Can Data Analytics Make Dangerous Intersections Safer?

A project in Bellevue, Wash., uses video data and machine learning to learn which streets and intersections are the most dangerous. The data is considered more reliable and less biased compared to traditional surveys.

by / September 20, 2019
A project in Bellevue, Wash., uses video data and machine learning to learn which streets and intersections are the most dangerous. This heart map image from an intersection in Bellevue aggregates multiple occurrences of close-call events between people walking and driving. Using this visual output, engineers and planners can identify where the most conflicts occur in the intersection. Brisk Synergies

Bellevue, Wash., located in the Seattle metro area, is undergoing a citywide review of near-miss incidents involving pedestrians, cyclists and other cars. Using images from its closed circuit video network, as well as high-level analytics and machine learning, the city wants to understand which streets and intersections are the most dangerous, and how they might be made safer.

Bellevue is partnering with the group Together for Safer Roads (TSR), which represents a coalition of private-sector companies, including Brisk Synergies, to conduct a comprehensive near-miss study from August to September where roughly half of the city’s network of 80 public video cameras will be used to gather some 34,000 hours of footage representing about 21 terabytes of data. The data will be processed by Brisk using artificial intelligence and machine learning to gain insights into “near-miss” incidents.

“This is the first network-wide traffic safety monitoring assessment of its kind,” said Franz Loewenherz, principal transportation planner for Bellevue. “Together, we’re using Bellevue’s existing traffic cameras at key intersections to identify the rate and severity of near-miss collisions between people driving, walking and bicycling."

Loewenherz is hopeful the data from the project will provide “predictive insights” on where and how collisions are happening, or could happen. That information can then be worked into the city’s Vision Zero Action Plan.

Technologically led near-miss studies like this one are promising, according to officials, because they offer information to help prevent accidents before they happen, rather than having transportation experts devise intervention efforts after an accident has occurred.

Crash and fatality data “is only one data point,” said Noah Budnick, TSR's senior director for programs and operations. “And if that’s all you’re relying on, you map those and it might not tell you very much. With that small of a data set patterns may not emerge. Having a data set that’s more recent, like from the last month, or six months, can really tell you what’s going on,” he added.

Also, live feed video footage is often more effective — and cheaper — than having human observers stand at an intersection, recording what they see, said Budnick.

“For me, the amazing thing about using technology and machine learning in this hit or miss stuff is that, back in the day, we used to do near-miss surveys where you’d go to a dangerous intersection with a clipboard and you’d survey who was walking by or biking by and ask them questions, or you’d stand there and observe,” said Budnick. “And while those are amazing community organizing strategies, they’re very labor intensive, and they’re very subjective.”

By collecting such a large cache of video footage, transportation officials get a vantage point that “is impartial, unbiased, and can really be a strong and reliable data set for planners at the city of Bellevue to then prioritize where they’re making improvements, and get ahead of where fatalities or serious injuries may occur,” Budnick added.

The data produced by the two-month study can be “searched, managed, and used to provide detailed information on traffic flow, speeds, and other vehicle conditions,” said Loewenherz. “This will allow a more rapid response to traffic incidents.”

Other cities are using technology to learn about traffic safety concerns. Las Vegas is rolling out a project across its downtown where sensors will detect wrong-way driving in the area’s many one-way streets. The idea is to gain data-supported insight into this problem to learn exactly where improved signage and other interventions should be put in place.

In Bellevue, transportation officials have not ruled out extending the project beyond its two-month window. The data and analytics it provides could be invaluable when it comes to evaluating the effectiveness of the city’s Vision Zero efforts, according to Loewenherz.

“Ultimately, we hope that this work will be of value to other cities, both in the U.S. and in other nations, as we work toward the common goal of zero fatalities and serious injuries on our roadways,” he added.

Skip Descant Staff Writer

Skip Descant writes about smart cities, the Internet of Things, transportation and other areas. He spent more than 12 years reporting for daily newspapers in Mississippi, Arkansas, Louisiana and California. He lives in downtown Sacramento.

Platforms & Programs