A Practical Analytical Review of Posture Estimation Technology for Fitness AI Coaching: Posture Correction, Repetition, and Range of Motion Analysis

This paper critically examines the applicability of markerless human posture estimation technology in real-world fitness coaching environments. Aiming to bridge the gap between academic research and commercialization, this review study focuses on the deployment readiness of four core functions: posture correction, repetition counting, range of motion analysis, and exercise recognition. Following the PRISMA protocol, we analyzed 43 recent studies published between 2020 and 2025 to deeply examine not only the achievements of each technology but also the fundamental limitations and challenges it faces in real-world settings. Our analysis reveals that despite advances in deep learning models, dataset bias, the absence of standardized evaluation criteria, poor generalization across diverse users and environments, and fragmented feature development remain key bottlenecks to technology commercialization. This paper proposes a concrete future research roadmap to overcome these limitations and build a reliable and scalable AI fitness coaching system beyond prototype-level research.