EMVA’s European Machine Vision Forum took place from September 4 – 6, 2019, in Lyon, France.

Being already the 4th edition of the EMVA addressing ‘Research meets Industry‘ the forum was held under the focal topic

Photonics and Machine Vision: Going Deep into Integration

Get more impressions and visit the EMVF-2019 Picture Gallery.

“Deep learning is not enough – overall system optimization is necessary.”

This headline summarizes the conclusion of the 4th European Machine Vision Forum, where research and industry met in Lyon from September 5 to 6, 2019.  In tune with the region around Lyon with Grenoble and St. Etienne, the main topic was the integration of photonics and machine vision: “Photonics and Machine Vision: Going Deep into Integration”.

The starting point for this main topic was a glimpse into nature. Over millions of years, biological vision systems have emerged which not only have a tremendous ability to learn, but have also evolved very different types of visual systems adapted to the task in terms of spatial resolution, spectral sensitivity, colour vision, motion perception and response time.

In the first keynote talk “The future of image sensing – More intelligence or more sensing?” Prof. Peter Seitz (Hamatsu Photonics, Soloturn) demonstrated how diverse information carries light, which can be made visible by suitable image sensor technology, but which cannot be detected by even the best learning methods without this option. The second presentation “Learning high-level reasoning in and from Images” by Prof. Christian Wolf, INSA, Lyon critically analysed the interpretation of machine learning. While until 2012 characteristics were determined empirically, the same will be done today with the structures of neural networks and tomorrow with the methodology of image processing. At the same time, he gave hands-on advice on how rewarding it is to integrate known contexts such as symmetries and invariances into the design of neural networks. The synthesis between photonics and machine vision was the central focus of the third keynote talk “The convergence of photonics and electronics: an opportunity for machine vision” by Dr. Francois Simoens, CEA-Leti Grenoble. He pointed out how the new technologies of photonics, sensor technology and electronics are merging and how the new opportunities of 3-D stacking of wafers enable complex vision systems with light sources, microoptics, sensor technology and electronics. As examples a complete LIDAR system on a chip with scanning laser light without mechanics (“Optical Phased Array”) was presented and a vision chip (“Smart Retina”) with integrated pre-processing.

The program featured a further 25 presentations and an accompanying exhibition. The participants honored the high quality of the presentations, the professional organization of the event and the numerous opportunities for exchanging ideas between research and industry. This encouraged both sides to take home fresh ideas for their projects. The fact that a European region was able to present itself at the “French Evening” also resulted in several new contacts at pan-European level.

Outlook on the 5th European Machine Vision Forum 2020

At the end of the event the 5th Forum was announced by the local Chair, Prof. Peter O’Brien. The forum will take place from 10-11 September 2020 at the Tyndall National Institute in Cork, Ireland with the focus on “Accuracy, Reliability and Limits of Machine Vision”. An ideal topic for dialogue between research and industry. The research can find out where the theoretical limits exist and reveal to industry what the current state of technology is today. Both partners together thus have a solid basis for jointly initiating new projects for the development of novel, powerful and at the same time cost-effective systems in areas where there is a corresponding market need. Since image processing is a cross-sector technology, the limits can be manifold: What is the highest possible image resolution? What limits the sensitivity and sharpness of image acquisition? What are the limits of different processes for 3D image acquisition? To what extent can the image data stream be reduced in order to obtain the information of interest? (Compressive Sensing). How tiny can image processing systems become? How reliable can image processing systems be? How much can the delay between image acquisition and evaluation result be reduced? What is the minimum power required for certain image processing operations? …

The topic was inspired by the high accuracy requirements of components that integrate photonic elements (light sources, micro-optics, sensors) and electronic parts. The Tyndall National Institute is hosting the European Photonic Packaging Pilot Line.

Author

Prof. Dr. Bernd Jähne

Bernd Jähne has been Professor of Physics at the Interdisciplinary Centre for Scientific Computing (IWR) since 1994 and Senior Professor since October 2018.  In 1995, he founded the Heidelberg Image Processing Forum and from 2008 to 2017 was coordinating director of the Heidelberg Collaboratory for Image Processing (HCI), an “Industry on Campus” project at the University of Heidelberg. Since 2008, he has been Chairman of the EMVA 1288 Standardization Committee and became member of the EMVA Board in 2015.