IMUAngle: Joint Angle Estimation with Inertial Sensors in Daily Activities

Publication Type Conference Paper
Authors Lena Uhlenberg, Swathi Hassan Gangaraju, Oliver Amft
Title IMUAngle: Joint Angle Estimation with Inertial Sensors in Daily Activities
Abstract We present a framework for IMU-based joint angle estimation during activities of daily living (ADL). Personalised musculoskeletal models were created from anthropometric data. Three sensor fusion algorithms were optimised to estimate orientation from IMU data and used as input for the simulation framework. Four ADLs, involving upper and lower limbs were simulated. Joint kinematics of IMU-based simulations were compared to optical marker-based simulations. Results for IMU-based simulations showed median RMSE of 0.8−15.5° for lower limbs and 1.5−33.9 ° for upper limbs. Median RMSE were 4.4°, 5.8°, 6.9°, 6.5° for ankle plantarflexion, knee-, hip flexion, and hip rotation, respectively. For upper limbs, elbow flexion showed best median RMSE ∼3.7°, whereas elevation angles (∼24.5°) and shoulder rotation (∼12.5°) performed worst. Increased RMSE at upper limbs was attributed to the degrees of freedom at the shoulder region compared to the hip. Overall, transversal plane movements (rotations) showed higher median RMSE compared to sagittal plane movements (flexion/extension). Optimisation of orientation estimators improved performance considerably depending on ADL (up to ∼20 °). Comparing sensor fusion algorithms, Madgwick and Mahony produced comparable joint kinematics, whereas the Extended Kalman Filter performance showed larger variability depending on the ADL. Our approach offers a realistic representation of joint kinematics and can be supported by optimising parameters of sensor fusion algorithms.
Date September 2022
Proceedings Title ISWC '22: Proceedings of the 2022 International Symposium on Wearable Computers
Conference Name ISWC' 22
Place Cambridge, United Kingdom
Publisher ACM
DOI 10.1145/3544794.3558470
Full Text PDF
Friedrich-Alexander-Universität Erlangen-Nürnberg