Support Vector Machine-Based Video Anomaly Detection Approaches

In this chapter, we introduce another classification method, named support vector machine, for individual abnormal behavior learning. First, we provide a general description of this supervised learning algorithm which consists of a training stage and a testing stage. More specifically, in the training stage, a set of labeled source outputs, that is, the labeled normal and abnormal image sequences of videos, are given and the foregrounds are segmented from the background into connected components using one or more of the object detection methods that have been described previously. This is followed by a description of a popular approach for object tracking, namely the Kalman filter, to generate a set of trajectory sequences, each for a foreground object, as a representation of its behavior. Each constituent behavior, together with its image label, is then supplied as an input to the Support Vector Machine learning algorithm described in this chapter so as for the classifier to learn the behavior patterns, that is, to generate the hyperspheres that separate the abnormal behaviors from the normal behaviors. Finally, in the testing stage, object movements in the newly coming videos are transformed into behavior trajectories and input into the trained classifier for anomaly detection. The section containing more advanced topics is marked with a *. If you are not interested in these materials, you can skip it. We conclude this chapter by the applications of the technique to an individual anomaly behavior detection task.