Foreground extraction based on anomaly detection

Abstract

A robust method to detect foreground regions in video sequences is presented. For each pixel, the time series of pixel intensity is represented as symbol sequences that are trained and modelled using the Markov model to obtain the anomaly detection threshold. The foreground extraction is performed by comparing the anomaly transition probability of each pixel and its detection threshold. The experimental results demonstrate that the proposed method outperforms the traditional methods. © The Institution of Engineering and Technology 2014.