A surface reconstruction neural network for absolute orientation problems

Abstract

The authors propose a neural network for representation and reconstruction of 2-D curves or 3-D surfaces of complex objects with application to absolute orientation problems of rigid bodies. The surface reconstruction network is trained by a set of roots (the points on the curve or the surface of the object) via forming a very steep cliff between the exterior and interior of the surface, with the training root points lying in the middle of the steep cliff. The Levenberg-Marquardt version of Gauss Newton optimization algorithm was used in the backpropagation learning to overcome the problem of local minima and to speed up the convergence of learning. This representation is then used to estimate the similarity transform parameters (rotation, translation, and scaling), frequently encountered in the absolute orientation problems of rigid bodies.<>