We are developing novel evidence selection & collection method based on Bayesian theorem for object recognition and pose estimation in real environment. Herein the evidence means features such as SIFT (Scale Invariant Feature Transform), line, color, robot motion and so on, for recognition or pose estimation of 3D object. The main advantage of this approach is to estimate probability more easily using Bayesian rule and to select an optimal set of evidences automatically. Therefore this strategy is able to help the robot take advantage of evidence selection automatically in real environment. And we expect that this strategy can apply to some other systems which are necessary for robust object or human recognition.

 * Cognitive Perception Engine (CPE)



* Automatic Evidence Selection



* Experimental Results