Markov Random Field-Based Video Anomaly Detection

In Chap. 3 , we saw one probabilistic approach for anomaly detection based on the intuition that anomalous patterns are usually associated with small probabilities. In this chapter, we see another method of probability estimation-based abnormal behavior detection called Markov random field. Basically, if the human behaviors can be segmented into a sequence of discrete states, this Markov model can be used as one of the most popular ways of representing the dependence among the states, giving rise to the state transition probabilities. Then, the anomaly detection will be to identify those sequences of behavior states with low transition probability, that is, those sequences of behavior states that rarely happen. We will look at the basic ideas behind Markov random field and describe its application to anomaly detection.