PGSD: an accurate 3D vision system using a projected grid for surface descriptions

This paper describes an active vision system using a projected grid which outputs the 3D coordinates and surface shape parameters of objects. Our work is based on the idea of making full use of the parametrization offered by the projected grid and hence all the proposed methods are curve-based. The basic and difficult problem of such a system is the correspondence problem solving. A robust and fast correspondence method is presented. The correspondence is established per curve without any ambiguity by using uniform and metric criteria in 3D space. The correspondence method relies on an accurate calibration system which proceeds by relaxation to calculate directly the sensor parameters. Two independent families of 3D curves are obtained which permit to calculate a parametrization of surfaces and to deduce a set of shape parameters. The 3D accuracy has been estimated on a manufactured object and the precision is one part in 5000 (or an average error of 0.1%). (C) 2000 Elsevier Science B.V. All rights reserved.