A Comparative Study of Detecting Violation of Face Mask and Social Distancing using Different Deep Learning Models

The growing pandemic caused by COVID-19 has been making headlines recently, causing a global disaster with an unavoidable spread. The WHO recommends wearing a face mask and maintaining physical distance. This research can assist in keeping track of social distance and it can also spot face masks. The entire system is set up in public settings. Images and videos are used by this system to identify persons. Frame by frame processing is done on the video. People are recognized by image processing with TensorFlow and OpenCV. MobileNetV2 and ResNet50 are used for detecting the face mask violation. It first recognizes the crowd’s presence before calculating the centroid distances between individuals using an object tracking algorithm and a distance algorithm. In order to assess the social distance between people, the Euclidean distance is typically employed to calculate how far off each pair of centroids of the bounding box is identified from one another. At a distance of six feet, the program can determine the number of pixels in an image. The detection models MobileNetV2 and ResNet50, respectively, obtained accuracy of 97% and 98% with the large dataset, which contains 12000 samples with and without masked faces.