Autonomous drone navigation in cities often stumbles because GNSS signals bounce off buildings and lose accuracy. A new arXiv preprint, MatchCityLoc, tackles this by turning to visual localization against detailed city models, specifically Level-of-Detail (LoD) models that capture building geometry.
The method works by matching surface normals—the directional orientation of building facades and rooftops—observed from the drone's camera to the normals rendered from the LoD model. This alignment yields a precise pose estimate without relying on satellite signals.
The paper is announced as a new preprint, and the abstract emphasizes the need for precise pose estimation in urban environments. As with any single-source preprint, the results have not yet been peer-reviewed, and the full methodology and evaluation details remain to be examined.