1) The social distancing detector was not effective at identifying potential violations of social distancing rules in the input video. I belive this is because the detector works when examining the full front or back of a person. It is unable to detect a person using their side profile and therefore unable to detect any violations.
2) The general approach of monitoring and displaying violations in public when and where they occur, I think, is a valid approach to enforce social distancing. This concept is analagous to signs displaying the speed of oncomming cars and warning them to slow down if they are over the speed limit. However, the model should be optimized in order to accurately discern families, possibly be identifying children and therefore not analyzing the group of people they are with. Effective enforcement would also mean using a method of getting the attention of offenders in order to communicate that they are in violation of social distancing guidlines. I would put up a camera that displays the output, so that people can see themselves and possibly others. This would contribute to the offenders feeling singled out and potentially publicly shamed by having their picture displayed and will cause them to increase their distance between others. A clear large red warning message that they are in violation would also incentivize offenders to follow the guidelines.
3) I would improve the camera angles to ensure that they are placed directly in front of high traffic areas, ensuring they capture images of people fully front facing. The most optimal locations would include doorways and waiting areas. I would impose the limitation of not saving any video after a certain amount of time, perhaps a week. This is to protect the privacy of those captured on camera. I believe the addition of any noise maker to get the attention of the offenders would be excessive and aggravating to all passing by and therefore it will not be included.