Many contagious and non-contagious diseases have affected humanity in recent centuries. Fever is one of the most common symptoms of these diseases. Reliable devices have been used to identify people's fever. However, such devices are not used efficiently in health barriers and put the lives of other people at risk when cases of contagious diseases. Furthermore, they require an operator and cannot identify multiple febrile people simultaneously, resulting in a long waiting line. Finally, such devices do not calculate the probability of fever (show the punctual temperature only), nor consider the passerby's skin condition. This paper describes a real-time computational method that identifies multiple febrile individuals and their respective regions of interest from a thermographic frame using convolutional neural networks; then calculates their temperature and estimates the skin emissivity according to their condition (dry, sweat, or with lotion), recalculates the temperature and display the probability of fever for each individual identified at the sanitary barrier. This method can be embedded in a graphic processing unit, enabling a robust, easy-to-operate, and more efficient screening system for febrile people than those currently in use.