The assessment of inertial odometry system performance in tracking upper limb kinematics

Human movement tracking is one of the most important topics in experimental biomechanics and robotics. In the last decades, visual-inertial odometry (VIO) systems achieved greater popularity for such applications even if a full characterization of their performance is still missing. In this context, this paper aims at assessing the performance of a VIO system for the measurement of upper limb kinematics by comparing them with optoelectronic (OS) and an inertial-based (IMU) systems. With this aim, a healthy young woman was asked to perform ten repetitions of (i) shoulder flex/extension; (ii) elbow flex/extension; and (iii) reaching tasks. All the movements were repeated three times. The experimental setup consisted in three VIO systems and three IMUs, placed on trunk, arm and forearm, along with eight optoelectronic cameras. The performance of the VIO system was computed both considering the overall setup and a reduced setup composed of only the trunk and the forearm system. The root mean square error of relative angles was used as a performance indicator. The accuracy in vector positioning estimation was also evaluated. VIO systems showed better performance in comparison to the IMU in terms of RMSE, with mean value equal to 2.1 degrees and 1.2 degrees for the elbow and shoulder joint, respectively. A substantial increment of RMSE was found when considering the reduced setup. The body displacement average error ranged from 4.7 to 6.2 cm when comparing OS and VIO. The low values of RMSE and body displacement tracking error encourage for further investigations on VIO system and the measurement of human kinematics.