In this Udacity project for the Self-Driving Car Nanodegree I built a deep convolutional neural networks to classify traffic signs. The model is built so it can decode traffic signs from natural images. The used data set is the German Traffic Sign Dataset and the obtained test set accuracy is 97%. I then tested the trained model on new images of traffic signs that I found in Vienna, Austria.
Details of all individual steps are provided directly in the notebook Traffic_Signs_Recognition.ipynb
This project requires Python 3.5 and the following Python libraries installed:
Run this command at the terminal prompt to install OpenCV. Useful for image processing:
conda install -c https://conda.anaconda.org/menpo opencv3
-
Download the dataset. You can download the pickled dataset in which we've already resized the images to 32x32 here.
-
Clone the project and start the notebook.
git clone https://github.com/udacity/traffic-signs
cd traffic-signs
jupyter notebook Traffic_Signs_Recognition.ipynb
- Instructions detailing the code are provided in the
Traffic_Signs_Recognition.ipynbnotebook as well or equivalently in report.html.