Abstract

3D reconstruction with missing data has been a very challenging computer vision task since the late 90s. This paper proposes the GPU-powered version of our weak-perspective Structure from Motion algorithm published in BMVC2008. Although this method is iterative, it is very rapid since all substeps in each iteration minimize the parameters optimally with respect to the reprojection error. We demonstrate here in both synthetic and real tests that the use of the GPU significantly reduces the time demand of the algorithm. Real-time 3D reconstruction is possible if the parameters of the optimization algorithm are set properly.

Paper available at IEEE.