by Deanna Zammit
This week, as the United States and Canada ease social distancing restrictions enacted to stem the spread of COVID-19, Patriot One Technologies announced the release of new video object recognition modules that are aimed at facilitating a return to normalcy while keeping people safe.
The health and safety detection modules—part of the PATSCAN Multi-Sensor Threat Detection Platform—work with off-the-shelf thermal and digital cameras to detect elevated body temperatures and facial mask compliance. Further, thanks to a $4.5 million co-investment from Canada’s Digital Technology Supercluster, the company will expand its team to fast track the refinement of that technology and develop advanced features that screen physical spaces for social distancing and aid in contact tracing.
Patriot One Magazine sat down with two team leaders—James Stewart, svp of video analytics, and senior product manager Matthew Carle to give us the details on how the new technology works and why it’s a step ahead of what’s already on the market.
Congratulations on your funding. Tell us a little bit about the health and safety modules you’re developing and how you’ll be putting this money to use.
James: What we’re developing is a pandemic management toolkit. There are three parts to it: thermal detection of elevated body temperatures, mask compliance detection, and contact tracing and social distancing monitoring.
Initial versions of the first two—elevated body temperature and mask detection—will be ready in our next release on Monday. The third component for social distancing and contact tracing will be available later this year. Like any new disruptive technology firm, we continually refine and improve our work. We’ll be hiring new team members with skills in AI, data science, software engineering, and project management to further develop these technologies.
Matt: Because we have a mature computer vision product with our PATSCAN Platform, we’re able to add multiple modules, versus just one feature. There are companies that are doing mask detection, and others doing social distancing or even thermal for elevated body temperature detection. Our goal is for the PATSCAN Platform to offer all four health and safety solutions—elevated body temperature, as well as facial mask, social distancing, and contact tracing detection—and even additional modules to support emerging pandemic management policies. Plus, we have our physical threat solutions including weapon and disturbance detection.
It’s an interesting addition for Patriot One Technologies, which entered the market as a weapons detection platform. What spurred your move into this health safety space?
James: We’re a threat detection platform. We’re career law enforcement and public safety professionals who are in this to protect people. COVID-19 is just another threat facing our global communities.
Matt: Right, and PATSCAN already does thermal screening and object recognition to find concealed weapons and identify violence. These health safety capabilities are new to our existing platform, but we’ve been developing the underlying AI pipelines and technology for years.
Critics have already noted thermal detection cameras measure external body temperature, not the internal temperature that indicates a fever. What does your technology do differently?
James: Cameras that can detect elevated body temperatures on their own are quite expensive, up to $25,000 CAD per camera. Our software enables elevated temperature detection using only cheaper, commodity-based thermal cameras that have too wide a variance in temperature accuracy to do that on their own. With our approach, we don’t need to know what an individual’s precise temperature reading is from the thermal camera. We learn what normal looks like when someone presents to the camera. When someone presents differently, we highlight them for secondary screening.
Matt: That’s right—we remove the dependence on absolute temperature which can vary by environment or even time of day.
Why should we trust AI over, say, a human being with a thermometer gun?
James: Well, we’ve been developing our AI pipelines for four years. We have several levels of evaluation that takes place throughout this pipeline. We track false positives and true positives at each level, along with the opposite, false negatives and true negatives. So, we might do a frame-based evaluation, and decide, “Okay, this model appears better.” And then we put it through another set of testing on actual real-life videos. And if it still looks like it’s a better model, we push it out of the lab, and into a real-life environment and monitor how our metrics perform in the wild before pushing it out to production. It’s a very scientific, staged process.
Most importantly, AI helps us sift through a massive amount of data and detect anomalies that trigger our platform to alert someone on site, like a security guard at the entrance of a building. Our platform, with all its sensors, is designed to provide enhanced and better information, so onsite security can respond appropriately when a threat presents itself.
Matt: Our thermal model is being tested at five locations: a university, a sports stadium, as well as a U.S. hospital, a law enforcement facility, and a business office. As with all of our products, we only push it out for general release when we’re comfortable that it’s delivering the high performance we demand.
Still, the term “artificial intelligence” still makes people nervous, particularly when it intersects with security.
Matt: Totally. In the AI companies of the world, you see there are two groups. There’s the “We use AI to replace people” group. That’s a very common message, which is scary, even for me. And then there are AIs that are built to support people. Patriot One is very much in that vein, using AI to crunch massive amounts of data to provide onsite security with information to protect people. We’re here to be part of the team, not to replace the team.
James: We’ve been criticized in the past by people saying, ‘‘We want to hook this right into our 911 system directly. Take the human out of the loop.” We tell them that that’s not appropriate. There will be false positives, which will alert police. Some of these same people then tell us that our competitor gets only one false positive, per 40 cameras, per 30 days. I respond how impossible that is with the current state-of-the-art. Even if this is possible someday, there should always be a human being making the decisions.
So then how should it work?
James: AI should be used as a force multiplier, allowing security and screening staff to do more to increase the flow of people, allowing people to get into the office and back to normal, versus having to stop for an invasive close contact measurement. If you can highlight anomalies, and highlight those individuals setting off an alert for a secondary screening, that’s valuable.
Matt: It’s the same on the PPE (personal protective equipment) side. It shouldn’t be about that you’re not wearing a mask, so therefore the door doesn’t open. This is really an opportunity for organizations to screen people entering a building quickly and efficiently. So when onsite security sees somebody without a mask, it’s their chance to step in and educate on why masks are important for the safety of everyone in the building.
So, we’ve discussed the thermal imaging and facial mask detection technologies. What about the social distancing and contact tracing portion of this solution?
James: That is future work. Once we identify a person with elevated body temperature and security has confirmed their fever, we want to be able to trace where they’ve been and who they’ve been in contact with during their visit. It’s a bit of a leadership piece, and potentially where this Canadian Innovative Supercluster Initiative could evolve to.
Matt: What’s important is that we always start with the threat, not the person. We don’t identify anyone unless a threat has been detected. Once a threat is detected, we can track using people’s clothing, so we can identify other people they have come into contact with. And that ability is recursive, because whoever they came into contact with, then we want to check who they came in contact with after that initial interaction, and so it goes.
Do you think people will welcome that kind of tracking, given the scope and breadth of the current pandemic?
James: At this level, we’re not tracking the person, we’re tracking the threat. Again, we don’t identify or store information on people: we only surface information pertaining to a threat our platform has detected, be it a physical threat like a gun or rifle, or invisible threat like elevated body temperature which could be virus related. The idea is, by tracking somebody by clothing, we can still keep people safe while staying on the correct side of the civil liberties and privacy discussion. It’s about balance.