VISION-BASED DISPLACEMENT ESTIMATION OF LARGE-SCALE INFRASTRUCTURE—A CASE STUDY

Abstract

Vision-based displacement measurement is promising for infrastructure applications because of its ability to cover traditionally hard-to-access regions, as well as its ability to enable simultaneous dense displacement measurements over large areas within a short time period. This paper proposes, demonstrates, and evaluates a visionbased strategy to estimate the displacement of large civil infrastructure under in-service loading. The first step of the strategy consists of conducting a photographic survey of the structure during a typical loading event. Next, a Kanade-Lucas-Tomasi (KLT) feature tracker is employed on photos to track displacements at various locations on the structure. Finally, computer vision-based techniques including camera motion compensation with 2D geometric transformations, correction for lens distortion, and localized histogram equalization of image intensity are implemented to tackle the inherent challenges with field-collected image data. This paper demonstrates the proposed vision-based strategy on a large, steel miter gate at the lock and dam on the Columbia River at The Dalles, OR. To evaluate the accuracy of the displacement estimation strategy, virtual images of a 3D photorealistic model of the gate with known displacements from a finite element analysis (FEA) model were considered. These displacements were then compared to the FEA displacements projected from the 3D model onto the image plane. The differences between these displacements, which ideally should be zero, directly indicate the error of the proposed strategy. Subsequently, displacements estimated from the field images are compared to those predicted by the FEA. Future work will investigate using these differences to generate high-density data for use in Bayesian model updating. The proposed approach is readily adapted for understanding the deformation of other large-scale civil infrastructure systems under inservice loading and preparing for a FEA model-updating schema.