Overall, I find the social distance detector to not be that effective at detecting violations. The detector most focused on the foreground, as it was difficult to distinguish people in the background, but it struggled with those closest to the camera. Four of the main people in the video remain in green squares the whole time, despite the fact that they are less than six feet from one another. However, I think the detector struggled with because of its depth perception. As the video is of a crowd and its taken at eye level, it is difficult for computer to determine the distance between two people in terms of depth. This video depcits the crowd at the removal of Richmond’s Stonewall Jackson monument, so it is similar to many crowds that have gathered across the country lately to protest. Most people were wearing masks and standing some distance apart, but it was still a packed area. I think it’s interesting that the detector seemed to have a harder time with the crowd than it does with people walking down the street, as a crowd presents greater concerns in terms of social distancing.
I think this method could be effective for both maintaining social distance as well as contact tracing. Setting up a live feed in busy areas could help limit social distancing violations, as in addition to the tally and red squares in the video, there could be some sort of noise triggered to indicate to people on the street that they need to separate. It could also be used for contract tracing. However, this brings up privacy concerns, as in order to effectively contract trace people’s faces must be seen. In fact, I felt a bit uncomfortable using a video I took of real people, although I don’t think anyone’s face is visible. It is especially concerning in this context. Although this is not footage of an actual protest, it is similar to one, and many people who protest fear their faces being captured by drones or cameras. Therefore, I think it would probably be best if this approach was solely used for experts to gain data as to how well social distancing is being enacted in certain areas in order to have a better estimation of the spread of new infections and determine appropriate action.
As I mentioned before, I think the detector would need an improved assessment of depth, as that is where it seems to struggle the most in determing violations. One way to decrease this problem would be to get footage from a higher vantage point. Additionally, as we discussed in class, I think it would be more effective to consider each person’s center and place them in a circle with a radius of six feet, as then the distance would be accurate from each angle. However, a bird’s eye view would probably be necessary to implement that properly.