Ben@3D
11-30-2011, 07:26 PM
Interesting read....
Will the other guy run a red light? MIT algorithm figures the odds, tells you to brake
By Bill Howard (http://www.extremetech.com/author/bhoward) on November 30, 2011 at 6:53 pm
http://www.extremetech.com/wp-content/uploads/2011/11/MIT-red-light-20111129173411-1_web-348x196.png
You’ve got a traffic light that’s just turned green. Some idiot approaching the intersection at right angles doesn’t appear to be slowing down. Will he stop or will he sail through on red and give you a chance to test your side air curtains? Researchers at MIT have bottled up a mathematical formula that predicts the odds the other driver will run a red light with 85% accuracy. That’s important because half the time there’s a red light fatality, it’s not the red light runner learning about Darwinism, but an innocent pedestrian or the car with the green light that pays the price.
To predict red light behavior, MIT’s car geeks looked into all sorts of variables. Where you or I might think, “tricked out ride … tinted windows … young driver … we’re in New Jersey … yep, he’s going through the light,” MIT tracked more concrete data and found it could confidently separate potential violators from compliant cars by tracking vehicle deceleration (or lack thereof) and distance from the traffic signal. For all this to work, says MIT professor Jonathan How, the algorithms need a new generation of smart cars with vehicle-to-vehicle (V2V) communications, or short range transponders that constantly report location, speed, direction, rate of acceleration, and brake status.
Only test cars have dedicated short range communications (DSRC) transponders now. Ford this year is road-testing prototype vehicles (http://www.extremetech.com/extreme/92532-ford-working-on-car-to-car-wireless-mesh-network-for-real-time-telemetry-government-use) with advanced WiFi communications systems. But, says How, if the benefits of V2V are clear, “You might have … a snowball effect where, much more rapidly than people envisioned, this [V2V] technology may be accepted.” As the cost of embedding telematics packages with cellphones approaches $100, WiFi-based DSRC might become similarly affordable, perhaps even cheaper. It may also have applicability to air traffic control, says How, who’s a professor of aeronautics at MIT.
One of How’s former students, Georges Aoude, created an algorithm to analyze vehicle data and got the number-crunching down to five milliseconds for a go-or-stop recommendation. That’s equal to three inches of travel by a car approaching the yellow-unto-red light at 30 mph (four inches if, more likely, he’s doing 45). Aoude and fellow MIT students Vishnu Desaraju and Lauren Stephens plotted their algorithm against data from a heavily instrumented intersection in Christianburg, Virginia, and found they could predict red-light runners 85% of the time, 15-20% better than existing algorithms, and with fewer false positives. “If you’re too pessimistic [issue too many warnings],” says How, “you start reporting there’s a problem when there really isn’t, and then very rapidly, the human’s going to push a button that turns this thing off.”
http://www.extremetech.com/wp-content/uploads/2011/11/P90086337_highRes-300x200.jpg (http://www.extremetech.com/wp-content/uploads/2011/11/P90086337_highRes.jpg)
The MIT researchers say the warning would best be delivered via a head-up display in the driver’s line of sight. BMW recently implemented a high-res, full-color HUD (http://www.extremetech.com/extreme/102723-could-bmws-full-color-head-up-display-replace-the-instrument-panel). Ford has been talking about a poor man’s almost-HUD created by mounting key instruments at the base of the windshield, above the steering wheel and almost in the driver’s line of sight, for a lot less than BMW’s $1,300 HUD. Honda has some cars with an upper instrument cluster just below the windshield.
Details on the red light algorithm will be in the journal IEEE Transactions on Intelligent Transportation Systems. Other MIT researchers recently reported on SignalGuru (http://www.extremetech.com/extreme/94118-how-a-smartphone-camera-can-cut-fuel-costs-20-at-rush-hour), a technology using a dashboard-mounted smartphone camera to predict the best driving speed to pace green lights going in your direction. Such a technology could cut city fuel consumption by 20% and the camera also eyeballs gas prices as pass stations.
It’s unclear how many lives would be saved each year if the red-light running algorithm makes its way into a future generation of cars with vehicle-to-vehicle communications. The most would be 700; last year there 32,788 US traffic fatalities (http://www.nhtsa.gov/PR/NHTSA-05-11). According to the National Highway Traffic Safety Administration, there are 2.3 million car crashes at intersections each year, with 7,000 deaths; 700-plus fatalities are the result of running red lights. But no predictive model is 100% accurate, drivers won’t always listen to good advice, and half the deaths are probably also logged as drunken driving or no-seat-belt fatalities as well. (Fatalities, like success, have many fathers.) Cut DUI fatalities with tougher enforcement and you’ll take a big bite out of red light fatalities; the MIT algorithm ought to cut the toll still further.
