The entrance of VISION 2016 fair, in Stuttgart

VISION 2016 in 9 keywords: from Embedded Vision to Deep Learning

by Pavel Dvorak

Last week I took part in VISION 2016 in Stuttgart, curious to learn the latest development in the sector of Machine Vision for further developing our research activities in Konica Minolta Laboratory Europe. The fair, held from 8th to 10th November, has been visited by almost 10,000 visitors from 58 countries, and I think that the hottest topics were especially related with Embedded Vision, High Speed cameras, 3D vision and Deep Learning.

Some pictures from VISION 2016

Overall, VISION 2016 was crowded of robots manipulating and picking components and smart cameras demonstrating high speed barcode reading and face recognition. Finally, Deep Learning has definitely invaded the sector of Machine Vision. Of course, there was plenty of other technologies related to machine vision, I have selected only those that were interesting to me as Computer Vision researcher and fitted to the portfolio of our organization.

VISION 2016 in 9 keywords

The following is a list to summarize many of the most interesting topics covered in VISION 2016:

1. Embedded Vision will play a major role in the future of Machine Vision. However, this does not mean that the standard way of processing the visual information will disappear. In a variety of use cases, there will still be a need for sensor fusion as it is becoming more and more important. In such scenarios, the embedded vision will rather play a role of an initial block of the whole processing pipeline and more complicated tasks will be performed directly in the camera hardware.
2. Reliability is a keyword that should be at the basis of future developments of all Embedded Vision components, and good lessons learnt in Machine Vision should be kept as guidelines.
3. Standardization, heavily discussed during the panel discussion organized by VDM Association, will enable especially Embedded Vision to become widely applied in several different contexts.
4. Easy programming tools for embedded vision systems should be made available. This point relates to the standardization. Making it easier to deploy algorithm developed in desktop environment will enable many companies to develop their own solutions in embedded vision and this will accelerate the whole business. Some of the tools for such rapid prototyping of the embedded systems are becoming available and were presented at the expo.
5. Complexity of embedded vision will increase and decrease simultaneously. It will become easier for end users so they will be able to easily apply solutions in different cases, but this of course will make life more difficult for developers.
6. Smart cameras are becoming smaller and smaller. I have seen several tries, quite successful I have to say, making the intelligent cameras small and cheap. As an example, I can provide an effort done within a European project called Eyes of Things building a prototype of smart camera with estimated price of units or tens of euros.
7. Robotics is for sure an application fields where many of the above-mentioned keywords will find applications, and this is apparent when we think of machine vision in factories as manufacturing or inspection components.
8. High-speed cameras are becoming more common and important for manufacturing since they can accelerate the visual inspection systems and quality control.
9. Deep Learning is definitely changing the industry. We may not be sure about how quickly this will happen, but the effects of this shift are already apparent: we see big investments in this subject and we hear many voices speaking about it. Within Vision 2016 deep learning has been an important buzzword mentioned in talks, panel discussions and across many booths of companies offering solutions based on it. However, there are still several machine vision problems, such as component measurement, that have to be currently solved in a different manner. This, of course, may be only a matter of time until some deep learning researchers will find a way to deal with this using deep learning.

So, I described the most interesting elements I have identified in the three days I was visiting the fair and I am also happy to mention some of the research institutions, whose activities I found interesting, such as German Fraunhofer IPA, Austrian AIT, French CEA and Swiss CSEM. Of course, plenty of other ideas and topics were presented and many other organizations took part in VISION 2016, but I won’t be going into details for all of them. Please, if you have any other keyword fitting in my list, I would be happy to receive your suggestions: feel free to contact me at research@konicaminolta.eu and via Linkedin private message.