A Novel Scene Matching Navigation System for UAVs Based on Vision/Inertial Fusion

Scene matching navigation system (SMNS) remains challenging in many navigation tasks, which rely heavily on accuracy, computational efficiency, and robustness. Due to the different generation conditions of the matching images, it is difficult for traditional methods to cover every aspect of the three navigation performances. This article aims at developing an accurate, fast, and robust SMNS based on vision/inertial fusion to provide complete navigation information for unmanned aerial vehicles (UAVs). Using the mechanization results of the low-cost micro-electro-mechanical system (MEMS), the proposed system first completes the georeferencing of the real-time aerial images, in which the projection errors are reduced greatly by introducing an optimized factor to the homography matrix. Then, applying a robust noise processing strategy, an improved feature extraction algorithm is designed to eliminate most of the features that vary with climate, time, and season, which lays a solid foundation for the accuracy of the following matching procedure. Under the framework of the SMNS, a novel matching strategy based on logic graphs is designed, which can facilitate the matching procedure. Eventually, by combining the mechanization results of the MEMS and the matching results of the SMNS, the proposed system can provide complete navigation results. Experiments in typical and complex scenarios are carried out, respectively, to verify the effectiveness and robustness of the proposed system. Experimental results demonstrate that the proposed SMNS possesses accuracy, computational efficiency, and robustness, which outperforms the state-of-the-art strategies [i.e., histogram of orientated phase congruency (HOPC), chanel feature of orientated gradient (CFOG), phase congruency (PC)] in terms of matching aerial and satellite images.