Event Modelling and Reasoning with Uncertain Information for Distributed Sensor Networks Jianbing Ma, Weiru Liu, and Paul Miller School of Electronics, Electrical Engineering and Computer Science, Queen’s University Belfast, Belfast BT7 1NN, UK {jma03,w.liu}@qub.ac.uk, p.miller@ecit.qub.ac.uk Abstract. CCTV and sensor based surveillance systems are part of our daily lives now in this modern society due to the advances in telecommunications tech- nology and the demand for better security. The analysis of sensor data produces semantic rich events describing activities and behaviours of objects being mon- itored. Three issues usually are associated with events descriptions. First, data could be collected from multiple sources (e.g., sensors, CCTVs, speedometers, etc). Second, descriptions about these data can be poor, inaccurate or uncertain when they are gathered from unreliable sensors or generated by analysis non- perfect algorithms. Third, in such systems, there is a need to incorporate domain specific knowledge, e.g., criminal statistics about certain areas or patterns, when making inferences. However, in the literature, these three phenomena are sel- dom considered in CCTV-based event composition models. To overcome these weaknesses, in this paper, we propose a general event modelling and reasoning model which can represent and reason with events from multiple sources includ- ing domain knowledge, integrating the Dempster-Shafer theory for dealing with uncertainty and incompleteness. We introduce a notion called event cluster to rep- resent uncertain and incomplete events induced from an observation. Event clus- ters are then used in the merging and inference process. Furthermore, we provide a method to calculate the mass values of events which use evidential mapping techniques. Keywords: Bus Surveillance; Active System; Event Composition; Event Reason- ing; Inference. 1 Introduction CCTV-based 1 surveillance is an inseparable part of our society now – everywhere we go we see CCTV cameras (e.g. [2, 11, 5, 13], etc). The role of such systems has shifted from purely passively recording information for forensics to proactively providing ana- lytical information about potential threats/dangers in real-time fashion. This shift poses 1 This paper is an extended version of [9] in which we have included a set of running examples, a method (summarized by an algorithm) to calculating mass values of events which uses ev- idential mapping techniques, and the newly introduced notion rule clusters. Furthermore, we also demonstrate in this paper how to interpret and use the background knowledge, or domain knowledge, which was only preliminarily introduced in [9].