Authors:
Polycarpo Souza Neto
;
Nicolas S. Pereira
and
George A. P. Thé
Affiliation:
Department of Teleinformatic Engineering, Federal University of Ceara, Fortaleza, Pici campus, Bl 725, Zip code 60455-970 and Brazil
Keyword(s):
Computer Vision, Iterative Closest Point, Point Cloud Registration, Point Cloud Sampling.
Related
Ontology
Subjects/Areas/Topics:
Image Processing
;
Informatics in Control, Automation and Robotics
;
Robotics and Automation
;
Vision, Recognition and Reconstruction
Abstract:
In 3D reconstruction applications, an important issue is the matching of point clouds corresponding to different perspectives of a given object in a scene. Traditionally, this problem is solved by the use of the Iterative Closest point (ICP) algorithm. In view of improving the efficiency of this technique, authors recently proposed a preprocessing step which works prior to the ICP algorithm and leads to faster matching. In this work, we provide some improvements in our technique and compare it with other 4 variations of sampling methods using a RMSE metric, an Euler angles analysis and a modification structural similarity (SSIM) based metric. Our experiments have been carried out on four different models from two different databases, and revealed that our cloud partitioning approach achieved more accurate cloud matching, in shorter time than the other techniques. Finally we tested the robustness of the technique adding noise and occlusion, obtaining, as in the other tests, superior p
erformance.
(More)