The identification of multiple U-turns in gait: Comparison of four trunk IMU-based methods

The identification of turns during walking allows for the segmentation into straight and turn walking bouts. Several IMU-based methods were developed to this purpose, however many of them were tested on specific subject population. In this study, we tested four methods for the identification of turns in walking tasks with multiple U-turns that did not exploit any a-priori knowledge of the turn occurrences. We evaluated their robustness by recording IMU data on healthy and pathological subjects (healthy elderly, stroke survivors, patients with Parkinson disease and choreic patients) walking at two different speeds along a closed loop formed by straight bouts and U-turns. Overall, all methods identified correctly the totality of the U-turns when elderly and Parkinsonian patients were analyzed. When stroke survivors and choreic patients were analyzed, U-turns were either missed or erroneously detected in a limited number of cases. The only method using the magnetometer signals was the best performing, highlighting the usefulness of the magnetometer when turns are being investigated.