Gait Analysis

Abstract

The analysis of dynamic posture and lower limb movement patterns is a critical technique used in medical fields such as diagnostics, rehabilitation, orthopedics, and neuroscience, as well as in sports science and forensics. Precision medicine is enhanced by the accurate and reliable identification of gait patterns and characteristics in clinical settings. Continuous monitoring and evaluation of these patterns enable more effective, individualized treatments and support predictive outcome assessments. Many physical and neurological diseases either cause abnormal gait patterns or manifest through gait abnormalities.Observational gait analysis remains widely used by clinicians because of its simplicity, accessibility, and low cost. However, the validity, reliability, specificity, and responsiveness of these qualitative methods are subjects of ongoing debate and scrutiny. This thesis presents a decision support system for clinicians, offering a comprehensive approach to gait analysis through the use of easily accessible RGB camera videos and computer vision models for pose estimation. By leveraging advanced algorithms to extract joint landmarks from video sequences of subjects walking, this system facilitates the identification of key gait events and the calculation of critical gait parameters. This innovative method provides clinicians with valuable insights into gait patterns, enhancing diagnostic accuracy and treatment planning. The initial phase involved setting up a camera system to capture RGB videos of subjects walking. These videos were processed using the Mediapipe pose estimation library to extract joint landmarks, providing a comprehensive dataset of the subjects’ movement patterns. To ensure the reliability of the data, we applied error reduction techniques, including filters and outlier removal methods, resulting in smooth and gap-filled landmark trajectories. The analysis was conducted using the GPJTAK gait dataset, which offered a diverse range of gait patterns for comprehensive evaluation. Subsequent analysis identified key gait events, such as toe-off and heel-strike frames. From these events, we calculated various gait parameters critical for diagnosing gait abnormalities. These parameters included step length, stance time, swing time, double support time, step distance, gait speed, knee flexion angles, and ankle and hip angles. We developed a cloud-based web application for physicians that provides decision support. This application allows users to upload videos and returns a comprehensive gait analysis report, facilitating more informed clinical decisions