This article was originally published to Technical.ly
Imagine for a moment that, every week, four to five commercial airplanes crashed in America.
In reality, a similar number of people die per week in traffic accidents, but, for the most part, those deaths don’t resonate with us in the same way.
“If that many airplanes were crashing every week, it would not be acceptable to people. Unfortunately, we’ve come to view traffic death as the ‘cost of doing business,” said Franz Loewenherz, principal transportation planner for the city of Bellevue in Washington state. “But it doesn’t have to be.”
Under his leadership, Bellevue has partnered with Microsoft to come up with a software solution that could reduce or eliminate traffic deaths. The software, developed under Microsoft Distinguished Scientist Dr. Victor Bahl, will eventually recognize locations where crashes are most likely to occur.
“This is an entirely different way of looking at safety,” Dr. Steven Lavrenz, technical programs specialist with the Institute of Transportation Engineers said. “We’re detecting events before they occur in order to ensure that they never do.”
How does it work?
In short, the software algorithm analyzes a city’s traffic footage and uses artificial intelligence to recognize traffic events known as near-misses. (Think: a car screeching to a halt to avoid hitting a pedestrian.) This strategy, known as surrogate safety analysis, uses near-misses as “surrogate events” in order to detect risk before actual crashes take place.
As patterns emerge, city transportation officials will gain the ability to identify where the highest risks for traffic accidents are around the city. Eventually, the goal is to build a database that city officials will be able to use to spot the riskiest areas on the streets.
But the software is not yet fully developed — it still has trouble differentiating between pedestrians and bicyclists. And the challenge is that the software cannot teach itself; it requires a human being to label an object multiple times before it can begin to recognize it on its own.
So, to bring this project across the finish line, developers are calling on the public to help by watching traffic footage and labeling people and bikes.
That’s where D.C. comes in.
Last week, Mayor Muriel Bowser announced that D.C. would be participating in this effort as part of the Vision Zero initiative, a campaign to end traffic fatalities in the District by 2024.
What exactly has Mayor Bowser signed us up for? Now that we are signed on, Microsoft will be granted access to the 130 closed-circuit traffic cameras that perch above our streetscape here in the District. Video footage of traffic will be fed into the software, and people from all over the United States will help by labeling objects for the machine to learn.
Eventually, D.C. city officials will have access to a wealth of information about high risk areas on our streets and be able to respond accordingly. While many of these solutions won’t be clear until the problems are identified, corrective actions could include adjusting traffic signal timing, reducing crossing distances for pedestrians, implementing roundabouts or conducting public education campaigns.
Artificial intelligence: driving the future
All of this represents the tip of an iceberg, with artificial intelligence set to play an increasing role in the way we transport ourselves in coming years. A great example of this, of course, is the autonomous car — but think beyond that. Here are four ways AI could help:
- Improve your driving. Dr. Victor Bahl says that as autonomous cars gain sophistication, they will be able to communicate with surrounding infrastructure to keep us safer. Take for example a recent demo in Hannover, Germany in which an autonomous car was able to brake safely for a pedestrian obscured from its view. How? By communicating with a traffic camera that had a clear view.
- Save you time. Artificial intelligence will also help motorists save time. An example of this includes Xerox Research Center Europe’s plans to develop software that recognizes available parking spaces on city streets via cameras attached to buses. Another is the Robotics Institute at Carnegie Mellon project in Pittsburgh to develop “smart” traffic lights that use artificial intelligence to communicate with one another, adapting to changing traffic conditions in order to increase efficiency at intersections. They’ve been shown to reduce travel time by 25 percent and idling time by over 40 percent.
- Improve public safety. Dr. Bahl points out that an intelligent, connected camera system could even aid in Amber Alert cases by sorting through traffic footage in real time to identify car color, make, and model and license plate numbers.
- Help navigate. It’s feasible that this technology could compete with Waze and Google Maps by using smart, connected cameras to issue real-time traffic alerts to help ease congestion. (And, Dr. Bahl points out, the less time spent on the road, the fewer opportunities for accidents.)
Where do we go from here?
Before the software can go mainstream, privacy concerns will need to be sorted out. According to Loewenherz, a number of cities interested in implementing this technology have had to put it on the backburner while working out issues with their existing video privacy policies.
But it is likely that, as the software develops and yields compelling results, more cities will be eager to sign on.