What's the likelihood we will see Speed Limit Sign recognition?

[bob_p]

Has there been any statement by Tesla on this? Or are we just guessing?

Traffic sign recognition is critical to get AP/FSD working correctly - so Tesla and all other manufacturers have to find a way to make this work - and if MobilEye has a blocking patent, then they'll have to either pay the license fee or find a way to work around the patent.

The bigger challenge probably isn't the patent - it's getting traffic sign recognition to correctly and reliably detect traffic signs & signals, made more difficult because of the lack of standardization and partial visibility due to obstructions or vehicles blocking view.

Recently Musk stated this was coming this year - doubt he would have made that statement unless they were close enough to getting this working (he probably has been using it in his car for a while).

 

[PhilDavid]

Tesla software can already do this but it just isn't put it because of the patent. Why would they pay royalties to their competitor MobilEye while FSD is nowhere near FSD 3?

My guess is that once Tesla FSD reaches level 3 they just discuss licensing the patent with MobilEye and likely MobilEye using patents from Tesla.

One good thing is that Tesla can deal with Intel, as they are the actual owner of these patents now.

 

[Arizona Willie]

It seems around where we live they are constantly changing speed limits with some changes in the 10mph range. These changes make Autopilot speed limit offsets not accurate and I would love to reliably know the correct speed limit by loking at the IC.

I heard something about MobileEye having a patent for this (surprised that patent is not subject to prior work, but that's a different topic altogether), but surely by now Tesla has patents of their own for autonomous driving that perhaps MobileEye might be infringing on.

How likely and when do you think we will see speed limit sign recognition? At the least, whenever a Tesla encounters a speed limit sign that is contrary to what is in the database, perhaps it can upload the video and someone can then manually update the speed limit in the map database to get around any patent issues or perhaps if the pattern recognition is done "offline" with the map database updated it gets around whatever patent that might be in the way right now?

Seems this is just a variant of OCR that has been around forever. The speed limit sign recognition would be so useful when driving in an unfamiliar area.

==========
I thought we already had it. I've noticed the speed limit on my screen changes right as I pass a speed limit sign if it is low enough. There's a school zone I go through and the speed limit sign is too high for the car to read. But most other times it changes just as I pass the signs.

 

[PhilDavid]

==========
I thought we already had it. I've noticed the speed limit on my screen changes right as I pass a speed limit sign if it is low enough. There's a school zone I go through and the speed limit sign is too high for the car to read. But most other times it changes just as I pass the signs.

A coincidence. I know it has not reflected any of the changed speed limits on roads close to where we live.

 

[emmz0r]

He stated in Autonomy Day that they don't want to rely on maps as much, so sign recognition is paramount.

What is the patent about exactly? The technique reading signs? I would assume they use ML for sign reading now, and not OCR / "classic" computer vision with edge detection and so forth.

 

[RDoc]

So Tesla would need to work around these patents to get speed limit sign recognition to work?

Blue Origin had a patent on barge landings too. The Patent Office will issue patents on a lot of things that the owner can't defend. It just means there's a court fight.

That patent also seems to have a specific algorithm in mind that is trying to match a sign to a model with a scoring system system used to determine if the object is a sign and if so, what kind of sign. That doesn't sound like a deep learning neural network.

 

[D.E.]

==========
I thought we already had it. I've noticed the speed limit on my screen changes right as I pass a speed limit sign if it is low enough. There's a school zone I go through and the speed limit sign is too high for the car to read. But most other times it changes just as I pass the signs.

There’s a speed limit database so the display will change as the car moves from one speed limit to another. The signs are also placed at the point of change. So you’ll often see the displayed speed limit change as you pass the road sign. That doesn’t mean the car is reading the sign. The problem is that the speed limit database is sometimes inaccurate, or local changes have been made. Reading the signs would be the best of both worlds. Computer version 1 does read signs, not version 2 or later.

 

[D.E.]

He stated in Autonomy Day that they don't want to rely on maps as much, so sign recognition is paramount.

What is the patent about exactly? The technique reading signs? I would assume they use ML for sign reading now, and not OCR / "classic" computer vision with edge detection and so forth.

You don’t need to assume, you can read the actual patent, there’s a link earlier in this thread.

 

[D.E.]

Road signs should have RFID tags. Then all cars could constantly display current, accurate, and updated information. There’d be no need for computers to try to read and decipher all these road signs. It would eliminate the errors, it’d eliminate the utility of this horrid patent.

 

[emmz0r]

You don’t need to assume, you can read the actual patent, there’s a link earlier in this thread.

I skimmed through it. They don't say exactly what they use to do it, so it's open for multiple methods. Wonder what VW uses on their e-Golfs.

 

[PrGrPa]

Speed limit sign recognition is pretty common on German cars. Some of that predates MobilEye I suspect.

I’d like to see sign recognition. What I’d like to see already is preemptive limit adherence: the car should go no faster than the posted limit from the start of the limit until the end. Now it zooms into the limited area and then slows down. Not good behaviour.

 

[DCEV]

Yeah I find it highly unlikely that this MobileEye "patent" would survive if challenged in court.

The best outcome is for Tesla to also have a robust portfolio of patents and they enter into a "no fee" cross licensing agreement with Intel so everyone can move forward. In the DC area, speed limit sign recognition is really essential. They have cameras almost every few blocks now and they change speed limits all the time so would be great to have some confidence that the speed limit on the IC is accurate.

