- By Michael Fickes
- March 1st, 2012
Video analytics arrived a number of years ago to great fanfare, and many campus public safety directors equipped their video surveillance systems with the new technology.
Distinct from old-fashioned motion detection that alarms on virtually any motion within a frame or frame partition, video analytics uses sophisticated algorithms and partitions to interpret shapes, sizes of shapes, and speeds. “The technology can identify shapes made by moving pixels as humans, vehicles, or packages,” says Alan Lipton, chief technology officer, intellectual property with Reston, VA-based Object-Video, Inc., a video analytics provider. “You can set the technology to alarm when a person crosses a threshold or loiters outside of a door.”
As valuable as analytics may sound to campus safety directors, not everyone found the technology useful, and some even gave up on it.
Industry observers say that safety directors with successful campus video analytics and surveillance programs selected appropriate applications, set up the cameras properly, held out realistic expectations and, perhaps most important, shifted the public safety department’s goals away from investigating crimes to prevention.
Appropriate Video Analytics Applications for Campuses
“It’s all about the correct application,” says Paul Timm, PSP, president of the Lemont, IL-based RETA Security, Inc., a security consultancy specializing in education. “If you select the wrong application, you will generate too many nuisance alarms.”
Since video analytics first came to market, vendors and successful users have learned more about what constitutes appropriate applications. “Analytics divide into two categories of applications,” says Leon Snyman, director of video analytics with DVTel, Inc., a video system manufacturer based in Richfield Park, NJ. “Behavioral analytics analyze what people are doing: a group of people converging on a person, a person with a gun, or a sudden vehicle stop.”
Behavioral analytics work in a controlled environment, continues Snyman. The real world, however, presents too many anomalies, which cause behavioral analytics to trigger nuisance alarms.
“On the other hand, analytics are good at detecting people entering an area and people loitering — at certain times of the day,” Snyman says. “For example, people typically move in and out of a campus research building throughout the day. Using the system to detect people entering would be useless. But it might be useful to detect people loitering outside of a science building where sensitive research is being carried out.”
“At night, you can set an interior camera to detect an intrusion — someone entering the building when it isn’t expected.”
ObjectVideo’s Alan Lipton looks at it this way: “Right now, video analytics are good at seeing what a few people are doing,” he says. “Two people have entered an empty room. One person is loitering outside a residence hall late at night. As the crowd gets bigger, it becomes more difficult to identify individual activities.”
As the number of people in the video frame increase, it becomes more difficult to identify individual actions, continues Lipton. Analytics can, however, report general movements of crowds. Analytics can alarm when a crowd is forming, dispersing, or even moving in an unexpected direction.
Positioning the Cameras
Observers also attribute the failure of some video surveillance and analytics systems to flawed camera positioning. “We’ve had customers complain that the analytics didn’t detect an event,” says Lipton. But when we reviewed the video, the sun was shining directly into the lens and washed out the video. An easy rule of thumb to follow is that if you can see it, so can the camera.”
If the sun is blinding you to the scene, it will blind the camera. Raise the camera up and aim it down to get the sun out of the scene. Careful, though: don’t let tree limbs or streetlights block the view.
If cameras are positioned properly with correct lighting, video analytics are 98 percent accurate, indicates DVTel’s Snyman. “A good video analytics algorithm may use parameters like size of objects, speed, and distance traveled within time,” he says. “So although it is possible to detect small movements, it is logical that when a camera has a wide field of view and is placed at the correct height, the analytic rules can be better tailored to discriminate between real and nuisance alarms.”
After positioning a camera, it is important to test it. If you want the camera to alarm upon seeing someone loitering late at night, conduct a trial run and make sure the analytics trigger the alarm. Adjust the camera until it does.
Campus safety directors often complain about the number of nuisance alarms caused by analytics. It’s unrealistic to expect only true alarms from video analytics. You can adjust the cameras in an appropriate application to limit nuisance alarms, say observers, but a security officer must make the final decision about which alarms to follow up on and which to ignore.
If a particular application produces too many alarms even after you’ve reconfigured the cameras, you may need to adjust the way you respond or scrap the application.
How many alarms are too many?
It depends upon your ability to respond. At the 140-acre Johns Hopkins University campus in Baltimore, Executive Director of Campus Safety and Security Edmund G. Skrodzki has been installing cameras with video analytics since 2005. The system currently monitors nearly 200 cameras and produces, on average, 3.7 alarms per minute, which may seem like a lot.
But Johns Hopkins security officers have undergone training in how to respond to alarms. “The average event-per-minute rate has remained fairly consistent with time,” Skrodzki says. “Our people are trained in how to react to each type of alert. They see the value of the technology and continue to approach this part of their function enthusiastically.”
The Analytics Mindset: Prevent First, Investigate Later
Security professionals experienced with the technology say that video analytics changes campus policing and security by adding a proactive, preventive function to what has always been a mostly deterrent and investigative function.
For instance, one reason that the Johns Hopkins rate of alarms per minute is so high is that Skrodzki wants to prevent campus bicycle thefts. He has cameras aimed at all the bike racks. “We have caught numerous people stealing bikes,” he says. “Often, they are looking at the camera while cutting the locks — they don’t believe the cameras are monitored.”
Between 2005 and 2008, the system cut bicycle thefts from 36 per year to three. Thefts have remained low ever since.
Cameras with analytics also monitor campus garages between 6:00 p.m. and 6:00 a.m., when few people are expected to be in the garage. Cameras monitor the alleys on campus and bordering the campus after 8:00 p.m. at night, again a time when no one is expected to be in the alleys.
“Cameras are a tool,” Skrodzki says. “You still need people nearby to respond in a timely manner to alarms. That’s important, and we train for it.”
What’s Next for Cameras and Video Analytics?
Video analytics providers aren’t standing still. They are constantly refining applications. Facial recognition, an application that has not worked well in the past, may be rounding into shape. While it may still be too expensive for colleges and universities, casinos have been using the technology with some success.
Similarly, people counting may become an important security application. If a building catches fire, for instance, a people-counting application can report how many people were in the building before the fire, how many came out when the alarm sounded, and how many are likely still inside.
Still another innovation that may soon come to campus is video analytics combined with audio analytics. Cameras with video analytics pick up a person walking in the alley behind a building and either breaking a window or washing it. Audio analytics can pick up the sound of glass breaking. The video and audio analytics together qualify the event for an alarm better than video alone.
These applications remain new and probably expensive. But they bear watching, because as providers refine the technology and a growing base of users drive down prices, they will add to the current capabilities of video analytics.