Real-time estimation of lower limb joint angles through inverse kinematics during walking using a scaled OpenSim model

Abstract

Background: Real-time estimates of three-dimensional (3D) joint angles and musculotendon forces can potentially enable rapid patient evaluation and biofeedback for gait retraining. We have recently demonstrated real-time estimation of musculotendon forces using our electromyography (EMG)-driven neuromusculoskeletal models. However, this needs real-time estimation of 3D joint angles using inverse kinematics (IK) in OpenSim software, which involves many processing steps and other software. This preclude it is real-time use. Thus we aimed to (i) develop real-time OpenSim IK procedures, (ii) integrate this within our EMG-driven neuromusculoskeletal models, and (iii) compare the real-time estimation of 3D joint angles and musculotendon forces to those using offline processing. This will ensure real-time estimates are not compromised by marker drop and other specific operations that enable real-time processing.

Method: Custom software was written in C++ to read 3D markers trajectories from Vicon motion capture system in real-time. The OpenSim IK algorithm and software was modified to accept these markers trajectories on a frame-by-frame basis. 3D gait data was acquired using a full body marker set (68 markers) for a single subject walking on a treadmill. A static calibration trial was used to initially scale the OpenSim model to the subject's segmental dimensions. During the walking trials the custom software was used to produce and save the IK joint angles, while the marker trajectories were also saved for subsequent offline processing. The IK joint angles from hip, knee, and ankle determined using the real-time and offline pathways were then compared using a modified coefficient of multiple correlation (CMC), which assessed the similarity of waveforms where a value of 1 indicates maximum similarity.

Results: We found similar hip, knee and ankle joint angle waveforms, for three consecutive gait cycles, with CMC's of 1 for the hip and knee, and 0.998 for the ankle. A delay between the subject's motion and the real-time calculation of joint angles was approximately 30 ms. The real-time estimation of musculotendon forces is currently being evaluated.

Discussion: Real-time IK produced estimates of the joint angles the same as those determined using the offline processing. The small variations present were due to the filtering of trajectories in the offline method that should not affect musculotendon forces estimates. This is the first software to calculate, in real-time, 3D joint angles using the OpenSim IK from an individually scaled anatomical model. This enables real-time estimation muscle and joint contact forces in real-time using our current EMG-driven neuromusculoskeletal models.