A mobile tool for static standing posture assessment using accurate 3D body reconstruction

Accurate posture assessment is essential for diagnosing and managing health issues related to postural disorders. Existing mobile applications rely on 2D imaging and analyzing reflective markers placed on anatomical landmarks of the human body without comprehensive view of body's posture. A mobile tool of 3D body reconstruction-based posture assessment is proposed in this paper. A smartphone is used to capture video circling around a static standing person in an approximate A-pose. Some specific multi-view body images using body orientation estimation are extracted from video. Then OpenPose and U2Net are employed to extract 2D joints and contours of each image. Camera poses are estimated using feature matching, and 3D joints are reconstructed from 2D joints and camera parameters. Multi-view projection contour consistency is used to iteratively optimize SMPL parameters for accurate 3D body reconstruction. From reconstructed 3D SMPL body, the required parameters for posture assessment including key 3D body points, joint angles, and spatial measurements, etc. can be easily obtained and calculated. Finally, assessment results are compared with artificial intelligence posture evaluation and correction system (APECS), and reliability is evaluated using Cohen's d and the Intraclass Correlation Coefficient (ICC). The proposed 3D body-based method achieved over 90% accuracy in identifying common postural abnormalities, such as uneven shoulders and forward head posture. The posture assessment results were in close agreement with existing method. 3D body reconstruction is demonstrated to be an effective method for posture assessment. Smartphone video-based posture assessment provides a user-friendly, visualized, abnormal posture screening tool.