Abstract

Abstract. 3D reconstruction of objects is a basic task in many fields, including surveying, engineering, entertainment and cultural heritage. The task is nowadays often accomplished with a laser scanner, which produces dense point clouds, but lacks accurate colour information, and lacks per-point accuracy measures. An obvious solution is to combine laser scanning with photogrammetric recording. In that context, the problem arises to register the two datasets, which feature large scale, translation and rotation differences. The absence of approximate registration parameters (3D translation, 3D rotation and scale) precludes the use of fine-registration methods such as ICP. Here, we present a method to register realistic photogrammetric and laser point clouds in a fully automated fashion. The proposed method decomposes the registration into a sequence of simpler steps: first, two rotation angles are determined by finding dominant surface normal directions, then the remaining parameters are found with RANSAC followed by ICP and scale refinement. These two steps are carried out at low resolution, before computing a precise final registration at higher resolution.

Highlights

  • With the emergence of highly automated recording techniques, automatic point cloud registration has attracted great interest in past years

  • Since iterative closest point algorithm (ICP) is a local method and requires good approximations to converge to a correct result, different methods have been proposed to find approximate registration parameters using point correspondences

  • It is inspired by the way a human operator interactively performs this task: since the operator has to work with a 2D view of the point cloud, he first rotates and translates the point clouds to get a similar perspective for both in which he can see similar shapes or features, adjusts the remaining parameters, and refines the registration

Read more

Summary

Introduction

With the emergence of highly automated recording techniques, automatic point cloud registration has attracted great interest in past years. Since ICP is a local method and requires good approximations to converge to a correct result, different methods have been proposed to find approximate registration parameters using point correspondences. Algorithms such as 3D SIFT (Li and Guskov, 2005), and SPIN images (Johnson and Hebert, 1997) extract features and match them to obtain correspondences between point clouds. A different approach is the 4-points congruent sets (4PCS) algorithm (Aiger et al, 2008), which does not attempt to extract features Rather, it randomly picks four coplanar points with large baselines and matches them based on their geometric configuration in order to obtain the registration parameters. The method requires a known scale, and that the point clouds have similar and relatively uniform point density. (Zinßer et al, 2005) proposed a method which incorporates a scale estimation into the ICP algorithm, but again is a local refinement which requires that the two models are already roughly aligned

Methods
Discussion
Conclusion

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call

Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.