Rollover Detection of Infants Using Posture Estimation Model

In predicting the posture of an infant during sleep, images taken by a surveillance camera are useful. The posture estimation of infants is designed to be used in a home environment and uses monocular camera images rather than special cameras such as distance cameras. In this paper, we compare the accuracy of infant posture estimation by two posture estimation models, OpenPose and Cascaded Pyramid Network(CPN). We also introduce a system for estimating infant’s sleep turn using the posture estimation results. The experimental results show that CPN can detect infants’ posture estimation with higher accuracy than OpenPose. In addition, the system is successfully used to detect the infant’s turning with high accuracy.