paper

Dissecting uncertainty handling techniques: Illustration on maritime anomaly detection

Volume Number:
13
Issue Number:
2
Pages:
Starting page
158
–
Ending page
178
Publication Date:
Publication Date
1 December 2018

paper Menu

Abstract

Detecting and classifying anomalies for Maritime Situation Awareness highly benefits from the combination of multiple sources, correlating their output for detecting inconsistencies in vessels’ behaviour. Adequate uncertainty representation and processing are crucial for this higher-level task where the operator analyses information in conjunction with background knowledge and context. This paper addresses the problem of performance criteria identification and definition for information fusion systems in their ability to handle uncertainty. In addition to the classical algorithmic performances such as accuracy, computational cost or timeliness, other aspects such as the interpretation, simplicity or expressiveness need to be considered in the design of the technique for uncertainty management for an improved synergy between the human and the system. The Uncertainty Representation and Reasoning Evaluation Framework (URREF) ontology aims at connecting these criteria to other uncertainty-related concepts. In this paper, we dissect six classical Uncertainty Representation and Reasoning Techniques (UR-RTs) in their basic form framed into three uncertainty models of probability, belief functions and fuzzy sets, and addressing a fusion problem for maritime anomaly detection. We introduce the Uncertainty Supports as a means to capture what is the carrier of uncertainty and distinguish between three types of supports, that are single variables, sets of variables and uncertainty representations. The latter type indeed captures second-order uncertainty. The different URRTs are qualitatively evaluated according to their expressiveness along the uncertainty supports, and quantitatively evaluated according their accuracy and conclusiveness (uncertainty and imprecision) when processing real AIS data with pseudo-synthetic anomalies. This study illustrates a possible use of the URREF for the assessment and comparison of uncertainty handling methods in fusion systems. The framework provides solid basic foundations for a formal assessment to guide further development and implementation of fusion schemes, as well as for the definition of associated criteria and measures of performance.

Country
FRA
Affiliation
CS Group - France
IEEE Region
Region 08 (Africa, Europe, Middle East)
Email