Seeing more hazards, preventing more incidents: Vinnie’s even smarter

    At Newmetrix (formerly Smartvid.io), we’re constantly working to train our AI, Vinnie, to recognize additional hazards on a jobsite, and we’ve recently added nearly 20 new tags that cover a wide array of hazards, from heavy equipment and unsafe ladder positioning, to improper body positioning and exposed holes.

    As a result, Vinnie can identify even more hazards within the progress photos or 360 site images that most construction organizations already collect to provide an even sharper picture of a jobite’s safety risk and provide even better data to improve the accuracy of predictive-based safety. Our goal is to enable construction companies to predict and stop accidents before they happen, so that everyone goes home safe to their families every evening.

    So, given the scope of the new tags we’ve introduced, it’s going to take more than one post to describe them. We’ll start with new tags that deal with work at height and falls.

    Identifying a wide range of falling hazards

    Falls are an important topic, because the U.S. Bureau of Labor Statistics found that falls are the leading causes of worker deaths in construction, accounting for just over one-third (34%). Not coincidentally, fall protection violations were the most frequent OSHA violation in 2019.

    Vinnie can already identify work at height through the tag “person at height,” which identifies people on lifts, ladders, roofs or near leading edges. But we wanted to expand on this area to provide additional information about violations related to work at height.

    New tags include:

    • Guardrail Issue: Identifies instances where a guardrail is missing and can also identify whether it has been installed correctly or is missing key components.
    • Fall Arrest Issue: Identifies whether a person lacks a fall protection system.
    • Scaffolding Issue: Identifies whether a scaffold has been set up correctly or if there are missing components.

    But that’s not all. We also wanted to provide a deeper look at ladders, one of the most common, dangerous and misused tools on a construction site. We trained Vinnie to help identify common mistakes when working with ladders, including:

    • Ladder Setup Issue: Vinnie can identify instances where a ladder is set up in a way that is dangerous and would make it unsafe to use. If the wrong type of ladder is being used for the job at hand or the ladder is placed on an unsafe surface or area, Vinnie will tag it.
    • Ladder Use Issue: Vinnie can also see workers actively using the ladder incorrectly.

    ladder-issue
    Vinnie detects "Ladder Use Issue" and "Ladder Setup Issue"

    We also trained Vinnie to understand stairways, which are inherently hazardous. If a stairway is misplaced or lacks integrity, it can lead to severe injuries and even death. According to OSHA, the majority of stairway related accidents are due to issues related to the stairway itself, such as missing rails, clutter, housekeeping issues or loose structural integrity. Vinnie can detect stairways on job sites in various scenarios, which enables Newmetrix to provide additional insights related to stairway health so safety managers can identify potential areas where fall risks might be heightened.

    Finally, exposed holes are another common source of fall hazards, as they are prominent on both indoor and outdoor sites where there is a lot of worker activity. Vinnie can now identify exposed holes without a cover, which opens up the item for further analysis to determine whether the hole lacks guardrails, barricades or signs. Vinnie can also identify whether the workers in the vicinity of the hole are wearing the fall arrest systems necessary to conduct work in a safe manner.

    The old tags just keep getting better, too

    But all these new tags and capabilities don’t tell the entire story of Vinnie’s evolution. Our AI is learning. This team actively and frequently retrains the underlying computer vision models on proprietary data sets to improve the “Vinnie Engine”. By using state of the art computer vision methods, the team feeds Vinnie new training data within new contexts that emerge over time, which improves the accuracy and sophistication of our core Safety Monitoring tags, including:

    • Identification of fall hazards
    • Standing Water
    • Housekeeping issues
    • Lack of PPE, and
    • Construction activities such as Excavation & Trenchwork and Rebar placement.

    Trench WorkVinnie detects "Trench Work " 

    Our continuous AI training and tag improvements enable our customers to consistently receive ever more robust insights from Vinnie and empowers our visual AI to recognize new situational cases. And it’s all in the service of a single goal: Improving the engine’s ability to detect potential risks on the jobsite so risk and safety personnel can make sure everyone goes home safe when the day is done.

    With these new tags and improvements to those Newmetrix already had in place, leadership can better understand fall protection performance, identify safety trends and provide mitigation and training to decrease the likelihood of falling incidents on site. Vinnie acts as another set of eyes, creating potentially thousands of new observations and catching additional hazards that human observers might miss in safety walks. All of which makes job sites safer so more workers go home safe every night. And with a better safety record backed up by the documentation Newmetrix can provide, construction companies can also make a strong case that they merit lower insurance premiums.

    In part two, we’ll talk about how Vinnie recognizes heavy equipment, worker positioning and even long-term hazards such as improper ergonomics. If you want to receive it directly to your inbox, be sure to subscribe to our blog for instant notifications!

    See how Newmetrix helped Lithko quadruple the number of safety observers on site, substantially increase the number of safety conversations and, as a key part of the company’s larger safety program, reduce recordable incidents by 20% in one year.



    Written by Ergeta Muca

    Ergeta Muca is a Product Manager at Newmetrix, with a focus in analytics. Ergeta earned her Masters degree in Applied Analytics from Columbia University. She also holds a Business degree from Boston University. Ergeta's previous professional experience include consulting at PwC, data science/analytics at AT&T, as well as research at Columbia & BU. Outside of work, Ergeta has a passion for the fine arts & literature. She enjoys cooking, traveling, dancing and live jazz music.

    View more posts by Ergeta Muca.

    Follow me:

    Related Posts

    Leave a Comment: