Automatic hierarchical registration of aerial and terrestrial image-based point clouds

Abstract

ABSTRACTPoint cloud registration has been a major research challenge in recent years. In this paper, a novel hierarchical method is proposed for registering Aerial and Terrestrial image-based Point Clouds (APC & TPC). This registration aims to yield a complete and dense coverage on both top and side faces in urban areas. The main challenges, however, arise from their heterogeneous views and very low overlap as well as the high discrepancies in scale and three-dimensional rotations between the World Coordinate System (WCS) of APC and the Camera Coordinate System (CCS) of TPC. The proposed method begins by an innovative TPC-sufficient method to solve two rotations of TPC around x- and y-axis. After that, in horizontal registration phase, 2D façade lines were extracted from both datasets; and accordingly a novel polar parameterised mathematical model was extended for simultaneous robust matching and parameter estimation. Finally, vertical registration was done through calculating the dominant vertical shift between the points of the ground planes extracted from both datasets. The evaluation results in two different modes, which were conducted on two different urban datasets, showed the efficiency of the proposed method (RMSE error of 0.12 m/0.10 m in horizontal/vertical direction).