Embedded World 2020 took place in Nuremberg, Germany under the difficult circumstances of COVID-19 having arrived in Europe, resulting in a significant cancellation rate of exhibitors and a much lower visitor density than in previous years. Most of the machine vision exhibits, however, opened their booths and even showed overall satisfaction of the lower quantity but equally good quality of the contacts.
One rather prominent exception was Basler who decided at the very last minute not to have personnel in attendance. Amongst the machine vision players present were Allied Vision, MVTec, Framos, Imago, ATD Electronique who shared a common booth with their parent company Macnica and booth partner technique, based in New Zealand. Irida Labs showed a live demo of an embedded vision system using deep learning for food identification and classification (shown here).
This years´ embedded vision award was presented to DC Vision Systems for their stereo vision processor DC-SVP with Linux running on the camera. The implemented stereo image processing pipeline enables measurement of the surrounding 3D geometry within the field of view directly on the camera in real-time.
Panel discussion on embedded vision
One highlight from a machine vision point of view was the panel discussion “Embedded Vision Everywhere!?”. Addressing the rise of embedded vision technology, it was stated by Dr. Christopher Scheubel from cubemos, a spin-off from Framos, that the consumer sector has finally triggered both a downward spiral of prices and at the same time an upward spiral of performance of embedded vision systems. Interesting enough, even after quite a few years that embedded vision has been flying around as a buzzword,it was still first to define amongst the panelists what embedded vision is. Everyone agreed that compactness and edge computing are two major characteristics, and the concepts of embedded vision systems had been there for decades starting with smart cameras.
Jason Carlson, CEO of congatec stated that pushed by the rise of edge computing (embedded) vision has become the single biggest data driver in the industry. Opinions parted, though on the necessity of low power consumption, which was seen as important by suppliers, but not mandatory for the application side in the panel, which was represented by CST and forklift maker Still/KION Group. To the contrary, Jan-Erik Schmitt from Vision Components argued that often power consumption is a topic for customers with stand-alone embedded vision systems, which may even be powered by solar cells.
It was further discussed whether AI will play an important role within embedded vision. Congatec’s Jason Carlson pointed out that AI cannot be deployed just for the sake of AI, but instead needs to deliver value where there was no value before. Also Christopher Scheubel said that AI can only be part of any vision solution, as it opens doors where classic machine vision technology stops. Also Jan-Erik Schmitt agreed that AI is not the magic stick but needs to be deployed in combination with established machine vision technology.
One clear statement made by the panelists was the almost unanimous call for further standardization of embedded vision systems to enhance interoperability and ease of use. In this context, GenICam and MIPI were mentioned as two existing standards that might play a leading future role. Standardization would be a customer expectation and, according to Gion-Pitschen Gross from Allied Vision Technologies a prerequisite for making cameras as easy as possible to operate. Bengt Abel from Still/KION Group even proposed an embedded vision system which is connected to an app store providing various plug and play applications.
Written by Andreas Breyer, Senior Editor, Germany, Novus Light Technologies Today