착용형 관성센서 기반 역동역학에서 분절-관절 벡터의 시변화가 관절 토크의 정확도에 미치는 영향

Abstract

Estimation of the joint kinetics such as joint torques based on wearable inertial sensors has received a lot of attention in biomechanics. The joint torques can be calculated via inverse dynamics when segment and joint kinematics as well as external force such as ground reaction force are available. While most of the kinematic information needed for the inverse dynamics can be provided using inertial sensor signals in real time, segment-to-joint (S2J) vectors are usually predetermined via a calibration procedure in which an S2J vector is a vector from one point on a segment to a joint center observed in a segment-attached coordinate system. Conventionally, S2J vectors are treated as constants as the human bodies are modeled as multi-link systems. However, it is obvious that they can be variated due to the effects of soft tissue artifacts. In this regard, constant S2J vectors may lead to inaccuracies of not only the position estimation but also joint torque estimation. This paper investigates the effects of applying time-variation of S2J vectors on estimation accuracy of joint torques in inertial sensor-based inverse dynamics. A method of determining the time-varying S2J vectors based on a learning procedure was applied to inverse dynamics. Experimental results showed that, for example, time-variation of S2J vectors improved the estimation accuracy of hip joint torque by more than 68.41%.