Trending machine vision and imaging topics of 2018
Deep learning, embedded vision, and polarization imaging were among the most popular machine vision topics in 2018. Learn more about some of the latest products and technologies involving these topics in this year-end article.
Throughout 2018, several areas have emerged, grown further, developed or evolved as hot topics. A number of them have been trending for years, while some have more recently grown in popularity. Some of these include 3D imaging, collaborative robots, deep learning (and the broader category of machine learnning and artificial intelligence), embedded vision, multispectral and hyperspectral imaging, and polarizationimaging.
As these topics have become increasingly popular, more products and technologies have been developed and released, and more of an emphasis has been placed on them both at trade shows and in media coverage. In this article, three of these trending vision topics will be examined, along with some of the most recent, relevant products that have been released within these areas.
Figure 1:A polarization camera is used to detect glossy and curved surfaces of an image and indicate that measurements in these areas might be invalid, both in the intensity image (right) and in the false color polarization (left).
Deep learning is an area of machine learning that enables computers to be trained and learn through architectures such as convolutional neural networks (CNNs). This has been, of course, a hot topic for a few years now. Yet still, more companies within the machine vision marketplace have dedicated a significant amount time and resources toward developing deep learningproducts.
Machine vision and imaging companies that have recently developed, announced, or updated their deep learning software products or tools include Cognex (VisionPro ViDi; Natick, MA, USA;www.cognex.com), Cyth Systems (Neural Vision; San Diego, CA, USA; www.cyth.com), EVT (EyeVision software; Karlsruhe, Germany; www.evt-web.com), Laon People (NAVI AI; Seongnam-si, South Korea; www.laonpeople.com), MathWorks (MATLAB; Natick, MA, USA; www.mathworks.com), Matrox Imaging (Matrox Design Assistant; Dorval, QC, Canada; www.matrox.com/imaging), MVTec (HALCON; Munich, Germany; www.mvtec.com), and SUALAB (suaKIT; Seoul; South Korea; www.sualab.com).
Several companies have also developed deep learning-related components, including cameras and frame grabbers. These include—but are not limited to—CEVA Inc. (NeuPro AI processors for deep learning; Mountain View, CA, USA;www.ceva-dsp.com), Euresys (EasyDeepLearning library; Angleur, Belgium; www.euresys.com), FLIR Integrated Imaging Solutions (Firefly deep learning camera; Richmond, BC, Canada; www.flir.com/mv), and Silicon Software (deepVCL frame grabbers; Mannheim, Germany; www.silicon.software).
Additionally, companies such as Amazon (Seattle, WA, USA;www.amazon.com), Ambarella (Santa Clara, CA, USA; www.ambarella.com), Google (Mountain View, CA, USA; www.google.com), Intel (Santa Clara, CA, USA;www.intel.com) and Microsoft (Redmond, WA, USA; www.microsoft.com) have recently released or announced products or platforms related to deep learningtechnology.
Application example: Deep learning promises automotive inspection improvements (http://bit.ly/VSD-DLA).
Polarized sensors and cameras can be deployed to uncover hidden material properties that are not detectable with conventional imaging technology. While not a new technology, polarization cameras and imaging components became significantly more mainstreamin 2018.
This can be attributed, at least in part, to the introduction of the IMX250MZR/MYR (monochrome, color) CMOS sensors from Sony (Tokyo, Japan;www.sony.com), which are 5.1 MPixel global shutter CMOS image sensors with a four-way polarized filter design, which consists of four separate polarizing filters – angled at 0°, 45°, 90°, and 135°–that have been arranged in a regular pattern across the sensor.
