I Spy With My Electronic Eye
WITH A POPULATION OF more than 1.2 million people, the city of Birmingham, Alabama, has had the typical urban crime problems, but it has been making strides in crime control through efforts such as the City Action Partnership, which has lowered crime in the downtown area by more than 60 percent. Now it is using cutting-edge video analytics technology on crime hot spots.
The city is unusual in that it contracts out its video monitoring to ION Interactive, which the company’s vice president, Richard Cruit, describes as “an end-to-end solutions provider…. We are not a regular integrator because they usually sell a specific line of products. We go out and source best-of-breed products. You can say we’re ‘equipment agnostic.’”
ION, though it is itself a service provider, is also the end user with regard to what it purchases to deliver that service, because, “We own the equipment—the cameras, the transmission lines, the servers,” says Cruit. “Our clients rarely even know what product we put in.”
In early 2008, ION worked with the Birmingham Police Department to place surveillance cameras in dozens of the city’s crime hot spots. These were determined not just by statistics but by ION representatives spending time with beat officers who pointed out locations that they suspected were prone to illegal activities such as drug dealing.
ION then installed pan-tilt-zoom cameras that were set up to send video feed to ION’s monitoring center over a proprietary network. The video is monitored, recorded, and stored. From the monitoring center, personnel can dispatch officers if any criminal activity is seen or suspected.
Almost immediately, the camera network began to catch a number of illegal activities. According to Cruit, the police department executives were “pleased because they didn’t have to take people off of [other] productive work.… They didn’t have to redirect resources towards folks sitting behind a bank of monitors watching cameras.”
Later in the year, ION became aware of a new video analytics product from BRS Labs of Houston, Texas, called AISight, which later received one of ASIS International’s Accolades Awards at the 55th Annual Seminar and Exhibits in Anaheim, California. What differentiates AISight from other analytics is that it is not rules-based; rather, it has the ability to learn from the visual input it analyzes, including human behaviors. Says Cruit, “We had been using a low-end analytic that simply told us the difference, say, between a tree blowing and the sun coming out versus a person or a car. Then we were introduced to AISight and saw the potential for using it with Birmingham’s crime surveillance cameras.”
The AISight system consists of four components: a video-analysis engine; a “machine learning engine,” which uses technology based on human cognition to build behavior pattern models; a system manager, which controls how AISight issues alerts and to whom; and an operator console, which provides operators with real-time views and notifications of alerts, displays live feeds for closer analysis, and manages the frame-by-frame observations of objects and behaviors from any standard desktop computer.
ION began a trial of AISight last April. “We brought their servers into our shop, and we picked out a number of test sites, all crime surveillance,” states Cruit. “Each camera fed a video stream into the AISight engine; and it began immediately to learn the scene minute by minute, hour by hour, day part by day part—because different things might happen at night than during the day. For example, there might be more foot traffic during the day and at night the foot traffic might draw down to nearly zero. If there is foot traffic in an area or people coming together to congregate [in the areas covered by the hot-spot cameras], there’s a high probability that it’s a drug deal. AISight helps us by saying ‘This is not what I normally see during this day part; come look and tell me if you want to see this again…and if this is something you want me to point out or not.’”
If the behavior is not one that the ION monitoring technicians are concerned with, telling AISight to ignore it is as simple as clicking a button on the computer screen.
Cruit says that the amount of time it took for each camera to learn about the environment varied from as short as a few hours for a fixed camera to as long as several weeks for a pan-tilt-zoom. “After that, it had built its memory and began to deliver alerts to us. And we are able to tell it, ‘Yeah, this is what we want to see,’ or ‘No, you can ignore that.’
“Our monitoring techs sit in front of a bank of monitors and a 16-foot screen on which are broadcast the hot cameras for that particular day part. The rest of the cameras stay down in the bank of monitors, and the cameras change depending on what day it is or what time of day it is, when we want the techs to be watching those scenes heavily,” explains Cruit.
BRS Labs worked with ION to make changes to the way that AISight alerts are perceived by the monitoring staff. “The camera feed is surrounded by a cool blue box. When there is an anomaly in the scene, the box changes color to yellow.
One area where ION saw an immediate improvement over the old analytic software was with left behind objects. “With in a number of days of being installed, AISight was watching a retail plaza. A kid stopped, put a backpack down and walked off. About 8 seconds later, we got an alert on it. We looked at a prevent clip and saw what had happened. It was a benign situation that time,” notes Cruit.
ION finished incorporating all the Birmingham hot-crime cameras with AISight in December 2009. “We are still making small software changes with BRS. There were no installation issues at all, but [there were] technical issues—as always— it’s software,” says Cruit. “What we’ve found, however, is that BRS is very responsive to us and to fixing anything we need fixed. It’s been a very stable product that’s proving quite valuable to us and to the Birmingham Police Department.”
(For more information: BRS Labs, phone: 713/590-5160; Web:www.brslabs.com; email:[email protected].)