Gait Analysis of Healthy Subjects in Real-world Environments through Video Input

Abstract

Gait, a person’s way of walking, can be negatively affected by neurological diseases such as Parkinson’s. When attempting to correct defects, it is useful for clinicians to create goals based on normal gait. However, data from the lab compared to data taken outside is hypothesized to create unrealistic targets due to the controlled nature of the lab. Most gait measurements are collected inside the lab because current methods of measuring gait are not convenient outside the lab. This research develops a versatile method of analyzing gait based on deep learning algorithms that require only video input. Using this method, the gait asymmetry in average populations both inside and outside the lab can be measured and compared, with the aim of establishing more relevant clinical goals. We found that in step length and stride time asymmetry, there was higher variance outside the lab compared to inside the lab. Stride time asymmetry also showed a higher average asymmetry outside the lab, though we did not see this in step length asymmetry. Finally, subjects tended to take larger steps inside the lab compared to outside.