Video-based facial discomfort analysis for infants

摘要: ABSTRACT Prematurely born infants receive special care in the Neonatal Intensive Care Unit (NICU), where various physiological parameters, such as heart rate, oxygen saturation and temperature are continuously monitored. However, there is no system for monitoring interpreting their facial expressions, most prominent discomfort indicator. In this paper, we present an experimental video automatic detection infants’ faces based on analysis of expressions. The proposed uses Active Appearance Model (AAM) to robustly track both global motion newborn’s face, well its inner features. detects by employing AAM representations face a frame-by-frame basis, using Support Vector Machine (SVM) classifier. Three contributions increase performance system. First, extract several histogram-based texture descriptors improve appearance representations. Second, fuse outputs individual SVM classifiers, which trained features with comp lementary qualities. Third, temporal behavior stability applying averaging filter classification outputs. Additionally, higher robustness, explore effect different image pre-processing algorithms correcting illumination conditions enhancement evaluate possible improvements. evaluated 15 videos 8 infants, yielding 0.98 AUC performance. As bonus, offers infant’s expressions when it left unattended additionally provides objective judgment discomfort. Keywords: analysis, detection, pain, infa nt