Simultaneous global inverse kinematics and geometric parameter identification of human skeletal model from motion capture data

• Published in: Mechanism and machine theory Vol. 74; pp. 274 - 284
• Main Authors: Ayusawa, Ko, Ikegami, Yosuke, Nakamura, Yoshihiko
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.04.2014
• Subjects: Computation; Geometric parameter identification; Human; Inverse kinematics; Mathematical models; Motion capture; Motion perception; Nonlinearity; Optimization; Parameter identification; Recursive; Geometric parameter identification; Motion capture; Inverse kinematics; Optimization
• ISSN: 0094-114X; 1873-3999
• DOI: 10.1016/j.mechmachtheory.2013.12.015

This paper presents a method to identify the geometric parameters of a human skeletal model from motion capture data. For the identification, the joint trajectories have to be computed from the data at the same time. The method solves both the inverse kinematics and the geometric parameter identification simultaneously. In the method, the parameters are modeled as the generalized coordinates of virtual mechanical joints. It solves the large-scale inverse kinematics to compute the generalized coordinates at all given time-series instances and the time-invariant virtual coordinates. Though the method is based on the nonlinear programing, the computation is accelerated by decomposed gradient computation based on recursive Newton–Euler formulation. The method was tested on a 34-DOF human skeletal model using a motion capture system, and the results of the geometric parameters and the time-series coordinates are to be shown. The method was also applied to obtain a subject-specific musculoskeletal model. •An identification method of the geometric parameters of a human model is presented.•The whole parameters can be identified from motion capture data.•The method is computationally fast by using a recent inverse kinematics technique.•The geometric parameters are modeled as virtual generalized coordinates.•It solves the generalized coordinates of all time instances and the virtual ones.