Validation of a Model-based Motion Reconstruction Method Developed in the REALMAN Project 2005-01-2743
Realistic motion reconstruction is the first step for ‘human like’ motion simulation by a digital human. In a recent European research project REALMAN (IST 2000-29357), a model-based motion reconstruction method from external marker trajectories was developed. It consists of two steps. The first one is to define a digital twin of a real subject using the technique of superimposing a digital human model upon at least two photos of different view, and identifying marker positions on it. In a second step, joint angles are estimated by using a kinematic model of the human body which is described using natural coordinates: coordinates of points and components of unit vectors for defining the joint locations. The model includes a detailed description of the torso, arms and legs, with simplified hands and feet. A total of 26 joints are used, connecting 27 rigid links, among them 6 located on the spine. The joint angles are calculated by minimizing the distance between markers located on the model and the experimentally measured marker positions, requiring the solution of a nonlinear constrained optimization problem arising from kinematic and driver constraints.
The purpose of this paper is to present a validation study of this motion reconstruction method. Its advantages and limitations will also be discussed.
Validation was made on two types of seated arm reaching movements carried out at INRETS: seat belt reaches and generic reaches. In total, 44 individual digital twins were defined. The comparison with measured anthropometric dimensions has shown that the accuracy of the digital twins depends on static posture used. Good results with an error of less than 30 mm in average were obtained using a standing ‘pharaon’ posture. However, a big difference was observed for the sitting height from the sitting posture. Its average difference reached to 44 mm. Concerning the validation of motion reconstruction, more than 90% of captured motions were successfully reconstructed for both seat belt and generic arm reach movements.