Video anomaly detection

We present a comprehensive overview of techniques for detecting rare or unusual patterns in transportation video. The anomaly detection problem is parsed into two stages: event encoding and anomaly detection model, and various approaches in each stage are presented. Anomaly detection models are broadly classified into structured versus unstructured and supervised versus unsupervised methods, based on how much information is known about normal and anomalous events during training. Spanning these categories, four flavors of models are presented: classification methods, hidden Markov models, contextual techniques, and sparsity models. As a recent promising approach in this domain, the sparsity model is presented in significant detail along with experimental results. The chapter ends with open problems and future research directions in the area.