Mobile-Based Fitness Form Enhancer: Real-Time Posture Analysis and Personalized Feedback Using YOLOv7-Pose and LLaMA
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Proper form in strength training is vital to optimize outcomes and avoid injury. Personal trainers provide guidance, yet the services may not be accessible to all, particularly for home training. In response, this paper introduces a mobile system for enhancing fitness form supported by real-time computer vision and natural language processing to track exercise posture and provide personalized feedback. The system utilizes YOLOv7-Pose, a fast pose estimation model, to detect joint keypoints from real-time video feed captured by a smartphone camera. In accordance with computed joint angles, the system labels the user’s posture and detects typical form errors. To produce actionable and goal-directed advice, a LLaMA-based large language model (LLM) is employed to synthesize the feedback according to the user’s BMI, body fat ratio, calorie consumption, and fitness goals such as fat loss, muscle gain, or recomposition. The architecture runs in near real-time through a mobile front-end and a web-based LLM API, providing meaningful feedback within 1.5–2 seconds. The initial benchmarks validate the responsiveness and accuracy of the system in detecting form deviations for a range of strength training exercises. The scalable and modular architecture provides an actionable solution for smart, on-the-go fitness coaching without dedicated hardware.
Fields of Study & Subjects
Human Pose and Action Recognition
Context-Aware Activity Recognition Systems
Physical Education and Training Studies
Publication details
- Journal / source
- International Conference on Computing, Communication, Control and Cyber-Physical Systems (I5CPS)
- Pages
- 1–6
- Publication date
- 26 February 2026
- Language
- EN