Planar and Linear Feature-Based Registration of Terrestrial Laser Scans with Minimum Overlap Using Photogrammetric Data

Abstract

Three Dimensional (3D) modeling is crucial for studying, analyzing, reconstructing, and documenting our environment, in general, and man-made structures, in particular. 3D modeling of structures of interest can be directly established by a Terrestrial Laser Scanner (TLS). Nevertheless, several TLS scans from different positions/orientations are necessary to have a complete coverage of the surveyed structure. Transformation of the collected scans into a common coordinate system through a registration procedure is necessary in order to have a meaningful 3D model of the structure in question. The registration process requires large overlap area among the TLS scans for reliable estimation of the transformation parameters. In this paper, the large overlap area requirement between the TLS scans is reduced/eliminated using a photogrammetric data, which can be economically and quickly collected, as additional information for the registration process. Different registration primitives can be used to estimate the transformation parameters among the different scans. Manmade structures are rich with linear and planar features. Moreover, linear and planar features can be reliably extracted from both photogrammetric data and TLS scans. Therefore, in this research, planar and linear features are chosen as the registration primitives. Quality control of the aligned scans is necessary to evaluate the registration results and to compare the performance of linear and planar feature as possible primitives. In this research, quality control of the registration results is qualitatively and quantitatively evaluated. The qualitative quality control of the registration outcome is conducted by plotting and visual inspection of registered laser scans. Quantitative quality analysis, on the other hand, is conducted by calculating the point-to-plane normal distances between the photogrammetrically reconstructed models and TLS scans after the registration process, and also between some of the TLS scans if there are any available overlap areas. In addition, the results from a variant of the commonly used ICP registration are compared with those derived from the proposed registration. Experimental results from a real dataset will show the feasibility of the proposed technique where less than 10 cm point-to-plane normal distance between the registered surfaces has been observed.