Intelligent video analytics, which have been on the scene for several years now, take surveillance to the next level. Rather than merely capturing video, the system can be programmed to alarm for certain suspicious or prohibited activities, such as leaving a package, walking the wrong way in a restricted location, or driving too fast.
Some of the analytics that companies are mastering are the identification of left behind objects and tripwire detection. Motion detection, though not technically an intelligent analytic, is also getting grouped into this category.
In general, intelligent video is still an early market. But proponents believe that intelligent video may eventually facilitate a sea change in the way CCTV is used, making it more of a proactive tool than a forensic one. “You could do it [analytics] in real time and it’s providing you actionable intelligence,” says Sandy Jones of Sandra Jones & Co., a security consulting firm. Goldberg says analytics can turn a false sense of security into “true security.”
As with many cutting-edge technologies, however, when purchasing a system that purports to offer intelligent video, it’s essential to sift through the hype.
Many analytics still have too many false alarms to be helpful, according to Nilsson of Axis Communications. “Ninety percent accuracy for some applications isn’t good enough,” Nilsson says. Those 10 percent false alarms add up to consumers not trusting applications.
There are exceptions to that, he says. In some cases, even just 90 percent accuracy is desirable over nothing. For example, he cites tripwire installations in subways that develop alerts for people going onto the tracks. “It’s much better than not catching anything today, because most of the time, no one is watching those cameras,” Nilsson says.
A trend cited by several integrators and others in the CCTV world is the continued move to put analytics at the “edge,” near or in the camera or at the encoder. The other option is to have analytics conducted at the “back end,” such as at the NVR or DVR.
There are several advantages to having analytics in the camera. First, they would be working on uncompressed video, giving the analytics’ algorithms more pixels with which to work. Second, the move of the analytics to the front end allows for less processing and storage power needed on the back end, according to Fenton.
Essentially, the decision of whether the footage is important can be made before the video is streamed over the network for live viewing or for storage. “There’s no point in recording 23 hours and 55 minutes of video with nothing going on,” says Alan Matchett, CPP, of Johnson Controls.
Vidient, for example, has a standalone device that is installed on the camera side of the network and carries out compression and analytics, among other functions. When an alert goes off, selected clips of live video can be streamed over the network, rather than all of the video. Thus, the system can serve as a real-time security tool but not take up as much network space as it would if all of the video had to be streamed constantly over the network.
There are some drawbacks to having analytics in the camera rather than in a “centralized architecture” back at the server, however. For one, each camera would need to be fitted with the software, and you would need licenses for all, according to Sarangan.
Additionally, although prices of intelligent analytics are noticeably decreasing, analytics do still increase the price of the camera, sometimes by hundreds or thousands of dollars. “I don’t think everyone can really afford to have every camera with analytics on,” says Sarangan. “They might pick the one door or the one point of entry or one location where they have noticed that there’s the highest crime or highest threat.” Sarangan says with that approach, some cameras are left out. In contrast, systems that use centralized analytics, all the data will be reviewed.
And most importantly, some experts just don’t think the cameras have enough processing power to adequately perform all the analytics a client might request.
One compromise is a combination solution that includes analytics in multiple spots, starting at the edge. Howes calls this the “winning” solution. He thinks optimized analytics will be done at the edge and then “as new analytic capabilities come along, they’re going to be perfected in servers.”
An example of analytics that Howes believes must stay server based for now is any type of facial recognition. To begin with, such biometric applications require the use of databases for comparisons. Additionally, certain analytics require more processing power than what is generally available in camera technology today. Some companies have moved to fix the processing problem, however. Goldberg says Vidient has added parallel processing power to its device.
Banerjee agrees that some analytics require too much processing power for cameras, but he argues that the market for those analytics is small. Banerjee adds that his company created cameras that can handle such applications as objects left behind or removed and people or objects moving in the wrong direction.
The capabilities of IP-connectivity, storage systems, and intelligent analytics will continue to evolve in the coming years. The challenge for security directors, as always, is to assess which advances in technology offer enough of a boost in system performance for their own application that they merit the investment.
Laura Spadanuta is assistant editor at Security Management.