Udacity Self-Driving Car Engineer Nanodegree

Traffic Sign Recognition using Deep Learning

Overview

This project creates and trains a deep convolutional neural network from scratch, with the task of classifying German traffic signs using the German Traffic Sign Dataset. The trained model is subsequently tested on German traffic signs found on the web, as well as US traffic signs.

Dependencies

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

• Jupyter
• NumPy
• SciPy
• scikit-learn
• TensorFlow
• PIL
• OpenCV

Dataset

This project uses the German Traffic Sign Dataset. A pickled version that has images resized to 32x32 is available here. The pickle file was used as the raw data for this project.

For dataset augmentation, run python data_aug.py

How to run

You can step through the Jupyter Notebook at Traffic_Signs_Recognition.ipynb, or you can read the HTML export of the notebook at Traffic_Signs_Recognition.html.

If you want to run the model training and prediction on sample images, in a stand-alone script, run python traffic_signs.py.

Pre-trained model

The model I have trained is saved in model.ckpt, as a TensorFlow model.