AI-Based Gym Tracker and Real-Time Posture Correction APP

Abstract

The growing popularity of home workouts has led to a surge in fitness applications; however, most lack real-time posture correction and personalized feedback, often resulting in poor form and risk of injury. To address these challenges, this paper presents PeakForm, an AI-based gym tracker that combines computer vision and deep learning to provide real-time workout analysis, posture correction, and personalized exercise recommendations. The system uses a pose estimation model, such as MediaPipe Pose, to detect 33 body landmarks from a live video stream and evaluate user posture during exercises like push-ups, squats, jumping jacks, and planks. A posture evaluation algorithm compares the user’s joint angles with reference standards, offering immediate feedback through visual and auditory cues when deviations occur. For static workouts, the system includes a countdown timer with a 3-second pre-start and a beep notification upon completion. Peak Form’s recommendation module generates personalized workout routines based on the user’s height, weight, and fitness goals. Implemented using Python, OpenCV, and TensorFlow, the system achieved an average accuracy of 95% in exercise detection and form evaluation during real-time testing. The results demonstrate that PeakForm effectively enhances user performance and safety while offering a trainer-like experience without wearable devices. This project represents a step toward intelligent, accessible, and personalized virtual fitness training for all users.