Safe Distance Monitoring for COVID-19 Using YOLOv3 Object Recognition Paradigm

The ongoing outbreak of the COVID-19 virus has caused a global catastrophe with its deadly spread. The risks of the virus can be reduced by following social distancing. Therefore, this work aims to come up with a deep-learning platform for community-level tracking. YOLOv3 object recognition paradigm is used to identify people in video sequences. The detection tool is designed to test whether people keep a safe distance with a video feed test. A pre-recorded video was used as an input and a pre-trained opensource object detection model based on the YOLOv3 paradigm was used to identify people. The acquisition model identifies people using the information binding box obtained. By using the distance formula (Euclidean), the centroid's pairwise distances of people in the detected bounding box are calculated. To evaluate social distance violations between individuals, we have used physical distance to pixels and set a threshold value. The infringement limit is established to assess whether the range of the breach violates the minimum public distance.