Convolutional Neural Networks Based Video Anomaly Detection Approaches

Video anomaly detection can be viewed as image anomaly detection with a temporal component since a video is composed of a time sequence of images. From this point of view, the only ‘new’ technique introduced in this chapter is a strategy to take advantage of this temporal correlation. However, due to the complex behavior patterns in video surveillance scenes and the susceptibility to interference from unrelated areas in the surveillance images, it is often difficult to efficiently and effectively select the right set of handcrafted features so as to accurately predict abnormal behaviors by traditional techniques described in the previous chapters since there are different situations in which video anomaly detection becomes necessary, each requiring a solution specific to its peculiar conditions.