Mounted to seat posts of bike commuters recruited by a local bike advocacy group, those cameras automatically recorded and analyzed hours of footage to document not only the location and prevalence of near-miss crashes along popular bikeways, but also the average speeds and passing distances of nearby drivers, and even the types of vehicles on each road.

Ultimately, the tech compiled all the data into a series of "stress scores" that Ann Arbor could overlay across its map alongside other data like crash statistics, bike counts, and cyclist testimonies to paint a compelling picture of where Ann ARbor's next bike paths should be built, or where existing paths should be outfitted with more protections.

Velo AI is also partnering with Pittsburgh's bike share provider, POGOH, to put the cameras onto some of its fleet, which they hope will help them capture data about biking conditions in less-affluent communities where shared bikes are frequently ridden. They've also received a grant from the federal Department of Transportation's Complete Streets AI pilot in August 2024, through the office's entire web presence has since been taken down following the new administration entering the office.

Velo AI isn't the only AI-powered bike camera, of course, with other companies using similar technology to things like help cyclists automatically record drivers who violate passing laws and receive real-time alerts about drivers who pass suddenly into their blind spots. CoPilot itself is also sold directly to riders who can use it to automatically alert drivers to their presence using extra-visible light cues, among other features.

The kinds of applications can sometimes prove controversial among sustainable transportation or those who can't afford the latest technologies and fear being blamed if they're victim of a crash—making pilot programs like Ann Arbor's an interesting evolution of how communities-at-large can benefit from rapidly advancing technologies like Velo AI's tools. The idea of using AI to super-charge data collection may very well be able to cut through the ongoing debates in this space, transforming technology into a tool that ultimately can better support the case for bollards and jersey barriers (rather than a replacement for them).

"We've used crash data previously to evaluate our projects and the impact on safety; we also have bike count data, which is helpful to some level," said Shelby Fergon, planning specialist for the Ann Arbor Downtown Development Authority. "But that near-miss data was somethign that we were lacking; it seemed like a really neat opportunity to quantify the experience that people were having in a way that we weren't able to previously."

When layere dwith qualitative data like interviews with riders themselves, Ann Arbor officials say Velo AI's insights have helped them put numbers and images to harrowing stories that cyclists have been telling them for years, making those experiences undeniable to even the most ardent skeptics. They say they've already seen results from sharing footage, maps, and analyses from the pilot in meetings with local politicians, and used those tools to make the case to the general public that the downtown bike network must be extended—even though all downtown addresses are alerady within two blocks of a separated bike lane.

Ann Arbor officials said they could envision AI bike cameras being used to complement other kinds of safety efforts, too, like putting real bike footage into PSAs, or helping driver's ed students see exactly how scary it is for a cyclist to be close-passed by a driver.

Source: Streetsblog