We present a new strategy for active vision-based localization and navigation of a mobile robot in a visual memory, i.e., within a previously-visited area represented as a large collection of images.