Video based detection of normal and anomalous behaviour of individuals

摘要

This PhD research has proposed novel computer vision and machine learning algorithms for the problem of video based anomalous event detection of individuals. Varieties of Hidden Markov Models were designed to model the temporal and spatial causalities of crowd behaviour. A Markov Random Field on top of a Gaussian Mixture Model is proposed to incorporate spatial context information during classification. Discriminative conditional random field methods are also proposed. Novel features are proposed to extract motion and appearance information. Most of the proposed approaches comprehensively outperform other techniques on publicly available datasets during the time of publications originating from the results.