A 1.8 Gpixel sensor has at 24 frames/s and 24 bit/pixel a data rate of nearly a TBit/s. So I would actually like to see, how they are handling such a data rate (inside of the drone). And how they actually get the data out of the drone.
And to look a bit more into this numbers ( and using the numbers from Backblaze [1]) it seems that the uncompressed video of a single drone over a year would cost $1.5e9, of course divided by the compression ratio. However for 14 days of video the storage cost would be in the ballpark of $60M per drone, that sounds actually quite possible ( especially given an additional factor of ~20 lower costs due to compression and a similar factor of ~20 for five years of HD development).
So this seems actually be technology which could on a timescale of five years be available at costs comparable to two or three police cars (plus the costs of the drone, which should be in a similar range). And I for one find this quite creepy.
The vast majority of the image doesn't change too much between frames.
I'm no image compressing genius but calculating only the deltas between frames and transmitting only those pixels that actually change by an order of magnitude (with some type of filter) should dramatically reduce bandwidth and storage size and cost by an order of magnitude.
Kind of like a large number of small git commits for slowly changing data - store large amounts of mostly duplicated data with progressive delta changes.
It would of course be compressed, but I have the least confidence in the estimate in the compression ratio which is achievable in such a setting. So I did only add it in a handwaving way in the end.
1.8 Gpixel sensors don't exist, due to physics. It's combining 368 sensors.
No one stores or transmits uncompressed video outside of RAM. 15Mbps is decent for 1080p24, so 1.8 Gpixel 24fps would be decent at 15Gbps. Assuming say 14 hours of daylight (drones can't stay up indefinitely and night vision has different requirements), that's no more than ~100 TB per mission or day. As for getting the data out, I'd assume they swap an array of 50 HDs or so.
The announcer in the video claims "a million terabytes of video a day" at 2:45. That sounds like science fiction and has to be off by several orders of magnitude.
Well, I guess the sensor could theoretically be made using the entirety of a wafer. But it would be useless since the lens needed for such a large sensor would be more than the size and weight of a large ground telescope.
And to look a bit more into this numbers ( and using the numbers from Backblaze [1]) it seems that the uncompressed video of a single drone over a year would cost $1.5e9, of course divided by the compression ratio. However for 14 days of video the storage cost would be in the ballpark of $60M per drone, that sounds actually quite possible ( especially given an additional factor of ~20 lower costs due to compression and a similar factor of ~20 for five years of HD development).
So this seems actually be technology which could on a timescale of five years be available at costs comparable to two or three police cars (plus the costs of the drone, which should be in a similar range). And I for one find this quite creepy.
[1]roughly $50/TB http://blog.backblaze.com/2013/02/20/180tb-of-good-vibration...