Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

Combining synthetic image acquisition and machine learning: Accelerated design and deployment of sorting systems

: Retzlaff, M.-G.; Richter, M.; Längle, Thomas; Beyerer, Jürgen; Dachsbacher, C.

Heizmann, M. (Hrsg.); Längle, Thomas (Hrsg.); Puente Leon, F. (Hrsg.):
Forum Bildverarbeitung 2016 : 01.-02. Dezember 2016, Karlsruhe
Karlsruhe: KIT Scientific Publishing, 2016
ISBN: 978-3-7315-0587-7
ISBN: 3-7315-0587-8
Forum Bildverarbeitung <2016, Karlsruhe>
Fraunhofer IOSB ()
optical inspection; computer graphics; image synthesis; procedural modeling; pattern recognition; classification; glass sorting

Machine learning methods can automate the design of large parts of an image processing pipeline in automated optical inspection (AOI) systems. However, these methods typically require an annotated sample of the objects under inspection, and creating such samples is still a manual and labor-intensive process. Synthetic image acquisition (SIA) can fill the gap to automate this step. SIA joins a physically-based image synthesis pipeline and procedural modeling techniques to recreate a physical image acquisition process. We show that, when the hardware parameters of a system are known, SIA can be used to train a classifier, which can then be used for the physical system. Timeconsuming manual acquisition and labeling of a training sample is no longer necessary. Evaluations in the domain of glass recycling demonstrate that the SIA approach performs on par with a classifier that was trained using a manually collected training set.