Machine Understanding of Mother-Infant Interaction: Using motion capture and eye tracking technology

Abstract

Parents’ interaction with their infant is crucial in the children's early cognitive development, but capturing and measuring these interactions is not a trivial task. Currently, a professional psychologist is required to watch the recording of parent and infant attentively to detect and label all interactions. While human is the one setting the baseline and standard for the encoding system, and hence the most reliable source of measurement, it is no doubt that relying on human is highly inefficient, and we shall aim at automating the pipeline using state-of-the-art technology. Therefore, in this thesis, an automatic workflow of capturing and analysing mother-infant interaction with the help of Machine Learning is demonstrated. To be specific, two focus modes of interaction in this thesis are the amount of mother's gaze on baby and the head-to-head distance between the pair.