🚦 Traffic Sign Vision

A deep learning-based project for traffic sign recognition using Keras and TensorFlow. It uses a custom dataset of traffic signs organized in a structured directory and labeled using a CSV file. This project demonstrates how to preprocess image data, train a Convolutional Neural Network (CNN), and evaluate model performance.

---

📁 Project Structure

traffic-sign-vision/
├── images/ # (Currently empty or unused)
├── data/
│ ├── train/ # Training images (can be organized by class folders)
│ ├── test/ # Test images
│ └── test.csv # CSV with filenames and labels
├── traffic_sign_recognizer.ipynb # Main Jupyter Notebook
└── labels.txt # List of traffic sign class names (optional)

---

🔧 Features

• Preprocessing with ImageDataGenerator
• Training/validation split with real image data
• CNN model for classification
• Evaluation metrics (loss & accuracy)
• Support for both folder-structured and CSV-labeled datasets

---

🧠 Model

A simple CNN model is used:

• Conv2D + MaxPooling
• Dropout
• Dense layers
• Softmax activation for multi-class output

---

🛠️ Technologies Used

• Python
• TensorFlow / Keras
• NumPy & Pandas
• Matplotlib
• Jupyter Notebook

---

🚀 How to Run

1. Clone the repository

   git clone https://github.com/yourusername/traffic-sign-vision.git
   cd traffic-sign-vision

`` 2. Install dependencies

pip install -r requirements.txt

3. Run the notebook Open traffic_sign_recognizer.ipynb in Jupyter or VS Code and run all cells step-by-step.

---

📊 Dataset Notes

• Training data is in data/train/

• Test data is in data/test/

• Labels for test data are found in data/test.csv

• Each image should be resized to (32, 32) before training.

---

✅ To-Do

• Add early stopping and learning rate scheduler

• Evaluate with test set

• Add confusion matrix and visualization

• Export model as .h5 or TFLite

• Improve model accuracy with data augmentation

🧪 Data Augmentation

To improve model generalization and performance on unseen data, the following augmentation techniques were applied using ImageDataGenerator:

• rotation_range=15
• width_shift_range=0.1
• height_shift_range=0.1
• zoom_range=0.2
• horizontal_flip=True
• rescale=1./255

These help the model become robust to real-world variations in traffic sign appearances.

---

👨‍💻 Author

Muawiya Amir
AI Student & Developer
YouTube: @Coding_Moves

---

📜 License

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

---

⭐ If you like this project, consider giving it a star!