Computer Sentry

The Johns Hopkins University (JHU) campus is in a residential area of Baltimore City. The area around the campus has a reputation for crime.

At 11:30 PM one recent night, a flat-screen monitor in the JHU security center lit up in response to an alarm signal sent from a camera located in an alley just inside the campus perimeter. Equipped with a new technology called video analytics, the camera sent an alarm upon detecting a slow-moving vehicle and a person on foot in the alley.

It was probably nothing, but the system had been designed to alarm on unusual things so that a security officer could take a closer look — just in case.

Noting the alarm, the officer on duty in the security center used the camera controls and zoomed in on the pedestrian to capture a good shot of his face. The she got a close-up of the license plate. Finally, she got a shot of the whole truck.

A few minutes later, two patrolling JHU security officers reported an armed robbery on Charles Street, not far from the alley. A pick-up truck had driven up to a woman on foot. A man had jumped out of the passenger’s side, pointed a gun, and stolen the woman’s purse.

The officer in the security center had just seen a pick-up truck and quickly checked it out. She ran the tag and called the police, who visited the truck’s owner. He said that he had loaned the truck to a friend. The police checked on the friend and discovered that he had a record. They pulled his photo and made up a photo-lineup. The robbery victim picked out the friend.

Over the next few days, two other victims of crimes in the same neighborhood identified the truck and the robber.

“The system solved three robberies,” said Edmund G. Skrodzki, executive director of security at JHU.

The JHU campus security department has installed 166 cameras on campus since 2005. Fiber-optic cable connects the cameras to the security center, where the video runs through a server equipped with video analytics, a technology that is also called behavior recognition. Supplied by Reston, VA-based Cernium, the JHU system is called Perceptrak.

“If we monitored video ourselves, we probably would have missed the truck,” Skrodzki said. “Studies show that people lose concentration after watching a security monitor for 15 or 20 minutes.”

But computers don’t lose concentration. A video analytics or behavior recognition system watches for activity it has been programmed to look for and alarms when the activity occurs.

The JHU system can alarm on 18 different behaviors, including a slow-moving person or vehicle, a fast-moving person or vehicle, a lurking person, two or more people approaching each other, a vehicle that stops suddenly, and so on.

The security department has programmed the system to recognize different kinds of behavior at different times of day. For example, the system can be set to recognize and alarm when a crowd forms. On a busy campus during the day, that would not be unusual. But it would be unusual at night in areas where you wouldn’t expect to see people. The security staff, drawing on years of experience with what happens when on campus, set the system appropriately.

How Does Software Recognize Behavior?
“Object identification is based on pixels,” said George Maroussis, a product manager with Verint, a video analytics provider headquartered in Melville, NY. “But don’t confuse this with old-fashioned motion detection. Motion detection alarmed when the pixels changed — which indicated motion.

“Today’s systems identify people as well as different kinds of vehicles by their shapes and relative sizes. Once the system identifies an object as a person or a vehicle, the system can track the object through the scene, which is composed of objects that don’t move — buildings, sidewalks, and so on.”

Complex mathematical formulas called algorithms underpin video analytic software applications that can identify and track objects. After more than a decade of refinement, video analytics or behavior recognition applications have developed powerful capabilities.

For example, Vidient Systems, Inc., a Santa Clara, CA-based developer of video analytic systems, recently introduced an application that automates object tracking. When the Vidient system detects an intruder, it directs a pan-tilt-zoom camera to point and zoom in on the intruder. Next, the system adjusts the pan-tilt-zoom controls to track the intruder’s movements. If the human shape disappears behind another object, the system automatically zooms out to a wider view, finds the intruder again and zooms back in. The camera will shoot a lot of close-up footage, ensuring good shots of the intruder’s face. The system tracks vehicles the same way, taking many close-up shots to capture a clear image of the license plates.

Are Colleges and Universities Adopting Video Analytics?
Steve Foley, senior vice president of Verint’s video division, hasn’t seen a major move by colleges and universities to adopt video analytics. “The first thing schools are doing is standing up video surveillance platforms,” Foley said. “Many campuses are ready to go because they are IP enabled.”

Traditional video surveillance systems use analog cameras connected by coaxial cable, with power supplies near the cameras, continued Foley. But colleges and universities typically have campus-wide Internet Protocol (IP) network systems. That means schools can leapfrog analog systems and go directly to digital cameras that can plug into the existing IP network and eliminate expensive cabling costs. The current generation of cameras can also acquire power from IP systems, eliminating expensive power supply costs as well.

Once schools have installed video surveillance systems, they can begin looking at video analytics, Foley said.

The success of video analytics on the campuses of early adopters like JHU will eventually promote their use. According to Skrodzki, the JHU system has shown encouraging results.

The system has reduced bicycle thefts, for instance. Skrodzki placed cameras near all the bicycle racks on campus. “Whenever anyone goes past the bike racks, we get an alert and an officer takes a look,” he said. “We do this 24/7. Before we started, in 2005, we had 36 bikes stolen; in 2008, thieves stole just three.”

In the past 26 months, the main campus and contiguous areas have seen no robberies. “Robberies do occur a couple blocks away,” Skrodzki said. “We have pushed robberies back away from campus.”