MIT Algorithm Predicts Red Light Runners
adeelarshad82 writes "Researchers at MIT have developed an algorithm that determines which drivers will run a red light, within one to two seconds before a potential collision. The research, based on 15,000 cars at a busy intersection, monitored various factors to determine which cars were were likely to run a red light. They found that their predictions were correct about 85 percent of the time, which is about 15-20 percent better than existing traffic prediction algorithms."
With this, its just a matter of time before these "predicted" red light runners are ticketed for their "pre-crime".... We slide further down the slope that Huxley warned us about....
THANK YOU, Edward Snowden!! Americans owe you a debt of gratitude (whether they know it or not..)
Traffic court division!
I got here through a series of tubes
If the car isn't slowing down, it's more likely to run the light.
Deep.
Have you read my blog lately?
DUH!
sigs are for losers (except to point out that sigs are for losers)
"For the technology to work, How said vehicles would need to be able to communicate with one another, wirelessly sending and receiving data like the car's speed and position."
This would require the red light runner to also be transmitting their speed and position.
I assume it simply selects BMWs?
AI at the Crossroads - predicting who is going to run a red light
Cars in MA routinely run a red lights because they are afraid of being tailended it they stop. Seriously.
I usually look when the light turns green to make sure no one is about to run it. I have a similar accuracy in determining when I shouldn't cross because someone is going a little too fast. I don't think it's the speed, because some people like locking up their brakes at the last second. I think even those people have their foot on the break and are ready to stop. Whereas runners won't be decelerating much or if at all, and may not even be looking at the red light. Just imagine you're driving on a straight flat piece of land through a green light. That's what those drivers will do. That kind of behavior. Only the light is red. If you watch for it, you'll see it every time.
My issue with this technology is that the dumber types will pick it up and think that the same idea can also be used to catch speeders, drunk drivers, etc, etc. So they'll demand those systems be built and offer stupid amounts of money for it to happen. When it does happen, and it just might, the accuracy will be low, but you'll still have to go to court to fight your way out of a DUI because the computer said you swerved more than a few inches once. They already managed to get the field sobriety test approved, which most people fail SOBER, especially beside a busy interstate in the cold at night with all those bright lights buzzing past at 70mph. You fail a field sobriety test, you're drunk. Period.
Job? I don't have time to get a job! Who will sit around and bitch about being broke and unemployed then?
If this is just used to add an additional delay to the cross traffic getting a green light, this could save a lot of lives.
Cars in MA routinely run a red lights because they are afraid of being tailended it they stop. Seriously.
Pretty amazing self-driving cars you've got there in MA.
I bet a human has even better accuracy at forecasting this. I don't think anyone needs an assistance from a computer to tell them what they are already seeing.
Is driver on mobile phone? Add 1
Is driver drinking coffee? Add 1
Is driver putting on makeup/shaving/combing hair? Add 1
Is driver having animated (you can see heads turning and arms waving about) discussion with passenger/children? Add 1
Is driver speeding? Add 1
If your score is 3 or higher then expect them to run the light, hope you are not in a crossing lane.
A feeling of having made the same mistake before: Deja Foobar
Predict that every car will not run the light. My prediction is correct much more than 85% of the time. Why aren't I in the news?
Hope the folks at Google adapt this to their self-driving car. Seems like a no-brainer.
Come to think of it, the whole interaction would be a no-brainer. Car detects red-light runner, car avoids red-light runner.
Does the red get held on the cross street longer ?
That just makes the red-runners life a lot more safer and encourages the behavior.
Does it trigger the 5ton metal barrier at the stop line ?
That make sit safer for the cross street and discourages the behavior. But we don't have the 5ton barriers.
It's good that they know fear. Not many inanimate objects are capable of that emotion. Although many can smell it.
MIT? The M in that TLA stands for Massachusetts. I've visited the state and seen the drivers, and I'm pretty sure that red light running there is mandatory. As near as I could figure, the law there is if you ever see the green or yellow as you approach the intersection, then the light is still considered green for you. I'll wait for research from somewhere else.
I'm an American. I love this country and the freedoms that we used to have.
"Even though your light might be green, it may recommend you not to go, because there are people behaving badly that you may not be aware of."