 

[DCEV]

Road signs should have RFID tags. Then all cars could constantly display current, accurate, and updated information. There’d be no need for computers to try to read and decipher all these road signs. It would eliminate the errors, it’d eliminate the utility of this horrid patent.

The bets way to avoid problems and ensure everything works everywhere is go buy the standards humans use and that is reading speed limit signs. They'll need to figure out a way to read speed limit signs one way or the other for FSD.

 

[evster]

Any news on this? I find it really strange that we don't here more about speed sign recognition and a Neural Net that perform this task? This should be one of the mandatory features towards FSD.

I know there is talk about traffic light recognition and stop sign recognition, but speed signs never mentioned?

This is really one of the annoying things I find with current FW for AP2 because the GPS-data with speed limits is not always accurate and there are often temporary speed limits that are not in the (latest) navigation db.

Anyone who know what we can expect and if Tesla has something in the works soon?

 

[ucmndd]

It's obvious at this point it's a legal problem, not a technology one, and Tesla is in no condition to pick a fight with Intel.

 

[KimVerska]

Not sure where I've read this but I've seen this mentioned several times. If this is a patent issue, I hope they figure out a solution soon, especially given Tesla's own patents and the fact that MobileEye is now owned by Intel.

Just researched very quickly and found this -- Not sure how well such a patent could stand up to scrutiny/challenge.

US20080137908A1 - Detecting and recognizing traffic signs - Google Patents
AUTONOMOUS VEHICLE SPEED CALIBRATION - Mobileye Vision Technologies Ltd.

The Mobile Eye patent has nothing to do with reading speed limit signs. It's all about calibrating actual speed:

"Systems and methods may calibrate an indicator of speed of an autonomous vehicle. In one implementation, a system may include at least one processor programmed to: receive from a camera at least a plurality of images representative of an environment of the vehicle; analyze the plurality of images to identify at least two recognized landmarks; determine, based on known locations of the two recognized landmarks, a value indicative of a distance between the at least two recognized landmarks; determine, based on an output of at least one sensor associated with the autonomous vehicle, a measured distance between the at least two landmarks; determine a correction factor for the at least one sensor based on a comparison of the value indicative of the distance between the at least to recognized landmarks and the measured distance between the at least two landmarks."

 

[bob_p]

FSD vehicles must be able to react to speed limit and traffic signs and signal, including manual signals provided by first responders.

Vehicles can't rely on stored speed limit data, otherwise the vehicles will never be able to react to current conditions, after conditions have changed (construction, road changes, emergencies, …).

Just reading the speed limit numbers off the signs is the easy part - the hard part is determining which of the speed limits actually apply to the vehicle, since there can be ambiguity in where the sign is placed and for which lanes/roads the speed limit is intended - this is especially a challenge in construction areas or when there are parallel roads (HOV lanes, main lanes, exit ramps, frontage roads, …).

Tesla can't deliver FSD until they work this out - they'll need to overcome both the technical and legal challenges.

For now, I wish there was a way to completely disable automatic speed adjustment - because the errors in the speed limit data cause too many unnecessary instances where the software wants to rapidly slow due to incorrect data.

 

[emmz0r]

Exactly, it would be impossible with FSD and no speed sign recognition. The map database is chronically out of date, and the GPS can be several meters off.

I had a very bad offset error on the gps and it got the speed data from off ramps, slowing the car down. Sign recognition needs to happen ASAP.

 

[bob_p]

The current AP software has two major issues:

• Incorrect speed limits causing unnecessary and rapid braking (in the middle of high speed traffic) when operating on AP (even only on TACC). And because the lane change is made without any audible or major visual warning (sometimes without any indication at all), the driver must be ready to override vehicle speed at all times - especially on roads with current or recent construction. Until they fix this problem - it would be extremely useful (and safer) to provide setting to disable automatic speed changes, or at least make a sound when the speed was changed by software.
• Flaws in the auto-lane changing rules that causes unnecessary lane changes or waits too long to change lanes when approaching an intersection or exit ramp. Fortunately, you can disable this by turning off NOAP and putting the software back into lane keeping mode.

These are both areas that must be fixed before Tesla can achieve their FSD goals.

 

[PrGrPa]

The current AP software has two major issues

• Flaws in the auto-lane changing rules that causes unnecessary lane changes or waits too long to change lanes when approaching an intersection or exit ramp. Fortunately, you can disable this by turning off NOAP and putting the software back into lane keeping mode.

These are both areas that must be fixed before Tesla can achieve their FSD goals.

The current AP software has two major issues:

• Incorrect speed limits causing unnecessary and rapid braking (in the middle of high speed traffic) when operating on AP (even only on TACC). And because the lane change is made without any audible or major visual warning (sometimes without any indication at all), the driver must be ready to override vehicle speed at all times - especially on roads with current or recent construction. Until they fix this problem - it would be extremely useful (and safer) to provide setting to disable automatic speed changes, or at least make a sound when the speed was changed by software.
• Flaws in the auto-lane changing rules that causes unnecessary lane changes or waits too long to change lanes when approaching an intersection or exit ramp. Fortunately, you can disable this by turning off NOAP and putting the software back into lane keeping mode.

These are both areas that must be fixed before Tesla can achieve their FSD goals.

just try that auto-Lane-change in the uk. 5 seconds and it times out. Not Tesla’s fault, they’re just complying with uk legislation and the 5-second rule. But it turns what was an amazing safety feature, the car waiting until it was really safe to move lanes, into a real headache of cheesing off Audi drivers who think 6ft is a safe stopping distance when they flash you to move out.