Social Distancing System

This machine learning-based tool detects and tracks social distancing violations in a given area. This system can be implemented on any video stream, like CC TV footage, to distinguish between people following social distancing and those not. The tool follows a three-step process to determine the social distancing detection status. Firstly, it detects humans in the frame using the pre-trained YOLOv3 model. Next, it calculates the distance between every person detected in the frame. Finally, it provides information on how many people violate social distancing rules.

The detection model provides bounding box information with centroid coordinates around every person detected, which are then used to measure the pairwise centroid distances between the detected bounding boxes. An approximation of physical distance to the pixel is used to determine whether the distance value violates the minimum social distance threshold. With the implementation of this system, people can be made aware of social distancing violations and take necessary precautions to protect themselves from COVID-19.