🏋️ AI Gym Trainer

An AI-powered gym training assistant that uses computer vision and pose estimation to count your exercise repetitions in real time. Currently tracks bicep curls — with the angle of the right arm detected frame-by-frame, the system counts every complete rep and displays a live progress bar.

---

✨ Features

• Real-time pose detection powered by MediaPipe
• Automatic rep counting — no buttons, no timers, just move
• Live progress bar showing how far through the current rep you are (0 → 100%)
• Rep counter panel displayed on screen
• Works with a webcam feed or a pre-recorded video file

---

🛠️ Technologies Used

Technology	Purpose
Python 3	Core language
OpenCV (cv2)	Video capture, drawing overlays
MediaPipe	Human pose landmark detection
NumPy	Angle-to-percentage interpolation

---

📋 Prerequisites

• Python 3.8 or higher
• A webcam or a compatible .mp4 video file

---

🚀 Installation

1. Clone the repository

   git clone https://github.com/Mpradeep-dev/AI_Trainer.git
   cd AI_Trainer

2. Install dependencies

   pip install opencv-python mediapipe numpy

---

▶️ Usage

Run with the sample video (default)

python main.py

By default main.py reads res/gym1.mp4.

Switch to webcam input

Open main.py and change line 5:

# Use webcam
vid = cv.VideoCapture(0)

# Use a video file
vid = cv.VideoCapture("res/gym1.mp4")

Controls

Key	Action
Esc	Quit

---

📁 Project Structure

AI_Trainer/
├── main.py          # Entry point — video loop, rep counting, UI overlays
├── pose_module.py   # Reusable pose detector class (MediaPipe wrapper)
├── res/
│   └── gym1.mp4     # Sample workout video
└── README.md

---

🧠 How It Works

1. Frame capture — each video frame is read and resized to 2× for better visibility.
2. Pose landmarks — MediaPipe identifies 33 body key-points per frame.
3. Angle calculation — the angle at the elbow (shoulder → elbow → wrist, landmarks 12-14-16) is computed using atan2.
4. Rep counting — the angle is interpolated to a 0–100% range. A full rep is counted when the arm goes from 0 % → 100 % → 0 % (or vice-versa).
5. Overlays drawn:
   • Right-side magenta bar = current rep progress.
   • Bottom-left green panel = total reps completed.

Pose Landmarks Used

12 → Right Shoulder
14 → Right Elbow  (vertex of the angle)
16 → Right Wrist

---

🔧 pose_module.py — Detector API

dec = pm.detector()

img  = dec.find_pose(img, draw=True)   # Draw skeleton on frame
lmks = dec.get_conn(img, draw=False)   # Returns list of [id, cx, cy]
angle = dec.find_angle(p1, p2, p3, img, draw=True)  # Angle at p2

---

🗺️ Roadmap / Future Improvements

• Support multiple exercises (squats, push-ups, shoulder press)
• Voice feedback announcements ("Rep 5 done!")
• Web or desktop GUI with session history
• Webcam auto-detection with file-path fallback
• Export rep data to CSV / JSON

---

📄 License

This project is open source and available under the MIT License.