Although OpenCV is already packaged with several cascade classifiers, there might be a need for detecting some particular object, or class of object, of your choice. Creating a custom cascade classifier is not straightforward since it requires thousands of images from which all the variance should be removed. For instance, if a classifier for faces is being created, all the images should have their eyes aligned. In this section, we will describe the process of creating a cascade classifier using OpenCV.

In order to train a cascade, some tools have been provided in OpenCV. They can be found in the opencv/build/x86/vc11/bin directory. The opencv_createsamples and opencv_traincascade executables are used for preparing a training dataset of positive samples and for generating the cascade classifier, respectively.

In order to give a good idea of the process, we have included files from UIUC Image Database for Car Detection, collected by Shivani Agarwal, Aatif Awan, and Dan Roth. These...