😷 Real-Time PPE Compliance Monitor

An AI-powered computer vision pipeline that detects face mask compliance in real-time. This project uses a modern Convolutional Neural Network (CNN) face detector paired with a fine-tuned Deep Learning classifier to accurately track and evaluate Personal Protective Equipment (PPE) compliance, even under heavy occlusion.

📸 Demo

Accurately detecting correctly worn KN95/Surgical masks.

Flagging non-compliance and incorrectly worn masks in real-time.

🚀 Tech Stack

• Python 3.12
• TensorFlow / Keras: Fine-tuned MobileNetV2 for mask classification.
• OpenCV 5.0.0: YuNet (CNN-based face detector) for real-time, robust face tracking.
• NumPy: Matrix and array manipulations.

🧠 How It Works

Traditional computer vision algorithms (like Haar Cascades) fail to detect faces when the nose and mouth are covered by masks. This project solves that bottleneck by using a two-step pipeline:

1. Face Detection (YuNet): A lightweight, native OpenCV CNN detects faces in the webcam feed regardless of steep angles, poor lighting, or heavy occlusion (hands, mugs, and masks).
2. Classification (MobileNetV2): The detected face ROI is dynamically padded, preprocessed, and fed into a custom-trained MobileNetV2 model to classify the crop into one of three states:
   • With Mask (Green)
   • Incorrect Mask (Yellow)
   • No Mask (Red)

🛠️ Installation & Setup

1. Clone the repository

git clone https://github.com/YOUR-USERNAME/RealTime-PPE-Compliance.git
cd RealTime-PPE-Compliance

2. Create a virtual environment (Recommended)

python3 -m venv venv
source venv/bin/activate  # On Windows use: venv\Scripts\activate

3. Install dependencies

pip install -r requirements.txt

4. Download the YuNet ONNX Model Because of GitHub file size limits, you must download the OpenCV YuNet weights directly:

curl -L -o face_detection_yunet_2026may.onnx https://github.com/opencv/opencv_zoo/raw/main/models/face_detection_yunet/face_detection_yunet_2026may.onnx

💻 Usage

Run the main application script to start the webcam feed:

python3 app.py

Press q on your keyboard to quit the video stream.

📁 Repository Structure

• model_training.ipynb: The Jupyter Notebook used to train and fine-tune the MobileNetV2 model on the FMD Dataset.
• app.py: The main OpenCV deployment script for real-time webcam inference.
• ppe_mask_model.keras: The saved weights of the trained classification model.
• requirements.txt: Python package dependencies.

👨‍💻 Author

Rudra Pratap Singh