-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
What's the likelihood we will see Speed Limit Sign recognition?
- Thread starter PhilDavid
- Start date
Somehow I can't get excited about the car's ability to read speed limit signs. Obviously such capability is not going to work in school zones (will the software read and understand the blinking yellow lights indicating a school zone - or understand the presence of a school crossing guard?). And, it will not make up for poor road maintenance where branches obscure or partially obscure the sign.
With some exceptions, the speed limit database utilized by Tesla seems to be accurate except on one local road where it posted 55 mph. The speed limit is 35 but there is a long stretch where no signs are posted.
With some exceptions, the speed limit database utilized by Tesla seems to be accurate except on one local road where it posted 55 mph. The speed limit is 35 but there is a long stretch where no signs are posted.
I found it to be very unreliable on a AP1 model S I had for a while. Displayed info was wrong or in many cases not displayed. Must be dependent to exact sign placement, etc.
Of course AP2 is crap. I'm still hoping someone can point to the source database and I can go right to that place and make the edits.
Once it does exist my guess is some sort of hybrid approach. Cameras plus a DB w/ some sort of analytics applied to understand bogus sign or fix the data for road changes.
Is this an AP2+ problem? I don't see any such issues in my AP1 car. Could you have a defective radar emitter or sensor?
Read earlier posts on this thread. MobileEye has a patent on reading speed limit signs.
I feel this is an example of a BS patent that should never have been awarded but that's another topic.
Reading speed limit signs is easy - and something the software is likely already doing.
Interpreting them is very difficult. Not only is there the challenge of time-based speed limits or speed limits triggered by warning lights, there's also the problem of determine for which lanes a speed limit is intended (complicated when there are HOV lanes, main lanes, exit/entry lanes, frontage road lanes adjacent to each other).
The current use of the onboard speed limit database, coupled with the "fleet speed" Tesla is evidently applying as an average of speed used by other Tesla vehicles on the same roads - that combination isn't working.
I have one route I take periodically for business where there is frequent unnecessary slow downs due to improper speed limit data. One road was under construction for years as lanes were added and is now completed - and that road has problems where the speed limit had been temporarily reduced during construction or where a mainlane is where a frontage road used to be located. The other road is still under construction - and has similar issues with incorrect speeds, and incorrectly using the speed limit for the frontage road - which has now been moved. These errors result in frequent sudden braking and sometimes automatically decreasing the TACC speed.
To be safe, the AP software must be able to operate under the current road conditions. Musk commented that AP systems using stored high resolution map data are doomed to failure because they rely on data which is almost guaranteed to be out-of-date. The same applies to Tesla's use of the onboard speed limit and onboard navigation maps - both of which will always be months or years out of date.
Plus, there will likely be a regulatory requirement - that FSD approval will come only if the manufacturer is able to implement the AP systems to strictly adhere to the traffic laws - which not only means interpreting the speed limit traffic signs/signals - but also obeying police and first responders providing manual traffic direction.
Will be interesting to see how this is all handled. Safe autonomous driving is much more than just maintaining speed/spacing and lanes...
Interpreting them is very difficult. Not only is there the challenge of time-based speed limits or speed limits triggered by warning lights, there's also the problem of determine for which lanes a speed limit is intended (complicated when there are HOV lanes, main lanes, exit/entry lanes, frontage road lanes adjacent to each other).
The current use of the onboard speed limit database, coupled with the "fleet speed" Tesla is evidently applying as an average of speed used by other Tesla vehicles on the same roads - that combination isn't working.
I have one route I take periodically for business where there is frequent unnecessary slow downs due to improper speed limit data. One road was under construction for years as lanes were added and is now completed - and that road has problems where the speed limit had been temporarily reduced during construction or where a mainlane is where a frontage road used to be located. The other road is still under construction - and has similar issues with incorrect speeds, and incorrectly using the speed limit for the frontage road - which has now been moved. These errors result in frequent sudden braking and sometimes automatically decreasing the TACC speed.
To be safe, the AP software must be able to operate under the current road conditions. Musk commented that AP systems using stored high resolution map data are doomed to failure because they rely on data which is almost guaranteed to be out-of-date. The same applies to Tesla's use of the onboard speed limit and onboard navigation maps - both of which will always be months or years out of date.
Plus, there will likely be a regulatory requirement - that FSD approval will come only if the manufacturer is able to implement the AP systems to strictly adhere to the traffic laws - which not only means interpreting the speed limit traffic signs/signals - but also obeying police and first responders providing manual traffic direction.
Will be interesting to see how this is all handled. Safe autonomous driving is much more than just maintaining speed/spacing and lanes...
I have not seen actual proof that it's a patent problem. They have to solve this before any FSD functionality will be released.
How will they do it?
Cheburashka
Active Member
Yeah even if they had a patent, that doesn't make it defensible.
Other AV companies are reading speed signs and not being sued by MobilEye.
Other AV companies are reading speed signs and not being sued by MobilEye.
There cannot be a patent on reading speed limit signs - just on the algorithms that do so. If you write your own wthout copying someone else's then that is original work. it'd be as basic as trying to patent the idea of using a system to stop a car - yes you can patent your version of brakes but can't stop someone else designing another functional braking system.
whitex
Well-Known Member
Do you have a link to that patent? I'm curious what they actually managed to patent. Using a computer to read a sign, entire computer vision field, etc?
ucmndd
Well-Known Member
US20080137908A1 - Detecting and recognizing traffic signs - Google Patents
It may well be a bullshit unenforceable patent but Tesla can’t afford to pick a legal fight with Intel right now.
brucet999
Active Member
But not very well. Mine sees the "End 35 MPH" sign at the south end of Bishop on Hwy 395 and won't allow TACC or AP1 to exceed 40mph (I have +5 set). Apparently it looks only at sign shape and the number.
I wonder with FSD on surface streets expected to be working within the next few months with traffic light and stop sign recognition, if the speeds limits will also be recognized by then. Surface streets tend to have speeds limits that can change often so hopefully we will see speed limit recognition soon!
The 2021 Ford Mach-E website is showing it will have speed sign recognition.
Tesla can't rely on speed limit databases or on "fleet speed" to determine vehicle speed, since neither reflects the current conditions - and Tesla would likely not be able to get approval for FSD unless the software obeys all of the traffic laws, including the posted speed limits.
At some point, if Tesla can't work around the Mobileye patents - then they should just suck it up - and pay a licensing fee for the patent...
Tesla can't rely on speed limit databases or on "fleet speed" to determine vehicle speed, since neither reflects the current conditions - and Tesla would likely not be able to get approval for FSD unless the software obeys all of the traffic laws, including the posted speed limits.
At some point, if Tesla can't work around the Mobileye patents - then they should just suck it up - and pay a licensing fee for the patent...
boonedocks
MS LR Blk/Blk 19”
Further that....many expressways in our area now have dynamic speed limit signs that change based on a multitude of reasons. My OLDAF AP1 used to respond and display the correct speed limit reading when passing them....
Also construction zones often have temporary speed limits. With AP1 cars you used to have some peace of mind seeing the speed limit on the IC was "the last seen" speed limit sign but I have no faith on the speed limit displayed on surface streets. There's a road here where the limit just went down from 45 to 35 but it is still shown has 45 in the car. So if you do 50 mph (5mph over the limit is customary here), you could get a ticket.
Then how do you explain AP 2.0+ not reading speed limit signs? They obviously have the technical capability to do this... AND this would be immensely useful for customers.
AP 2.x is operating at 100% (with reduced frame rate and image size) already. No room/ processing power for sign recognition.
Kiwiadventure
Member
Similar threads
