Automatic Generation of a Subject Specific Upper Body Model From Motion Data

Abstract

This paper details an automated process to create a robotic model of a subject’s upper body using motion analysis data of a subject performing simple range of motion (RoM) tasks. The upper body model was created by calculating subject specific kinematics using functional joint center (FJC) methods, this makes the model highly accurate. The subjects’ kinematics were then used to find robotic parameters. This allowed the robotic model to be calculated directly from motion analysis data. The RoM tasks provide the joint motion necessary to ensure the accuracy of the FJC method. Model creation was tested using five healthy adult male subjects, with data collected using an eight camera Vicon© (Oxford, UK) motion analysis system. Common anthropometric measures were also taken manually for comparison to the FJC kinematic measures calculated from marker position data. The algorithms successfully generated models for each subject based on the recorded RoM task data. Analysis of the generated model parameters relative to the manual measures was performed to determine the correlations. Methods for replacing model parameters extracted from the motion analysis data with hand measurements are presented. The accuracy of the model generating algorithm was tested by reconstructing motion using the parameters and joint angles extracted from the RoM tasks data, correlated manual measurements, and height based correlations from literature data. Error was defined as the average difference between the recorded position and reconstructed positions and orientations of the hand. For all of the tested subjects the model generated using the RoM tasks data showed least average error over the tested trials. Each of the tested results were significantly different in position error with the FJC generated model being the most accurate, followed by the correlated measurement data, and finally the height based calculations. No difference was found between the end effector orientation of generated models. The models developed in this study will be used with additional subject tasks in order to better predict human motion.