Infant Posture Assessment Based on Rotational Keypoint Detection

• Published in: 2021 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) pp. 1546 - 1550
• Main Authors: Zhao, Xuyang, Takata, Shogo, Fukumori, Kosuke, Tanaka, Toshihisa
• Format: Conference Proceeding
• Language: English
• Published: APSIPA 14.12.2021
• Subjects: Adaptation models; Computational modeling; Data models; Estimation; Feeds; Information processing; Visualization

The risk of an infant's developing neuromotor im-pairment is primarily diagnosis through visual examination by clinical experts. Due to the visual diagnosis is a very professional process, many infants at risk are not detected in time and develop into more severe conditions. There are some posture assessment diagnosis methods based on image/video data, but the performance will be affected by the infant's complex and diverse posture. In this paper, we propose a simple but effective method to improve infant posture assessment performance. Before the posture assessment, we first judge the direction of the infant's body in the image and calculate the inclination angle of the direction, then adjust the angle of the image to correct the infant's body orientation to upright. Last, the adjusted image will be feed into a model for posture assessment. Experimental results demonstrate that our proposed angle adjustment method can further improve the performance of the posture assessment for infants.