I agree with most of your points. However, I think it would be easier to use machine learning and computer vision to solve this problem. The profit would come i

I agree with most of your points. However, I think it would be easier to use machine learning and computer vision to solve this problem. The profit would come in the form of reduced traffic which would mean more throughput to local businesses. FedEx would profit, trucking would profit, mom-and-pop shops with prime retail locations would profit. Hence, the municipalities would profit from surge in tax revenue. As far as processing goes, the red-light cameras get recorded somewhere, right? Why not double up those servers, put an RPI with a camera on each traffic light, and then beam the data back to the server room for processing. With machine learning and a lot of patience the system would be able to improve itself over time. I have always believed that this problem could be solved with about $200 worth of hardware at the traffic light, a cheap mobile plan, and a central server location (wherever they host the red-light footage from cameras that already exist). Now, if anyone has a million bucks laying around I can get to work right away. ;)