Efficient measurement for surface displacement of slope retaining structures with 3D point cloud semantic modeling

Abstract

Slope point clouds directly reconstructed based on multi-view images contain many complex backgrounds, leading to inefficient post-processing operations and unreliable displacement measurements. A novel image mask-guided semantic modelling approach is proposed for reconstructing point cloud model of slope retaining structures from multi-view images containing complex scenes by UAVs. The core of the approach mainly contains three steps: slope localization based on YOLOv8, segmentation based on DeepLabv3+, and reconstruction using SfM-MVS based on mask-guided mechanism. The reconstruction results show that the proposed approach takes nearly half the time and has higher semantic accuracy than traditional post-processing strategies. The reconstructed point cloud semantic model of the retaining structure overcomes the limitations of the traditional single-point displacement monitoring methods, and can be used for the global displacement measurement of the retaining structure, reflecting the stability of the slope through the multi-temporal macroscopic displacement changes, and providing basis for the maintenance decision of the slope.