In the previous chapters, we learned about a number of algorithms for object detection and tracking. We learned how to use color-based algorithms, such as Mean Shift and CAM Shift, in conjunction with histograms and back-projection images to locate an object in an image with incredible speed. We also learned about template matching and how it can be used to find objects with a known template of pixels in an image. All of these algorithms rely in one way or another on image properties, such as brightness or color, that are easily affected by a change in lighting of the environment. Based on these facts, we moved on to learn about algorithms that are based on knowledge about significant areas in an image, called keypoints or features. We learned about many edge- and keypoint-detection algorithms and how to extract descriptors for those keypoints. We also learned about descriptor matchers and how to detect an object in an image using good matches of descriptors extracted from an image of the object of interest and the scene where we're looking for that object.

In this chapter, we're going to take one big step forward and learn about algorithms that can be used to extract a model from a large number of images of an object, and later use that model to detect an object in an image or simply classify an image. Such algorithms are the meeting point of machine learning algorithms and computer vision algorithms. Anyone familiar with artificial intelligence and machine learning algorithms in general will have an easy time proceeding with this chapter, even if they are not fluent in the exact algorithms and examples presented in this chapter. However, those who are totally new to such concepts will probably need to grab another book, preferably about machine learning, to familiarize themselves with algorithms, such as support vector machines (SVM), artificial neural networks (ANN), cascade classification, and deep learning, which we'll be learning about in this chapter.

In this chapter, we'll look at the following:

• How to train and use SVM for classification
• Using HOG and SVM for image classification
• How to train and use ANN for prediction
• How to train and use Haar or LBP cascade classifiers for real-time object detection
• How to use pre-trained models from third-party deep learning frameworks