Volume XLII-4/W16
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-4/W16, 425–432, 2019
https://doi.org/10.5194/isprs-archives-XLII-4-W16-425-2019
© Author(s) 2019. This work is distributed under
the Creative Commons Attribution 4.0 License.
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-4/W16, 425–432, 2019
https://doi.org/10.5194/isprs-archives-XLII-4-W16-425-2019
© Author(s) 2019. This work is distributed under
the Creative Commons Attribution 4.0 License.

  01 Oct 2019

01 Oct 2019

ORIENTATION-BASED PAIRWISE COARSE REGISTRATION WITH MARKERLESS TERRESTRIAL LASER SCANS

S. N. Mohd Isa1, S. A. Abdul Shukor1, N. A. Rahim1, I. Maarof2, Z. R. Yahya3, A. Zakaria1, A. H. Abdullah1, and R. Wong4 S. N. Mohd Isa et al.
  • 1School of Mechatronic Engineering, Universiti Malaysia Perlis, 02600 Arau, Perlis, Malaysia
  • 2Faculty of Architecture, Planning and Surveying, UiTM, 02600 Arau, Perlis, Malaysia
  • 3Institute of Engineering Mathematics, Universiti Malaysia Perlis, 02600 Arau, Perlis, Malaysia
  • 4Geodelta System Sdn. Bhd., 22, Jalan SS20/11, Damansara Utama, 47400 Petaling Jaya, Selangor, Malaysia

Keywords: Point Cloud, Terrestrial Laser Scanner, Pairwise Registration, 3D Keypoints, Heritage Building, Keypoint Extraction

Abstract. In this paper, pairwise coarse registration is presented using real world point cloud data obtained by terrestrial laser scanner and without information on reference marker on the scene. The challenge in the data is because of multi-scanning which caused large data size in millions of points due to limited range about the scene generated from side view. Furthermore, the data have a low percentage of overlapping between two scans, and the point cloud data were acquired from structures with geometrical symmetry which leads to minimal transformation during registration process. To process the data, 3D Harris keypoint is used and coarse registration is done by Iterative Closest Point (ICP). Different sampling methods were applied in order to evaluate processing time for further analysis on different voxel grid size. Then, Root Means Squared Error (RMSE) is used to determine the accuracy of the approach and to study its relation to relative orientation of scan by pairwise registration. The results show that the grid average downsampling method gives shorter processing time with reasonable RMSE in finding the exact scan pair. It can also be seen that grid step size is having an inverse relationship with downsampling points. This setting is used to test on smaller overlapping data set of other heritage building. Evaluation on relative orientation is studied from transformation parameter for both data set, where Data set I, which higher overlapping data gives better accuracy which may be due to the small distance between the two point clouds compared to Data set II.