 |
 |
 |
Smart surveillance
A UR computer science researcher has found a way to make intelligent surveillance cameras-a technology with applications in homeland security.
Randal Nelson, associate professor of computer science, has developed software to help surveillance cameras find out whether something has changed in a particular environment.
"It observes an environment, sees if anybody is moving in and out of it and, if they are, it looks to see if it has changed anything," Nelson said. "If they have, it goes in for a closer look; for example, if you want to know if somebody stole a computer out of your office or stole something out of your warehouse."
Nelson worked with a couple of people on the project who assisted in testing the software. The main goal was to make machines stand in for humans in observation tasks, Nelson said.
He started by experimenting with object recognition jobs for the camera. Based on the complexity of the item, the software may need one to several photos of the object from various angles. With those images, the software matches the new object it sees with a database to determine its identity.
The software was initially written for a smart home, to help seniors keep track of where they left personal items. At one point, a prototype of the system was running at the Center for Future Health at URMC.
"In that case we had more artificial intelligence stuff I had written, which is a program that can identify objects by their shape, so it would take a picture of what had changed and send it off to the object recognition part," Nelson explained. "A person would have access to this on a touch screen, so they would go to the screen and touch the picture and if it knew where it was it would say 'You put your glasses on the kitchen table.'"
Nelson's work attracted local firm PL E-Communications LLC, which plans to develop the technology for video cameras in security applications. For example, linked cameras might be used to exchange information about objects like a suspicious package in an airport. UR has licensed the technology to the company.
"We are hoping to make this technology do things that were long thought impossible," CEO Paul Simpson said in a statement. "Making things more secure without the need to have a human operator on hand every second."
Nelson mainly develops the software himself and is working on related projects.
"I am interested in doing connections for natural language. The system would say, 'Hey, there is a guy who is walking funny and pausing in places he shouldn't,'" Nelson said. "Or 'A man left a briefcase behind and drove off in a green car'-rather than saying 'Take a look at this tape,' which I guess is fine, if you have time to look at the tape.
"But if you had a lot of tapes, you would want to prioritize, so you would look at a written description or it could tell you, too, if you wanted it to speak."
