@dgatwood
The car actually does know what time it is, and what day it is.
The problem with those signs isn't knowing what time of day it is. The problem is:
- Is school in session that day? Is it a teacher in-service day, a holiday, spring break, fall break, summer break, Thanksgiving or Christmas break, etc.?
- Do the cameras provide enough resolution to reliably read the small text on the signs every time at whatever angle it happens to be at when the car sees the sign?
- Is the font used for the sign one that the text recognition can reliably read?
- Is there other text that has to be understood by the car to know what the numbers mean (e.g. No parking from 6-11 A.M. on the first Tuesday of each month)?
I've done a little bit of work with computer vision on a much simpler level, involving reading text literally drawn on the screen of a cell phone using screenshots, and even with that level of consistency, it can misread text periodically — particularly small text or when the background isn't completely clean.
Given that we still haven't reached a point where Tesla cars can reliably tell the difference between truck speed limits and car speed limits because of how many different ways those signs are designed, I'd guess we're many, many years (and maybe decades) away from the level of conversational semantic text interpretation required to read and interpret thousands of slightly different conditional speed limit and parking signs in real time, at least within the resource constraints of a computer system that can realistically fit inside a moving vehicle.
Add to that the problem of knowing whether it is actually a school day or not (which means some kind of Internet query against the school system's website, coupled with a level of text comprehension bordering on AGI), and at least when it comes to a general solution, the problem is likely intractable.
So the question is do you want to wait two decades for this to
maybe work generally, or do you want to put in a big "Speed Limit 15 When Flashing" sign that computer vision can plausibly recognize and interpret correctly 100% of the time with today's technology?