Cylindrical and Polygonal Object Modelling and its use in LiDAR Calibration and Point Cloud Registration

Abstract

LiDAR systems are optical metrological instruments that capture surfaces of objects as highly redundant sets of discrete points (known as point clouds) in a 3D coordinate system from which spatial information can be extracted to support many applications. The accuracy of the LiDAR measurements can be improved by performing appropriate calibration. In this thesis, two novel cylinder-based calibration methods are presented for recovering interior systematic errors of different types of the LiDAR systems. The first method is a cylinder-based self-calibration technique which primarily uses point clouds of vertical cylindrical features captured from several static scan locations. The method is suitable for LiDAR systems with time-invariant errors. The second calibration method is for multi-beam spinning LiDAR systems. It allows frequent, repeated calibrations to be performed in either static or kinematic scanning mode for recovering the time-varying interior errors. For the calibration in kinematic mode, roadside power and lamp poles are used as the calibration references. In addition to these, a new geometric model of octagonal lamp poles along with a new model-driven point cloud registration method is proposed. The new geometric model uses the rotational symmetry property of the polygon to overcome the challenges of modelling the polygonal periphery using a single equation instead of piecewise functions. The proposed registration method is based on the model and requires only point clouds of a single octagonal lamp pole as registration primitives, and no actual overlap between the point clouds of the pole captured at different scan locations is needed. Each proposed method was individually verified with several real datasets, and most of them were also tested with some simulated datasets. The results suggest that all the proposed methods are practical and also offer improvements compared to the existing methods. The main contribution of this work is that many cylindrical and polygonal objects already exist in the built environment can now be used for sensor calibration, point cloud registration, and some other potential new applications as discussed in the last chapter of this thesis.