Registering Pointclouds of Polyhedral Buildings to 2d Maps

This paper presents a method for automated pointcloud-to-map registration using a plane matching technique. The registration is based on estimating a transformation between a set of planes inferred from the map and their corresponding planes extracted from the pointcloud. A plane matching algorithm is developed to search for corresponding planes in the pointcloud and map coordinate systems, and estimate the transformation between the corresponding planes. The search for correspondences takes advantage of an initiate-and-extend strategy to avoid high computational cost of an extensive search. The search strategy is further strengthened by using a linear model for the estimation of the transformation. The plane matching algorithm is shown to perform robustly in the presence of outlier and missing planes, and achieve accuracies in the order of centimetres as the mean distance between the transformed pointcloud planes and the map planes.