An easy framework for an accurate estimation of the Structure From Motion

In this paper we introduce a simple framework for the Structure From Motion problem resolution using the classical and general formulation of the problem. We propose a two steps algorithm that does not impose any constraints on the scene. The purpose of the first step is the extraction of a convenient “Projective Reconstruction” using a simple method. Indeed, it firstly initializes the projective calibration of the scene and then refines it using non-linear optimization. The extracted reconstruction is then integrated in the second step consisting on the upgrade of the reconstruction to a metric one; this is done through the research of the camera intrinsic parameters necessary for the upgrade stage. A multi-stage auto-calibration algorithm is proposed to iteratively estimate each parameter alone and finally refines all of them at once. Once these parameters are obtained, the 3D Euclidean reconstruction of the scene is extracted using a publicly available Multi-View Stereo software.