In this project, we use TensorFlow to build a neural network to classify road signs based on an image of those signs. To do so, we’ll need a labeled dataset: a collection of images that have already been categorized by the road sign represented in them. For this project, we’ll use the German Traffic Sign Recognition Benchmark (GTSRB) dataset, which contains thousands of images of 43 different kinds of road signs.
The downloaded dataset will a zip file named gtsrb.zip. After unzipping this file, the gtsrb directory will have 43 subdirectories in this dataset. Each numbered subdirectory represents a different category (a different type of road sign). Within each traffic sign’s directory is a collection of images of that type of traffic sign.
The base model consists of the following layers:
- Conv2D layer
- MaxPool layer
- Flatten layer
- Output layer
In further iterations, mulitple Conv2D layers and a fully connecter layer were added to improve model accuracy. The model accuracy results are compared in the python notebook.