Development of a Stereo Vision-Based Method for Tri-Axial Displacement Measurement of Landslide Surface Cracks

Abstract

Surface cracks are critical macroscopic indicators of landslide initiation and progressive deformation, yet most monitoring methods emphasize crack width while ignoring vertical, tensile, and shear displacement data. To overcome this, a stereo vision-based method is proposed for noncontact tri-axial displacement monitoring of landslide surface cracks. The method integrates a binocular camera with synchronized triggering, a mounting unit, and an image-processing algorithm. The image-processing algorithm combines semi-global block matching (SGBM)-based disparity reconstruction for vertical displacement, Canny edge detection with medial-axis transform (MAT)-based centerline extraction for tensile displacement, and marker recognition and tracking with coordinate transformation for shear displacement. In addition, laboratory validation, optimization, and environmental sensitivity tests were conducted. The results demonstrate that the proposed stereo vision method can reliably and linearly measure vertical, tensile, and shear crack displacements with high accuracy, with measurements closely following the 1:1 reference trend across the tested ranges. When the object distance was 100–120 cm, all displacement components achieved their optimal measurement accuracy, whereas errors increased at both shorter and longer object distances. Environmental sensitivity tests identified a soil moisture content (SMC) of 40%, an illuminance of 2000 lx, and a dynamic shadow ratio below 60% as favorable conditions, while tensile displacement was more sensitive than vertical displacement to moisture and lighting variations.