© Ekkasit919 | Dreamstime.com
Computer Vision

Computer Vision Is Slow to Catch On, But the Technology Is Improving

June 18, 2021
Initial-use cases like computer-vision-guided robots are beginning to join the production floor.

Over the last five years, we’ve seen a gradual increase in exposure to industrial computer vision applications. Improvements to the image quality, size and cost of simple cameras, along with continuous improvements in machine learning algorithms, sensor price points, and bandwidth in cloud and edge computing, have started to unlock the potential of computer vision in industry.

In manufacturing, we’re starting to see initial-use cases like computer-vision-guided robots joining the production floor, enabling the 24/7 production line and freeing up human capacity for more creative work. But most companies in heavy-asset industries are lagging in comparison to the consumer market in terms of broad computer vision application.

This is surprising considering vision data is anticipated to be a major driver of Industry 4.0, a transformation that is expected to deliver between $1.2 to $3.7 trillion in potential value by 2025 worldwide, according to McKinsey. That’s a great deal of potential and promise, so why aren’t we seeing more industrial computer vision applications?

The lag in industrial computer vision adoption can generally be attributed to two main issues. First, there is a lack of understanding of how computer vision-- can add value to the business. Second, there is an inability to integrate and contextualize data at scale and in real-time.

How computer vision adds business value in manufacturing

As additional types of image data become more readily available, potential computer vision use cases are expanding and their operational value is increasingly more obvious. 

3D quality control with AI-powered defect detection: Computer-vision systems can take multiple images of an item as it moves through production, to produce a 3D model that is immune to lighting, contrast, and distance issues found in 2D images. Pair this with artificial intelligence that can quickly learn how detailed, intricate components are supposed to look, and manufacturers can improve production quality while reducing quality control efforts; a combination that reduces overall operational costs exponentially.

Remote monitoring for advanced predictive maintenance: Computer vision allows manufacturers to leverage more cameras and robots to increase the volume and quality of inspection data capture. Deep learning models can then be used to detect maintenance needs and identify developing issues before they arise. Predictive maintenance is perhaps the highest-value computer vision use case—able to deliver significant economic, environmental, and human-safety benefits.

Ensure employee safety and enforce security standards: Computer vision can replace manual site monitoring processes that tend to be costly and error-prone. Instead, a computer vision system can offer constant, site-wide, AI-driven monitoring that immediately alerts employees of potential danger and reports all compliance violations to the respective manager. 

Dealing with the data is essential for success

Even as the understanding of the business value from computer vision applications grows, there is still the issue of soiled and incompatible data sources. Without a clear data vision and the right data management strategy, computer vision applications cannot drive operational value. To realize the opportunity, we need to start collecting and utilizing the data at our disposal.

Step one is to liberate and centralize siloed data so that you can cross-reference images with other data sources, such as batch quality and sensor data. Step two is to automate data preparation and analysis so that you can collect, contextualize and make the vision data actionable.

Only actionable vision data can be put to use for asset surveillance and monitoring, the development of 3D models, or to enable augmented reality and industrial artificial intelligence.

While the rise of cheaper, more powerful hardware is driving interest in computer vision applications, an “industrial data-ops” mindset is needed to truly unlock their power. Industrial data-ops is the discipline of breaking down silos and optimizing the broad availability and usability of industrial data. The data is ready and waiting, we just need to make it do more if we want to open the eyes of our industrial operations.

Francois Laborie runs Cognite's operations in North America, including offices in Houston and Austin, Texas. Previously, he managed Cognite’s overall marketing activities as chief marketing officer. He spent the preceding 11 years with software company Vizrt, where he served as COO and CCO, responsible for all commercial and operational activities. Laborie has a Ph.D. from the Toulouse Computer Science Research Institute.

Sponsored Recommendations

Voice your opinion!

To join the conversation, and become an exclusive member of IndustryWeek, create an account today!