Military Hand Signal Detector

A real-time AI-powered military hand signal detection system using your webcam. Built with MediaPipe, OpenCV, and Scikit-learn. Detects tactical hand signals and announces them out loud using text-to-speech.

---

Demo

Point your hand at the camera and the system will detect and announce the signal in real time.

---

Signals Supported

Signal	Description
Stop	Flat hand raised, palm facing forward
Move Out	Index finger pointing forward, arm swinging
Enemy	Fist with index finger pointing downward
Down	Palm facing down, pushing toward ground
Come Here	Arm extended, fingers curling toward yourself

---

How It Works

1. MediaPipe detects 21 hand landmark points (x, y, z) per hand from your webcam feed
2. Those landmark coordinates are fed into a Random Forest classifier
3. The classifier predicts which military signal is being shown
4. pyttsx3 announces the detected signal out loud in real time

---

Tech Stack

• Python 3
• OpenCV — webcam feed and visual overlay
• MediaPipe — hand landmark detection
• Scikit-learn — Random Forest classifier
• pyttsx3 — text-to-speech output

---

Getting Started

1. Clone the repo

git clone https://github.com/zayd100/MILITARY-HAND-SIGNALS-CAMERA-DETECTOR.git
cd MILITARY-HAND-SIGNALS-CAMERA-DETECTOR

2. Create and activate a virtual environment

py -m venv venv
.\venv\Scripts\activate

3. Install dependencies

pip install opencv-python mediapipe scikit-learn numpy pyttsx3

4. Download the MediaPipe hand landmark model

curl -o hand_landmarker.task https://storage.googleapis.com/mediapipe-models/hand_landmarker/hand_landmarker/float16/1/hand_landmarker.task

---

Project Structure

MILITARY-HAND-SIGNALS-CAMERA-DETECTOR/
│
├── data/                  # CSV files of recorded hand landmarks
├── collect_data.py        # Record your own hand signal training data
├── train_model.py         # Train the AI classifier on your data
├── detect.py              # Run real-time detection with voice output
└── README.md

---

Usage

Step 1 — Collect training data

py collect_data.py

• Pick a signal number
• Perform the signal in front of your webcam
• Press SPACE to capture 200 samples
• Repeat for all 5 signals

Step 2 — Train the model

py train_model.py

Trains a Random Forest model on your captured data and saves it as model.pkl.

Step 3 — Run real-time detection

py detect.py

Your webcam opens and the system detects and announces signals live.

---

Configuration

In detect.py you can adjust:

COOLDOWN_SECONDS = 1.5  # how often it speaks when a signal is held

Lower this value for faster announcements.

---

Requirements

• Windows 10/11
• Python 3.10+
• Webcam
• hand_landmarker.task file (downloaded separately — see setup)

---

Future Ideas

• Add more signals (wedge formation, speed up, slow down)
• Full body pose detection using MediaPipe Pose
• Log detected signals with timestamps
• Build a GUI interface
• Add support for two-handed signals

---

Author

Zayd — @zayd100

---

License

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