Human Pose Estimation for Medical Applications: A Vision-Based Review

Abstract

Human pose estimation (HPE) has emerged as a key enabling technology for markerless human movement analysis in medical and healthcare settings, Advances in deep learning and computer vision have enabled accurate estimation of body joint configurations from monocular RGB images and videos, offering an accessible alternative to traditional marker based motion capture systems, This review provides a comprehensive and clinically oriented survey of human pose estimation methods with a specific focus on medical applications, including data analysis, physical rehabilitation, surgical assistance, fall risk assessment and remote patient monitoring. We systematically review pose representation models, learning paradigms, data sets, and evaluation metrics from a medical perspective, emphasizing validation against clinical gold standards. Current challenges related to accuracy, robustness, generalization, privacy, and regulatory adoption are discussed, and future research directions toward reliable clinical deployment are outlined. This work aims to bridge the gap between algorithmic developments in pose estimation and their translation into practical medical tools by providing a structured, clinically driven analysis and highlighting key limitations, trade-offs, and real-world deployment considerations that are often underexplored in existing survey studies