ABSTRACT
A video camera is a versatile sensor for monitoring and diagnosing. It is so affordable that many people have more than one on their phones, cars, or computers. In this chapter, we combine visualization and analytic algorithms to make video analytics an intuitive task. The goal is to extract dynamic features in videos for ambient diagnostic applications. We will study the following:
• Moving object detection • Object tracking • Shape description • Video summarization • Activity recognition • Video magnification
Given a set of video frames and assuming the camera is stationary, how do we extract the moving foreground and remove the static background? Figure 7.1 shows a desirable scenario, where the moving objects are extracted as binary blobs. A binary blob is the simplest visual form of a shape description. It is so important that intelligence
analysts often refer to video analytics as “blobology.” Here, we introduce three basic background subtraction methods: frame subtraction, approximate median, and Gaussian mixture model (GMM).138
