A Vision-based UAV Landslide Monitoring Technique Using Multi-source Pose Fusion and Its Application

Key Points

• This research aims to develop a low-cost, high-accuracy monitoring technique for landslide detection using UAVs.
• Developed a visual monitoring technique using multi-source pose fusion and global beam optimization.
• Utilized YOLOv8 for target recognition and sub-pixel point location to capture 2D image coordinates accurately.
• Implemented adaptive weighting strategy to fuse UAV poses with high-frequency visual poses for joint optimization.
• Achieved stable absolute displacement measurement error of 3.4-3.8 cm under multi-angle flight mode.
• Significantly outperformed baseline methods reliant on UAV poses or pure visual motion recovery structure.
• Successfully maintained low costs and high flexibility while enabling centimeter-level deformation monitoring.