IMAGE PROCESSING FOR LABORATORY GEOTECHNICAL MEASUREMENTS

Abstract

This research is an effort to apply image processing techniques for the non-contact determination of soil movement. A programming language called Labview (Laboratory Virtual Instrumentation Engineering Workbench) was used to develop computer codes for measuring the displacement in 2D and 3D of specimens in a triaxial apparatus. The first program was developed for 2D measurement using two cameras that pointed in parallel to the specimens. In this stage, a calibration procedure was established for the correction of distort images. The pyramidal optical flow algorithm (Lucas and Kanade, 1981) was used to trace the positions of interest points, from which later derived displacements. The results were compared with the measurements obtained from LVDTs and were well agreed. For the 3D measuring system, stereo vision processes consisted of stereo calibration, stereo rectification, stereo correspondence and 3D re-projection were performed before applying the optical flow algorithm. The 3D system was verified by an isotropic compression test of a cylindrical sponge in the triaxial apparatus. The estimated displacement and discharge of water were compared with the measured ones and were well agreed. After programs had been developed and tested, a shearing test on a real soil was made by the triaxial apparatus. From the experiment it was found that the binocular stereo vision developed has a good potential in making measurements. The repeatability and the resolution of the developed system were in the order of 0.006 cm and 0.004 cm, respectively. Since the developed system could determine deformation from only the area where the cameras can view, the current hardware configuration was not allowed for determining the discharge of water from un-symmetrical deformation. However, the developed system can be extended readily by adding more cameras so that photos can be taken from the missing view directions.