Bellevue, WA, plans to use AI to leverage cameras for safety

by Michael Brenneis

Agencies that aspire to achieve zero traffic fatalities need to know where to invest for the biggest crash reductions. Advances in artificial intelligence (AI) are allowing DOTs to leverage their existing camera technology in order to extract large quantities of data that can then inform decisions about how to improve or control intersections. The city of Bellevue, WA, recently announced a plan to study footage from its traffic cameras in order to “analyze the correlation between past collisions” and near misses, according to a press release.

Most cities don’t have the budget to retrofit their entire transportation network to complete streets standards. Out of necessity they prioritize the redesign of the locations where the most crashes happen. But crash data is sparse, and some crashes go unreported, so it can take time to build a robust data set.  Near-miss data helps to paint a more nuanced picture of road safety, but it has historically been costly to collect. With a near-miss video analytics system up and running, as proposed in Bellevue, cities won’t need to wait for a number of crashes to occur before making changes to dangerous intersections. The Video Analytics Towards Vision Zero Partnership is a collaboration of the city of Bellevue, Brisk SynergiesTogether for Safer Roads, and PacTrans–University of Washington.

Once the AI system is trained, it will be able to extract near-miss data from the hours of traffic camera footage already collected, at a rate well beyond what humans can reasonably be expected to identify. Detroit is piloting a similar system in an effort to make intersections safer. It consists of cameras and AI that can make modifications to traffic signal controls in real time.

Automated data collection has the advantage of consistency, and minimal vulnerability to human bias. Bellevue plans to train the AI using crowdsourcing, which can benefit from the perspective of diverse volunteers, and has the potential to be economical. Aggregating multiple independent responses may also avoid some of the pitfalls of crowdsourcing, such as wrong answers or manipulation.

All the data in the world will not improve the survivability of roads without good policy to guide their redesign. Bellevue’s pursuit of Vision Zero, which puts safety ahead of speed, is laudable. Facilitating the implementation of policies that reduce VMT, prioritize transit, and make active modes more attractive and safer is what justifies collecting all of this data.

Michael Brenneis is an Associate Researcher at SSTI.