Optics.org
Optics.org
daily coverage of the optics & photonics industry and the markets that it serves
News
Menu
Applications

Embedded systems and machine learning present disruptive forces at Vision show

09 Nov 2016

Machine vision industry is on the cusp of a transition to chip-level technology, hears the Stuttgart trade show.

Machine vision, a technology that until now has been largely restricted to factory production lines, is poised to enter a whole range of new applications with the development of embedded systems at the chip level.

Alongside the impact of Industry 4.0, and the growing adoption of thermal and hyperspectral imaging, that was one of the key messages on the trade show floor in Stuttgart, as the Vision 2016 event got under way.

Lower cost point
Executives from some of the sector’s key players discussed the opportunities and challenges presented by embedded vision in a panel session featuring logic chip designer ARM Holdings, whose presence indicates the significance of the trend.

In essence, an embedded vision system performs much the same function as the more familiar setups based around a camera, illumination systems and optics, all hooked up to a PC. The key difference is the size and cost of technology, and the potential to scale up volumes dramatically.

Arndt Bake, chief marketing officer at the industrial camera firm Basler, explained in the panel session that embedded systems offer a much lower cost point for machine vision because they are typically based around only the light sensor, rather than a full camera.

However, Basler also released its camera-based “PowerPack” development kit for embedded vision at the Stuttgart show. It combines a 5 megapixel camera with FPGA-based image capture, lens and processing board, and is described as a “stepping stone” to custom developments of more compact systems.

Although fully embedded systems tailored to specific applications are expected to demand greater up-front challenges in the form of much higher development costs, the panellists broadly agreed that the far more compact, lower-power, and “lean” architectural approach would open up significant new markets for machine vision through scale and cost.

Beyond the factory
ARM’s VP of embedded marketing Richard York said during the panel session that the technology would become applicable to entirely new markets, for example in retail logistics. “[Embedded vision] opens up tremendous opportunities beyond the factory,” he said, with Olaf Munkelt – chairman of the VDMA’s machine vision group – suggesting that the transition to embedded systems was already under way.

However, there will not be a wholesale transition from PC-based vision to embedded systems – factories will still require PC-based systems to deliver the kind of always-on, industrial-level reliability and flexibility that they have come to expect.

Where the difference will be seen is in high-volume applications that can support the kind of high development costs and low cost points that will be demanded. In Bake’s view, two different types of customers must be served: those in the factory with their demanding technical requirements, and those wanting low-cost, embedded technology that can be deployed and used by non-specialists.

And York sees a major opportunity arising for any company that is able to develop and offer conventional and embedded versions of its software to enable a seamless transition between the two formats.

Machine learning: not to be ignored
One related trend is the parallel emergence of machine learning, something that ARM’s York believes will “transform” machine vision. “The fusion of machine learning and conventional imaging algorithms is the key thing,” he said. “Those who ignore machine learning do so at their peril.”

York added sensor fusion to the list of “transformational” technologies set to impact the sector, for example in automotive applications, where billions of research dollars are being invested in the development of new automated functions. He said that ARM would be doing all it could to make its CPUs more applicable to the machine vision space, as embedded applications begin to proliferate.

One area where embedded visions systems appear to be lacking at the moment is in standardization. With the rapid pace of development, that is not entirely surprising, although as the key organizations within the machine vision community work to remedy that situation it should only be a temporary effect.

• The European Machine Vision Association (EMVA) and Vision show organizers are to host the very first European Embedded Vision Conference - at the same Messe Stuttgart location as the Vision 2016 event - in October next year.

Running 12-13 October, the show includes an exhibition and will target developers and users of embedded vision systems from all industries and application sectors. The program committee involves the North American Embedded Vision Alliance.

EMVA's general secretary Thomas Lübkemeier said: "Worldwide, embedded vision is making inroads. Interest is growing among a multitude of industries on what can be realized with these intelligent systems.

"Be it automotive, automated guided vehicle systems, drones, the medical sector, automation, sports and entertainment or even marketing - embedded vision systems take care that tasks are performed automatically, stand-alone, fast and smart."

Diffraction InternationalSchaefter und Kirchhoff GmbHEdmund OpticsBristol Instruments, Inc.Ealing UGBRD Optical Co., LtdSACHER LASERTECHNIK GMBH
Copyright © 2019 SPIE EuropeDesigned by Kestrel Web Services