A Generalized Structure from Motion Framework for Central Projection Cameras

Abstract

This paper presents a novel Structure from Motion (SfM) framework designed for central projection cameras. The goal is to support future large scale multi-view 3D reconstruction algorithms. We believe that these algorithms will be able to benefit from several different sources of visual information. Accordingly, SfM approaches will need to handle this variety of image sources, such as perspective, wide-angle and spherical images. However, this issue has not yet been addressed. Current state of the art techniques are not able to handle heterogeneous images simultaneously. Therefore, we introduce SPHERA, a generalized SfM framework designed for central projection cameras. By adopting the unit sphere as underlying model it is possible to treat single effective viewpoint cameras in a unified way. We validate our framework on synthetic and real datasets. Results show that SPHERA is a powerful framework to support upcoming algorithms and applications on large scale 3D reconstruction.