抄録
This paper describes a virtual-environment edge-potential eigenspace method. It utilizes CG technique to synthesize images as a substitute for taking images of the real world. All images, i.e. images for learning and testing input images, are changed into representative colored images, and, then, into edge potential images. These two processes make the parametric eigenspace method practical since it dispenses with taking many pictures corresponding to the locations to be estimated. We apply the proposed method to self-localization and pose estimation of a mobile robot and examine its performance.
