X

Piecing together smart cameras at DemoFall 09

Today's "smart" surveillance systems are really rather dumb, according to experts in recognition technology.

Rafe Needleman Former Editor at Large
Rafe Needleman reviews mobile apps and products for fun, and picks startups apart when he gets bored. He has evaluated thousands of new companies, most of which have since gone out of business.
Rafe Needleman
3 min read
The Viaas camera Third Iris

At Demo on Tuesday, Third Iris pitched Viaas, a video-monitoring system for business that's simple to install and use. Plug the cameras into just an Ethernet cable (if it's enabled for Power-over-Ethernet, that is), log into the Viaas Web site, and you can get your own business surveillance system up and running in a snap.

From a business perspective, Viaas follows the mobile phone model: the cameras, which sell for a low price ($199.95) considering their high-end sensors, are subsidized by the monthly fee you pay to access them, $29.95, or more if you want to have access to historical video beyond the default four-week archive. Each camera in your system incurs a monthly fee.

It seems like a solid offering for small businesses or even some residential installations that need remote access to cameras. The system has some smarts, too: The video service detects motion, and can send alerts based on that. But the image recognition system is very broad, as competitive products are. Even with motion detection, reviewing the "tapes" for a particular person or type of incident would be tedious.

Chris Shipley (left) with pattern recognition experts, from left: Donna Dubinsky and Jamie Niemasik of Numenta, Rob Haitani of Vitamin D, and Dick Lyon of Google. Rafe Needleman/CNET

Wednesday at the conference, we heard from three companies doing much more sophisticated pattern recognition. These emerging technologies could make systems like the Viaas more valuable.

The Vitamin D technology can pick people out of crowded backgrounds. Screenshot by Rafe Needleman/CNET

First, Numenta CEO (and Palm co-founder) Donna Dubinsky showed off her company's use of "cortical algorithms" to do sophisticated and rapid visual pattern learning. The demo showed how the system could tell a chair from a sofa, not based on programmed rules but rather human training and feedback: You tell the system what's what, and it learns. But from a security perspective, the more impressive demo followed Dubinksy's. It was from Vitamin D's Rob Haitani. This company's system can recognize shapes of people and things in real time, or as a video is being reviewed. You could, for example, find all clips of someone leaving a storeroom, and if that gave you too many returns, you could filter out the video to show only people who then walked out the building exit door next to it. See the cool video demo.

The Viaas video-monitoring system I opened this story with does not record sound, just video. Third Iris CEO Steve Roskowski told me there are legal issues around storing peoples' conversations. But he also said he wished he had put mics in the cameras anyway. The third science demo Wednesday, after the two video demos, from Google's Dick Lyon, showed how some of the same types of pattern recognition that work on video can apply to audio recordings.

Lyon's demo was not about speech recognition but rather creating and recognizing the fingerprints of certain sounds--a particular bird call, or glass breaking, or an engine starting. As Dubinsky said during a discussion after all the science demos, the best results in real-world recognition systems, and by extension, surveillance, will come when we have audio and video recognition engines reinforcing each other, the same way humans and other animals process their sensory inputs.

Dick Lyon sees a world of audio search. Screenshot by Rafe Needleman/CNET

See also: Avaak's consumer-grade camera system (which has no motion detection at all), launched at the March 2009 Demo conference, and a my Real Deal podcast on security cameras.