Please use this identifier to cite or link to this item:
|Title:||MM: The Markov Model Tool for Image Reviews|
|Authors:||LOMBARDI Paolo; VERSINO CRISTINA; GONCALVES JOAO; HEPPLESTON Matthew; TOURIN Laurent|
|Citation:||Proceedings of the 49th Annual Meeting of the INMM - Institute for Nuclear Materials Management p. 1-8|
|Publisher:||INMM - Institute for Nuclear Materials Management|
|Type:||Articles in periodicals and books|
|Abstract:||In nuclear plants, flasks of material undergo a processing governed by a rather regular sequence of stages. This regularity provides motivation for modeling the temporal succession of Safeguards-relevant events on the basis of past review reports annotated by inspectors and re-using this knowledge to assist future reviews. This paper discusses a tool, called Markov Model (MM), which uses time and sequence information of events to assist inspectors during image reviews. In previous work, we described the mathematics of the MM approach to reviewing images. The MM tool is now part of the Safeguards Review Station (SRS), a software prototype which includes several image filters stemming from R&D activities. SRS shares the same look-and-feel and event visualization modalities as the state-of-the-art review software used in Safeguards. This design choice was used to make the evaluation of the MM review tool independent of human interfacing aspects. We tested the MM tool on multiple image sets taken from different nuclear sites. Since SRS implements Safeguards reference Scene Change Detection (SCD) algorithm, we measured MM¿s performance by taking SCD as the basis for comparison. Furthermore, since SCD is an effective filter, we ran MM on the subset of images first selected by SCD. During the tests it became clear that in certain MBAs the video surveillance conditions (e.g. positioning of cameras) and, especially, the actual nuclear process well suit to the MM approach, leading to a reduction in the number of images to review (45-90% reduction in experiments on flask processing). Conversely, in some other plants, the assumptions made to define the underlying MM were not satisfied over time, this required an extension of the time parameter in the model to capture these events. (e.g., to account for rare events where the timing of an event in part of the sequence was significantly longer than expected). The paper presents the latest results and discusses the philosophy behind the creation of the data reduction filters and, in particular, explains how filter underlying assumptions reflect the field operation reality/characteristics.|
|JRC Institute:||Institute for the Protection and Security of the Citizen|
Files in This Item:
There are no files associated with this item.
Items in repository are protected by copyright, with all rights reserved, unless otherwise indicated.