Recognition System Considering Incompleteness and Inaccuracy of Observation, and Contradiction between Observations
Terumoto Komori, Norihiko Kato, Yoshihiko Nomura, Hirokazu Matsui · The proceedings of the JSME annual meeting · 2000
There sometimes occur problems in observational data : (1) "inaccuracy", especially, due to outliner data, (2) "incompleteness" that data don't satisfy sufficient conditions. In addition to these, there occurs another problem, when integrating multiple pieces of information, i.e., "contradiction" between the information. These problems with observations, furthermore, result in characteristic problems with recognition : (1) "uncertainty", (2) "ignorance", and (3) "contradiction", respectively. These problems degrade credibility of recognition system. This paper proposes following methods for these problems : (1) a calculation method of a feature value considering the outliner observations (2) a quantification method of the "ignorance" by applying the Dempster-Shafer theory (3) another quantification method of the "contradiction" with upper and lower bounds.