An AI-Driven Multi-Modal Framework for Exercise Form Detection and Correction in Strength Training

Integration of intelligent systems into gym and fitness environments has gained considerable attention in recent years based on the development of computer vision and deep learning methods. Accurate identification of the exercise type is critical to minimize injuries and maximize workout efficiency. This study suggests an extensive multi-stage framework for the classification of gym exercises, leveraging the strengths of YOLOv8 for human and object detection, MediaPipe for pose estimation, and EfficientNetV2 for feature extraction and classification. The system first detects the human and equipment used in the exercise using YOLOv8. Later, MediaPipe infers joint landmarks and computes joint angles and evaluates the correctness or otherwise of an exercise posture in relation to established criteria. Pose data that is pre-processed is later fed to EfficientNetV2, from which spatial features are extracted and the final classification helped. The presented pipeline supports real-time execution as well as accuracy in exercising postures' differentiation. This approach is efficient in managing poses variations of human, lighting situations, and gym exercise environments. Utilizing strong deep learning structures and pose estimation architectures, the system provides excellent performance for exercise analysis in a gym setting. This contribution extends the growing field of AI-enhanced fitness tracking through a flexible, accurate, and scalable exercise posture assessment and classification solution.