Integration of IMU-based Motion Tracking Algorithms into Wearable Devices for Human Joint Angle Estimation

Motion capture technologies generate real-time data that dynamically represents the position and orientation of a human body in three-dimensional (3D) space. In the clinical medicine and rehabilitation fields, the electromyography (EMG) is commonly used for interpreting patients’ muscle conditions but does not give information about the objective performance of movement execution. This technique can be combined with motion capture technologies to track patient improvement and guide therapy. Since the 1980s, many technologies have been tested to track human motion. Visual marker-based tracking systems are considered the gold standard in this field. However, they are expensive and restrict the analysis to a laboratory setting. Inertial measurement units (IMU) are now at the center of the research to overcome these problems. They are light, affordable, and wearable devices that combine accelerometers and gyroscopes. In order to estimate the orientation of an IMU, inertial data can be merged through sensor fusion algorithms. These algorithms integrate gyroscope data and correct the value obtained by observing accelerometer data. The resulting IMU orientation can be expressed as quaternions or Euler’s angles. This thesis project aims to integrate an IMU-based motion tracking system into the embedded device designed by Rossi et al. for EMG acquisition. Beyond this analog front-end for bio-signal acquisition, this board contains an IMU module (LSM6DSO32) and a microcontroller (AmbiqMicro Apollo3 Blue). Firstly, the Serial Peripheral Interface (SPI) protocol has been implemented to enable communication between the IMU and the microcontroller, while the Universal Asynchronous Receiver-Transmitter (UART) communication protocol has been used to exchange data and commands between the user interface and the microcontroller. Then, an algorithm to calibrate the IMU has been implemented in the firmware to improve the accuracy of the sensors. A Graphical User Interface (GUI) has been implemented in MATLAB programming language to allow the user to control the system and visualize output data. A validation protocol has been performed by comparing the angles obtained from the relative position of two IMUs with those measured by a modular absolute encoder. Three different Sensor fusion algorithms have been tested and compared in terms of execution time, hardware memory usage, and errors in angle estimation. The comparison of angles obtained from each algorithm has been performed for seven subjects at three different movement velocities and two different starting positions. Under best conditions, we obtained errors of: 9,11°±3,78° for the Madgwick algorithm, 8,77°±3,81° for the complementary filter, and 16,00°±2,78° for the Extended Kalman filter, results that are in agreement with what is reported in the literature.