A computer vision application that uses webcam to count exercise repetitions in real-time. This project features two distinct detection approaches: a simple motion-based counter and a AI-powered pose estimation counter.
The application running in Pose Estimation mode, showing the live camera feed, detected skeleton, and rep count.
This project was born from a simple problem: losing count of repetitions during high-intensity workouts like burpees. I would often put my full focus on how my muscles felt and my form. AutoRep solves this by using a webcam to automate the counting process, allowing the user to focus on form and effort.
The project was developed in two stages:
-
Approach 1 (Motion Detection): A proof-of-concept that uses simple frame-differencing to detect motion spikes. While fast, it's sensitive to background noise and can't distinguish between different types of movements.
-
Approach 2 (Pose Estimation): Uses Google's MediaPipe library to track the user's body joints and display live on screen. It counts reps based on the user's actual posture and form.
This is demo of Approach 1.
This is demo of Approach 2.
Checkout the Technical Report for more info.