Do Something Great • January 09, 2009
"Better Eyes"
Professor James Davis is developing a new computerized surveillance system that could help law enforcement around the state.
!["Better Eyes"](https://webarchive.library.unt.edu/eot2008/20090116080227im_/http://www.osu.edu/images/rotate/2009/surveillance/main.jpg)
To view flash video, this browser needs the Flash 8 (or higher) plug-in
Transcript
James Davis: So the camera's actually located on the rooftop and can adjust its pan and tilt to follow the person through the scene. But it also has a zoom adjustment that tries to keep the person the same height all the way through the scene. So right now, they're moving pretty far away here.
So typical surveillance cameras have pan, tilt, zoom capability, but the problem is that most people have to try to re-orient and move the camera around the scene to look at different areas. And, furthermore, the actual camera view that comes out of this, that they show on a TV monitor is actually quite small compared to the actual view coverage area that the camera could potentially see. So, in our system, we've developed a method to automatically generate a wide-angle panorama for the camera and with this you can actually use it as a broader context for this camera so that when they look at the smaller view, now they can place it within this larger region.
Our algorithm is designed such that a person could actually point or click anywhere in this panorama and have the camera automatically point and move to that particular target. So to accommodate multiple surveillance cameras, we take the panorama generated from each camera and place them down onto a large Google map aerial image. And the nice thing about this is that people can look at a single image and see all the areas that can be covered by these cameras. Furthermore, with our system, now the person could actually select or click any location in this Google map and all available cameras in the are can now point and orient towards this location.
So, currently in my research lab, I employ graduate students and undergraduate students. I started off as an undergraduate researcher myself and it's great to have them in the lab. The environment is very dynamic now because we have all levels of students and mentoring happening between the grad students and undergrads and it just turns out to be a very valuable experience.
Matt Nedrich: Most of the work I do involves the smart camera controls and some of the interface for them. The atmosphere here in the lab's awesome. All the people I work with are very helpful. It's been a great learning experience.
James Davis: To deal with the issues of privacy, as we have our own surveillance system here at OSU, we pretty much deal with how people move, where they're existing, but we don't spend any effort on trying to do face recognition or any types of identification. So we try to look at, again, what you're trying to do rather than, you know, who you are. Joe, Mary, or Sue; it's the activity we're mainly concerned about.
This is the new theme in surveillance is how do we detect the abnormal behavior that exists in the scene? So, the biggest issue is how do we learn the trends that are happening? Right now in surveillance systems that exist you can go in and you can select a little region in an image and say, "anytime someone enters this zone, let me know about it" and an alarm would be sent to a security guard. But we'd like to up it to the next level. Rather than annotating all the video and putting hard and fast rules in there, could it just try to sit there and observe the scene over time? Days, weeks, months, even years. And try to figure out what's the typical things that happen in the scene and if I see anything atypical let somebody know about it.
There are several things that are going on in the state, from violent crimes to Department of Homeland Security that we're trying to now deploy it across several cities where a person at a central location could, theoretically, now go look at all individual cities across the state, in real time, all over the place. And try to coordinate the activities that are happening.
With a team of students, Professor James Davis is working to create computerized “smart” surveillance, with a system of networked video cameras that would allow surveillance officers to observe a wide area quickly and efficiently.
Traditional cameras pan, tilt, and zoom. But when surveillance operators look through these cameras, they get only a tiny image that some call a "soda straw" view of the world. And as they move the camera around, they can easily lose a sense of where they are looking within a larger context.
In Davis' system, cameras take snapshots from every direction. The result is a seamless panorama, used to create a 360-degree, high-resolution view.
Once that field of view is displayed on a computer screen, operators can click a mouse anywhere within it, and the camera will pan and tilt to that location for a live shot. Or, they could draw a line on the screen to track a particular route--a certain street, for instance.
The networked system of cameras allows security officials to "follow a person's trajectory seamlessly," Davis says.
The next step is determining who should be followed. Davis wants the system to be able to pick up on trends--when an area is busy, for instance, or how a lost person looks differently than someone who's acting suspiciously.
"Right now in surveillance systems that exist you can go in and you can select a little region in an image and say, 'Anytime someone enters this zone, let me know about it,' and an alarm would be sent to a security guard," Davis says. "But we'd like to up it to the next level."
“We care what you do, not who you are. We aim to analyze and model the behavior patterns of people and vehicles moving through the scene, rather than attempting to determine the identity of people. ”
—James Davis, Ohio State professor of computer science and engineering
"Could it just try to sit there and observe the scene over time? Days, weeks, months, even years. And try to figure out what's the typical things that happen in the scene, and if I see anything atypical, let somebody know about it?"
Davis says the surveillance system is concerned with actions, not people.
"To deal with the issues of privacy, we pretty much deal with how people move," he says. "We don't spend any effort on trying to do face recognition or any types of identification. So we try to look at, again, what you're trying to do rather than, who you are, Joe, Mary, or Sue. It's the activity we're mainly concerned about."
Davis is looking at testing the system through the state. Law enforcement could link video cameras around the major cities, map video panoramas to publicly available aerial maps, and use their software to provide a higher level of “location awareness” for surveillance, he says.
Davis' lab includes three Ohio State students: doctoral student Karthik Sankaranarayanan, who is funded by the National Science Foundation, and two undergraduate researchers, Matt Nedrich and Karl Salva.
"The atmosphere here in the lab's awesome," Nedrich says. "All the people I work with are very helpful. It's been a great learning experience."
Says Davis: "I started off as an undergraduate researcher myself and it's great to have them in the lab. The environment is very dynamic now, because we have all levels of students and mentoring happening between the grad students and undergrads, and it just turns out to be a very valuable experience."