Engineering an Open-source Motion Capture Platform to Characterize Abnormal Gaits for Resource-limited Environments

Abstract

Gait abnormalities, a leading cause of death for older adults, are on the rise with the increasing global median age. While existing technologies can detect and quantify these abnormalities, their limited use in low-resource settings due to high costs and limited interoperability poses a severe challenge. This thesis overcame that barrier by introducing a novel approach: an open-source motion capture tool. The paper details creating a color-based marker detection system that utilizes a Kalman filter for tracking and the methods for achieving 3D marker coordination using a stereovision setup. The paper determined the necessary rotation matrices to obtain physiologically relevant angles utilizing a series of markers placed at anatomically relevant reference points. The motion capture system was validated by characterizing and classifying four gaits at three different speeds across two participants. The paper employed the motion capture system to assess the key features of normal, crouch, spastic, and steppage gaits in both the time and frequency domains. The paper performed spline interpolation, principal component analysis (PCA), linear discriminant analysis (LDA), logistic regression, support vector machine (SVM), and k-nearest neighbor (KNN). The paper found that nonlinear statistical methods slightly outperformed linear ones when classifying gait. The paper showed that using the frequency domain can mitigate the impact of temporal noise during windowing and provide a more effective approach to classifying speed. The paper demonstrated that individually trained models have little generalizability, but increasing the population over which model training occurs can mitigate this concern.