YOLO v11 based PPE detection system

This repository contains the implementation of YOLOv11 fine-tuned for detecting Personal Protective Equipment (PPE) in construction sites. The model is trained using a Roboflow dataset containing 1,206 images annotated with five classes:

• 🟢 helmet
• 🔴 no-helmet
• 🟡 no-vest
• 👷 person
• 🦺 vest

The model has been hyperparameter tuned to achieve high precision and recall, making it ideal for real-time detection of safety compliance.

📂 Dataset

The dataset is sourced from Roboflow and consists of:

• Train set: train/images
• Validation set: valid/images
• Test set: test/images
• Number of Classes: 5
• Class Labels: ['helmet', 'no-helmet', 'no-vest', 'person', 'vest']

The dataset is configured in data.yaml as follows:

train: train/images
val: valid/images
test: test/images

nc: 5
names: ['helmet', 'no-helmet', 'no-vest', 'person', 'vest']

---

🚀 Installation & Setup

1️⃣ Clone the repository

git clone https://github.com/yourusername/YOLOv11-PPE-Detection.git
cd YOLOv11-PPE-Detection

2️⃣ Install dependencies

Make sure you have Python 3.8+ and pip installed, then run:

pip install -r requirements.txt

If using a GPU, install torch and torchvision accordingly and requirements.txt assumes you already have torch installed.

3️⃣ Download the dataset

• Extract or place the dataset inside datasets/dataset/
• Ensure data.yaml is correctly referenced in the scripts.

---

🏋️ Training the Model

We have already place best.pt which a fine tuned model, but if you want re-train then: To train YOLOv11 from scratch or fine-tune on the PPE dataset, run(Ensure proper dataset configuration):

python train.py

The training script:

• Loads the YOLOv11-nano model (yolo11n.pt)
• Performs hyperparameter tuning for 300 iterations
• Saves the best-trained weights in runs/detect/train6/best.pt

Best Hyperparameters Found:

lr0: 0.00919
lrf: 0.01
momentum: 0.86723
weight_decay: 0.00041
warmup_epochs: 2.54
warmup_momentum: 0.731
box: 0.198
cls: 0.427
dfl: 1.23
mosaic: 0.607
fliplr: 0.499

---

📈 Model Performance

The best trained model was evaluated, and the following results were obtained:

🟢 Precision-Recall Curve

🔵 Precision vs Confidence Curve

🔴 Recall vs Confidence Curve

🟡 F1-Confidence Curve

🏗️ Confusion Matrix

🔹 Normalized Confusion Matrix

Key Metrics:

• mAP@0.5: 0.887
• mAP@0.5:0.95: 0.504
• Precision: 0.94 (for helmet)
• Recall: 0.80 (for helmet)

---

🖥️ Running the Detection API

A Flask-based API is provided in main.py for real-time video inference.

1️⃣ Start the API Server

python main.py

2️⃣ View the Live Detection Stream

Open a browser and visit:

http://127.0.0.1:5000/

3️⃣ Start the Detection Process

Upload a video file from directory and Start Predictions

4️⃣ Stop the Detection Process

5️⃣ Click 'X' and upload a new file for new Detection

---

📈 Model / Application Results

Precditions on Image

Application Image

📜 Notes & Future Work

• The model can be further improved using data augmentation.
• Future improvements include model pruning for edge device deployment.

---