A Practical Introduction to Computer Vision with OpenCV

9 Video

Cameras supply video feeds, but in computer vision they are often used simply to acquire occasional static images from the stream of available video data (for example, in industrial inspection applications). However, as we have progressed from the factory to less constrained environments, we have found that video provides us with much more useful information through the analysis of what is changing in the video scene. Some of the most common such techniques are used for the detection of moving objects (see Section 9.1) within video streams. In this domain, we typically use static cameras (i.e. cameras which are fixed in place and do not pan, tilt or zoom) and hence a background image/model can be derived in a relatively straightforward manner from the video (see Section 9.1.4). Typically, these techniques allow us to identify moving/changing pixels within the scene but have difficulties translating this to moving objects as moving objects frequently overlap, resulting in confusion about which object is which. As a result, another major topic in video analysis is the visual tracking of objects within a scene without use of a background model (see Section 9.2). We finish this chapter by looking in Section 9.3 at performance analysis in video, which can be quite different from the performance analysis for images described in Section 8.6.

9.1 Moving Object Detection

Motion detection is used extensively to analyse video sequences containing moving objects such as cars and ...