Advancements in guidance systems for high-speed aircraft based on computer vision methods

This article explores modern computer vision algorithms applied to the terminal guidance of aerial vehicles. The focus is on approaches that implement 3D reference-to-scene matching for final orientation. Two methods are considered: comparison of a partially reconstructed 3D model with a reference model, and projection of the reference model into image space for subsequent comparison. Both deterministic and neural network-based algorithms for 3D model reconstruction are discussed in the context of terminal guidance. Particular attention is given to image matching algorithms under dynamically changing environmental conditions. The article also highlights a navigation approach based on terrain map and real-scene matching using neural network architectures, such as GeoGAN.