Approach
The first step is to collect data with the choice of a database in the public domain. We need a large database of international traffic signs and a small database of Quebec traffic signs.
Figure 1: German database
Figure 2: Quebec database
For the choice of the two models we stopped on a convolutional network to process the images and then finish with transfer learning.
We will therefore do a preliminary training of the convolutional network with the large database of international signs and then finish the training with transfer learning with the small database of Quebec signs. To evaluate the classification system, we have the classification rate then the confusion matrix.