ABSTRACT
Reinforced soil retaining wall (RSW) frequently collapses due to various causes such as drainage problems owing to heavy rainfall and increase in earth pressure owing to surface load. Therefore, we need technology that can continuously monitor the behavior of the retaining wall. In general, since a depth image between an image and an object cannot be extracted from an image taken from a monocular vision, it is not possible to accurately analyze the three-dimensional behavior of the real space. The behavior of the reinforced soil retaining wall is defined as facing displacement and settlement based on the collapse mechanism of the reinforced soil retaining wall. Mask R-CNN was applied to detect and track the RSW block before and after the behavior, and performances of matching and displacement calculation were evaluated. The errors were distributed in 2.12 mm in laboratory RSW experiment.
