Optical Flow-Based Weighted Magnitude and Direction Histograms for the Detection of Abnormal Visual Events Using Combined Classifier

Related Work

First thing in suspicious event detection is to extract features which can robustly describe the scene statistics e.g. low level features and high level features. After extracting features from the scene, event modelling or classification of data is done based on extracted features. In event modelling, algorithm learns the behavior or pattern of extracted features and classify whether scene contains an anomalous event or not. Event modeling is generally known as machine learning and can be classified into three major categories: supervised techniques, semi-supervised techniques and unsupervised techniques.

The anomaly detection based on supervised technique requires the labeling of samples for both normal samples and abnormal samples to train the model and give prediction on test samples. These methods are generally train model for specific abnormal state whose features are previously known or set, such as ’U’ turn detection in traffic surveillance scene (Zen & Ricci, 2011; Z et al., 2005) .