Gait-cycle segmentation method based on lower-trunk acceleration signals and dynamic time warping.

Gait analysis is the systematic study of human walking. The analysis of gait signals from the lower trunk, acquired through accelerometers, begins with the proper identification of gait cycles. The goal of this work is to supplement gait-event based segmentation methods, tested for unimpaired and impaired populations, so that their need to calibrate or rely on pre-defined thresholds is overcome, and to implement strategies that reduce step-detection errors. A new system for the automatic extraction and analysis of gait cycles from acceleration signals of the lower trunk, combining knowledge from previous strategies with a dynamic time warping function, is presented. Performance was tested on gait signals from public databases. Sensitivities in step detection above 99.95% were achieved, with a positive predictive value of 100.00%. Step-correction strategies reduced the number of incorrect detections from 57 to 3 of 7056 steps. Bland-Altman plots and equivalence tests performed on cycle times by the proposed method and selected references showed good agreement, with mean differences below 0.003 s, and percent errors of 2%. This method may give place to a research tool for the automatic analysis of signals from subjects in a variety of cases.