Implementación de un sistema de reconocimiento de distanciamiento social como medida preventiva para covid 19 usando deep learning

Abstract

COVID-19 is a global problem, since it originated in 2019, no solution has been found, this keeps the world population on alert to prevent contagion, as there is no effective solution, the recommendations of the authorities and Health organizations are the use of masks, maintain two meters of social distance, frequent hand washing, avoid frequent contact of the mouth or nose. Of all these measures, social distancing is something that cannot be controlled correctly, because in places where there is a lot of people shopping centers, pharmacies, hospitals, etc. correct social distancing is not achieved. That is why the idea of implementing a system for the recognition of social distancing through the use of Deep Learning arose. This research describes the state of the art of the current situation of COVID-19 worldwide, the solutions that have been implemented in the area of Artificial Intelligence to detect, predict and prevent COVID-19, and the solutions dedicated to the control of the social distancing. Then a social distancing system is created using an object people detector to track them and check if they are at a safe distance. The YOLO version 3 object detection model previously trained with the COCO image set was chosen, this model offers good performance and fairly accurate predictions. To eliminate weak detections, the nonmaximal suppression is applied to the prediction results and then the coordinates of the bounding boxes are extracted and the distance calculations for the centroids of each object found in the image are performed.