/Traffic-Sign-Classifier

Classify road signs using a deep convolutional neural network.

Primary LanguageJupyter Notebook

Traffic-Sign-Classifier

Please refer to: Self-Driving Car Engineer Nanodegree - Udacity, Term 1 - Project 2

Udacity - Self-Driving Car NanoDegree

Overview

This project uses a deep convolutional neural network to classify traffic signs. The CNN model takes as input the German Traffic Sign Dataset and understands road signs from images. The notebook shows the code as well full results.

Dependencies

The code is stored in a Jupyter Notebook and requires Python 3.5. Please refer to the project website for implementation details.

Preprocessing

The project pipeline includes functions to transform poor colors in high contrast grey images. E.g.:

  • No passing

Combined Image

  • Speed limit

Combined Image

Results

At the end, with this CNN model you can obtain the following accuracy:

  • training set: near 100%
  • validation set: 94%
  • test set: 85%