Towards deformation monitoring with terrestrial laser scanning based on external calibration and feature matching methods

摘要

Terrestrial laser scanning (TLS) plays an important role in deformation monitoring tasks. Data quality assessment is vital to data processing, and is influenced by various error sources. In outdoor applications of TLS, systematic errors sometimes can't be removed even if a calibration is carried out before the scanning process, because atmospherical and object related characteristics may influence the performance of scanners. Additionally, not only the errors between the datum and scanner stations, but also errors between scanner stations may affect data quality. Based on classic methodology, an extended methodology has been developed for deformation monitoring with TLS. The following four aspects have been focused on: 1. Errors from atmospherical and object related characteristics influence data quality, especially in long range outdoor measurements by TLS. Calibration models related to incidence angle, beam wander and atmospherical refraction are introduced and discussed for data correction. 2. Feature matching methods have the advantages of lossless contact to object surfaces and a larger amount of correspondences when compared with using artificial targets. Images can be generated by the reflectance information of point clouds. Due to the constant relationship between the images and the point clouds, 3D matches can be determined from the collected matching pairs. Both 3D matches and artificial targets are used as identical targets for the transformation process within the registration and georeferencing process. 3. A combined model can be used by inserting the external error models into the similarity transformation model. Transformation parameters and calibration parameters are then solved simultaneously. Variance components estimation and equivalent weight matrices are introduced in a robust estimation. 4. In a uniform framework, the object surfaces are divided into small blocks and each block is estimated as a representative point. Based on representative points, rigid body movements (6 parameter transformation) and similarity transformations (7 parameter transformation) are performed to correct errors between multiple scans in the process of fine registration. An Iterative Closest Point (ICP) method is used to compute the 6 transformation parameters, and a Singular Value Decomposition (SVD) algorithm is utilized to estimate the 7 transformation parameters. Results from these two approaches are compared. This extended methodology was utilized in the monitoring of a dam surface in the Harz mountains with TLS. Eling (2009) demonstrated the existence of systematic errors with a magnitude of approximately 5 mm between scanner stations, even when the calibrations of the vertical angles and the distances were undertaken before the scanning process. The proposed extended methodology reduces these systematic errors by approximately 3.2 mm. The a posterior variance factor is decreased by approximately 3 times.