PoseScape: Pose-based Analysis System for Long-term Observation Studies

Abstract

Designers employ various design methods to observe people in their daily lives, to uncover rich information about their behaviors. However, manually analyzing and annotating long-term data such as data from video can be time-consuming. In this paper, we propose PoseScape, an automated long-term posture analysis system which uses markerless motion capturing, to help designers understand the patterns of the poses in use and to conduct ergonomic assessments. The system clusters postures using K-means clustering to reveal their varieties and visualizes the frequency, the duration, and the transitions of postures to help designers better understand human behaviors. For an early evaluation, we applied our tool to analyze sitting patterns from an in-the-wild perspective. We collected 3D postures of people working with a tablet PC on a sofa for two hours and compared analysis results from both PoseScape and design researchers. We identified a near resemblance between those two results, though with a subtle difference. We discuss the results and the future implications of pose recognition systems on the design of everyday things.