Using Model-Based Feature Extraction for Uncalibrated Visual Guided Grasping

verfasst von

Oliver Hornung, Bodo Heimann

Abstract

In this paper an approach to the feature correspondence problem for image based visual servoing systems is presented. Using an object model calculated off-line, robust image features of a partial scene reconstruction are extracted for tracking by model registration. The registration between model and reconstruction is processed in two stages. The first stage copes with the problem of coarse structural alignment while in the second stage an alternation between reconstruction scale optimization and point registration is adopted for finetuning. As a consequence the reconstruction does not have to be calculated by calibrated cameras which makes the approach suitable for systems with only uncertain intrinsic and extrinsic camera parameters. Experimental results verify the effectiveness of the approach applied to a grasping task with an uncalibrated zoom camera and a 7 DOF uncalibrated manipulator in an eye-in-hand configuration.

Organisationseinheit(en)

Institut für Mechatronische Systeme

Typ

Aufsatz in Konferenzband

Seiten

436-441

Anzahl der Seiten

6

Publikationsdatum

2004

Publikationsstatus

Veröffentlicht

Peer-reviewed

Ja