Abstract: We present new test results for our active objectrecognition algorithms. The algorithms are used toclassify and estimate the pose of objects in differentstable rest positions and automatically re-position thecamera if the class or pose of an object is ambiguousin a given image. Multiple object views are now used indetermining both the final object class and poseestimate; previously, multiple views were used forclassification only. A feature space trajectory (FST)in eigenspace is used to represent 3D distorted viewsof an object. FSTs are constructed using imagesrendered from solid models. We discuss lighting andmaterial settings for photorealism in the renderingprocess. The FSTs are analyzed to determine the camerapositions that best resolve ambiguities. Real objectsare recognized from intensity images using the FSTrepresentation derived from rendered imagery.!17
展开▼
