Autonomous Localization and Navigation for Drones Over Rivers

Unmanned aerial vehicles (UAVs) are increasingly used for environmental monitoring and riverine studies; however, their dependence on Global Positioning System (GPS) localization is often compromised in forested areas, under bridges, or above highly dynamic water surfaces. Conventional vision-based navigation methods struggle with sparse textures and transient disturbances, leading to unstable positioning and drift. This paper presents a robust real-time visual–inertial navigation framework that integrates neural network–based optical flow estimation (NeuFlow v2), the Dynamic Divided Method (DDM), and inertial measurement unit (IMU) data through a Kalman Filter (KF). The DDM adaptively segments the optical-flow field according to water-surface dynamics and local texture stability, suppressing transient artifacts such as ripples and reflections, while the KF stabilizes state estimation by compensating for outliers and short-term measurement loss. Field experiments at Kangqiao, Green River, and Wu River sites demonstrate that DDM-enhanced visual odometry (VO) significantly improves robustness in low-texture and dynamic conditions. The fused framework combining NeuFlow v2, DDM, and KF achieves high trajectory similarity with GPS benchmarks, reduced drift, and enhanced spatial consistency. These findings validate a practical and effective solution for accurate UAV localization and autonomous environmental monitoring in GPS-degraded riverine environments.