Assessing children's outdoor thermal comfort with facial expression recognition: An efficient approach using machine learning

Abstract

In this study, children's physiological indices and facial expressions during activities with distinct intensities in outdoor open spaces were real-timely measured. Correspondingly, meteorological measurements and questionnaire surveys were conducted to explore change laws of children's physiological feedback, facial expressions, and subjective perception. Then, a predictive model of children's outdoor thermal sensations was built with the coupling relations of physiological indices and facial expressions. Results showed that the skin temperature on the face tended to increase as thermal stress increased but gradually declined with increasing activity intensity. Mean ear skin temperature (Tear) was the most sensitive, followed by mean cheek temperature (Tcheek). When conducting moderate and vigorous intensity activities, the heart rate (HR) increased significantly within 5 min of exercise and then stabilized. The HR increase in vigorous intensity activities was greater than that during moderate intensity activities. With an increase in thermal stress and activity intensity, the proportion of negative emotions (NegEmo) in facial expressions gradually increased. Sadness and disgust expressions both increased significantly, followed by fear and anger. Thus, Tear and NegEmo expressed facially were important indices to predict children's outdoor thermal sensation. Finally, the predictive model for children's outdoor thermal sensation based on facial expressions, Tear, and HR allowed for non-invasive data collection was built integrating machine learning, and its accuracy reached as high as 97.1 %. Our results propose a faster feedback method for children's thermal perception based on facial expression recognition, which can real-timely monitor thermal comfort for children, ensuring their outdoor thermal health and safety.