LiDAR Point Cloud Descriptor for UAM Place Recognition with Point Cloud Map

Accurate localization is a critical element for the successful and safe operation of Urban Air Mobility (UAM). In this study, we present a method for UAM place recognition that utilizes point cloud map (PCM) data and a virtual LiDAR sensor model. The PCM-based approach enables the creation of a virtual descriptor database (VDD) for place recognition. To generate descriptors invariant to translation and rotation, we introduce a region of interest sampling method and a feature point detection approach, effectively minimizing altitude influence. We also outline a technique for creating translation and rotation invariant descriptors through the integration of robust feature extraction methods. Furthermore, we conduct an experiment utilizing a game engine-based UAM simulator to validate the proposed method. PCM and VDD are generated through the simulator, and a quantitative analysis of descriptors and place recognition is subsequently carried out.