In this chapter, we trained a multiclass classifier to recognize traffic signs from the GTSRB database. We discussed the basics of supervised learning, explored the intricacies of feature extraction, and extended SVMs so that they can be used for multiclass classification.
Notably, we left out some details along the way, such as attempting to fine-tune the hyperparameters of the learning algorithm. When we restrict the traffic sign dataset to only 10 classes, the default values of the various function arguments along the way, seem to be sufficient for performing exceptionally well (just look at the perfect score achieved with the HOG feature and the one-vs-one strategy). With this functional setup and a good understanding of the underlying methodology, you can now try to classify the entire GTSRB dataset! It is definitely worth taking a look at their website, where you will find classification results for a variety of classifiers. Maybe, your own approach will soon be added to the...
