Monday, January 23, 2006

Police store DNA records of 24,000 innocent kids

Police store DNA records of 24,000 innocent kids provides an alarming overview of police surveillance activity in the UK. As the article says "Britain already has the largest network of CCTV cameras in the world", and they intend to go further. The main issue in the article is DNA profiling, where it's known the police have DNA records of 24,000 young people who have never been even accused of anything. Where there's great room for concern, the article claims "most people are already resigned to the whole population having its DNA held in police and government databases". SIGH

The article details several projects which, taken together, amount to the same end goal as the Total Information Awareness System (TIA). Since that's in the UK, this would be the British equivalent to TIA. And perhaps since some aspects of British and American government are joined at the hip, maybe MI5/6 and the NSA/CIA/DOD are working together to implement TIA?

You, the reader, may have believed the TIA project was shut down. What happened is that one of the sub-projects in TIA (it was either FutureMap or Wargaming the Asymmetric Environment) was unveiled to the public, and it offended officialdom, and enough of them went on a harumphing storm that the DoD made a show of shutting down TIA. But it's obvious that most of the subprojects to TIA continued on being funded, as they would be very useful to a surveillance system.

Now, let's go through the projects reported by The Register and align them with the TIA projects which I recorded back in 2002.

DNA Profiling: There isn't a direct analogue to a TIA project. One problem with DNA profiling, however, is the expense of doing DNA testing. Doing a full sequencing of the DNA in a tissue sample currently costs many thousands of dollars. Perhaps forensic DNA testing doesn't do full sequencing, but it still isn't going to be cheap. Be aware, though, that the government is funding research to decrease the cost of DNA testing.

The currently understood methods of DNA testing involve tissue culturing steps. For example, they take a tissue sample, grow a tissue culture in a laboratory, then kill the tissue sample, and study it under microscopes. That means identifying someone from DNA in a tissue sample will take several days.

The Bio-Surveillance project in TIA is very different from any DNA profiling system.

CCTV cameras on the roads and streets: Associated with this is number plate recognition, and face recognition. This is clearly associated with Human ID at a Distance with the addition of number plate recognition.

Implementation involves installing closed-circuit TV cameras (CCTV) in desired areas. The more cameras, the more intrusive and comprehensive is the monitoring system. You might think "oh, monitoring TV is manual, there's going to be a human looking at every video feed". No, if you have a city full of video cameras, there's no way a staff of humans can effectively scan them all. The cameras might be ignored most of the time, and only used when a call arives in the neighborhood of a given camera.

But what's possible is for the video feeds from these cameras to be analyzed by computers. That's what the Human ID at a Distance project is intended to be doing. And face recognition systems have been under test for several years. For example a public test was performed at a Super Bowl game a couple years ago, but apparently the test results were very disappointing.

The technique is to use image analysis to look for patterns that let the computer software zero in on the information of interest. The problem is computers are rather dumb. Human bodies have zillions of years of evolution to our image recognition hardware, and we recognize such things very readily. But computers have only 60 years of development (or so) behind them, and all they can deal with are numbers. It may look like they deal with words, and pictures, and sounds, and movies, and whatnot, but the software developers have to invent ways to turn all that into streams of numbers, because computers only know how to deal with numbers. This makes image analysis tricky.

Take face recognition. There's probably some pattern of pixels that usually indicate a human face. There's a small range of skin pigmentations, and then the shape of a face is generally the same with two eyes, nose, mouth, etc. It's made tricky because faces have a lot of variability, even when there's a lot of similarity. Recognizing license plates (number plates, as the British call them) would be simpler. There's a fairly well known set of colors to look for, and you know ahead of time what shape the license plate will have, and its location on a car.

Once you've got software that can reliably recognize what you're looking for ... car license plates ... you connect a fleet of computers to the CCTV cameras. They'd constantly be looking for motion, when they see motion the software looks for a license plate, and records whatever it sees. Assuming reliable software the computers can register all cars that pass by a CCTV camera.

The next step is to install multiple cameras throughout the city. Each camera gets connected to the computers. The computers register each car that passes by each camera. Or, if reliable face recognition software exists, the computers can also register each pedestrian as they pass by each camera. Hence, these computers could easily track the movement of every car or every pedestrian where-ever they (we) go in the city.