Automatic recognition of gait phases from accelerations of leg segments

The development of robust control for restoration of gait, as well as the assessment of the achieved gait, was the main motivation for starting this project. Recent neurophysiology findings suggest that mimicking temporal characteristics of the biological control could lead to effective control of the assistive system used in rehabilitation of the walking. This ultimately leads to the idea of using the finite-state model of gait. Today, low cost, easy to mount, micromachined accelerometers are available; hence, we decided to study the applicability of these sensors for creating of gait model. The detailed analysis of the data recorded with many sensors resulted in the conclusion that four accelerometers mounted on the foot, shank and thigh provide a detailed temporal description of the gait. The automatic recognition of phases was done by introducing the coding of three discrete values: −1, 0 and 1. The automatic coding was based on the thresholds (low values around zero). The verification of the finite state model was done by comparing the timings of gait events obtained from force sensing resistors mounted in the shoes, and the timings obtained from accelerations. The measure of the success was the correlation between the gait phases determined: the correlation was greater than 90%, which is within the range of variability of the gait events in normal walking.