Even with a green light I may not go because there are people behaving badly that I am aware of. Don't need a traffic light for that - when in traffic, always keep your eyes open (and traffic lights aren't fail-proof either). For those dumb-asses that rely solely on what a traffic light tells them, let Darwin do the work.
Some driver assistance here might not be bad, but I'd hope it would be based on something more generic than "detected a driver that's going to run a red light".
Using this technique plus normal(?) traffic cameras, police could pretty quickly build up a list of habitual red light runners. (Even if they didnt actually "run" the light, they would be put under suspicion). They could be put on a list for "random" pull overs. (This presumes that video cameras with auto license plate reading are present and functioning on police car dashboards. I'm not sure this is the case nationally, but when I was in Denver recently a police car pulled over my friend because the computer had her license plate on a list. Call it the "Do Not Drive List").
This is nothing compared to when face recognition systems become widely prevalent. That'll blow away the "big-brother" predictions made by science fiction films like minority report (which used retina scans). Of course, WE'VE been supplying the government with tons and tons of this pre-edited, organized data tagged data. Thanks Facebook! (which is another reason why I don't use it). Call it the "Do Not Walk List".
That coupled with national biometrics programs (India, Afghanistan) and GPS tracking in every smartphone (Carrier IQ) and warrantless tapping/tracking of American citizens (war on terror) means we are rapidly heading towards a world where your government CAN know where you are at every moment. Whether or not they WILL know where you are is up to the battles over privacy information.
you need a blood test or the desktop sized tester.
We largely solved that by implementing a tailgating law. Basically if you run into somebody's tailgate you're responsible unless you can prove that it's their fault. It does have issues from time to time like when somebody swerves in front of you and slams on the brakes, but ultimately most of those problems aren't there if you keep adequate space ahead of you.
I RTFA, and no algorithm was shown. As for the 15% error, did anyone consider cargo? Without seeing the math, how are we compelled to even consider that this was nothing more than a SWAG(scientific wild @$$ guess). Me thinks this is nothing more than a variation of a Markov Chain. This would make a good lab project for a second year student on the west coast.
(snap m.i.t.)
Why don't they just override the red light's runners controls so that the car is stopped automagically for the red light?
That would prevent 100% of red light running, negating the need for the cross traffic's warning system. Not to mention it would eliminate the associated costs of red light running (assuming all cars have the required wifi control system that is required for the detection to work.)
Great comment, did you read the NYTimes article about stereotypes of various car drivers? I think it went something like this:
BMW - Arrogant, spoiled (this perception went way up after the official's son who killed someone while driving a BMW)
Mercedes - for older people
Audi - powerful (don't mess with the driver. This is because many officials drive this)
American cars I seem to remember have a pretty good reputation. Who knew? ;). But I guess they've been getting better.
Cars in MA routinely run a red lights because they are afraid of being tailended it they stop. Seriously.
This is why MIT did the study in Virginia.
Years ago I was sitting at light. The light turns green and the driver in front of me starts going oblivious to the car that's sailing down the road and clearly not intending on stopping for the red. So this guy slams right into the guy in front of me.
This was back when a light would turn green almost immediately after the intersecting street's light would go red. Drivers in my city are notorious for flaunting the rules and generally driving like jerks, but it was rare to have someone go through a red because people were aware of the risk involved.
Then at some point in the past 10-15 years traffic engineers got the idea to delay the interval between one light turning red and the next going green. So now there's a good 2+ second delay where all lights are red.
What has been the side-effect of this change? Now people brazenly blow through red lights. And the thing is that I've seen it happen everywhere, upscale and low-income areas alike. I've seen lines of 3-5 cars keep on going through when the light had clearly gone red. It's so bad sometimes that there are still cars in the intersection after the other light has already turned green, and this is with the aforementioned delay.
But yeah, it's pretty easy to spot the ones who aren't going to stop. They're the ones still moving at a good clip and making no attempt to slow down and stop.
