EFFICIENT TIBIOFEMORAL JOINT FORCE ESTIMATION USING SIMPLIFIED MOTION CAPTURE SYSTEM AND MACHINE LEARNING

A machine learning approach is employed to reduce the number of inertial measurement units (IMUs) required to estimate the force acting on the tibiofemoral joint during straight-line walking motion. Gait analysis measurements are first collected over 50 gait cycles using a motion capture system consisting of seven IMUs placed on the torso, left thigh, right thigh, left shank, right shank, left foot, and right foot. Machine learning is then applied to 5% of this data to build a kinematic model to estimate the relationships between these body segments during the different stages of the gait cycle. Having constructed this model, the force acting on the tibiofemoral joint is estimated using a force equilibrium equation. A simplified motion capture system is constructed consisting of just four IMUs placed on the thighs and shanks. The data acquired from these sensors during gait motion are input to the kinematic model, which predicts the motions of the torso and feet and subsequently, the tibiofemoral joint force. The correlation coefficient between the measured and estimated angular acceleration data reaches as much as 0.9724. Moreover, the estimated tibiofemoral force is in good agreement with that calculated using the full data capture system with seven IMUs. Overall, the results show that the proposed system achieves an accuracy comparable to that of traditional gait analysis methods while significantly reducing the number of IMUs. It thus minimizes the impact of the motion capture system on gait motion and enhances the practicality of gait analysis for real-world applications.