Infant Facial Expression Analysis Towards Defecation Detection

Infants need constant care from parents or caregivers, and cannot express themselves verbally. Therefore, automatic infant facial expression recognition system plays an important role in reducing the care burden. As a non-negligible part of baby care, automatic defecation detection based on infant facial expressions has not been investigated yet. This detection is especially important for young parents with little caring experience. In this paper, we built an Infant Expression Database of 465 images from 419 infants, including four categories: happiness, pain, defecation and sleep. Based on this database, we proposed a shallow network structure containing only three convolutional layers for infant facial expression recognition, which achieved an average accuracy of 84.93% for these four categories. Moreover, the recognition results showed that the network has a strong generalization ability, with 81.82% detection accuracy of infant defecation.