This is why I'm somewhat supportive of stop light cameras. It's not like speed cameras which don't really target the real problem, aggressive or careless driving. Going through a red light poses real danger and is a clear example of reckless driving. Of course, I realize that stop light cameras are abused as well; one popular tactic being to shorten the yellow in order to boost the number of offenders. Otherwise running red lights is a persistent problem I don't really see anyone addressing. Probably because it involves more effort and brings in less revenue than going after speeders.
not impressed. it's just a simple calculation of distance and speed. show me something more difficult
This seems like a half baked solution to one aspect of a bad idea. At least it's not as bad a "solution" as installing red light enforcement cameras everywhere. Intersections are just plain bad. Yellows are often too short for a variety of reasons, and that is the number one cause of red light running. After improving the signal timing, which shouldn't be hard, roundabouts may be the most practical alternative.
Then there's the interchange, which is unfortunately very expensive. Yet it's crazy the way we spend millions on a limited access highway, and then go only halfway, and make the crossroads stop for the exits. For a more exotic idea, what if the vehicles did the bridging, instead of the road? Have vehicles be long enough to span at least 2 columns at all times, then make an overpass with just columns, no bridge deck. Could make interchanges cheap enough that we would never use intersections, and never need traffic lights.
Intellectual Property is a monopolistic, selfish, and defective concept. It is "tyranny over the mind of man"
and red-light runners can go off some sweet jumps.
Reminds me of a toy car kit I had as a kid.
to filter out the bad drivers before they go for their test. Have a stop light before the door and see who runs the yellow light.
by TheSpoom (715771) Uncaring Linux user here. I have nothing to add to this but please continue. *munches popcorn*
I'm wondering what an MIT team was doing studying traffic at the home of Virginia Tech? Were the two cooperating, or was there some sort of one-ups-man-ship going on?
I assume their plan is that 2 seconds is plenty of time to avert an accident by hitting the guy who was going to run the red light with some sort of rocket? This is MIT we're talking about, so I'm going to assume some sort of rocket is involved somehow.
I'm trying to teach myself to set people on fire with my mind... Is it hot in here?
It does have issues from time to time like when somebody swerves in front of you and slams on the brakes, but ultimately most of those problems aren't there if you keep adequate space ahead of you.
In Britain there's been a spate of people swerving in front of other cars and slamming on the brakes so they can make a big insurance claim, thanks to the belief that the person who hit them must have been driving too close.
Laws largely affect the sensible, law-abiding people who'd mostly act sensibly anyway. The idiots don't care because... they're idiots.
How complicated is that algorithm? bool willRunRedLight = speed > 0 && light == red; I jest, I jest.
I haven't thought of anything clever to put here, but then again most of you haven't either.
If it works, even with "only" 85% accuracy, what you can do is make it so that when somebody is coming who is likely to run it, DELAY the green light in the opposite directions for a couple seconds, instead of turning it green the instance the other one turns red. Could save many lives.
New York Yankees logo (sticker on car window, cap on driver's head, etc.)
I'm 200 miles upstate and this algorithm has a high degree of reliability here.
You missed an option.
Delay the cross-green briefly to prevent a collision... and tell the next stoplight that the red light runner is going toward that it should give that car a red light well before he gets there. You run a red light here, and you're likely to be fourth in line waiting for the red at the next light. It is easier to do that where lights are mostly sensor-driven than where strict grid timing needs to be followed.
You're right. Apply oil to all intersections and people will be more careful at intersections. Well done.
And what, exactly, is the value of doing that?
they fine people at 85%
By making the amber light last just slightly shorter, again!
I would love to see what it is that predicts these actions, and see if it says anything about whats going on.
I was hoping for the same info. I am left to figure that it is the obvious items, like approaching the intersection above the speed limit, accelerating at the yellow, perhaps a recent lane change.
Interesting to me is when I am in traffic and I think "That guy wants to cut me off" or "This guy wants to get the jump on me when the light turns green". Sometimes I can point to a behavior and say 'there's the tip-off', but often as not I just get a feel that they're driving aggressively. The automotive body language is (perhaps) emitted and read unconsciously.
Some info about that would have been interesting.
-- "Oh. This guy again."
Delay the green so opposing traffic doesn't get killed, but let the yellow go red as normal
Photograph the miscreant running the red light, and ticket them
Voila, increased safety, deterrence, and revenue. Of course, the system should be regularly audited to ensure they're not shortening yellows to catch more drivers, I hate that crap.
Never let a lack of data get in the way of a good rant.
Link
...for cross traffic and photograph and ticket the car running the red. Still safe, and you bust the jerks who don't think they should have to follow traffic rules.
