SCC-Loc: A Unified Semantic Cascade Consensus Framework for UAV Thermal Geo-Localization
Researchers have developed SCC-Loc, a new framework for Unmanned Aerial Vehicle (UAV) thermal geo-localization in environments where GPS is unavailable. This system addresses the challenges posed by thermal-visible modality gaps by employing a unified Semantic-Cascade-Consensus approach. SCC-Loc utilizes a shared DINOv2 backbone for efficient global retrieval and matching, incorporating modules for adaptive viewport alignment, texture-structure filtering, and reliability-aware position selection to achieve highly accurate, zero-shot localization. The framework establishes a new state-of-the-art, achieving a mean localization error of 9.37 meters and a 7.6-fold improvement in accuracy within a 5-meter threshold, supported by a new dataset called Thermal-UAV. AI