Wearable inertial sensor based parametric calibration of lower-limb kinematics

Abstract

In this study, the calibrations of the lower-limb kinematics based on the five wearable IMU sensors were performed to reduce the inherent kinematic errors due to the both incorrect measurements of the kinematic parameters and the uncertainties of the geometric modeling of the lower-limb joints. Unlike the actual body structure, the ankle and knee joint of the lower-limb part are generally modeled as pure revolute joints of two or three degrees of freedom with some assumptions. However, it is obvious that these mismatches in the simplified geometrical modeling of the joint in the lower-limb kinematics lead to uncertainties in the model-based 3D motion tracking of human,body such as walking velocity"estimation, gait analysis, etc. As main research contributions, we intended to examine exactly which error factors are responsible for the kinematic uncertainties that arises from the 3D motion tracking based on the lower limb kinematics and wearable IMU sensors. The lower-limb kinematics model in this research assumes that the ankle, knee, and hip joints are all three degrees of freedom, and that the three principle axes at each joint may not intersect at a common point. That is, the kinematic uncertainties included in the joint model will be confirmed by performing a kinematic calibration taking into account the orientation of joint axes and joint offsets as kinematic error parameters as well as the case of only calibrating the link lengths. This study also uses a new calibration plate with seven sole-shaped jigs and five wearable IMU sensors to measure the actual pose of the right foot, which is the distal end of the lower-limb kinematics with respect to the left foot. In order to secure an in-depth insight into the wearable IMU sensor based calibrations of the lower-limb kinematics, the seven sole-shaped jigs were designed to independently allow the right foot on those jigs to have predefined 3-dof poses. Finally, an iterative least square method is adopted to minimize the error between the nominal kinematic model-based pose and the actual poses of the right foot with respect to the left foot on the calibration plate. In an experiment with one subject, the reduction rate of the position error depends on the measurement of the initial kinematic parameters, but the RMS position error of the right foot relative to the left foot was improved by 336.1% from 36.5 mm to 11.3 mm. (C) 2017 Elsevier B.V. All rights reserved.