Model-based determination of object position and orientation without matching

This paper presents an algorithm for computing, from a single image, the 3D position and orientation of an object whose 3D shape is known. This algorithm does not require 'matching' of the object with its image. The 3D position and orientation of the object is determined by computing the hypothetical '3D motion' of the object from a known position to the position of the observed image. The motion is iteratively estimated and updated from the values of 'observables' globally defined over the model image and the observed image. Various technical issues are discussed, and examples based on real images are shown, The technique can be applied to various manufacturing processes by industrial robots where the positioning control of machine parts is not very precise.