Machine vision challenges for system integrators

Vision system integrators are looking to stay a step ahead of emerging technologies such as deep learning and other smart technologies.

By Winn Hardin, AIA September 21, 2018

Today’s vision system integrators can choose from a plethora of products: off-the-shelf lenses, lighting, cameras, camera interface boards, and software. And the choices keep evolving as vendors introduce more affordable products with added functionality to replace systems that once would have cost thousands of dollars.

"One of the biggest drivers for this change is the commoditization of 2-D machine vision offerings," said Markus Tarin, president and CEO of MoviMed and MoviTHERM. "Smart sensors and smart cameras, as well as configurable vision systems, have largely eliminated the need for machine vision system development, with most common applications now being accomplished with off-the-shelf plug-and-play technology. Sophisticated and capable machine vision integrators now find themselves in in a position where it becomes increasingly difficult to add value to a common 2-D machine vision system. And some vendors of these systems are underlining this situation by marketing their configurable machine vision systems directly to the end customer."

John Salls, president of Vision ICS, Inc., also recognizes the progress over the last decade with smart cameras becoming more functional and lighting companies offering a broader range of products. However, just as software becomes more powerful and prices keep falling, Salls sees an issue with the interconnection and standardization of software packages.

"Different companies use different terminology for the same things," he said. "Even standardized communications like Ethernet have huge variation from company to company, and there is not really a push for open [software] standards in the vision industry."

Lighting, like other products in the machine vision lineup, needs careful attention from the integrator. "In every machine vision system, it is critical to realize the best possible image, and this is where lighting is key to create strong contrast definition of the features being inspected," said Earl Yardley, director of Industrial Vision Systems.

"There is a vast range of generic off-the-shelf machine vision lights that are adequate for the majority of applications, but these standard offerings should not be the limiting factor," Yardley said.

To address more complex applications, Industrial Vision Systems develops lighting specifically for a project. Recent examples include high-intensity UV lighting, segmented ring-lights integrated into a compact robot inspection head, and custom form-factor backlights.

"Producing these custom solutions creates extra design time at the front end of the project, but this is outweighed by the time saving when it comes to programming the system with a fully optimized image," Yardley said.

Delivering on 3-D and deep learning

While many of today’s vision products can meet the needs of most applications, system integrators must stay a step ahead as technologies and customer demand evolve. In the 3-D imaging market, for instance, Tarin points out hardware innovation precedes software innovation.

"Although there are a number of 3-D sensors and cameras available, such as laser triangulation, time of flight, stereoscopic sensors for pseudo random pattern generators, and others, there is a large gap in the development tool chain to allow for rapid system development," Tarin said.

Original equipment manufacturers (OEMs) currently use open-standard 3-D sensors or cameras and program their application from scratch, or use "closed" systems with configurable tools often come with a cost-prohibitive price tag, Tarin explained.

"Perhaps what is required is a 3-D sensor or camera with programmable FPGA for high-speed onboard image processing to enable a non-FPGA programmer to deploy 3-D image processing algorithms all in one package," Tarin said.

Another technology peeking its head into the machine vision market is artificial intelligence, and more specifically, deep learning—the ability of computers to learn from experience. At this early stage, the biggest challenge is separating hype from reality. "AI and deep learning algorithms sometimes overpromise a solution to every difficult-to-solve machine vision problem," Tarin said.

Tarin noted that deep learning, while beneficial, only goes so far.

"While it is true that machine vision applications are already benefiting from the deployment of deep learning algorithms, these are far from providing a silver bullet," Tarin said. "This is especially apparent when one compares the effort necessary trying to achieve greater than 99% accuracy compared to traditional programming efforts. Nonetheless, this technology definitely has its place and will continue to gain importance over the next few years."

Eyeing emerging technology

While many systems integration challenges have been drastically reduced by the availability and increasing affordability of smart cameras that embed lighting, software, and I/O interfaces, emerging technologies will pose more puzzles for machine vision to solve. In developing multispectral imaging systems, for example, specialized lighting will be required to illuminate products at specific wavelengths. For hyperspectral imaging, broadband LED illumination will replace the current halogen-based systems.

In data fusion applications, where a number of different sensors—ultrasound, visible, infrared, and lidar—are used, sophisticated imaging software will need to be tailored to perform efficiently on high-performance graphics processors. With the advent of ever faster CMOS-based high-speed cameras, system integrators will have to support optical networks to transfer data from cameras to computers. Tying this together, edge-based vision systems will be required to work in conjunction with cloud-based computing systems to tie analyzed captured data with factory management and robotics systems to fully automate manufacturing processes.

Since machine vision system integrators are problem solvers by nature, they’ll do what they’ve always done: overcome any obstacles in order to find the best solution for the application.

Winn Hardin is contributing editor for AIA. This article originally appeared in Vision Online. AIA is a part of the Association for Advancing Automation (A3), a CFE Media content partner. Edited by Chris Vavra, production editor, CFE Media,


Keywords: machine vision, system integration, deep learning

  • System integrators have many solutions to machine vision challenges that are cheaper and more efficient than before, but the number of available choices can be overwhelming.
  • System integrators must stay a step ahead as technologies and customer demand evolve in a market where hardware innovation precedes software innovation.

Consider this

What other machine vision developments will have a major impact on system integrators?

Original content can be found at