Project Overview
This project is a computer vision-based fitness assistant designed to analyze a video of knee bend exercises. It automatically detects exercise scenes, filters out frame fluctuations, and employs pose estimation to track the knee joint angle. The system counts successful repetitions, provides real-time feedback, and ensures correct exercise form, making it a valuable tool for fitness enthusiasts
The main features and context of the project are as follows:
1. Scene Detection:
The project uses the scenedetect library to automatically detect different scenes in the input video based on content changes.
2. Frame Filtering:
It filters the video feed for sudden frame fluctuations or transitions, ensuring that only stable frames are processed for pose estimation.
3. Pose Estimation:
It employs the Mediapipe library for pose estimation, tracking key landmarks on the person's body, including the knee joint.
4. Rep Counting:
The code counts the number of successful knee bends (reps) based on the angle of the knee joint. It measures the time elapsed during each rep to determine if it was executed correctly.
5. User Feedback:
The project provides real-time feedback to the user, indicating whether they need to keep their knee bent or not.
6. Visual Interface:
It displays the video feed with overlays showing the detected landmarks, knee angle, rep count, and timer.
Here is a decision diagram for the process :
