Video Analytics and Vision Zero

vision-zero-analytics

It's been a couple of years since we've written about Vision Zero (1, 2), but hardly a week goes by that we aren't reminded of the importance of it. As Portland's streets get busier, they also get more deadly. Last year, 50 people lost their lives in traffic incidents. (For some reason, PBOT's Vision Zero map only shows 45.) Either way, I don't think anyone would disagree that it's far too many.

Thankfully for Portland, tax revenue is finally available to start addressing some of its most dangerous streets, and construction on 17 safety projects will begin this year. And while traffic engineers are always striving for better and safer designs, we still have a lot to learn about how to create urban environments where traffic violence isn't a given.

Enter Microsoft's Video Analytics toward Vision Zero. It's a project that seeks to crowdsource data about traffic in intersections to help machines (i.e., computers) "learn" patterns and influences that can then be used by designers to come up with safer street configurations. Anyone with a browser and a bit of time on her hands can help by identifying traffic elements (pedestrians, bicycles, cars, buses, etc.) in videos taken at intersections across North America. All you do is draw boxes around traffic elements and make sure they track as the video proceeds. It's an interesting meditation on the idiosyncrasies of perspective and low-resolution image capture. I recommend putting on some good music. When I did it, I got a ten-second clip of rush hour traffic on the Key Bridge in Washington, D.C. I tracked 36 discrete traffic elements, which took a little more than an hour. Traffic was more sparse in the next clip that came up, so your results may vary.

I don't know if machines can learn enough to keep us from killing each other in traffic. But often these crashes are referred to as "accidents," as if people can't help but make these fatal mistakes. Perhaps there are more structural changes that can be made to mediate our inherent fallibility. We already know what a factor speed is, and we're trying to do more about that. Microsoft wants to see if the robots can help us figure out what else to try.

Ady Leverette was a designer and a principal at Fat Pencil Studio between 2011 and 2018.