Reptrics

Many people working out alone struggle to track their reps accurately and maintain proper form. Poor form can lead to ineffective workouts or even injury, but most fitness apps only count reps manually and don’t provide feedback on movement quality. I wanted to explore whether computer vision could make workout tracking smarter.

To solve this, I built a workout assistant that uses computer vision to automatically detect and count exercise repetitions. Using MediaPipe’s pose estimation, the app tracks key body landmarks from a camera feed and analyzes joint angles during specific exercises. A rep is only counted when the full movement is completed with correct form, helping users build better habits and avoid reinforcing improper technique.

The project was built using MediaPipe for pose detection and a custom rep-detection system that evaluates motion patterns and joint positions to determine when a repetition begins, ends, and whether it meets the required form criteria.

One of the biggest challenges was making rep detection reliable. Small variations in camera angle, body positioning, and movement speed can make it difficult to consistently detect correct form. I had to experiment with different thresholds for joint angles and movement ranges to accurately identify valid repetitions.

Through this project I learned a lot about real-time computer vision, pose tracking, and how noisy human movement data can be. It was also a great experience translating raw pose data into meaningful feedback that could help users improve their workouts.

Built With

• computervision
• mediapipe
• opencv
• tkinter