PublicaHier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.
Anticipation-preprocessing for object pose detection
|Informationstechnische Gesellschaft -ITG-, Information Technology Society of VDE; Verband Deutscher Maschinen- und Anlagenbau e.V. -VDMA-, Fachverband Robotik und Automation, Frankfurt/Main; International Federation of Robotics; Deutsche Gesellschaft für Robotik -DGR-:|
ISR/ROBOTIK 2010, Proceedings for the joint conference of ISR 2010, 41st International Symposium on Robotics und ROBOTIK 2010, 6th German Conference on Robotics : 7-9 June 2010 - Parallel to AUTOMATICA
Berlin: VDE-Verlag, 2010
|International Symposium on Robotics (ISR) <41, 2010, Munich>|
German Conference on Robotics (ROBOTIK) <6, 2010, Munich>
Internationale Fachmesse für Automation und Mechatronik (Automatica) <4, 2010, Munich>
|Fraunhofer IPA ()|
| preprocessing; Lageerkennung; bin-picking; Bildverarbeitung; Objekterkennung; Algorithmus|
This paper shows a new approach to significantly improve the runtime behaviour of existing object pose detection algorithms. The idea of anticipation-preprocessing is to reduce possible object poses up to a minimum by limiting the solution space. The assumption is that a smaller solution space comes along with a much faster evaluation. Anticipation-preprocessing starts with the most obvious features of sensory input. The treatment of features takes gestalt law of organization into account. Based on that, the procedure is to find new features and to keep recombining them. With the aid of the achieved feature sets a conclusion about a solution subspace can be drawn. Anticipation-preprocessing is inspired by human visual perception, whereof several theories exist. Comparing to one of the most famous by Marr, memorization influences the entire process of the new approach and not just the final matching. The new approach was tested with an object pose detection algorithm calledCoherent-Distance-Characteristics (abbr.: CDC). CDC Object Pose Detection is a view based approach which uses compact object description and still provides strong reliability. It has been originally developed for bin-picking applications by the authors. CDC works on spatial depth data. It was shown that the use of just one or two features already results in a runtime reduction of about 90 percent.