Computer Vision and Imaging in Intelligent Transportation Systems

9Video Anomaly Detection

Raja Bala¹ and Vishal Monga²

¹ Samsung Research America, Richardson, TX, USA

² Pennsylvania State University, University Park, PA, USA

9.1 Introduction

The ability to detect anomalous or unusual events has numerous applications in the transportation domain, including identifying traffic violations, accidents, unsafe driver behavior, street crime, and other dangerous and suspicious activities [1]. In many surveillance settings, anomaly detection requires significant human intervention, and is hence not scalable to high volumes of video footage. Thus a large fraction of transportation video is simply stored without review. Automatic and reliable detection of anomalies from natural traffic scenarios is clearly of benefit. This problem is challenging for several reasons. First, video captured in natural urban settings often contains a large amount of clutter amid a complex dynamically varying scene. Second, since anomalies are by definition rare events, it is difficult to obtain a sufficiently large number of samples to accurately characterize anomalous events and behavior. In this chapter, we present a framework for video anomaly detection, provide a brief survey of state‐of‐the‐art methods, and elaborate on a few selected techniques that attempt to address the aforementioned challenges. Last, directions for future investigations in this topic are pondered. Note that anomaly detection can be thought of as a general framework that includes as special cases applications ...