3D pose computation in robot vision applications

Abstract

The paper proposes a 3D reconstruction technique suitable in robot vision applications, based on the correspondence of image features established at a training step and the same image features extracted at the execution step. The result is a Euclidean transform to be used by robot head for reorientation at execution step. A closed form solution is proposed for (R, t) transform to initialize the non-linear optimization procedure. For applications where the object to be processed does not rotate around X or Y axis this can be the final solution.