Open AccessObject reconstruction from multiple uncontrolled viewpoints
Author(s) -
Daniek Joubert
Publication year - 2011
Publication title -
suid-afrikaanse tydskrif vir natuurwetenskap en tegnologie/die suid-afrikaanse tydskrif vir natuurwetenskap en tegnologie
Language(s) - English
Resource type - Journals
eISSN - 2222-4173
pISSN - 0254-3486
DOI - 10.4102/satnt.v30i1.262
Subject(s) - bundle adjustment , computer vision , artificial intelligence , triangulation , viewpoints , object (grammar) , feature (linguistics) , sequence (biology) , 3d reconstruction , computer science , structure from motion , motion (physics) , geography , image (mathematics) , cartography , biology , art , visual arts , linguistics , philosophy , genetics
Rekonstruksie van voorwerpe is ’n belangrike probleem in rekenaarvisie en het baie toepassings.
Dit kan byvoorbeeld in robot-navigasie, vlugtige prototipering, toegevoegde werklikheid
(‘augmented reality’) en die skep van virtuele wêrelde (soos Google Street View) gebruik word.
We present a feature-based approach to reconstruct unknown objects in 3D, using only a sequence of images from multiple views. The camera motion between consecutive images is estimated and a reconstruction is obtained by triangulation. This approach is susceptible to errors, and we investigate bundle adjustment as a viable remedy.