I bought a Raspberry Pi a while back without much of a plan for what it would ultimately work on. After getting annoyed by a local traffic issue outside my house, I decided to go all in and see if I could write up a set of software that would:

  1. Record video of people driving cars on my street
  2. Detect the sets of frames within which a car is present, and convert the video to a GIF
  3. Record metadata about the movement of the detected object (the car), and generate a dataset of car transits
  4. Host a website that allows people to view the GIFs and judge whether or not cars are stopping at a stop sign
  5. Continuously train machine learning models that optimize on the votes people have volunteered.

The results of this are here, and the entire library for this project, along with detailed instructions on how to implement a device like this, are here. Also, it was featured on HackADay!