Real-Time Terrain-Aided Navigation for UAVs in Dense Urban Environments via Dynamic Feature Fusion and Adaptive Kalman Filtering

**Abstract:** This research details a novel Real-Time Terrain-Aided Navigation (TRN) system for Unmanned Aerial Vehicles (UAVs) operating within complex, GPS-denied urban environments. Our approach, Dynamic Feature Fusion and Adaptive Kalman Filtering (DFF-AKF), integrates LiDAR point cloud data, visual odometry, and inertial measurement unit (IMU) data streams through a dynamically weighted feature fusion strategy. A key innovation is an adaptive Kalman filter implementation that adjusts process and measurement noise parameters based on real-time environment density and visual cue reliability, enabling robust navigation even with intermittent visual or LiDAR data. This system demonstrates superior accuracy and robustness compared to traditional TRN methods, offering a viable solution for urban package delivery, search and rescue operations, and infrastructure inspection with minimal reliance on external positioning systems.