Fast and Accurate Algorithms for Projective Multi-Image Structure from Motion

We describe algorithms for multi{image structure from motion from tracked point features. The algorithms are essentially linear, work for any motion of moderate size, and give accuracies similar to those of a maximum{likelihood estimate. They are as fast as the Sturm/Triggs approach and give better results. We show experimentally that the Sturm/Triggs approach often fails for linear camera motions, especially for forward or backward motions. For the common situation where the calibration is xed and approximately known, we experimentally compare the projective versions of our algorithms to mixed projective/Euclidean strategies. We clarify the nature of dominant{plane compensation, showing that it can be considered as a small{translation approximation rather than an approximation that the scene is planar. We show that algorithms can accurately recover the (projected) inverse depths and homographies despite projective covariance.