Visual Navigation Technology Based on AI Heterogeneous Image Matching and Multimodal Data Fusion

With the development and maturity of computer vision technology, visual sensors have become standard equipment for various types of UAVs. Introducing visual navigation assistance on the basis of traditional inertial/GNSS integrated navigation, and obtaining carrier pose information from the rich environmental information and carrier motion information provided by visual images, can effectively solve the problem of high-precision navigation in GNSS-denied scenarios. A novel multimodal navigation enhancement technology integrating vision, IMU and GIS map is proposed in this paper. The AI model for heterogeneous image matching (SuperPoint+SuperGlue) is used to achieve accurate matching between real-time aerial images and reference satellite images, and Kalman filtering is combined to complete multimodal data fusion. Experimental results show that in low-altitude environments of 100~1 000 m with GNSS signal failure, the system positioning accuracy error is less than 5 m (RMSE).