Goal-directed perception in multisensor object recognition systems

The authors discuss the problem of selecting a sensor and its location in order to observe some object feature, with the goal of verifying an object-location hypothesis or refining its estimated location. The contribution of a sensor observation as a function which relates the involved geometric feature and sensor capabilities with its impact in the reduction of uncertainty in the location of the object is analytically characterized.This allows comparison of the potential benefit of using different sensors for a perception task, computation of a sensor location such that this benefit is maximal. Specific examples are presented for the observation of edges using two types of sensors: camera-in-hand 2-D vision and mobile proximity.<
>