PHIM: heterologous remote sensing image matching method based on PC-Harris for UAV visual navigation

The navigation system is essential for unmanned aerial vehicles (UAVs), providing necessary support for flight control, trajectory planning, etc. However, the widely used global satellite navigation system (GNSS) is susceptible to complex electromagnetic environments. Visual navigation based on scene matching, as a complementary, could realize UAV positioning in satellite-denied environments. This paper proposes a robust heterogeneous remote sensing image matching method (PHIM). Instead of traditional intensity information, Harris detection combined with phase congruency (PC) is employed for feature detection. During the feature extraction stage, a multi-layer Gaussian pyramid is built, Log-Gabor filtering is utilized, and a feature main orientation map is generated. Then PHIM employs the commonly used nearest neighbor (NN) matching and the fast sample consensus (FSC) method to eliminate mismatched points. Finally, the homography matrix is utilized to correct the matched point pairs for localization. The experimental results on practical data show that PHIM has effective performances on heterologous images, achieving maximum correct match pairs in all scenes. Showing precise spatial alignment, the average value on the improved root mean square error (RMSE) is 28.5844, 14.16% lower than PIIFD. On the average running time, PHIM costs 19.28% less than PIIFD. PHIM can accurately recover image pair relationships, which is significant for UAV positioning in satellite-denied environments.