New York's cyclists and bus-riders are certain they're being slowed and endangered by an epidemic of illegal lane-obstructions from delivery vehicles, taxis and Ubers, but policymakers have refused to do anything about it, saying that the evidence is all anaecdotal.
So a machine learning researcher/cycling advocate named Alex Bell created an image classifier that is trained to recognize when a bike- and/or bus-lane is illegally obstructed (sourcecode on Github), producing the first hard data on the phenomenon. He's analyzed the footage from a single traffic camera in his Harlem neighborhood, one St. Nicholas Avenue between 145th and 146th Streets, and found that the bus-stop there is blocked 57% of the time, while the bike-lanes are obstructed 40% of the time.
"Everyone keeps talking about the bus lanes being blocked, and buses being so slow I could just say, 'Oh they are so blocked, it's so bad,' — but that's not very helpful," Mr. Bell said in an interview. "I thought to myself, 'How could you show to everyone that bus stops and bike lanes are routinely blocked?' "
…The enforcement that does exist through the use of cameras is slim: Just 12 of the city's 317 bus routes have cameras mounted on objects like streetlights that are similar to red-light cameras. Any plan to add more such cameras must be approved by the State Legislature.
Our Camera [Alex Bell/Github]
Bus Lane Blocked, He Trained His Computer to Catch Scofflaws [Sarah Maslin Nir/New York Times]