Transform Cellphones Into a CCTV Swarm

holy_calamity writes "Swiss researchers have developed java software that has bluetooth-capable camera phones form a distributed camera network. Each phone shares information on visual events with its neighbours and can work out the spatial position of phones around it (pdf). The software will become open source sometime next year, and the creators say it could be used to make a quick and dirty surveillance system. 'The phones currently use the average speed people walk to guess the distances between themselves, based on how long people take to move from one phone's view to another's. In testing, the system determined the distances between each phone with about 95% accuracy. They were placed 4 metres apart, making it accurate to about 20 centimetres. In future, recording the speed at which objects pass by would make more accurate judgments possible.'"

The real question at hand here.

[blhack]

But does it run skullbocks?

Re:concert-recording on the cheap

[Anonymous Coward]

If it is possible to [apples], it ought to be possible to [oranges].

Sorry, but speaking as someone with a degree in electrical engineering who spent the better part of a decade studying this stuff, it just doesn't work like that. You can use padding tricks increase the [false] resolution of the spectrum you're dealing with, but you can't recover signal that you failed to record in the first place. See also: Nyquist-Shannon sampling theorem [wikipedia.org].

For a simpler analogy, it's like using 16-bit registers to record 32 bit integers. No matter how many 16 bit registers you use or how you combine them, you're not going to recover the upper 16 bits -- they're lost because you didn't record them.

Re:Open source surveillance

[VE3MTM]

There's a name for this: it's called "Sousveillance [wikipedia.org]".

Re:concert-recording on the cheap

[matfud]

Yes that true, but only for one source. If you have multiple 8khz signals whose sample points are not synchronized then you can combine them to improve the overall frequency range obtainable. However this would increase the final achivable frequency in proportion to the log of the number of sampling devices (under ideal situations) so you would need a fair few sources (If I remeber correctly). Mobile phones would probably be quite far from this ideal as they
a) would be physically seperated so you'd have to perform some correlation first to remove the arbitrary time delays from the audio source to the phones and this would remove some of the resolution
b) would not be sampling at the optimum times wrt each other (perfectly interleaved sampling).

A similar techique is used for images. An 8 bit camera can record, at best, 256 levels of grey. If you take multiple images of the same scene and average them together you can increase the effective number of grey levels you can reconstruct while also removing the effects of thermal noise. Doing similar with audio would not use averaging but would instead interleave the samples from the out of synch microphone ADC's.

matfud

Re:concert-recording on the cheap

[fenodyree]

That comparison is flawed.
To use a car analogy: You can construct a complete working car from a junkyard provided that the cars in the junkyard have different problems and conversely different working parts.
As the way multiple low resolution images of the same seen create a higher resolution is primarily based upon the (usually accurate) assumption that the images will be lit differently such that where information is lacking in one image, it _is_ in another, and vise versa.

However, with the phones, the same set of frequencies (3,400 Hz) and going to be dropped by bandpass filtering.