Abstract
Alzheimer's disease (AD) is a progressive neurodegenerative disease impacting older adults' cognitive and functional abilities. Early detection in the mild cognitive impairment (MCI) stage is vital for timely interventions to slow down the disease progression to AD. This study introduces a novel MCI detection that emphasizes accessibility and ease of use, utilizing a regular camera and pose estimation. Using the OpenPose algorithm, we analyzed 25 body joints during walking and extracted 48 gait features, identifying 17 key features that significantly distinguish MCI from healthy controls (HC). Our approach, combining statistical analysis, signal processing, and a machine learning model using a support vector machine, achieved an accuracy and F-score of 86.81% and 82.35%, respectively. This confirms the effectiveness of everyday camera data and pose estimation in detecting significant gait differences between MCI and HC, offering an easy, cost-effective solution for early MCI detection and monitoring in non-clinical settings. It removes the barriers of sophisticated equipment and specialized expertise, paving the way for practical remote monitoring and AD early intervention.