Real-Time Semantics-Based Detection of Suspicious Activities in Public Spaces

Abstract

Behaviour recognition and video understanding are core components of video surveillance and its real life applications. Recently there has been much effort to devise automated real-time high accuracy video surveillance systems. In this paper, we introduce an approach that detects semantic behaviours based on object and inter-object motion features. A number of interesting types of behaviour have been selected to demonstrate the capabilities of this approach. These types of behaviour are relevant to and most commonly encountered in public transportation systems such as abandoned and stolen luggage, fighting, fainting, and loitering. Using standard public datasets, the experimental results here demonstrate the effectiveness and low computational complexity of this approach, and its superiority to approaches described in some other work.