University life is closely related to students’ emotional health; scientific evidence shows that it significantly influences academic performance and personal development. Tracking classic emotions such as joy, disgust, anger, fear and sadness can trigger anxiety, stress and depression, negatively affecting concentration, memory and motivation, essential elements for academic success. This paper presents an alternative for the detection of stress using low-resolution thermal images obtained with a FLIR-Lepton®  Thermal Camera. The methodology is based on the identification of thermal facial regions of interest (ROIs), combining the Histogram of Oriented Gradients (HOG) and Filter Based-Kalman algorithms to adjust the temperature of the thermal images, together with a Support Vector Machine classifier. In addition, a process based on stress indices and thresholds is used to associate high, medium and low stress levels from anomalies detected in the thermal ROIs. The diagnostic model was tested on fifty subjects, each subjected to the five basic emotions, generating a total of 500 thermal images and a classic classroom scene (5-min interview). The results show an accuracy of over 80% in the classification of emotions and 86.6% in the recognition of stress-related emotional disturbances during a university day.