A Method for Enhancing the Positioning Accuracy and Robustness of Indoor Inspection by Unmanned Aerial Vehicles through Multi-modal Fusion

To address the challenges faced by UAV-based automatic inspection in complex indoor substation environments-such as the absence of satellite positioning signals and interference from dynamic obstacles-this paper proposes an indoor 3D mapping and high-precision localization method based on UAVs. The approach integrates multi-scale SIFT feature extraction, Difference of Gaussian keypoint detection, and descriptor matching, and constructs a complete indoor environment model through graph-optimized SLAM. By building both image pyramids and map pyramids, the system enables robust feature point detection and matching across multiple scales, significantly improving the accuracy and robustness of pose estimation. A graph optimization framework is introduced to globally refine the feature graph, effectively reducing accumulated drift and enhancing map consistency. Experiments were conducted in a typical indoor substation environment using a DJI Mavic 3 drone equipped with a vision sensor. Multiple rounds of field tests were carried out, and the system was integrated into a self-developed UAV inspection situational monitoring platform to achieve real-time visualization of localization and mapping results. Under challenging conditions including lighting variation, occlusion, and frame loss, the system maintained stable performance, achieving an average Absolute Pose Error (APE) of less than $\mathbf{1 0 ~ c m}$. These results validate the proposed method’s high robustness and practical value in complex indoor environments.