Learning OpenCV 3

Chapter 17. Tracking

Concepts in Tracking

When we are dealing with a video source, as opposed to individual still images, we often have a particular object or objects that we would like to follow through the visual field. In previous chapters, we saw various ways we might use to isolate a particular shape, such as a person or an automobile, on a frame-by-frame basis. We also saw how such objects could be represented as collections of keypoints, and how those keypoints could be related between different images or different frames in a video stream.

In practice, the general problem of tracking in computer vision appears in one of two forms. Either we are tracking objects that we have already identified, or we are tracking unknown objects and, in many cases, identifying them based on their motion. Though it is often possible to identify an object in a frame using techniques from previous chapters, such as moments or color histograms, on many occasions we will need to analyze the motion itself in order to infer the existence or the nature of the objects in which we are interested.

In the previous chapter, we studied keypoints and descriptors, which form the basis of sparse optical flow. In this chapter we will introduce several techniques that can be applied to dense optical flow.  An optical flow result is said to be dense if it applies to every pixel in a given region.

In addition to tracking, there is the problem of modeling. Modeling helps us address the fact that tracking techniques, ...