Read more at MIT (http://web.mit.edu/newsoffice/2011/driving-algorithm-1130.html), or download a video to see the algorithm in action (http://acl.mit.edu/ITSC10TAM.mov/ITSC10TAM.mp4)
From:
Will the other guy run a red light? MIT algorithm figures the odds, tells you to brake | ExtremeTech (http://www.extremetech.com/extreme/107395-will-the-other-guy-run-a-red-light-mit-algorithm-figures-the-odds-tells-you-to-brake?utm_source=feedburner&utm_medium=feed&utm_campaign=Feed%3A+ziffdavis%2Fextremetech+%28Ex tremetech%29&utm_content=Google+Feedfetcher)
Will the other guy run a red light? MIT algorithm figures the odds, tells you to brake
By Bill Howard (http://www.extremetech.com/author/bhoward) on November 30, 2011 at 6:53 pm
http://www.extremetech.com/wp-content/uploads/2011/11/MIT-red-light-20111129173411-1_web-348x196.png
You’ve got a traffic light that’s just turned green. Some idiot approaching the intersection at right angles doesn’t appear to be slowing down. Will he stop or will he sail through on red and give you a chance to test your side air curtains? Researchers at MIT have bottled up a mathematical formula that predicts the odds the other driver will run a red light with 85% accuracy. That’s important because half the time there’s a red light fatality, it’s not the red light runner learning about Darwinism, but an innocent pedestrian or the car with the green light that pays the price.
To predict red light behavior, MIT’s car geeks looked into all sorts of variables. Where you or I might think, “tricked out ride … tinted windows … young driver … we’re in New Jersey … yep, he’s going through the light,” MIT tracked more concrete data and found it could confidently separate potential violators from compliant cars by tracking vehicle deceleration (or lack thereof) and distance from the traffic signal. For all this to work, says MIT professor Jonathan How, the algorithms need a new generation of smart cars with vehicle-to-vehicle (V2V) communications, or short range transponders that constantly report location, speed, direction, rate of acceleration, and brake status.
Only test cars have dedicated short range communications (DSRC) transponders now. Ford this year is road-testing prototype vehicles (http://www.extremetech.com/extreme/92532-ford-working-on-car-to-car-wireless-mesh-network-for-real-time-telemetry-government-use) with advanced WiFi communications systems. But, says How, if the benefits of V2V are clear, “You might have … a snowball effect where, much more rapidly than people envisioned, this [V2V] technology may be accepted.” As the cost of embedding telematics packages with cellphones approaches $100, WiFi-based DSRC might become similarly affordable, perhaps even cheaper. It may also have applicability to air traffic control, says How, who’s a professor of aeronautics at MIT.
One of How’s former students, Georges Aoude, created an algorithm to analyze vehicle data and got the number-crunching down to five milliseconds for a go-or-stop recommendation. That’s equal to three inches of travel by a car approaching the yellow-unto-red light at 30 mph (four inches if, more likely, he’s doing 45). Aoude and fellow MIT students Vishnu Desaraju and Lauren Stephens plotted their algorithm against data from a heavily instrumented intersection in Christianburg, Virginia, and found they could predict red-light runners 85% of the time, 15-20% better than existing algorithms, and with fewer false positives. “If you’re too pessimistic [issue too many warnings],” says How, “you start reporting there’s a problem when there really isn’t, and then very rapidly, the human’s going to push a button that turns this thing off.”
http://www.extremetech.com/wp-content/uploads/2011/11/P90086337_highRes-300x200.jpg (http://www.extremetech.com/wp-content/uploads/2011/11/P90086337_highRes.jpg)
The MIT researchers say the warning would best be delivered via a head-up display in the driver’s line of sight. BMW recently implemented a high-res, full-color HUD (http://www.extremetech.com/extreme/102723-could-bmws-full-color-head-up-display-replace-the-instrument-panel). Ford has been talking about a poor man’s almost-HUD created by mounting key instruments at the base of the windshield, above the steering wheel and almost in the driver’s line of sight, for a lot less than BMW’s $1,300 HUD. Honda has some cars with an upper instrument cluster just below the windshield.
Details on the red light algorithm will be in the journal IEEE Transactions on Intelligent Transportation Systems. Other MIT researchers recently reported on SignalGuru (http://www.extremetech.com/extreme/94118-how-a-smartphone-camera-can-cut-fuel-costs-20-at-rush-hour), a technology using a dashboard-mounted smartphone camera to predict the best driving speed to pace green lights going in your direction. Such a technology could cut city fuel consumption by 20% and the camera also eyeballs gas prices as pass stations.
It’s unclear how many lives would be saved each year if the red-light running algorithm makes its way into a future generation of cars with vehicle-to-vehicle communications. The most would be 700; last year there 32,788 US traffic fatalities (http://www.nhtsa.gov/PR/NHTSA-05-11). According to the National Highway Traffic Safety Administration, there are 2.3 million car crashes at intersections each year, with 7,000 deaths; 700-plus fatalities are the result of running red lights. But no predictive model is 100% accurate, drivers won’t always listen to good advice, and half the deaths are probably also logged as drunken driving or no-seat-belt fatalities as well. (Fatalities, like success, have many fathers.) Cut DUI fatalities with tougher enforcement and you’ll take a big bite out of red light fatalities; the MIT algorithm ought to cut the toll still further.
Read more at MIT (http://web.mit.edu/newsoffice/2011/driving-algorithm-1130.html), or download a video to see the algorithm in action (http://acl.mit.edu/ITSC10TAM.mov/ITSC10TAM.mp4)
From:
Will the other guy run a red light? MIT algorithm figures the odds, tells you to brake | ExtremeTech (http://www.extremetech.com/extreme/107395-will-the-other-guy-run-a-red-light-mit-algorithm-figures-the-odds-tells-you-to-brake?utm_source=feedburner&utm_medium=feed&utm_campaign=Feed%3A+ziffdavis%2Fextremetech+%28Ex tremetech%29&utm_content=Google+Feedfetcher)