During this COVID-19 pandemic , many industries/organisation are required to maintain social distancing among their workers. so we felt to built a system to overcome this situation using our knowledge.
What it does
Several CCTV cameras capturing different area of industries are fed to our system . Then the videos are processed ,the humans are detected . And the area in which the people are not maintaining social distancing are alerted along with the video output of showing red boxes around the people who are violating social distancing.
How I built it
We completed this project in three parts. They are as follows -Human Detection -Distance detection between detected human -Highlighting or alerting the hummus who are not following social distancing
In brief Firstly we detected the Human from the video. For detection of human, we using pretrained Neural Network.,specially Mobilenet SSD i.e. region-based neural network. After the detection of humans,we highlighted the output using boxes around the respective humans.Then we calculated the distance between them.We specified the distance for safety as 5 feet from each person to person. If the distance between any pair of people is greater than 5 feet then ok for which the human highlighted with ‘Green Boxes’.If the distance between any pair of people is less than 3 feet then, these both people are highlighted with ‘Red Boxes’.Here we specified threshold distance as 5 feet, but this can be changed accordingly and monitored.
Challenges I ran into
Firstly this project was useful only when this all computed in real-time. For which all human detection , distance detection, and alerting should not lagging. For this have do make algorithm computationally less expensive. In addition to that, we have to have accuracy very good. So adjusting these two thighs were difficult but we were successful and for computation per frame in a video it is taking around 1 sec. Another thing was all video was captured by CCTV which was fitted to a top. So the calculation of actual distance was a task.
What I learned
We learnt how to implement different concepts together in one project.Also by doing this project we got to know how to ML can be applied in real world especially Today
What's next for Real Time Social Distancing Tracker for COVID
we are trying to recognize person individually who is violating social distancing and alerting them