I know something about DSP (in audio), but not nearly to the depth that you seem to have. But I still think higher resolution cameras could help a lot in two ways. The most obvious is that pixels can be combined to form a higher quality image that is at the same resolution as those provided by the current cameras. The same processing with better quality images should yield real benefits.
But when we drive, we are not constantly looking at what is happening a quarter mile away. We look occasionally and retain what we have observed until we get the chance to look again. Tesla could do something similar. Think of it as two threads of processing, one a very high frame rate of low-res images (formed by pixel binning), the other a much lower frame rate of native high-res images.
No real argument in what you're saying, but there's this thing with Shannon.
For those of you who don't know, Shannon is known as the Father of Information Theory. Amongst other things, he related bit flipping in a channel to Entropy. As in thermodynamics. And the 2nd law of Thermo, which says that, in a closed system, things proceed to a disordered state and no going back. (As I remember, the three laws of Thermodynamics are: 1) you can't win, 2) you can't break even, and 3) the game is rigged.)
How this relates to moving information: There's this plot, somewhere, that says that if one has a certain signal to noise ratio in a channel of fixed bandwidth, then there's a maximum data rate in that channel. It's the
Shannon-Hartly theorem. What's interesting about this theorem, and the plot, is it gives an upper limit that is truly hard and fast: You
can't transmit data faster than yea on that plot.
What it doesn't say is how to get to that limit
. Put the worlds largest collection of CRC-based forward error correction codes on that data one is launching into that noisy channel? One will
approach that limit, but won't pass it. Apply one's earth-shaking algorithm that nobody's ever thought about before? Great! One has moved things forward - and one is closer to that limit, but one won't pass it. It's
that kind of thermodynamics.
This idea in information processing (and, yeah, that includes image recognition, moving objects, etc.) applies
everywhere. So, if one has a high-bandwidth, non-noisy channel, it's just amazing how much information can get passed and used. A high-bandwidth,
noisy channel: I'm a-thinking that that describes what Tesla's trying to do.
Just so we're clear: This isn't saying that Tesla can't get autonomous driving down pat. We wetware types do it all the time. But the algorithms that we use to do this are obtuse and are part of current research. And, naturally, good old evolution and cut-and-try has Done Things in eliminating Things That Don't Work (Didn't spot that tiger? Now you can check on your failures from the inside. Didn't spot that tasty fruit? Now you get to starve. Etcetera.) What actually works.. Well, people
do get into accidents all the time, so there's a decent idea that we aren't, collectively, quite, up against any Shannon's limits.. at least with the algorithms that we're running inside.
Musk has been pretty adamant that Tesla is going to come up with a self-driving car that an order of magnitude or better. I strongly suspect that these statements aren't being done in a vacuum: Very likely, somebody has done enough of the theoretical work to show that the Shannon limit is big enough to do the job. The tricky bit is getting algorithms in silicon and neural networking hardware that surpass what our wetware can do.
This whole argument about Shannon limits and information processing is why I tend to disbelieve the occasional poster who shows up and says, without much evidence other than Gut Feeling, that it'll be years or never before self-driving cars roam the highways. That's a human-centric view that Humans are Best At Driving, Period. I don't think so: There's plenty of hardware and other living entities out there that are better at various tasks than we are, so, where's the proof that self-driving isn't a task that nothing else can do? (Strawman argument, I realize, but, still.)
