Generalized method to validate social distancing using median angle proximity methodology

Social distancing in industries as well as all public places due to the breakdown of pandemic of COVID-19 is currently of global interest. An advanced global solution for social distancing monitoring among the group of people in real time using computer vision and deep learning is proposed. A state of art method, Median angle proximity method is proposed on the coordinates obtained from the YOLOv3 (You only look once) algorithm to determine the predefined threshold resulting in the social distancing violating bounding boxes. The violating bounding boxes are then saved as single images using deep sort method. All the parts of the work are done using opensource programming language. The work has met all the requirements planned and is considered to be ready for deployment in real time with an average accuracy of 85%