Tracking the Body Center of Mass during simulated daily activities: a sensor fusion approach with barometric and inertial data

Monitoring the body's center of mass (BCoM) during real-world activities is vital in biomechanics and rehabilitation. It helps identify musculoskeletal disorders, assess energy use, and evaluate balance and mobility. An inertial measurement unit (IMU) can measure vertical BCoM movement by integrating accelerometer data, but this method can be inaccurate over long periods due to accelerometer biases and initial assumptions. To address these issues, this thesis integrates a barometer, traditionally used as an altimeter in avionics, into a wearable multi-sensor system that includes an IMU. The barometer is incorporated at both the hardware and firmware levels, and its readings are fused with the IMU data using sensor fusion techniques. This innovative approach aims to develop and validate methods for accurately estimating vertical displacement in individuals. Twenty healthy subjects participated in two trials mimicking daily life in a laboratory setting. Politecnico di Torino's ethical committee approved the study (Prot. 27213/2024). Tasks included Sit to Stand (StS) at normal and slower speeds to simulate mobility issues, lying on a mattress, going up and down the stairs, and squatting. Subjects wore the wearable system on their lower back. Stereophotogrammetry (SP) was used as a reference. BCoM displacement reconstruction involved two steps. First, transitions were identified through the first derivative of the barometric signal to locate the start and end of StS, lying, stairs, and squatting. Second, height changes during transitions were estimated using four methods: barometer alone (BAR), accelerometer alone (ACC), and combined barometer and accelerometer with a linear (LKF) or extended Kalman filter (EKF). Transitions are crucial as they mark points where vertical velocity is supposed to be null, facilitating the integration of accelerometer data. Height is considered constant between transitions, simplifying the process. Transitions detection was evaluated using the F1-score, sensitivity, and measuring delay relative to SP for correctly identified transitions. Height reconstruction was assessed via Root Mean Square Error (RMSE) over the entire trial and absolute deviation during transitions. An average F1-score of 90% was observed for transition detection, with a sensitivity of 99% for transitions associated with StS, slow StS, and 100% for lying. Transitions were detected with a mean delay of 12 ms. For height reconstruction, the RMSE values (average ± standard deviation) were (8±3) cm for BAR, (19±12) cm for LKF, and (33±19) cm for both ACC and EKF. To evaluate whether the errors were influenced by transition types, techniques, tests, and subjects, reconstructed height error distributions within each group were statistically compared. Results indicated significant differences between techniques and transition types. However, no significant differences were observed between tests and subjects for each method. The derivative method performed well on StS and lying but poorly on stairs and squats. BAR provided the most accurate reconstruction because transitions were identified with the barometric signal, making it essential to determine these integration intervals with the accelerometer to enhance sensor fusion accuracy. In conclusion, integrating the barometer and IMU shows potential for accurately measuring height changes, benefiting biomechanical analysis and energy consumption assessment, and enhancing the understanding and evaluation of human movement in real-world settings.