Posture estimation for motorcycleriders using IMU-based systems

Sammanfattning

This thesis investigated whether a wearable real-time pipeline based on inertialmeasurement units could estimate the upper-body posture of motorcycle riderswith sufficient accuracy, low latency, and portability for practical use. A prototy-pe was developed with three sensors placed on the motorcycle, the lower back,and the upper back, and implemented on a Raspberry Pi. The system synchro-nized sensor readings through a multiplexer, fused accelerometer and gyroscopedata with a Madgwick filter, transformed the resulting orientations to the motor-cycles reference frame, and computed both lower-back inclination and relativetrunk motion. The prototype was evaluated through latency measurements, sta-tic orientation tests, yaw-drift experiments, and comparison with a camera-basedreference for one rotation component. The results showed a stable update rateof about 50 Hz and a mean local processing latency of 17.94 ms, well below the100 ms requirement. For the validated component, the root-mean-square errorwas 2.03° for lower-back inclination and 7.35° for relative segment angle. Initialcalibration combined with continuous drift compensation also reduced yaw driftsubstantially compared with an uncalibrated signal. The study concludes that theproposed pipeline is a feasible and portable prototype for real-time posture esti-mation, but that full three-dimensional accuracy remains insufficiently validatedand that long-term yaw estimation is limited by the absence of a magnetometer.