Human pose estimation in library and information science: An exploratory case study of joint attention during virtual storytimes
TL;DR
Analysis of a child’s nonverbal behaviors during a virtual storytime hosted by a public library suggests that interactive components—such as singing, movement activities, and read-alouds—were associated with demonstrations of inferred joint attention behaviors.
Abstract
Human Pose Estimation (HPE), a computer vision technique that detects and tracks human body movements through skeletal keypoints, has been widely used in fields such as healthcare, sports, and education but remains largely unexplored in library and information science (LIS). This exploratory case study examines the use of HPE to analyze a child’s nonverbal behaviors—specifically inferred joint attention—during a virtual storytime (VST) hosted by a public library. A recorded live VST session was analyzed using the OpenPose framework and custom Python scripts to extract and quantify body keypoints related to head and eye positioning. The 1372-second session was segmented into 10 chronological storytime activities, including read-alouds, songs, movement-based activities, and a craft activity. Across the session, the child exhibited 1781 physical movements, with 81.8% corresponding with the librarian’s prompts or storytime activities, indicating coordinated or supported inferred joint attention. Results suggest that interactive components—such as singing, movement activities, and read-alouds—were associated with demonstrations of inferred joint attention behaviors. The findings highlight the potential for computational methods to advance research on early literacy programming and participant behavioral interaction in library environments.