Development of a Computer Vision-Based Three-Dimensional Reconstruction Method for Volume-Change Measurement of Unsaturated Soils during Triaxial Testing

Abstract

Problems associated with unsaturated soils are ubiquitous in the U.S., where expansive and collapsible soils are some of the most widely distributed and costly geologic hazards. Solving these widespread geohazards requires a fundamental understanding of the constitutive behavior of unsaturated soils. In the past six decades, the suction-controlled triaxial test has been established as a standard approach to characterizing constitutive behavior for unsaturated soils. However, this type of test requires costly test equipment and time-consuming testing processes. To overcome these limitations, a photogrammetry-based method has been developed recently to measure the global and localized volume-changes of unsaturated soils during triaxial test. However, this method relies on software to detect coded targets, which often requires tedious manual correction of incorrectly coded target detection information. To address the limitation of the photogrammetry-based method, this study developed a photogrammetric computer vision-based approach for automatic target recognition and 3D reconstruction for volume-changes measurement of unsaturated soils in triaxial tests. Deep learning method was used to improve the accuracy and efficiency of coded target recognition. A photogrammetric computer vision method and ray tracing technique were then developed and validated to reconstruct the three-dimensional models of soil specimen.