Models and Methods for Anomaly Detection in Video Surveillance
Ernesto Cruz-Esquivel, Zobeida Jezabel Guzman-Zavaleta · 2024
Video surveillance is essential for monitoring and recording activities in current infrastructure, both in cities and private spaces. Such systems aim to improve security, safety, operational efficiency, and risk reduction. The rapid adoption of video surveillance systems brings with it the challenge of analyzing large amounts of videos in real time. While supervised learning techniques have been widely employed in video surveillance, anomaly detection requires unsupervised or semi-supervised learning approaches to identify unusual or unexpected events or behaviors. This chapter describes learning techniques that help in detecting anomalies for video surveillance applications, including the use of Convolutional Autoencoders, Variational Autoencoders, Long Short-Term Memory networks, and Generative Adversarial Networks (GAN). Additionally, this chapter includes descriptions of the theoretical aspects of the selected models using supervised and semi-supervised learning, as well as some examples of the methods introduced, along with their application in alleviating the urgent necessity for automatic, effective, and efficient detection methods.