Share Email Print

Proceedings Paper

Multiscale registration algorithm for alignment of meshes
Format Member Price Non-Member Price
PDF $17.00 $21.00

Paper Abstract

Taking a multi-resolution approach, this research work proposes an effective algorithm for aligning a pair of scans obtained by scanning an object's surface from two adjacent views. This algorithm first encases each scan in the pair with an array of cubes of equal and fixed size. For each scan in the pair a surrogate scan is created by the centroids of the cubes that encase the scan. The Gaussian curvatures of points across the surrogate scan pair are compared to find the surrogate corresponding points. If the difference between the Gaussian curvatures of any two points on the surrogate scan pair is less than a predetermined threshold, then those two points are accepted as a pair of surrogate corresponding points. The rotation and translation values between the surrogate scan pair are determined by using a set of surrogate corresponding points. Using the same rotation and translation values the original scan pairs are aligned. The resulting registration (or alignment) error is computed to check the accuracy of the scan alignment. When the registration error becomes acceptably small, the algorithm is terminated. Otherwise the above process is continued with cubes of smaller and smaller sizes until the algorithm is terminated. However at each finer resolution the search space for finding the surrogate corresponding points is restricted to the regions in the neighborhood of the surrogate points that were at found at the preceding coarser level. The surrogate corresponding points, as the resolution becomes finer and finer, converge to the true corresponding points on the original scans. This approach offers three main benefits: it improves the chances of finding the true corresponding points on the scans, minimize the adverse effects of noise in the scans, and reduce the computational load for finding the corresponding points.

Paper Details

Date Published: 4 March 2004
PDF: 7 pages
Proc. SPIE 5263, Intelligent Manufacturing, (4 March 2004); doi: 10.1117/12.539994
Show Author Affiliations
Srikanth Vadde, Northeastern Univ. (United States)
Sagar V. Kamarthi, Northeastern Univ. (United States)
Surendra M. Gupta, Northeastern Univ. (United States)

Published in SPIE Proceedings Vol. 5263:
Intelligent Manufacturing
Bhaskaran Gopalakrishnan; Peter E. Orban; Angappa Gunasekaran, Editor(s)

© SPIE. Terms of Use
Back to Top