We present a method for automatically registering multi-view range images. For each range image in the data set, corresponding point sets of its feature points are established, on the rest of the scans, by matching regional point descriptors. Then we build a statistic model to estimate the likelihood of the image overlapping with each of the rest ones to select candidate overlapping range images. Candidate overlapping pairs of range images are verified and registered by enforcing geometric constraints on point matches and ICP algorithm. Finally the absolute positions of all range images are established and fine-tuned. Experiment result demonstrates efficiency of the method in picking up potential overlapping images, and establishing initial transformation between overlapping pairs. The proposed method offers a good solution to automatically building complete 3D model of objects from unordered 3D scan data sets.

Date Published: 13 October 2008
PDF: 7 pages
Proc. SPIE 7129, Seventh International Symposium on Instrumentation and Control Technology: Optoelectronic Technology and Instruments, Control Theory and Automation, and Space Exploration, 71290Q (13 October 2008); doi: 10.1117/12.807637

Published in SPIE Proceedings Vol. 7129:
Seventh International Symposium on Instrumentation and Control Technology: Optoelectronic Technology and Instruments, Control Theory and Automation, and Space Exploration
Jiancheng Fang; Zhongyu Wang, Editor(s)