Microsoft looks to stop bike crashes before they happen
Microsoft engineers and City of Bellevue planners have a sci-fi inspired strategy for curbing bike and pedestrian injuries on city streets: By using video analytics, they want to predict and prevent crashes before they happen.
“This is like ‘Minority Report,’ ” said Bellevue senior transportation planner Franz Loewenherz, referring to the 2002 film in which Tom Cruise preemptively stops crime. “We’re trying to get out in front of the collisions. We can take a corrective measure before someone gets hurt.”
The project employs the city’s existing traffic cameras used by police to investigate crimes and transportation officials to optimize the timing of traffic lights in this city just east of Seattle.
The Microsoft scientists are developing software for analyzing the footage, identifying whether a car, bike or pedestrian is using a street or sidewalk, their rate of speed and their trajectory. They’re writing algorithms to look for potential collisions and near misses in order to identify dangerous intersections and roads.
“If we nail it, this is huge,” said Victor Bahl, director of Microsoft Research’s Mobility and Networking Research. “Once we get the basic problems solved and get everybody excited, we open up the floodgates” to tackling more safety and infrastructure problems and “creating a next-generation traffic management system.”
The collaboration fits into the city’s pedestrian and bicycle safety initiative that launched in 2009. And earlier this year, Bellevue joined more than 200 cities participating in the “Mayors’ Challenge for Safer People and Safer Streets,” a yearlong campaign led by the U.S. Department of Transportation that aims to reduce traffic-related injuries and death.
The project started this past spring and is in the proof-of-concept phase. Microsoft researchers have experimented with video collected on their Redmond campus and are testing footage from a limited number of Bellevue’s cameras.
As the effort expands, the city has 200 cameras that could provide video.
Bahl was inspired to tap video cameras as a tool for addressing public safety following a recent sabbatical spent partly in London, where there are tens of thousands of surveillance cameras. At the same time, Loewenherz was exploring ways to make some gains in his city’s safety initiative.
“It just so happened that we met and said, ‘Let’s see if we can help each other,’ ” Bahl said.
He enthusiastically describes all the ways in which the project is challenging, from deciding which trajectories should be flagged as near crashes, to figuring out the most efficient way to network the video footage, to rapidly and affordably analyzing the data.
“That’s why it’s exciting for us,” Bahl said. “If it stresses every part of the system, and we can solve it, then that’s good.”
His group has “a decent number of Ph.D.s” working on the project, Bahl said, and he hopes to have a working system running within the next 12 months. From there, the dream is to develop technology that could be used by municipalities and governments internationally.
In addition to identifying collision close calls, the project will provide city planners with basic data on the number of cyclists and walkers using different roads. It will provide stats that are laborious to collect through most other means, but yield essential information.
While there is national data showing that the number of bike and pedestrian fatalities have increased since 2009, and Washington-state statistics indicating a steady rate or slight decline in serious injuries for cyclists, hyper-local data are a key driver for regional projects.
“When you go to justify a major investment for pedestrians and bikes, it gets really hard because you don’t have good data,” Loewenherz said.
The analysis can also provide a sense of how well the drivers, bikers and walkers are following traffic rules: Are bikes and cars routinely speeding, do vehicles stop and yield according to law, are pedestrians crossing at crosswalks or jaywalking mid-street?
Folks from the city and Microsoft are mindful of privacy concerns around the videos, and assure the public that they’re not being personally tracked.
“We’re not watching, we’re counting,” Loewenherz said. “Anyone’s individual identity is being held as private.”
Bellevue officials are also turning to crowd sourcing to identify transportation problem areas.
They created a WikiMap and are encouraging people to mark roads and infrastructure that they believe are dangerous. Since the site launched in September, people have tagged more than 1,200 points. The map will be up through October.
The non-profit Cascade Bicycle Club supports the overall bike-safety initiative and their volunteers have been out spreading the word about the effort, and the WikiMap in particular.
“We’re seeing a lot of positive movement around bicycling,” said McKayla Dunfey, the group’s Eastside policy and government affairs coordinator. “We’re reaching this tipping point.”
These efforts and others underway should bring Bellevue closer to “Vision Zero” — a strategy originating in Sweden that works toward zero deaths and serious injuries from traffic collisions.
The approach argues that the injuries are preventable through better street designs, education and less potential for human error.
“We want to reduce the burden on human beings who are tasked with the responsibility of safety on the roads,” Bahl said. Through video analytics, “we are trying to find all the interesting places for them to look at.”