作者: Michal Havlena , Tomáš Pajdla , Kurt Cornelis
DOI:
关键词: Structure from motion 、 RANSAC 、 3D reconstruction 、 Pixel 、 Omnidirectional antenna 、 Computer science 、 Camera resectioning 、 Field of view 、 Tracking (particle physics) 、 Artificial intelligence 、 Computer vision
摘要: This paper deals with a step towards 3D reconstruction system for city modeling from omnidirectional video sequences using structure motion together stereo constraints. We concentrate on two issues. First, we show how the tracking and paradigm were adapted to use images taken by lenses 180 degrees field of view. concerns mainly camera calibration transforming pixel locations into rays solving minimal problem 3D-to-2D matches RANSAC. Secondly, compare results additional constraints when these are not used that they needed make stable. Performance is demonstrated sequence 870 acquired while driving in city.