View on GitHub

Mini-Project

Mini-Project

My_Video

Was your social distance detector effective at detecting potential violations? Are you able to describe how the distance detector is applying its calculations of either being safe or noting a violation?

It was effective to an extent but there were many instances where it failed to highlight visually obvious violations. It seems to be using a range where the variable ‘dist’ should not fall between: (0 < dist < 0.25). Anything falling within this range is a violation and anything outside this range ‘dist > 0.25’ is not a violation.

Do you think this approach would be effective for estimating new infections in real time?

I think it may be useful in contact tracing as it would help identify people who were inclose contact and violated social distancing requirements. I dont think this approach would be effective for estimating new infections in real-time as not everyone who violates social distance requirements is necessarily in close contact with someone who carries the virus. If this were used in a way where all violators were marked as potential new infections, it may end up inflating the estimated infection numbers which is bad.

How would you implement such an approach in response to the COVID-19 pandemic we are currently experiencing?

If i had to use this approach i would modify the script so that violators would be highlighted red only if they are in close contact with each for a minimum time limit e.g. (1-5 minutes). As it is right now the detector highlights potential violators immediately which to me isn’t a practical design.

What limitations or improvements might you include in order to improve your proposed design?

I would limit the detector from painting a box red unless a set minimum time limit has been met. It is not very good at recognizing people standing in a straight line so miltiple camera angles would be a necessary addition to improve the detector’s performance. It is also not very good at detecting violations when people are standing very close to the camera whether it’s a small or large group. To remedy this cameras should be placed adequately far away enough from pedestrian traffic in a way that no person/group can get too close to the camera.