Processing Groups of People Is the New Challenge

Guest: Arun Vemury, Program Manager, S&T Biometrics and Identity Technology Center

[00:00:00] Arun Vemury: Most people going through these security checkpoints proceed by themselves. But about 40% of people travel with friends or family. So the intent here is how do we help make sure that this stuff works, the way people actually travel.

[00:00:12] Dave: Welcome to Tech Speak, a mini episode of the Technologically Speaking Podcast. I'm Dave, sound editor for S&T. In September, our podcast team attended the 2022 Biometric Technology Rally at our Maryland Test Facility in Upper Marlboro, Maryland, and they had a great time seeing several facial recognition technologies up close and personal. At the start of this mini episode, you heard repeat guest, Arun Vemury, Director of S&T's, Biometric and Identity Technology Center, speak briefly about the challenges being addressed by this year's Tech Rally. Now let's hear Arun talk about what exactly a Tech Rally is.

[00:00:45] Arun Vemury: Today is the first day of the DHS S&T Biometric Technology Rally for 2022. We're actually putting together an industry challenge where we, give a challenging scenario to industry. We tell them how we want the technology to perform, and we let them go at it. We give them a hard problem and we let them innovate and come up with creative solutions to solve a hard problem. And today, the use case we're going after is this thing called high throughput, where we have a large number of people who we need to have their identity verified before they go into let's say secure or a sterile area. In general, we're trying to do things where people are screened in less than three seconds per person. Ideally it should be under six seconds per person, but really we're pushing for that three-second mark. And really what we're trying to do is also just give industry room to innovate. We give them a volume and they can put whatever set of technologies they want in there.

[00:01:37] Arun Vemury: And the intention is as people walk and enter that space, their biometrics are collected and their identity is verified before they leave. And all of this happens in under three seconds.

[00:01:45] Dave: This was the fifth annual Biometric Technology Rally for S&T. Each one aims to address new and emerging challenges in identity verification. Here Arun describes the focus of this year's event.

[00:01:56] Arun Vemury: For the last couple of years, our rallies have been focused on this high throughput use case where we are trying to get people through the process as quickly as possible, make it as easy and intuitive to use as possible. However, in the past it's always been one person at a time. the challenge there is that's great. And most travelers or most people going through these security checkpoints proceed by themselves. But about 40% of people travel with friends or family. So, the intent here is how do we help make sure that this stuff works the way people actually travel and they move.

[00:02:25] Arun Vemury: So, the challenge here that's used, that's new and different, as there were processing groups of people, small groups of people. So the industry participants who are working on the rally don't know if they're going to see one person or 12 people. But every time a group comes through, they're supposed to figure out how many people are in the group. Take good photos of each one of those people and send the best possible photo back to our matching systems. It's not trivial to figure out how many people are in the group, which face is the best face, and make sure you're sending the best image back to the biometric system to try to get a good matching. So, we're trying out a lot of new things here that are relatively unique. It'll be interesting to see how industry kind of rises to the occasion. And that'll help us figure out maybe is this something that might be, really operationally viable two or three years from now?

[00:03:11] Arun Vemury: Or is this something that might require five or six years? Something that might take a little bit more time. This helps DHS stakeholders understand what the state of the art of the technology really is, so that they can make better choices and plan because that might help them figure out what use cases make sense, what use cases don't make sense. So with this rally, you know, we are looking at a couple of new things. So, one is this whole idea of the privacy measurement as part of the Rally. How do we make sure that we're collecting images of people who are walking into a space and they're specifically opting in, but we're also respecting the privacy of people who specifically opt out. And they might be processed right next to people who opt in. These cameras have wide fields of view, you know, they have a wide screen.

[00:03:53] Arun Vemury: How do we make sure that we're only collecting the faces of some people, but not other people? So, we're trying to make sure that the systems are actually designed for that purpose with privacy built in. Another thing is this whole idea of group processing, being able to process a whole group of people. Maybe some are taller, maybe some are shorter. You might have tall people in front of short people. How do the systems accommodate the way people move together and travel? And then finally, as always, we focus a lot on the questions of equitability and fairness. Do the technologies really accommodate the diversity of people who are using them for things based off of gender, skin tone, race, ethnicity? We wanna make sure the technologies work for everyone.

[00:04:31] Dave: During the event, our podcast team had the rare opportunity to test the facial recognition technology and see it in action with their very own eyes. Let's listen as Arun gives our team a demonstration of the technology and walks them through how it works.

[00:04:44] Arun Vemury: During the Rally, one of the things we do is we do basically side by side testing where each company has their own designated space, and we have volunteers interact with each one of the systems, and we evaluate how well the systems work with those volunteers. Does it take photos of everyone who it's supposed to take a photo of? Does it not take photos of people who've opted out and walked around the system? When it fails or when there are errors, are they attributable to people who have like maybe certain demographic characteristics or something else? Why? Why do the technologies work sometimes and not work others?

[00:05:16] Arun Vemury: And the intention here is really to make technologies that don't have errors. We want this stuff to work to be easy to use and be incredibly intuitive for the normal traveling public to interact with. So, in each one of these systems, we have a different system mostly from different companies. In some cases, we have two submissions from one company. But it's really about trying out different technologies, different configurations, and see which ones work and which ones, maybe needs some additional refinement. This system looks like it has a user interface where it's got a screen above it where it's welcoming people to come through. It's also attracting attention so that as people look at that, there's also a camera nearby so that you get a nice posed face. As you walk through, it's figuring out how many people are in the field of view. Are you opting into the system? Are you in the area where you're definitely gonna walk through? Or are you in this area around the system where you're not supposed to walk through? So, you see on that screen, like on the bottom left-hand corner, you see those blue curtains.

[00:06:12] Arun Vemury: If you're in the blue curtain, it's not supposed to take your photo. And if you're not close enough, it's not supposed to take your photo. That's the whole point is it's not gonna take your photo. You're not close enough. But as you get closer, and I'm taking steps forward here closer and closer to the point where it's within the specific range in this case, maybe within four or five feet, that it'll actually locate my face and finally actually take my photo.

[00:06:34] Arun Vemury: And the other thing too, so in that case, you just saw that my, it took my photo as soon as I crossed that line, but it didn't take my photo before then. And if I walked over here to the side and I was going through an opt out process and I don't like facial recognition, I don't want it taking my face. I can come over here and as I turn that corner and come around here, I'm in that blue curtain area, so it's not gonna take my photo. But the theory here is we want people who opt in to go through and basically not have to do much to not have a lot of interaction, follow a lot of instructions because most people are busy and they're just trying to complete their day. So here, if I walk through, I'm not really doing anything in particular. It still collects my photo and it works and it takes that really nice photo and sends it back to the matcher.

[00:07:18] Dave: This technology is so innovative. I wish I had the opportunity to test it myself. Maybe next year. But for now, hearing from Arun twice now has definitely changed my perspective on airport screenings. You can learn more about S&T's work with facial recognition technology by checking out season one, episode four of Technologically Speaking, the Three-legged Stool. Thanks for listening, and be sure to follow us at DHS SCITECH DHS, S C T E C H. Bye.