Recently, the estimation of joint kinetics such as joint force and moment using wearable inertial sensors has received great attention in biomechanics. Generally, the joint force and moment are calculated though inverse dynamics using segment kinematic data, ground reaction force, and moment. However, this approach has problems such as estimation error of kinematic data and soft tissue artifacts, which can lead to inaccuracy of joint forces and moments in inverse dynamics. This study aimed to apply a recurrent neural network (RNN) instead of inverse dynamics to joint force and moment estimation. The proposed RNN could receive signals from inertial sensors and force plate as input vector and output lower extremity joints forces and moments. As the proposed method does not depend on inverse dynamics, it is independent of the inaccuracy problem of the conventional method. Experimental results showed that the estimation performance of hip joint moment of the proposed RNN was improved by 66.4% compared to that of the inverse dynamics-based method.

Relative position estimation between body segments is one essential process for inertial sensor-based human motion analysis. Conventionally, the relative position was calculated through a constant segment to joint (S2J) vector and the orientation of the segment, assuming that the segment was rigid. However, the S2J vector is deformed by soft tissue artifact (STA) of the segment. In a previous study, in order to handle the above problem, Lee and Lee proposed the relative position estimation method using time-varying S2J vectors based on inertial sensor signals. Here, time-varying S2J vectors were determined through the joint flexion angle using regression. However, it was not appropriate to consider only the flexion angle as a deformation-related variable. In addition, regression has limitations in considering complex joint motion. This paper proposed artificial neural network models to compensate for the STA by considering all three-axis motion of the joint. A verification test was conducted for lower body segments. Experimental results showed that the proposed method was superior to the previous method. For pelvis-to-foot relative position estimation, averaged root mean squared error of the previous method was 17.38 mm, while that of the proposed method was 12.71 mm.

One of the problems in inverse dynamics calculation for the inertial measurement unit (IMU)-based joint force and torque estimation is the amplified signal noises of segment kinematic data mainly due to the differentiation procedure and segmental soft tissue artifacts. In order to deal with this problem, appropriate filtering methods are often recommended for signal enhancement. Conventionally, a low-pass filter (LPF) is widely used for the kinematic data. However, the zero-phase LPF requires post-processing, while the real-time LPF causes an unignorable time lag. For this reason, it is inappropriate to use the LPF for real-time joint torque estimation. This paper proposes a Kalman filter (KF) for inverse dynamics of IMUbased joint torque estimation in real time without any time lag, while utilizing the smoothing capability of the KF. Experimental results showed that the proposed KF outperformed a real-time LPF in the estimation accuracy of hip joint force and torque during jogging on the spot by 100 and 29%, respectively. Although the proposed KF requires the process of adjusting covariance according to the dynamic conditions, it can be expected to improve the estimation performance in the field where joint force and torque need to be estimated in real time.