Constraint-based object recognition in multisensor systems

A general procedure to compute and validate geometric constraints between uncertain geometric features is presented. Uncertain geometric information is represented using the symmetries and perturbation model. The procedure obtains the geometric relations between any pair of geometric elements in a systematic way, with an explicit consideration of the uncertainty due to the use of different sensors. It is shown how to use information regarding geometric relations, not only in the validation of hypotheses during the recognition process, but also for the selection of observations that allow the determination of the object location.<
>