- My code seemed to be quite effective in allocating the correct body figures within the video
I provided and recognized quickly whether figures were becoming to close or were within the
parameters of separation.
- I think that this tool can be used as research tool for the COVID-19 outbreak and for better coverage
among heavily dense areas. Though, I do not think that this tool is a definitive tool for saying that
the outbreak of COVID-19 will or will not happen. Yes, it can be used as a resource for studying possible
outbreaks, but not providing 100% accuracy predictions, only as a secondary source of information due to
immense amount of variables that contribute to viral outbreaks. I believe by putting this algorithm into action
on more densly populated college campuses would be more beneficial then William and Mary as our school is more
rural than others such as University of Virginia. Secondly, I believe that this could also play a possible
role in studies of urban health studies to help track human interaction in cities such as Detroit, New York, and
Los Angeles. By creating an initial study with this tool in largely populated and dense cities, researches
can better help government entities allocate resources towards health and safety.
- One thing that I found to be a possible flaw in the system was the angle in which you see the movement of figures
accross the screen. Inaccurate information can come about when running the simulation without possible
parameters incoded. When the camera is at a certain angle, the depth perception of the algorithm can be
scewed. For example, in the video I had run, my algorithm located numerous hockey players on the ice skating beside
each other and indicated that there was a detection of social distance violations, on the other hand, with a camera angle
below 90 degrees, the algorithm saw two figures that might be next to each other but were really further than feet
in front of one another. Of course to fix this, the use of a camera at a higher angle greater than 70 degrees could highly
improve the output predictions but this would require positioning of a camera in the rafter of the sporting arena or on a
skyscraper if installed in an urban area. Also, if you have people that are moving rather quickly, the amount of violations
occurs quickly and you are unable to capture the amount. If a counter was encoded into the algorithm to display each violation
on the screen increasing as the video played then it may help researchers calculate the amount of activity based on density of
population.