VisionArmor: PPE Detection for Construction Workers

Overview

VisionArmor is a Streamlit-based web application that detects Personal Protective Equipment (PPE) in construction workers using a deep learning model trained on a construction site safety dataset.

Live Deployment

Access the deployed application here: VisionArmor

Dataset

The model is trained on the Construction Site Safety Image Dataset from Kaggle: Dataset Link

Repository

GitHub Repository: VisionArmor

Directory Structure

visionarmor/
│── dataset/
│   ├── test/
│   ├── train/
│   ├── valid/
│   ├── README.dataset.txt
│   ├── README.roboflow.txt
│
│── models/
│   ├── best.pt  # Trained YOLO model for PPE detection
│
│── runs/
│   ├── detect/  # Output detection results
│
│── app.py       # Main Streamlit app
│── index.html   # Frontend UI
│── packages.txt # Dependencies
│── README.md    # This file
│── requirements.txt  # Python dependencies
│── visionarmor-ppe-prediction.ipynb  # Model training notebook

Model

• The model used is YOLO (You Only Look Once) for object detection.
• The best-performing model is stored as best.pt.
• The model detects various PPE like helmets, vests, and gloves.

How to Run Locally

1. Clone the repository:

   git clone https://github.com/HarshitSavanur/VisionArmor-AI_Based_Worksite_Protection.git
   cd visionarmor

2. Install dependencies:

   pip install -r requirements.txt

3. Run the Streamlit app:

   streamlit run app.py

Future Improvements

• Improve model accuracy by using a larger dataset.
• Deploy the model on more robust cloud services.
• Add more PPE categories and worker posture detection.