SENSOR FUSION INTEGRATING CONTEXTUAL INFORMATION

Abstract

The application of multi-sensor fusion, which aims at recogizing a state among a set of hypotheses for object classification, is of major interest as regards the performance improvement brought by the sensor complementarily. Nevertheless this needs to take into account the more accurate as possible information and take advantage of the statistical learning of the previous measurements acquired by sensors. The classical probabilistic fusion methods lack of performance when the previous learning is not representative of the real measurements provided by sensors. The theory of evidence is then introduced to face this disadvantage by integrating a further information which is the context of the sensor acquisitions. In this paper, we propose a formalism of modeling of the sensor reliability to the context that leads to two methods of integration when all the hypotheses, associated to the objects of the scene acquired by sensors, are previously learnt: the first one amounts to integrate this further information in the fusion rule as degrees of trust and the second models the sensor reliability directly as mass functions. These two methods are based on the theory of fuzzy events. Afterwards, we are interested in the evolvement of these two methods in the case where the previous learning is unavailable for a hypothesis associated to an object of the scene and compare these two methods in order to deduce a global method of contextual information integration in the fusion process.