Automatic event detection of abandoned and stolen objects in real-time surveillance

This paper describes an approach aimed at automatic identification of events of abandoned and stolen objects in variety environment. Our approach mainly include three steps of data processing: the first processing phrase is object extraction, involving a dual-time background subtraction algorithm which dynamically updates two sets of background. Then, extracted objects are classified as static or dynamic objects and human or non-human objects. Finally, a decision-making model is employed to calculate a confidence score for the classification about event, and an alarm will be automatically triggered if the certainty of action is higher than a pre-defined threshold. Also, the robustness and efficiency of the method was tested on our real time video surveillance system and evaluated by public database such as AVSS 2007 datasets.