Virtual GPS for UAV Navigation in GPS-Denied Environments: Sensor Fusion of Inertial Navigation System and Google Maps-Based Virtual GPS Using Kalman Filtering

Abstract

Modern unmanned aerial vehicles (UAV) have become essential tools across many fields yet they share one major weakness: a complete dependency on Global Positioning System (GPS) signals for accurate positioning. This dependency often makes UAVs invulnerable because GPS signals can become unavailable due to satellite positions, jamming, or failures etc. In such cases, UAVs can lose their accurate navigation capability, which often results in vehicle loss. This research concentrates on a solution to overcome these limitations by using Virtual GPS approach. The proposed system provides accurate global positioning capabilities without requiring satellite-based signals. The approach replaces GPS integration with an innovative visual-inertial localization system that uses publicly available satellite imagery from Google Maps as a global reference dataset.

The simulation is built around three integrated components. First, a virtual UAV camera view is generated from Google Maps satellite imagery using geometric transformations. Second, features are extracted from both the virtual UAV camera and satellite imagery using the Speeded-Up Robust Features (SURF) algorithm. Feature matching between the two image sources provides input to the Random Sample Consensus (RANSAC) algorithm to establish reliable correspondences while systematically rejecting false matches. The validated correspondences enable the computation of similarity transformations, which directly yield estimates for the UAV's global position and heading. The third component is sensor fusion, which provides a continuous localization solution by integrating the low-frequency visual position estimates with high-frequency inertial measurements through an Extended Kalman Filter (EKF). The EKF maintains a six-dimensional state vector, and visual measurements are validated using Mahalanobis distance calculations to reject outliers. The control component uses Nonlinear Model Predictive Control (NMPC) for trajectory tracking. The NMPC algorithm computes optimal control inputs by minimizing cost functions that penalize tracking errors, and it adaptively manipulates control weights based on position and heading errors to maintain optimal performance.

Simulations using various trajectories demonstrate that the proposed system is effective and reliable. The results indicate that the system provides stable and accurate localization even when subjected to sensor noise, IMU bias, and temporary visual ambiguities. This work represents a major step forward in making UAVs truly independent of GPS by combining computer vision, statistical estimation theory, and optimal control. Therefore, it offers a practical solution that could enhance UAV operations in places where GPS signals are unavailable or unreliable.