Vision-Based Gait Analysis for Neurodegenerative Disorders Detection

Abstract

Parkinson’s Disease (PD) is a debilitating neurodegenerative disorder that affects a significant portion of aging population. Early detection of PD symptoms is crucial to prevent the progression of the disease. Research has revealed that gait attributes can provide valuable insights into PD symptoms. The gait acquisition techniques used in current research can be broadly divided into two categories: vision-based and sensor-based. The markerless vision-based classification model has become a prominent research trend due to its simplicity, low cost and patient comfort. In this study, we propose a novel markerless vision-based approach to obtain gait features from participants' gait videos. A dataset containing gait videos from normal subjects and PD patients were collected, along with a control group of 25 healthy adults. The participants were requested to perform a Timed Up and Go (TUG) test, during which their walking sequences were recorded using two smartphones positioned at different angles, namely side and front. A multi-person pose estimator is used to estimate human skeletal joint points from the collected gait videos. Different gait features associated with PD including stride length, number of steps taken during turn, turning duration, speed and cadence are derived from these key point information to perform PD detection. Experimental results show that the proposed solution achieves an accuracy of 89.39%. The study's findings demonstrate the potential of markerless vision-based gait acquisition techniques for early detection of PD symptoms.