Bridge Column Maximum Drift Estimation via Computer Vision

Abstract

This paper considers the viability of applying computer vision techniques for estimating the peak experienced seismic displacement of damaged bridge columns, for use in postdisaster triage assessment. The primary objective of the associated study was to determine if there is a statistically robust relationship between peak seismic displacement and damage observations extracted from two-dimensional (2D) images captured after an event. To this end, correlations were developed using images and experimental test data from lateral-load tests performed on a series of reinforced concrete bridge columns. Computer vision algorithms based on a combination of image segmentation, feature extraction, and nonlinear regression analysis were used to estimate peak drift. The results presented in this paper indicate strong correlations between parameterized crack patterns and experienced structural displacement, regardless of the position of the camera. Key findings include the necessity of using nonlinear machine learning–based regression analyses, as well as the need to model spall damage at high drift levels. It was also found that large variations in camera lighting or column design can inhibit estimation accuracy.