Camera-robot transform for vision-guided tracking in a manufacturing work cell

The problem of camera calibration from the perspective of hand-eye integration (henceforth referred to as the Camera-Robot (CR) problem), is addressed in this paper. Mapping results obtained from a least-squares fit using pseudo-inverse technique and a three layer neural network are compared. The calibration matrix is developed to map the image coordinates of an IRI D256 vision processor equipped with a CCD camera directly on to the coordinates for an IBM 7540 SCARA manipulator. One transformation is obtained by performing a least-squares fit using pseudo-inverse technique on a set of one hundred data points which relates two-dimensional (2D) image coordinates to corresponding twodimensional robot coordinates. The CR problem is also approached by using the same data points on a neural network. The results not only demonstrate the ability of neural networks to 'learn' the transformation to a reasonable accuracy, but also from the basis for a relatively simple method of adaptive self-calibration of robot-vision systems. In a broader sense, the proposed method can be used to calibrate a variety of robotic sensors that are typically used in a flexible manufacturing environment. © 1992 Kluwer Academic Publishers.

Duke Scholars

Author Devendra P. Garg Thomas Lord Department of Mechanical Engineering and Materia ...