Computer Vision for Quantitative Gait Analysis and Movement Assessment in Healthcare

Clinical gait analysis is essential for assessing human movement to support diagnosis, prognosis and rehabilitation monitoring. Access to quantitative gait assessment remains limited because traditional motion capture systems are expensive and complex, restricting availability for many patients and clinicians. Video-based movement analysis using computer vision (CV) and artificial intelligence (AI) offers a promising way to improve accessibility using everyday devices. However, current CV approaches lack precision in joint-angle measurement and normative reference standards, and depend heavily on large clinically annotated datasets that are scarce, difficult to obtain and constrained by privacy requirements. This thesis develops an accessible and clinically aligned video-based framework for gait analysis. It examines how deep learning and CV can support clinicians in detecting movement abnormalities while offering interpretable spatial and temporal information. Five interconnected studies establish the components required for such a system. First, two clinically annotated datasets, the Gait Abnormality Video Dataset (GAVD) and the Clinical Abnormality Simulated Dataset (CASD), were created to support robust training and evaluation. Second, normative gait parameters for ten major joints were constructed from pose estimation and clinical input, enabling quantitative detection of deviations beyond simple classification. Third, a synthetic data pipeline generated biomechanically realistic gait sequences to address data scarcity, privacy and demographic diversity. Fourth, an unsupervised framework reduced reliance on manual labelling while preserving clinically meaningful structure. Finally, these components were integrated into MoveMetrics, a proposed clinical gait analysis system informed by usability evidence from clinicians, researchers and patients. Through case-based demonstrations, MoveMetrics illustrates potential applications in telehealth, patient education and scalable gait screening. While the framework shows promise for improving access to gait assessment, further validation across diverse populations and clinical settings is required to ensure reliability and generalisability.