Image analysis for detecting mask compliance

A computer implemented method for automatic training of a classifier for determining mask wearing compliance based on a thermal image, comprising:

• obtaining, for each of a plurality of subjects, a sequence of thermal images of the respective subject wearing a mask, obtained over a time interval;
• for each sequence of thermal images, analyzing the respective sequence of thermal images to identify a threshold thermal image of the sequence, where in thermal images prior to the threshold thermal image an indication of an estimated temperature of a region of a face of the respective subject is increasing, and in thermal images after the threshold thermal image the indication of the estimated temperature of the region of the face of the respective subject reached a maximal value and remained constant thereafter;
• labelling thermal images of the sequence prior to the threshold thermal image with a label denoting lack of mask wearing compliance, and labelling thermal images of the sequence after the threshold thermal image with a label denoting proper mask wearing compliance;
• creating a training dataset including, for each of the plurality of subjects, the sequence of thermal images with corresponding labels denoting lack of mask wearing compliance or proper mask wearing compliance;
• and
• training a classifier using the training dataset.