So Osama bin Laden is on the lam. Six feet, four inches tall, wearing a beard, a turban, and a $25 million price on his head, how long can it take to catch this guy? Possibly a very long time, simply because our methods of finding fugitives are so inefficient and antiquated. The most effective tool we have -� that $25 million reward �- is as old as money, itself. All our modern forensic technology, our fingerprints and DNA testing, our chemical analyses and digital wiretaps, are close to useless for finding one man in a crowded world. Even vaunted biometric technologies like face recognition aren't nearly as effective as their proponents claim. They can often be defeated by a pair of sunglasses or by a fake beard. But that may be about to change with the coming widespread use of a whole new approach to video surveillance. Like it or not, we are entering the era of the smart video camera, and these babies could mean the end for this bin Laden and for the next bin Laden, too.

The smart cameras I am writing about come from Oxford Microdevices, a little high tech company in Monroe, Connecticut, and from the fertile mind of Oxford's founder, Steve Morton. As always, I own no stock and have no business interest in Oxford. I just take the guy to lunch.

Steve's smart cameras are useful for identifying people from their size, shape, features, voices, even behaviors (find the man with the limp or the girl who picks pockets �- it's all the same, really). Steve's cameras use inexpensive but very powerful Digital Signal Processor chips to do this identification and the beauty of the system is in placing the intelligence in the camera, itself. Instead of gathering image data for central analysis, you just tell the smart camera who you are looking for (providing the facial, voice, or behavior equivalent of a computer virus signature) and the camera speaks-up when the target walks by. No rent-a-cops watching thousands of video screens between bites of donut. The camera does all the work.

Biometric face recognition as it is generally done is a process that does not scale well and is fraught with noisy data. You typically take a picture of someone and send it down the wire either in compressed or uncompressed form to a powerful computer that does the actual face recognition, comparing each sample image against pictures of bad guys. That's how it was done at the Super Bowl and at the Olympic games, and how it is presently being done in airports and public places all over the world. If compression is used in the camera, information is lost from the image �- information that can never be regained and used by the face recognition computer. Compression has inherent losses, and those losses degrade the precision of recognition. But it isn't any better without compression. In that case, the signal is degraded by line losses and interference picked up between the camera and computer. But by placing the intelligence in the camera, face recognition takes place before any compression or transmission.

That's not possible on current systems, where even when the image gets to that central computer, the system still doesn't scale. Face recognition computers can analyze tens of thousands of faces per second, but that isn't nearly enough for large venues. For high accuracy, a face has to be recognized hundreds or thousands of times, not just once. Multiply that times the 50,000 to 100,000 people at a football game or in a major airport, and no centralized system is big enough.

In contrast, Oxford's system using smart cameras scales linearly. Every time you add a camera, you are adding the processing power needed to serve that camera. And these cameras are cheap, costing no more than a few hundred dollars despite the fact that each runs at more than 50 gigaflops -� 50 billion floating-point operations per second. With the proper network architecture, this massive distributed processing power can be used also for searching video archives. Just send recorded video back down the line to 10,000 or 20,000 cameras and ask them to search retrospectively for some person or behavior. Years of recorded video can be analyzed in minutes using this technique. Crimes will be solved.

There are millions of closed-circuit cameras that can be replaced with Steve's smart cameras, often at substantial savings in labor. They'll bring to airports, post offices, cash machines and to large sports events a new tool for finding bombers and anthrax poisoners, purse snatchers and perverts. Post-September 11th, this change is going to happen, I know it, if not with products from Oxford, then with similar products from some competitor. It is just a grim reality of our digital age. This is a product that will definitely save lives.

But it also scares the pants off me. It recalls Orwell's Big Brother; we could quickly have a system that knows at every moment where every person is and what they are saying or doing. I don't even like giving out my cellphone number, so imagine how I feel about being watched 24/7. It's great for saving kids from predators at shopping malls, but it also strips away the last semblance of privacy any of us have. In time, we'll have to come to grips with these issues, to decide where such cameras can and can't be used. But they will be used, I know that.

Recognizing that, we must be very thoughtful about how these new systems will be regulated.

There is an irony here that the last time I wrote about Steve Morton, it was because he had used the same DSP chips -� his AX36 family -� to make guns with built-in fingerprint recognition that could be fired only by their owners. Those guns never made it to market because the firearms industry didn't like them. Now the same folks who hated smart guns will probably love smart cameras, which are bound to be a big success.