Estimating Thorax and Shoulder Motion Using Magnetic-Free Quaternion-Based Functional Sensor-To-Segment Calibration.

Wearable rehabilitation robots rely on accurate sensing of body motion. While 9-axis inertial sensors are commonly used to measure motion, signal interpretation can be challenging for upper-limb rehabilitation due to complex and unpredictable joint movements. Other studies have addressed these inaccuracies by attaching three sensor units and using magnetometer data. However, these sensors are often used near actuators that produce ferromagnetic disturbances. Therefore, the objective of this work is to develop a methodology for estimating thorax and shoulder motion using only the orientation data from two sensors, estimated by the internal fusion of accelerometer and gyroscope data. The proposed methodology involves a functional sensor-to-segment calibration that includes performing Principal Component Analysis on the data obtained during specific functional movements to identify the primary axes of rotation. The calibration aligns the sensor coordinate system with the anatomical reference frame of the body segment. Furthermore, the methodology estimates the rotation between the global coordinate systems of the sensor units. Through an experimental evaluation and a comparison with a reference sensor system, the tracking error was $4.07^{\circ}-5.14^{\circ}$ for shoulder orientation and $1.28^{\circ}-3.88^{\circ}$ for thorax orientation. The results provide a solution for tracking thorax and shoulder motion, without using the magnetometer, supporting the development of upper-limb rehabilitation robots.