In this paper, we present a method for infrastructure-free localization of Automated Guided Vehicles (AGVs) in a warehouse environment. To accomplish this, our approach leverages 3D data for both mapping and feature segmentation. First, a 3D reconstruction of the warehouse is created to extract salient natural features - in this case the shelving uprights - as landmarks. Next, the map-based localization approach leverages 3D LIDAR to enable 3D feature-to-landmark matching which minimizes the potential for data association errors. In our experiments in a representative warehouse environment, we demonstrated a localization accuracy of approximately 2cm without the use of retroreflector targets. Furthermore, 100% of visible landmarks were detected and there were no false positives.