Using Machine Learning and TensorFlow to Recognize Traffic Signs

by Sophia TurolDecember 16, 2016
Using neural networks, road signs can be recognized right on the go while driving a car.

This time, TensorFlow meetup in Silicon Valley offered the attendees a practical session aimed at enabling a neural network to classify traffic signs.


Technologies used

Waleed Abdulla, a founder and CEO at Ninua, delivered a hands-on training on applying machine learning to recognize traffic signs in a video shot from a moving car. Under the tutorial, Waleed demonstrated how to build a neural network from scratch and enable it to classify traffic signs.

For that purpose, the speaker utilized the following technology stack:



In the course of the tutorial, Waleed also gave some tips on how to:

  1. Parse and load the training data. Though the images are saved in the uncommon .ppm format, it can be solved with the scikit library.
  2. Handling images of different size. To verify data range and catch bugs early, Waleed suggested printing the min() and max() values on resizing images.
  3. Evaluation. When visualizing the results, one has to remember to use a validation data set and measure the accuracy of a model.

Find the source code for this demo at this Jupyter notebook or read Waleed’s article on the topic.

Join our group to get informed about the upcoming events.


Want details? Watch the videos!

This is the demo presented during the meetup.


Another session by Waleed Abdulla was focused on the same topic, but contained slightly other details.


Further reading


About the expert

Waleed Abdulla is a founder and CEO at Ninua. As a software engineer, he is interested in deep learning and web development. Currently, Waleed is running a tech startup in Mountain View, building SymphonyTools—a social media management dashboard for businesses.


The post is written by Sophia Turol with assistance from Alex Khizhniak.