems and global positioning systems data.

摘要

A Light Detection and Ranging system (LIDAR) is used to collect 3D data that can be used for terrain modeling. The LIDAR gathers scans and these scans are then merged together to map a terrain. This thesis discusses the merging problem, and how the time to correctly merge or register point clouds can be reduced dramatically.; The Iterative Closest Point (ICP) algorithm is used to register and merge together the eight scans taken by the 3D LIDAR. As a part of this step, a method called Sphere Outlier Removal (SOR) was formulated to accurately identify outliers and inliers, a necessary prerequisite to using the ICP algorithm. A K-Dimensional (KD) Tree was implemented to dramatically increase the speed of merging two scans together. The RANdom Sampling Consensus (RANSAC) algorithm was introduced and a new algorithm, a variation of RANSAC, called Random Sampling (RSAP), is developed and is shown to increase speed of the registration process by an order of magnitude.