A longtime staple of the manufacturing sector, machine vision is the foundation that underpins many emerging technologies, such as robotics, artificial intelligence (AI) and smart glasses, and is estimated to reach a near 100 million installed base in 2025. In the early days, industrial machine vision relied on rule-based programming that involved a lot of human expertise.
The recent emergence of deep learning (DL) technology has propelled the automation of such processes. DL models manage to pick up trends and abnormalities in large image and video sets, under a low level of human involvement. “The machine vision landscape has gone through multiple iterations, and the top three players have demonstrated their strength in keeping up with the technology evolution,” said Lian Jye Su, a Principal Analyst at ABI Research.
Targeting growing niche markets such as robotics and autonomous vehicles will generate new revenue growth beyond traditional machine vision markets like electronics and automotive manufacturing. Another interesting development is the rise of independent machine vision cameras with built-in edge artificial intelligence (AI) processors. The conventional machine vision system has a more centralized processing architecture based on industrial PC.
“However, as the use cases for machine vision expand into mobile devices and surveillance systems, a new generation of machine vision systems are designed to work as a standalone system, supported by onboard storage for AI-based machine vision models. This has led to the emergence of an AI solution that can auto-generate images and perform self-learning, without the need of large amounts of user data, as in the case of startup Laon People,” concluded Su.
Su tells IndustryWeek three primary concerns exist as manufacturer begin integrating AI into the computer vision space:
- Uncertainty: Currently all existing machine vision workflows are quite standardized and follow certain templates. Switching to AI in machine vision may require new type of data collection process, new setup and hardware equipment, which brings uncertainty to daily workflow.
- Relearning and retooling: In order to effectively deploy AI in machine vision, both engineers and vendors need to relearn their new skillsets and toolkits. This will take some time and it is challenging given the continuous operation and the stringent operational requirements in factory settings.
- Culture: People are generally reluctant to break what is not broken. Conventional machine vision has been working well in traditional scenarios like guidance, identification and inspection
However, despite concerns, AI is crucial in new use cases, such as optical character reading, texture and material classification as well as complex inspection and segmentation, explains Su.
“Deep learning-based machine vision can also incorporate data collected from various sensors, including LiDAR, radar, ultrasound and magnetic field sensors,” he says. “The rich set of data will provide further insight into other aspects of production processes. Deep learning algorithms deployed for machine vision can pick up unexpected product abnormalities or defects, providing flexibility and valuable insights to manufacturers.”
