Selective attention in multisensor object recognition

Presents a method to characterize the relevance of a pairing between a sensor observation and a model feature in determining the identity and location of the object to which it belongs. This relevance measurement expresses how much information the pairing contributes to support the object location hypothesis, and can be computed for any type of geometric element, obtained by any type of sensor. It takes into account the characteristics of the involved geometric feature, and the capabilities of the sensor that obtains the observation. The authors propose a set of strategies for the generation and verification of hypotheses that use the relevance measurement in the selection of the observations and model features that are to be processed first, so that plausible hypotheses can be produced promptly.<
>