Hybrid Feature-Based Visual SLAM with GPS Fusion: Multi-Descriptor Stereo Matching for Robust UAV Navigation

The study presents an innovative method which uses visual SLAM technology to enable Unmanned Aerial Vehicles (UAVs) to achieve simultaneous localization and mapping through its dual capability of processing stereo infrared visual data and Global Positioning System (GPS) data, which it accomplishes through its development of a new multi-descriptor feature matching method. The visual SLAM system experiences two main difficulties which include scale uncertainty together with trajectory drift, while GPS systems lose their ability to determine location because of signal interference together with multipath distortion. The implementation of three feature descriptors which include Oriented FAST and Rotated BRIEF (ORB) and Binary Robust Independent Elementary Features (BRIEF) and ScaleInvariant Feature Transform (SIFT) will help us achieve better matching results across different operational environments. The methodology includes Random Sample Consensus (RANSAC) for eliminating outliers and an adjustable Kalman filter for merging sensor information. Practical field tests confirm the system's effectiveness, yielding an 81.7 % inlier rate (versus 68.2 % for ORB alone), decreasing Absolute Trajectory Error by 42 % relative to vision-based odometry, and enhancing accuracy by 47 % compared to independent GPS operation under σ=0.5 ∼m simulated noise. Operating at 43.2 ms per frame (23.2 Hz), the system enables real-Time autonomous UAV guidance in GPSrestricted or visually demanding situations.