Automomous validation of proper mask wearing

A method for autonomous validation of a safety rule compliant wearing of a mask, the method comprises:

• detecting, by a processing circuit and within sensed information, at least a part of a face of a person;
• determining that a coverage of a mask worn by the person is proper when finding that the a region of the sensed information that corresponds to the face of the person (a) does not comprises a nose and a mouth of the person, and (b) comprises at least one other facial landmark that differs from the nose and the mouth of the person;
• wherein the determining is executed by extracting localized facial recognition features by (a) using a convolutional neural network (CNN) pre-trained for learning the facial recognition features to extract a 3D grid of activations that are obtained as output of one of the convolutional layers as an activations map, (b) obtaining such activations maps for multiple faces, in which occlusion augmentations are imposed on the faces on random locations, and (c) training a classifier for each grid point of the 3D grid that predicts whether the grid point of the 3D grid is occluded or not;
• estimating, based on sensed information representing the mask, whether the mask material is adapted to prevent a progress of saliva droplets;
• and
• deciding, that the person properly wears the mask in a safety rule compliant manner when the coverage of the mask is proper, and the mask is adapted to prevent progress of the saliva droplets;
• and
• responding to the deciding by populating a database and transmitting an alert.

A non-transitory computer readable medium for autonomous validation of a safety rule compliant wearing of a mask, the non-transitory computer readable medium that stores instructions for:

• searching, by a processing circuit and in sensed information, for maskable facial landmarks;
• determining that a the mask has a proper coverage when finding that all the maskable facial landmarks are masked;
• wherein the determining is executed by extracting localized facial recognition features by (a) using a convolutional neural network (CNN) pre-trained for learning the facial recognition features to extract a 3D grid of activations that are obtained as output of one of the convolutional layers as an activations map, (b) obtaining such activations maps for multiple faces, in which occlusion augmentations are imposed on the faces on random locations, and (c) training a classifier for each grid point of the 3D grid that predicts whether the grid point of the 3D grid is occluded or not;
• estimating, based on sensed information representing the mask, whether the mask, when the mask has a the proper coverage, is adapted to prevent a progress of saliva droplets;
• deciding, that the person properly wears the mask in a safety rule compliant manner when the mask has a the proper coverage and is adapted to prevent progress of the saliva droplets;
• and
• responding to the deciding by populating a database, transmitting an alert.