September 19, 2008

Life, Digitized

Inspired by my shoddy memory, ever-advancing game graphics, Steve Mann, the notion of augmented reality and alternate reality games, I decided to take historical digital storage data and extrapolate its growth. What sort of applications will be possible when $100 worth of digital storage will be large enough to store full-resolution stereo video+audio recordings of a human's lifetime of vision and hearing? When will we be able to conveniently store, in a computer, our entire seen and heard experience of life?

This is a muti-part calculation. First we need to calculate the size of a single frame of video with a resolution at the upper bounds of that visible by a healthy human eye. John Russ's book The Image Processing Handbook provides some insight. At a half-metre distance, the smallest noticeable detail of a high-contrast image is about 0.1mm. That's about 64,516dpi.

According to this site, normal human stereoscopic vision, when staring straight ahead has a 100 degree field of view (peripheral vision is about 180 degrees). For simplicity's sake, let's assume fixed vision for now. We're dealing with a display surface that's 1.2 meters squared:

1.2 meters is about 47", so this display has 143 megapixels. some estimates are much higher, but even this one could be overkill because human vision is always in motion, building higher-resolution composites through high-speed collage work. As the previously linked site says, vision is more like painting than photography. Let's stick with 143 megapixels for now.

So this hypothetical system has two 143 megapixel video channels. Let's make them both 48hz recordings -- double the standard cinema framerate for really smooth recordings when the cameras are moving. In terms of data storage requirements, we're looking at 143 000 000 pixels x 32bits/pixel x 0.125 bytes/bit x 2 channels x 48 frames/second = 55 gigabytes/second. That's a lot of data. Let's tack on some lossless video compressions like MSULab, which give us a compression factor of 2.75. So now we're looking at 20 gigs/second.

Assuming audio recorded in 128 bit MP3 format, we're looking at a negligible amount of data next to the high-resolution video, so we can safely ignore it in our calculations.

A single human lifetime of recordings would amount to 20 gigs/second x 60 seconds x 60 minutes x 16 hours (an approximate waking day) x 365 days x 80 years (an approximate lifetime) = 33 638 400 terabytes. Thirty million terabytes.

According to my extrapolation of historical storage prices, a hard drive this large will cost $100 in 2038, when I'm 61. Cool.

With a bit of ingenuity, better compression ratios and a tolerance for lower-resolution video graphics, this kind of system should be affordable much much sooner. In fact, 1 megapixel stereoscopic video cameras already exist for anyone interested in getting a head-start.

No comments :