Probability-Based Video Anomaly Detection Approaches

In Chap. 1 , we look at a general description of techniques that we will talk about in this book. In this chapter, under the assumption that we are given no more information but a source that generates a video sequence, that is, a sequence of correlated images, at least correlated within each of their local time periods that incorporate the structure in the data, we look at a probability-based video anomaly detection technique. First, we give a formal definition of anomalies such technique aims to discover. We next introduce a number of pixel-level ways for generating a model for a background scene, which is the normality for the anomaly detection. Then, according to the definition, we retain those rarely happening anomalies as the foreground in order to decrease the amount of video data for anomaly detection in later processing. We will intersperse the implementations with Python and OpenCV after the descriptions of some of the techniques. Finally, we provide a brief coverage of more advanced background modeling topics which arouse great interests and are extensively studied currently in the field.