Software soups up surveillance

Print | Close this window

http://news.cnet.co.uk/software/0,39029694,49249027,00.htm

26 January 2006

Michael Kanellos

There was a good reason why British authorities were able to identify the suspected London bombers last year more quickly than usual from surveillance tapes.

They assigned scores of officers to review the video footage from all of the cameras, said Tim Ross, co-founder of San Francisco start-up 3VR Security. Most surveillance operators don't have that manpower -- often, there's just a couple of pairs of eyes looking at pictures. To fill that gap, 3VR has come up with a software alternative that lets building owners and security officers rapidly comb footage from video cameras.

"Surveillance video is the largest mass of unstructured data out there, but if you bring structure to it, you can build new capabilities," said Ross, who noted 3VR wasn't involved with the 7 July effort. "There are three incidents a decade where you can afford to" get a large number of officers to watch.

Rather than have security guards watch archived video footage in a linear fashion, the software breaks up the footage into key scenes and creates an index of searchable thumbnails. Once it has labelled a person -- let's say, as visitor number 1430 -- the software's facial-recognition engine will pull up other scenes featuring that individual or similar-looking people.

If that sounds conceptually similar to TiVo's navigation and chopping up of video, there's a good reason for that. One of 3VR's other co-founders, Bob Vallone, used to be TiVo's vice president of engineering.

The company, along with competitors such as IBM and Pixlogic, believes that a ripe market exists for marrying information technology to physical security systems like cameras.

Simply put, current surveillance techniques aren't very efficient. In late 2002, Security Solutions, an Australian magazine devoted to physical security, reviewed a series of studies conducted at different military installations worldwide. It found that after approximately 12 minutes of continuously viewing two monitors, operators will miss up to 45 per cent of scene activity, and the figure climbs as time goes on. To find footage of a suspect, guards have to sift through hours of video amassed by several cameras.

Billions of hours of security video is shot every week in the US, Ross said. After all, each one of those cameras in casinos, banks and convenience stores captures action 24 hours a day, 365 days a year. In London, public security cameras have been pervasive for years. There's one camera for every 14 people in the city, according to analyst firm Frost & Sullivan, and the number is climbing.

"In the future, digital video recorders that convert analogue images into digital format for transmission over an IP network are likely to be a critical component of video surveillance solutions," according to a June report from Frost & Sullivan, which estimated that the surveillance software will grow by 23.4 per cent a year to $670.7m by 2011.

In close-up
3VR's technology essentially tries to filter out the noise and irrelevant data captured by those cameras. In traditional systems, security officers monitor ongoing activity through a bank of monitors. Under 3VR's system, only key scenes -- when someone opens the front door, when people leave a conference room -- are shown to the guard. Each scene is delivered as a static image; when clicked, it delivers a few seconds of motion. (The entire video is stored in company databases, of course.)

These thumbnails are also presented on a single screen. Presentation in this manner reduces the chances that crucial moments will be missed, Ross said.

The software also tries to enhance image quality around the face, by compressing the data less there than in other parts of the picture, Ross said. That technique is designed to make it easier to achieve exact matches between different images of a face in a search.

3VR's technology relies on pattern recognition, or pulling together a group of similar face captures, and compares them with a picture in a database. The software is meant to help track an individual, rather than identify him or her for access to a building.

The company sells its technology as software alone or bundled with servers. The price comes to roughly $1,000 (£550) per security camera.

Clients include Raytheon, Morgan Stanley and the US Coast Guard, Ross said. He noted that financial institutions are adopting this kind of tool to curb fraud, using it to create better watch lists. If someone cashes a bad cheque in a branch in California, the bank can extract a picture of the person who passed the cheque from that branch's video stream and send it to every other outlet via the corporate network. Some clients advertise that their systems make it easier to identify individuals in video footage, Ross said, while others keep it quiet.

Hotels use it for security purposes too. At the Talbott Hotel in Chicago, a guest reported a room break-in, Ross said. Security officials quickly skimmed security highlights dished up by 3VR's software and found images of a person trying to open several different rooms in the hotel. They soon captured the thief, who turned out to be a visitor brought into the hotel by another guest.

Visit reviews.cnet.co.uk for in-depth reviews of many more products