Flutterby™! : Idea OTM 2004-04-22 18:41:03.746636+00

Idea OTM

2004-04-22 18:41:03.746636+00 by Dan Lyke 7 comments

If I knew a lawyer who could get me a patent idea of the moment:

In the recent trip there were plenty of roads that had a lot of different speed limits. In the desert, a road that's 65MPH, dropping to 35MPH, then back up to 55MPH and down to 45MPH as it goes through towns and different terrain is the norm. I didn't often miss the slow-downs as I knew to expect 'em, but I did sometimes miss the speed-ups, and it'd be nice to have something that goes "bing" when it notices that the speed has dropped from 45 to 25.

I know that the FCC has reserved some frequencies for automated road info, but that's going to be years from being deployed.

It occurs to me that the code to recognize a speed limit sign is fairly simple. Point a cheap camera out the right side of the car, find a white blotch in between two vertical lines, wait for it to get to the right size and then compare it with reference images. You could probably do all of that in one FPGA, maybe even on an 8 bit microcontroller depending on your camera interface. Cheap webcams are $20 retail, you probably want to feed the rows and columns off the CCD straight into the FPGA, so this thing could be built for peanuts.

Then add to your in-car nav system a little "current speed limit" display. In fact, you could even tie it to your cruise control so that the car slows automatically when you come into town.

Great idea, not sure I'd want the car to slow automagically when that 18 wheeler is on my rear bumper, but having the speedometer display a difference of actual speed to last seen/recognized speed limit sign would be very useful (and a [ping] sound when the signs change. As the signs are high contrast and pretty uniform across the country, this makes lots more sense than 'smart signs'. I'd pay $200 easily for such a device, as I have found radar detectors to be more of an incentive to speed then my normal judgment of road/traffic conditions and karmically perceived 'bear' density. I've gotten TWO speeding tickets in the last 20 years, both by 'speed traps' with a sudden unexpected change in speed, with a cop sitting right past the speed change signage.

The car already has lots of circumstances under which it'll disengage cruise control, so a "bing" and then, say, two seconds later readjusting the cruise control down to the target speed makes sense to me. That gives the driver plenty of time to override, and doesn't do something silly like hit the anti-lock brakes hard to drop the car down to the target speed at .95G or anything.

I had that idea when I lived in Nevada - not the cruise control bit, since I'd never used cruise control at the time, but the camera that tells you what the last speed limit sign said. I sketched out a design for the image-recognition algorithm, even, but I don't know enough about electronics to build a cheap prototype, and couldn't afford the parts for a computer-based version. The hitch, I thought, was going to be figuring out how to make it fail gracefully. Most residential areas don't have speed limit signs, for example; if your speed-limit-meter keeps reading 65 mph even after you've been crawling around in a 15 mph residential area for the last ten minutes, you're going to stop paying attention to it. In order to be useful, it has to be reliable, and there has to be some way for the driver to judge its accuracy. How long since it last saw a sign, for example? I thought maybe the brightness of the display should peak whenever the gadget saw a sign, then fade at some rate proportional to the number of velocity changes per unit time. This would mimic the probability that the speed limit had actually changed since the last time you passed a sign: more turns and stops/starts means city driving, fewer turns and smooth speed means highway driving. It would also tend to alert you when passing speed limit signs, since the brightness would flare up.

Actually, Mars, if all you're going to do with the sign is display the current speed, you could even just capture a picture of the last thing that looked like a sign, maybe with a ring buffer and a "miles back" indicator. No image recognition necessary, all you have to do is find two edges with the right aspect ratio and no color.

Like the idea of the fading bit, but I think with the "bing" it would work as an adjunct to my seeing, not a replacement.

I also think that maybe another button on my cruise control, one which does "set to current posted speed" would be the right way to put the driver in the loop. Hear "bing", press a button, accelerate or slack off until current speed is hit.

As if I needed another project, this might warrant hanging a video camera out the window and seeing what I can slap together on the ferry.... after I ship the first usable version of this photo album software.

>> The hitch, I thought, was going to be figuring out how to make it fail gracefully.

Build it to do the intial job first and after that's done you start refining it to recognize things like you pointed out.

The biggest problem I see on the reliability end is that speed limit signs are often blocked from view on interstate highways by trucks.

The price of the hardware may be coming way down for building something like this. I was talking to some vendors a couple of months ago about the feasability of a new sensor they were toying with marketing, a low end vision system. This sensor would cost anywhere from $75 to $200 and could possibly be used for this application. If that's the case most of the software would be built into the sensor. It ought to be fairly easy at that point to graft it onto a display. Tapping into the factory cruise control could be a little more difficult.

Larry: Yep. That sounds like the price range. The more I think about this, the most expensive bit of it would be the display (I'm thinking one of the Noritake VFDs, mainly because I've got one on my workbench and it's way bright and super crisp), but if there's image recognition rather than just intelligent cropping then that could be a two-digit LED.

Hmmm... Anyone out there with ties to the automotive industry feel like funding a startup (or offer the right connections and business guidance, and it might even be self-fundable)? I'm sure I can show you a whole bunch of other cool ideas in this space that could keep a small R&D group busy for a few years, and that'd give all sorts of high-margin reasons to up the price of that luxury SUV.