Machine vision market goes to the next level

Edge computing, costs considerations and machine learning will influence the development of machine vision in the next five years. Beyond these, big data processing and yield analytics will also play a key role.

Machine vision in manufacturing “will continue to be [focused on] quality control and automation, areas that are well established by traditional industrial machine vision player and techniques, and that will be further supported by machine learning-based machine vision innovation,” said Lian Jye Su, principal analyst at ABI Research.

Besides cameras, machine learning-based machine vision incorporates data collected from various sensors, including LiDAR, radar, ultrasound and magnetic field sensors. This rich set of data then provides further insights into other aspects of the production process.

“As compared to conventional machine vision which can only detect product defects and quality issues which can be defined by humans, machine learning algorithms deployed for machine vision can go even further,” said Su.

“These algorithms can pick up unexpected product abnormalities or defects, providing flexibility and valuable insights to manufacturers.”

Some startups are also working on extending the power of machine vision to other applications in manufacturing. Su mentioned names like Instrumental, Landing.ai and micropsia industries. These firms have developed products that leverage image data collected from cameras to venture into other smart manufacturing services, such as big data processing, process optimization and yield analytics.

The next stage in machine vision

As machine vision technology advances, costs could also come down. “As costs decrease, cameras or pixel arrays are finding their way into places where analog and discrete sensors used to be,” said Frank Lamb, the Founder of Automation Consulting.

“As Industrial IoT becomes more widespread, I see smaller pixel arrays being used for more custom applications to save money. This is in opposition to buying packaged systems for major vision vendors,” said Lamb.

Ambarella, a computer vision semiconductor design company, expects the next key technology in machine vision to be edge computing.

Current machine vision systems use PCs for processing. This is suitable for situations such as sites with a fixed assembly line where the main function of the camera system is to scan packages zipping by. However, if the camera can process the data itself, rather than streaming the video to another device for analysis, it could bring another level of mobility to the warehouse, such as warehouse robots or mobile scanning systems.

“We believe that the new generation of technologically-advanced, low-power computer vision chips -- capable of performing machine vision at the edge -- will increase intelligence in mobile robots and scanners, enable use cases that were not possible before, and improve efficiencies in manufacturing workflow,” said Jerome Gigot, Senior Director of Marketing at Ambarella.

“In the next five years, the machine vision market will move towards AI (artificial intelligence)-based technologies,” said Su.

More cameras could be deployed for urban infrastructure purposes, and on commercial and consumer vehicles and robots, to enable automation and augmentation of work processes, Su suggested. Machine vision may also be seen in smart cities, automotive and mapping, logistics and warehousing.

Don’t forget customer education

“Educating customers on what machine vision can and can’t do, and how to maintain their systems is a challenge,” said Lamb. “People tend to think you can install a system and it will solve all of their problems and work 100 percent of the time, this is simply not true.”

For an existing factories, how to embrace Industry 4.0 and the latest machine vision technologies -- including accommodating automation systems and data exchanges from the internet of things -- will also be a challenge.

“If the customer isn’t willing to learn how the system works before it is installed and, and isn’t willing to invest in maintaining and adjusting when necessary, the application will not be successful,” said Lamb.