A project to crowd-source data on Arlington, VA's frequently-blocked bike lanes.
I want to help!Are you sick of bike lanes blocked by cars, shuttles buses and delivery vans? Help gather action-able, verifiable data on how frequently Arlington's bike lanes are blocked so we can make a strong case to the police for pro-active enforcement.
All you need to do is look at a photo of a block that has a bike lane and tell us it appears to be blocked or not.
You can do it from your phone, from your iPad or from your PC. Whether you can do 100 or just 1, every little bit helps. Do it while you wait for the bus, while drinking a beer on your patio or as penance anytime you fail to be a PAL. Let's get a solid answer to the question - Just how frequently are Arlington's bike lanes blocked?
Let's see some bike lanes!Crystal Drive: 2016-09-21 | 2016-09-28 | 2016-10-05 | 2016-10-12
Clarendon Blvd & Wayne: 2016-09-21 | 2016-09-28 | 2016-10-05
Hayes St & 12th: 2016-09-20 | 2016-09-21
The system consists of two pieces. The first is an application that saves a screenshot of various Arlington County Traffic Cameras every minute along with the date and time that it was saved. The second is this website, which displays those screenshots to users of the site and records whether the bike lane is blocked.
The site treats each screenshot as "unsure" until at least two different people agree that the lane is either blocked or unblocked. If people disagree, the system continues to show it to new users until there is a majority who agree.
This SEEMS like maybe the kind of problem that a computer would be good at, but I'm not computer vision expert. If you are (or would like some data to learn with) I have created two sets of training images that are already classified by humans that you could potentially use to train a neural network (or something similar) to do this work. If you get something that works decently, please let me know!
The data is a simple zip file with two folders in it - "blocked" and "notblocked". All of the photos in the blocked folder have been classified by humans as having a blocked bike lane. You get the idea.
The "single camera" set contains all the classified images I have from the single camera that I have the most data for. It contains about 3,600 images and is about 500mb in size. The "all cameras" set contains all the classified images that I have across all cameras. It contains about 7,600 images and is about 1.5GB in size.
If you come up with something that works, please let me know!
Download Training Images - Single Camera Download Training Images - All Cameras