A practical new approach to 3D scene recovery

An efficient new approach to 3D scene reconstruction from an uncalibrated image sequence without anything known about the scene and the cameras is proposed. Present algorithms cannot deal with cameras with aspect ratio deviated from 1.0 well. Camera intrinsic parameters are skillfully handled in the cost function of the self-calibration phase to dispose the variable aspect ratio. Compared with existing algorithms, both intrinsic parameters of cameras and feature points of the scene, can be recovered efficiently, even for the cameras with aspect ratio greatly deviated from 1.0, like 0.6 and 2.3 in practice. Both synthetic and real data have been used to test the proposed method. The method is verified to be more accurate and practical than state-of-the-art algorithms, even if the aspect ratio is 2.5. Moreover, the reconstruction does not require any further non-linear optimization for general applications.