yudhv / traffic-sign-classifier

A CNN model that has 97.5% testing accuracy on the GTSRB (German Traffic Signs) dataset.

Traffic Sign Classifier

Overview

In this project, I use a convolutional neural network to classify traffic signs. I train a model so it can decode traffic signs from natural images by using the German Traffic Sign Dataset. After the model is trained, I test it on new images of traffic signs both from a testing dataset, and also some random traffic signs I find near my home in India.

Dependencies

This project requires Python 3.5 and the following Python libraries installed:

• Jupyter
• NumPy
• SciPy
• scikit-learn
• TensorFlow
• Matplotlib
• Pandas (Optional)

Run this command at the terminal prompt to install OpenCV. Useful for image processing:

• conda install -c https://conda.anaconda.org/menpo opencv3

Project

The entire project has been built and described in detail in an IPython notebook which can be found in the root of this folder, named Traffic_Sign_Classifier.ipynb.

I've saved the notebook as the HTML file as well, named report.html, just in case :)