Date: June 2021.
Source: ASME Journal of Computing and Information Science in Engineering (JCISE), 21(3): 031006. https://doi.org/10.1115/1.4049445.
Research Summary: A high-fidelity digital representation of the human body is a key enabler for integrating humans into a digital twin, and the hand is a particularly challenging part of the body to model due to posture deviations across collected scans. This paper proposes a posture-invariant statistical shape model (SSM) of the human hand, built from 59 real-world 3D hand scans, the majority of which were captured on a dynamic-3D/4D 3dMDhand system, which can record at 7fps, but used here for static scanning, with the balance captured on a prototype close-range photogrammetry hand scanner using a Raspberry Pi, Python, and Agisoft architecture. The scans are first spatially aligned using a Möbius sphere-based algorithm, and an articulated skeleton, 20 bone segments and 16 joints, is embedded in each scan, allowing all scans to be aligned to a common posture using linear blend skinning (LBS). Three dimensionality reduction methods, Principal Component Analysis (PCA), kernel-PCA, and Independent Component Analysis (ICA), are evaluated for constructing the SSM based on compactness, generalization ability, and specificity, with the PCA-based model ultimately selected. Leave-one-out validation shows the resulting model fits a given real-world 3D hand scan to an accuracy of 1.21 ± 0.14 mm, and the posture-corrected SSM meaningfully outperforms an equivalent model built without posture correction, supporting its use in applications such as human-integrated digital twins, virtual reality, and personalized product design.
Article: Posture-Invariant 3D Human Hand Statistical Shape Model.
Authors: Yusheng Yang, Tianyun Yuan, Toon Huysmans, Willemijn S Elkhuizen, Farzam Tajdari, Yu Song. Faculty of Industrial Design Engineering, Delft University of Technology, Delft, South Holland, 2628CE, The Netherlands.