The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

Research Paper #573

Title:A 3D Shape Attractor
Date: 1992
Presented:Submitted to the 23rd International Symposium on Industrial Robotics, Barcelola, 1992
Abstract:A three-dimensional shape attractor tries to match a 3D "template" model to a 2D image of a part. The 3D template is projected onto the image plane and the outline polygon found. This is compared with the image and a least-squares technique is used to estimate the "best" way to rotate, translate and scale the template so that it fits the image. The template is rotated, scaled and translated and its new projected image is compared with the part outline once again. Although the estimate of the "best rotation" is crude, the errors are self-limiting and iteration ensures that an exact solution is quickly found. The 3D shape attractor is robust and its output is a transformation, not a similarity measure. This makes it particularly suitable for the control of robot motion. A robot programming example demonstrates how the algorithm can be used in part acquisition. The computer model of the part carries "handles" which are used to indicate where the image of the robot hand should be so the hand can pick up the part.

[Search These Pages] [DAI Home Page] [Comment]