A Graph-Optimization-Based tightly coupled Multi-Source positioning method for UAVs in GNSS-Denied environments

To address UAV positioning in GNSS-denied environments, this study proposes a tightly coupled IMU–vision–UWB localization method. A structured visual feature optimization strategy is first developed to improve feature uniformity and robustness through adaptive thresholding, density balancing, and grayscale-gradient fusion. A graph-optimization-based fusion framework is then constructed, incorporating UWB–IMU joint initialization and UWB differential-residual-enhanced global constraints. With an adaptive sliding window and IMU pre-integration compensation, the framework improves localization accuracy and real-time performance while alleviating temporal asynchrony. Finally, experimental results show that the optimized visual features increase ORB matched points by 11.8% and improve accuracy by 5% over ORB, while achieving better distribution uniformity than SIFT and ORB. On the VIRAL dataset, front-end replacement experiments demonstrate that the Structured Visual Feature Optimization Method shows good engineering practicality within a fixed back-end framework and the proposed location method achieves the lowest RMSE across all nine sequences, demonstrating superior overall localization accuracy and trajectory-level robustness in real-world UAV scenarios. In real flight experiments, it attains RMSEs of 0.480 m under high-dynamic conditions and 0.286 m under steady-state conditions. Ablation results show that the adaptive sliding window improves frame rate, while the other modules enhance accuracy. Comparative experiments confirm that the proposed method achieves a better trade-off between accuracy and efficiency in both high-dynamic and steady-state scenarios.