When you use Google to search for an image, you may not realize the categorization behind it is made possible by techniques called machine learning and computer vision (CV). These technologies enable the computer to “see” what’s within an image and then analyze and quantify the information they find.
Example of a computer vision engine finding objects in a picture. (Image: Google)
What is the potential application of CV in the AEC industry?
Every week, millions of pictures and videos are captured on projects worldwide, with many of them critical to tracking progress, resolving claims, identifying quality issues, etc. Take the example of safety, which is priority #1 in construction, with the OSHA reporting approximately 1 in 10 workers injured on the jobsite each year. Dedicated safety experts can’t visit every site. But what if there was a way for these computer vision engines to review images for evidence of elevated levels of risk?
Procore-integrated photo and video management platform, Smartvid.io, recently partnered with Engineering News-Record (ENR) to explore how computer vision tools can improve safety in construction. In the first-ever AI for Construction Safety demonstration, Smartvid.io’s AI engine, nicknamed “VINNIE,” worked alongside human safety experts to analyze site images and “flag” any photos that contained potential safety risks. To prepare, VINNIE was trained by analyzing tens of thousands of site pictures for “classes” of objects––a common CV practice––that might point to the presence of safety infractions, such as workers missing hard hats and safety vests.
The results were startling; VINNIE analyzed 1,080 images in a few minutes, while human review of the same data took over 5 hours!
Smartvid.io AI engine “VINNIE” detecting field personnel missing safety vests. (Image courtesy of ENR)
As the first demonstrations of AI for construction safety, these two milestones also herald the potential computer vision has across other roles and sectors.
The power of such automated tools is their ability to derive “order from chaos”––and often, once trained, without much human effort. These technologies are also adaptable to the content and classification methods they’re fed; computer vision engines can be trained to find and analyze almost anything, from workers standing too close to scaffolding edges to pictures showing only a specific color. Anyone whose role makes use of photos and videos can benefit, including:
- Safety teams who want to know when images of potentially risky situations are coming up on their projects
that need faster and easier ways to find the photo and video content they need
Marketing teams that need to search for key imagery by what’s in the image (and don’t need to request it from other teams)
IT Professionals that need to leverage photos and videos across different parts of the company
Training pros that need access to content to establish best practices
Whatever your role is, ask yourself these questions to get a sense of the power of automation:
What could I do if I had a set of autonomous eyes, always looking at my photos and videos?
What if I could train it to search for anything I wanted?
What would I ask it to find?
The possibilities really are endless. To see some of them in action, check out the Procore-Smartvid.io integration! It takes less than 90 seconds to connect Smartvid.io’s engine to your photos in Procore and start analyzing them!