As for your assertion that fining people has no influence on their behavior, I'm not buying it. A couple speeding tickets definitely got me to slow down. By not fining people you'd actually be rewarding people for illicit behavior: Don't feel like stopping? No problem, we'll just hold the light for you.
Never let a lack of data get in the way of a good rant.
that could photograph them running the red and ticket them. Honestly, it's not that hard.
Never let a lack of data get in the way of a good rant.
To put it another way, in the state of Massachusetts I think I could write a better algorithm just by predicting that everyone approaching the intersection will run the light if they can. Assuming that you don't count the ones that actually go through when the light is still green as "wrong", I think the above will do better than 85%. If you allow me to add that they will stop if there is a car stopped in each lane in front of them (but will swerve around any stopped car in their lane if another lane is open) then it will do even better, even though I lose a few percent for the rare cars that do actually stop.
They likely only got 85% accuracy because they felt the need to predict that some cars would stop.
I'm an American. I love this country and the freedoms that we used to have.
One road agency in SE Michigan is already installing 'dilemma zone detectors' in at least one intersection. If you are traveling at a certain speed at some distance before an intersection (depending on traffic density), the signal will not switch from a green phase to yellow.
Not really linked to the story, but it gives me a chance to relate this tale of idiocy.
A friend was pulled over by police for running through an intersection just as it turned from amber to red instead of stopping. She said that because the car behind was tailgating (business as usual), to have stopped would have caused an accident, so in her judgement it was safer to continue. That cut no ice and she was booked. Eventually they let her go and then pulled out behind her and followed. At the next intersection, the lights were on amber so this time she stopped. The police patrol car ploughed straight into her rear. They booked her again (she was naturally livid) but elected to take the matter to court. She was cleared of all charges including the original offence and the police had to pay all costs. Justice.
Sometimes the right thing to do is to press on on amber - I usually stop but only if there's time to do it without the half-asleep moron behind rearranging the back of your car.
There is most likely a "sweet spot" where you are not too close to the car in front of you from a safety perspective, but there's still not enough space for a dickhead to swerve in front.
FC Closer
Still need to extend the yellows, far to many municipalities are shortening them, to increase revenues from automatic fines. Getting a fine is a reason to pay a lawyer to fix it, they are cheaper in my state than the fines alone forget what it does to your insurance rates.
No sir I dont like it.
Do you really want YOUR ambulance to have to wait at every light between your house and the hospital? What about the Fire trucks? Police?
Maybe you're thinking "Hey, we can let them broadcast some sort of signal to retract the barriers automatically!" Well shit, son. Now you're talking about real-time life saving equipment - You're asking Emergency Services drivers to go blasting straight for the barriers on the assumption that they will close in time. That means at least 5 nine's (99.999%) reliability. How much do you think it's gonna cost to design, test, install, and maintain that kind of system?
And what happens when someone slams into one of these barriers? Do you think it will still be able to retract afterwards? You've just permanently closed that road until a work crew can come out and clear it -- or at least recertify it for 5 nines.
Fabulous.
I could just see that.
Yellow light. Car hits the gas. Crosses into the intersection as the light hits red. Steel plate up.
Choices:
(a) Slam into steel plate.
(b) Sit in the intersection. (It'll happen, someone will gun it and then second guess themselves; blocking the flow of traffic, including your thin bony ass.)
I don't like the plate idea or the moat idea, but I do like the original thinking.
1, Are people who get up late for work.
2, Parents doing the school run.
3, Women putting on makeup on the way to work.
4, People using cell phones and or sat navs.
5, Keystone cops
6, Drunk drivers
7, People trying to make it back home for dinner.
8, People who think that they think they are playing Gran Turismo or need for speed.
That would cover a more accurate eight legged groove machine.
If the cars can communicate and they stop for cars that are highly probably to run a red light, the fastest car with no intention of stopping will be given the right of way?
This can be used for a fine generator. First the system figures out who is most likely to run a red light and then it presents that driver with constant red lights, all of which have cameras installed.
This should help increase the income from fines...
For the technology to work, How said vehicles would need to be able to communicate with one another, wirelessly sending and receiving data like the car's speed and position.
If you like to run red lights, why would you let your car advertise its speed and position?