Since Sony introduced the sensors, numerous companies have announced cameras based on the IMX250MZR/MYR, including Allied Vision (Stadtroda, Germany;www.alliedvision.com), Baumer (Radeberg, Germany; www.baumer.com), FLIR Systems, JAI (San Jose, CA, USA; www.jai.com), LUCID Vision Labs (Richmond, BC, Canada; www.thinklucid.com), Matrix Vision (Oppenweiler, Germany; www.matrix-vision.com), Photonfocus (Lachen, Switzerland www.photonfocus.com), Pixelink (Ottawa, ON, Canada; www.pixelink.com), Teledyne DALSA (Waterloo, ON, Canada; www.teledynedalsa.com), and Sony itself (Sony Image Sensing Solutions; Weybridge, UK; www.image-sensing-solutions.eu).
Other companies that have recently released polarization cameras include Teledyne DALSA (Piranha4 line scan polarization camera), Photron (Crysta 2D polarization camera; San Diego, CA, USA;www.photron.com), and 4D Technology Corporation (PolarCam snapshot micropolarizer camera; Tucson, AZ, USA; www.4dtechnology.com). By the time this has been published, I presume that additional companies will have announced polarization cameras as well.
Application example:Researchers deploy polarization camera for carbon fiber inspection (Figure 1, http://bit.ly/VSD-CFI, Fraunhofer IIS image).
The argument could be made that embedded vision has been the single most popular topic of the year, and really, the past few years.
As embedded vision system cost, size, and power consumption decrease over time, machine vision and image processing will proliferate into thousands of new applications. Numerous machine vision camera companies have realized this and have expanded into or increased their presence in the embedded vision marketplace. Listing every single company developing embedded vision cameras would be a redundant task, but to refresh our readers’ minds, some companies that have recently released them include Alkeria (Navacchio, Italy;www.alkeria.com), Allied Vision, Basler (Ahrensburg, Germany; www.baslerweb.com), Baumer, D3 Engineering (Rochester, NY, USA; www.d3engineering.com), e-con Systems (Chennai, India; www.e-consystems.com), FLIR IIS, IDS Imaging Development Systems (Obersulm, Germany; www.ids-imaging.com), Jadak (Syracuse, NY, USA; www.jadak.com), LUCID Vision Labs, Matrix Vision, Omron Microscan (Renton, WA, USA; www.microscan.com), OpenMV (Atlanta, GA, USA; www.openmv.io), The Imaging Source (Bremen, Germany; www.theimagingsource.com), Sony, XIMEA (Münster, Germany; www.ximea.com), and Vision Components (Ettlingen, Germany; www.vision-components.com).
Additionally, several companies—over the past year-plus—have released embedded vision products such as PCs, boards, development kits, sensors, graphics processing units, controllers, and software. Some of these companies include Allied Vision, Ambarella, AAEON (New Taipei City, Taiwan;www.aaeon.com), Basler, Cadence (San Jose, CA, USA; www.cadence.com), CEVA Inc., Critical Link (Syracuse, NY, USA; www.criticallink.com), EPIX (Buffalo Grove, IL, USA; www.epixinc.com), FRAMOS (Taufkirchen; Germany; www.framos.com), Logic Supply (Burlington, VT, USA; www.logicsupply.com), Matrox Imaging, MVTec, Neousys Technology (Taipei, Taiwan; www.neousys-tech.com), NVIDIA (Santa Clara, CA, USA; www.nvidia.com), Sony, Vecow (New Taipei City, Taiwan; www.vecow.com), and Xilinx (San Jose, CA, USA; www.xilinx.com).
David Dechow, Principal Vision Systems Architect, Integro Technologies (Salisbury, NC, USA;www.integro-tech.com), told Vision Systems Design that, “There has been an explosion in the availability of embedded computers. A few years ago, we talked about embedded computers as being ‘industrial PCs’ – which were very cutting edge, but now we are seeing a trend in both single-board and system solutions that are targeting machine vision and, in combination, deeplearning.”
He added, “It’s exciting to be able to get a computer that has everything needed for machine vision and deep learning, in a package that is an industrialized or an embeddedprocessor.”
Application example:Low-cost components make portable fundus cameras a reality (http://bit.ly/VSD-FUND).