Speed Invariant Gait Event Identification using Dynamic Time Warping

Identification of gait events such as heel strike, toe off etc. is performed to derive significant gait features. To be effective in a real-world scenario, a gait event detection system is required to identify the events regardless of the change in a person’s gait pattern due to variation in speed. Analysis of robust gait event detection systems towards varying speeds is less explored in literature. A speed-invariant event detection method using Dynamic Time Warping (DTW) metric and an improvement of the DTW based method using Intrinsic Mode Function (IMF) features are proposed. The efficacy of the proposed methods is validated with MAREA accelerometer gait dataset. To examine speed invariability, each method is trained and tested with gait data of different speeds. The average mean absolute error is measured for predicting the events during experimental analysis. The DTW distances between same events are observed to be less than the distances between unlike events, irrespective of the speed. Experimental analysis on both treadmill and outdoor overground data suggests the invariability of the proposed methods with respect to speed. The events identified using only two accelerometers can be useful towards developing an economical clinical gait analysis system for health-care applications.