Terrestrial Laser Scanning Technology from Calibration to Registration with Respect to Deformation Monitoring
For several years now, terrestrial laser scanning (TLS) has been applied to metrology in geodesy and survey engineering. Since TLS is able to determine the spatial coordinates of a remote object using laser light and can obtain a huge high-resolution three-dimensional (3D) data set for an object of interest from just a single scan, it has become a standard surveying procedure in architecture, engineering, and construction, with a wide range of applications.
However, the application of TLS has some limitations. On the one hand, information from TLS is always expressed as a vast point cloud of 3D coordinates with a relatively random distribution on the scanning object's surface, and therefore it is not possible to extract the exact coordinates of a position of interest directly from the raw point cloud. On the other hand, the accuracy of TLS results is limited by a variety of factors, including uncompensated instrument biases, target surface structure and material properties, atmospheric effects, residual deviations of point cloud registration, among others. Thus, the first two objectives of this thesis focus on target identification and calibration in TLS. These two steps belong to the pre-processing of the point cloud. In the second half of this thesis, I will be concerned with post-processing of point clouds and the application of TLS. The topic most often discussed with regard to point clouds is 3D reconstruction (e.g., the digital city). The key technique in this area is rigid registration of point clouds (i.e., the assumption that two or more 3D point clouds are related by a rigid transformation). The goal of rigid registration is to align two or more point clouds that have been captured from different scanning perspectives. This thesis will go into the issue of rigid registration and provide a new algorithm to carry out this process. The final objective in this thesis is non-rigid registration. This is one of the state-of-the-art approaches in the post-processing of point clouds and has real potential for enabling TLS to handle the problems of deformation monitoring. Specifically, in this chapter of the thesis, deformation related information is not based on a single point from different epochs. Moreover, based on non-rigid registration, a real 3D deformed model can become a reality, with the ability to provide clear benefits compared with the 2.5D deformed models used in recent projects.
This dissertation is based on four scientific publications, which have been framed by an introduction and a concluding chapter. Publication 1 focuses on target identification in TLS. In this paper, we investigate the use of normal information from all kinds of scanners to calculate the center of a target. In order to verify the generality of the results, we exploit A4 paper targets in the tests, instead of the special material targets supplied with particular brands of scanners. Publication 2 describes TLS calibration. Based on a systematic error model and a stochastic model, we derive different criteria for estimating the precision of each additional parameter, as well as correlations between parameters. After having generated such criteria, we search for different configurations to satisfy these requirements. Publications 3 and 4 concentrate on registration issues. Rigid and non-rigid cases are discussed separately in these papers. In the rigid case, we provide a variant least squares 3D surface-matching algorithm to deal with different kinds of observational errors in both source and target point clouds and further reveal the possibility of analyzing the error behavior of each point cloud. In the non-rigid case, we display the potential of the four-point congruent set algorithm to generate correspondence from deformed surfaces. An automatic method to execute non-rigid registration is proposed in publication 4, in which we further show the potential of the proposed method for 3D reconstruction and deformation monitoring.