AI-Based Personalized Fitness Coaching System Using Human Pose Estimation and Adaptive Workout Recommendation

Regular physical activity is an important part of a healthy lifestyle; however, getting individualized coaching is difficult due to the costs involved, the time required, and geographical limitations.Most digital fitness apps use pre-set workout routines, manually entered information, or data collected from wearable sensors, and they rarely assess how well someone executes an exercise.In this paper, we describe the development of a closed-loop camera-based fitness coaching system that combines human pose estimation with an adaptive workout recommendation engine.Our proposed system will extract body landmarks from a single RGB camera, then convert these landmarks into geometric, spatial, and temporal descriptors.These descriptors will allow the system to recognize the type of exercise being performed, count the number of repetitions, and evaluate the quality of execution using an easily understandable weighted-deviation score.The performance descriptors created from this process will be input into a hybrid recommender along with the user's profile and workout history.The hybrid recommender will consist of a rule-based safety layer and a machine learning ranking model.We have also adopted a privacy-aware architecture where the video is processed locally, and only the derived metrics are saved.We have outlined a detailed experimental protocol that includes classification metrics, repetition-count error, inference latency, robustness testing, and an ablation study to determine the importance of pose-derived form quality for personalization.This paper reports the design and evaluation methodology of our proposed prototype; we expect to report the empirical results once the experimental phase is completed.