There is a perception, perhaps fed by high-tech crime-busting TV shows, that surveillance cameras are always in the right spot and that they produce pictures of such high resolution you can count the whiskers on the bad guy's face.
The reality is something else. Cameras break. They may produce fuzzy pictures, or they're pointed the wrong way. But the system is evolving, and two years after ABC7 first reported on 'the intelligent iris,' we thought we would return for part two.
There are thousands of cameras now capable of sending live feeds to Chicago's Office of Emergency Management and Communication (OEMC). Human eyes can't watch it all. And if a camera does happen to record a crime, it may take hours and hours to wade through miles of tape. Video analytics can reduce that chore to minutes.
"It can cut 12 hours of man hours down to 20 minutes with one person as opposed to three people sitting there at various computers," said Nick Beaton, OEMC Operations Center commander.
Analytics allows you to tell the system what to look for. Say, for example someone has just robbed a bank in the South Loop. Witnesses see him get into a Navy blue min-van, and he heads north on Dearborn.
"You say it was at 10 a.m…so we'll do the search for about the ten minutes thereafter," said Rick Kjeldsen, IBM.
Rick Kjeldsen helped develop this system. He enters time frame, basic vehicle description, and calls up what the cameras have recorded.
"So I'm gonna select the cameras in the vicinity that if they headed north on Dearborn," said Kieldsen.
The cameras at OEMC are recording about 2,000 vehicles per hour, but with those basic search parameters the system gives us a manageable number of thumbnails matching the description of the bad guy's car.
"When I click on an individual thumbnail, it'll pre-roll. In other words, it'll back up a few seconds for this particular frame of video and show me that sequence," said Kieldsen.
Kieldsen tries a few potential hits and then a few blocks north on Dearborn, here we are at about 10:06.
"And I stop it right there," said Kieldsen.
Depending on camera angle and resolution, enhancing the video has the potential to produce the license plate.
The would-be criminal heads north. A camera at the river would normally have provided a head-on shot, but earlier in the day for some reason it was re-positioned. Still, we are caught again, turning east on Illinois.
"We're going underground. They'll never be able to see us," said ABC7's Paul Meincke.
True in that short stretch, but as soon as we come above ground, another camera has a nice picture. Here we are later on Lower Wacker, and earlier at the beginning of our mission. Granted, the bad guy is not going to be seated at OEMC confirming his route of travel, and he did happen to make his getaway in an area where there are lots of cameras. But if this were the real deal, there would be multiple analysts working the analytics both here at and police headquarters where they have access to the same video.
"I've been here and watched the dynamics of it and watched it play out where they actually followed it camera to camera," said Commander Steve Culeris, Chicago Police.
Analytics is a tool that's evolving rapidly, and the near future will bring more automation.
"Once camera one for example finds what it's looking for, it might alert camera two and say at such and such a time here's what I found. It's coming in your direction. I'm anticipating it'll be in your field of view at such and such a time. See if you see this," said Roger Rahayem, IBM.
"This is not a be all and end all. That's the thing. Analytics is not going to replace anything. We still need to have the people down there with the eyes looking at it. The best thing that's ever going to work is human interaction," said Kieldsen.
We don't want to give the misimpression that video analytics is designed just to follow cars. It's far more than that. For instance, cameras trained on sensitive areas can signal alerts if someone places a bag or object, and then leaves. Electronic trip wires that only the camera and computer see can alert law enforcement if someone enters a zone that's supposed to be secure.
