Skip to content
Using synthetic data in combination with Deep Learning, to determine if a system can be made that will be able to recognise and classify correctly real traffic signs.
Jupyter Notebook Python
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.
.ipynb_checkpoints Updated for GTSRB Sep 24, 2017
German_signs Original and new data Feb 6, 2018
Google_search_backgrounds Updated for GTSRB Sep 24, 2017
Traffic_Signs_Templates/Images Original and new data Feb 6, 2018
figures Figures Sep 5, 2017
val_set Original and new data Feb 6, 2018
CNN-Classifier.ipynb Updated for GTSRB Sep 24, 2017
Dataset_creator.ipynb Updated for GTSRB Sep 24, 2017
LICENSE Create LICENSE Jun 2, 2017 Update May 28, 2018 Blurred data Sep 25, 2017 Visualisations & system v1.1 Aug 23, 2017

The Driver's Assistant: Utilising Synthetic Data Generation and Deep Learning for Traffic Sign Classification


A main concern of road sign recognition systems has been the use of a large dataset obtained through research and manually finding sufficient data in order to conduct the necessary experiments and optimally, create a model with high accuracy rates. The purpose of this dissertation is the introduction of a traffic sign recognition system , which is based on synthetic data generation methods (SGTD), that is not bound to a specific set of road signs as occurring in a country or region, but can gen- erate training examples given a defined set of templates. This work can significantly decrease the time spent by researches or system developers in order to acquire data and allows the reallocation of the main project focus on the construction of the pre- dictive model. Although, the training dataset is synthetically created, the variations in visual appearance between examples, such as differences in illumination and ro- tation, are reflected on each image to best represent road signs in real world settings and have been implemented using Computer Vision methods. The main classifier that the system is based on is a Convolutional Neural Network (CNN) that can achieve close to state-of-the art accuracy results when trained on actual data.

alt text

The first row shows the generated examples while the second are real-world signs


Git is required to download and install the repo. You can open Terminal (for Linux and Mac) or cmd (for Windows) and follow these commands:

$ sudo apt-get update
$ sudo apt-get install git
$ git clone

Note that the synthetic dataset SGTSD will need aprox. 10Gb and the sample set used for training will be close to 1Gb. Please allow sufficient storing space when running the generator on a system.




Alexandros Stergiou

a.g.stergiou at

Any queries or suggestions are much appreciated!

You can’t perform that action at this time.