Regression-Based Video Anomaly Detection Approaches

Videos are a type of time-series data, represent the primary source of information for surveillance applications and are available in large amounts but in most cases contain little or no annotation for supervised learning. This chapter introduces the state-of-the-art deep learning-based methods that make use of the past history of the video data to provide a prediction of the pattern for the future video images and use the difference between the predicted pattern and the real image for video anomaly detection. We will look at a number of recurrent neural network approaches principally designed for the prediction of future video images. Because we use the past history of the video sequence in a predictive manner to determine its future patterns, such schemes are also called the predictive or regression based schemes.