In the previous section, we explained how it is possible to build an OpenCV cascade of classifiers using positive and negative samples of a class of objects. We will now examine the basic steps of the learning algorithm that is used to train this cascade. Our cascade has been trained using the Haar features that were described in the introductory section of this recipe; however, as you will see, any other simple feature can be used to build a boosted cascade. Since the theory and principles of boosted learning are pretty complex, we will not cover all of their aspects in this recipe; interested readers should refer to the articles listed in the last section.
Let's first restate that there are two core ideas behind the cascade ...
