VISTA: Monocular Segmentation-Based Mapping for Appearance and View-Invariant Global Localization