Visual-Inertial Sensor Fusion and OpenSim Based Body Pose Estimation.

Body pose estimation is crucial for many human motion-related applications. Stable and low-cost methods for body pose estimation are highly desirable for daily life usage. In this paper, we propose a sensor fusion method for upper body pose estimation. A monocular camera and several IMU-marker sensor modules are integrated to achieve a visual-inertial sensor system. ArUco markers are attached to the IMUs to remove the usage of magnetometers. The raw IMU data are firstly corrected with pre-calibrated intrinsic parameters and then fused with the marker poses detected from the images via an extended Kalman filter. The driftless sensor orientation and marker trajectory are then imported into the OpenSim software to compute body states using the inverse kinematics approach. Experiments were conducted on human subjects to validate the effectiveness of the method. Movements used in the Fugl-Meyer assessment procedure are adopted during the experiments. The estimation results are compared with the joint states from the optical motion capture system showing good accuracy. The proposed method shows the potential to be applied in clinical assessment to reduce the efforts of professional physicians.