Analysing an Identity Information Fusion Algorithm Based on Evidence Theory
Abstract
In this paper, we analyse an identification algorithm in the evidence theory framework. The identification algorithm is composed of four main steps: (1) sensor reports are transformed into initial Basic Probability Assignments, (2) the successive BPAs are combined through Dempster's rule, (3) the resulting BPAs are approximated to avoid algorithm explosion, and (4) in parallel to step (3) a decision is taken on the identification/classification of an object from a database based on the maximum of pignistic probability criterion. The identification algorithm is applied to a Direct Fleet Support scenario where ESM reports are fused to identify six targets among a possibility of 142 in the database. As a basis for the analysis, we observe the behaviour of (1) the pignistic probability of the singletons of the database, used as decision rule, (2) the distance between a BPA and a solution (ground truth), (3) the distance between an approximated BPA and a non-approximated one, and (4) the non-specificity of the BPA.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 2004
- Accession Number
- ADA428563
Entities
People
- Alexandre Jouan
- Anne-laure Jousselme
- Eloi Bosse
Organizations
- DRDC Valcartier