@book{JRC32075, editor = {}, address = {}, year = {2006}, author = {Versino C and Stringa E and Goncalves J and Heppleston M and Tourin L}, isbn = {}, abstract = {The increased use of surveillance systems, including unattended remote monitoring, results in a large amount of data to be reviewed by inspectors at headquarters. The review task is made difficult by the size and the variety of the data streams. There is thus interest in the development of tools for a ‘Safeguards Review Station’ (SRS) aiming at easing the interpretation of incoming and archived Safeguards data. To date, several tools support the review only of individual sensor data (e.g. images, radiation measures, etc.). The CRISP review module of the RADAR system is an exception, as it provides a time-integrated platform to review nucleonic data streams collected at several points of interest. However, CRISP is limited to radiation sensors and the detection of one event by one sensor is independent of what the other sensors are ‘suggesting’. In this context, we are following a two-step approach for the development of the SRS. Firstly, and when possible, we improve existing filters associated to individual sensors to detect Safeguards-relevant events with higher precision. Along this line, this paper presents such an improvement over an image review filter: a state-of-the-art scene change detection algorithm is augmented with two different search-by-content techniques applied to images to detect precisely typical classes of Safeguards-relevant events. We present test results on real Safeguards images. Secondly, we propose a novel way to review all sensors’ data in an integrated way based on a Hidden Markov Model framework. The core idea is that, while the detection of events starts from the individual sensors in parallel, this first layer of detection can be ‘revised’ by the SRS by considering the plausibility of the sequence of events detected by the collection of sensors. This plausibility check would be based on prior knowledge of typical sequences of events taking place in a given plant. This knowledge could be derived, for instance, by analysing archives of past review reports for a given plant. }, title = {Towards the Integrated Review of Safeguards Surveillance Data}, url = {}, volume = {}, number = {}, journal = {}, pages = {1-9}, issn = {}, publisher = {European Commission}, doi = {} }