Research on Improved Indoor Localization Algorithm for UAVs Based on VINS

To address the limitations of UAV localization accuracy in complex indoor environments, an improved vision-inertial fusion-based UAV indoor positioning method was proposed. The conventional Vision-aided Inertial Navigation System (VINS)-Fusion algorithm was constrained by performance limitations under weak texture and dynamic lighting conditions, leading to degraded system positioning accuracy. To enhance positioning performance, Shi-Tomasi feature points were replaced with ORB (Oriented FAST and Rotated BRIEF) feature points. The binary BRIEF descriptor was employed to improve feature distinctiveness, and the gray centroid method was applied to assign orientation information, ensuring rotational invariance and enhancing image matching robustness in high-speed motion scenarios. Furthermore, a deep learning-driven Feature Booster technique was innovatively introduced to optimize feature descriptors. Through self-enhancement and interactive enhancement mechanisms, the correlations among features were fully exploited, effectively improving descriptor distinctiveness. Finally, a UAV hardware platform based on the RK3588 chip was developed, and a motion capture system was utilized to fine-tune UAV attitude control parameters. Simulations using the EuRoc dataset and real-world indoor flight experiments were conducted to verify the positioning performance and feasibility of the improved algorithm. Experimental results demonstrated that, in complex indoor environments, the root mean square error (RMSE) of the improved algorithm was reduced by 12.24% compared to the original VINS-Fusion algorithm, achieving higher positioning accuracy under dynamic lighting and sparse texture conditions.