A comparative study of TERCOM-EKF and ViSensorRF for UAV localization in GNSS-denied environments

• A comparative evaluation of TERCOM-EKF (Terrain Contour Matching with Extended Kalman Filter) and ViSensorRF (visual–inertial sensor fusion with Random Forest regression) for UAV localization in GNSS-denied environments. • ViSensorRF achieves accuracy improvements of 94–99.8% with sub-5 m RMSE across multiple trajectories. • TERCOM-EKF demonstrates faster performance, making it suitable for embedded and real-time UAV applications. • The study highlights the trade-off between localization accuracy and computational efficiency in UAV navigation. • Hybrid strategies are identified as a promising direction, combining rapid TERCOM-EKF updates with periodic ViSensorRF corrections to balance efficiency and precision in UAV deployment. Accurate localization is essential for Unmanned Aerial Vehicles (UAVs) in mission-critical applications such as military operations, search and rescue, and environmental monitoring. While Global Navigation Satellite Systems (GNSS) provide reliable positioning in open environments, GNSS-denied conditions remain challenging, such as in obstructed or hostile environments. Various localization strategies have been developed using visual and/or non-visual sensors such as LiDAR, barometers, Inertial Measurement Units (IMU), and cameras, with recent advances in machine learning. However, there is a lack of comparative studies that analyze the trade-offs between learning-based and conventional estimation approaches in such conditions. This study bridges that gap by comparing two fundamentally different methods: a Terrain Contour Matching algorithm-Extended Kalman Filter framework (TERCOM-EKF), and Visual-Inertial Sensor Fusion with Random Forest Regression (ViSensorRF), a machine learning approach based on Random Forest Regression (RFR). Both methods are evaluated in a GNSS-denied simulation environment using PX4, Robot Operating System (ROS) 2, Gazebo, and QGroundControl over a digital elevation model of the Taif region in Saudi Arabia. Results demonstrate that ViSensorRF achieves superior localization accuracy, with a root mean square error (RMSE) ranging between approximately 1 to 5 meters across multiple trajectories, but requires significantly higher execution time of 1.92 s. In contrast, TERCOM-EKF offers faster computation of 0.04 s, making it suitable for real-time deployment in resource-constrained UAV systems. These findings highlight a trade-off between localization precision and computational efficiency, offering valuable insights for UAV deployment in GNSS-denied environments.