Vision Based Fitness Analysis and Adaptive Health Management Using Deep Learning and Large Language Models

Abstract

In the past few years, a growing understanding of health and fitness has highlighted the need for intelligent systems that offer personalized and adaptive recommendations. This work describes a vision-based fitness analysis and adaptive health management system that utilizes computer vision and generative artificial intelligence to provide end-to-end fitness recommendations. The system classifies full-body images provided by the user into well-known male and female body types recognized from health and fitness publications. If the user identifies as a female, the classification options are round, diamond, triangle, hourglass, rectangle, pear, and inverted triangle. If the user identifies as a male, the classification options will include inverted triangle, trapezoid, rectangle, round, hourglass, and triangle. After the body structure is determined, users are asked to provide self-reported demographic and physiological data: age, sex, height, weight, and health goals. This aggregate data will then identify a tailored health plan, structured workout splits, and nutrition answers appropriate to the user&s;s characteristics. By combining visual body structure identification with adaptive plan generation, the system reduces barriers, not needing all user data, and concerns about relying on a professional expert consultation. The anthropometric aspects combined with visual recognition contextualize the scales of measurement allowing the system to show scalability potential for AI-based personal health management. (measured scientifically) The framework presented demonstrates the potential for intelligent health systems to engage with participants, encourage sustainable health and fitness practices, and ultimately support healthier lifestyles.