I think the social distance detector was very effective! In my video clip, there’s a short segment of 2 people walking towards each other on a sidewalk, and you can see the detector turn red as they approach one another. I will say there were some people together, whose detector should have been red, when they were actually shown as green. However, I think this is due to the quality and camera angle of the video. It should be a more aerial view.
Yes! I think this is definitly helpful in improving COVID-19 response. I think this code should be applied to videos taken in grocery stores and Costco, to see if the measures they have taken (implementing “paths” throughout the store) actually help maintain social distancing. Additionally, this detector may be helpful in deciding what protocols to take when reopening schools and colleges.
I think one limitation is definitly the camera angle. The code won’t run well if the camera isn’t slighlty from “higher up”. It’ll still work, but it may classify some interactions as maintaining distance, because the depth perception is unclear. An improvement that would be interesting to make is to distinguish between strangers ans family. For example, if a child and their mother were on camera next to each other, they would be classified as breaking social distanicing. However, if they have been quarantining with one another, then the only interactions that need to be classified is their distance to other people. Of course, this is a huge challenge, but I think it would be a great improvement!