Advancement of vision-based structural displacement measurement techniques for field applications

Abstract

Displacement is one of the most important physical quantities for understanding the health of a structural system and displacement calculation is central to structural damage assessment. With the rapid development of computer vision, vision cameras have been used for structural displacement measurement. However, the target must be manually determined, and it is difficult to apply during an earthquake. In addition, it is sensitive to the intensity of illumination, so there is a limit to continuous displacement estimation. This dissertation aims to expand the applicability of vision-based structural displacement estimation, and three different techniques are proposed: (1) An automated Region of Interest (ROI) selection technique is developed. The image frames that capture larger movements of the surrounding areas were selected, and the features in the selected frames were grouped using clustering algorithms. The feature group with consistent movement and high density was finally selected as the optimum ROI. (2) A seismic-induced permanent displacement estimation technique combining acceleration and computer vision measurements is proposed. Before an earthquake, the vision camera and accelerometer are vibrated in a controlled manner using a shaker on the target building. Then, a scale factor map, which converts a translation on an image (in pixels) into a physical displacement, is automatically generated. After the earthquake, a displacement map is constructed to estimate the relative movement of the target building to the natural targets, and then clustered using Gaussian mixture modeling. Finally, undeformed portions of the natural targets are retrieved to estimate the permanent displacement of the target building with respect to the distribution of the natural targets’ movement. (3) A continuous structural displacement estimation technique is developed by combining measurements from an accelerometer, vision, and Infra-Red (IR) cameras collocated at the displacement estimation point of a target structure. In initial calibration step, the IR camera temperature is optimized by comparing the filtered acceleration-based displacement with the IR-based displacement. After that, in the continuous displacement measurement step, a scale factor and optimum temperature are applied, and vision-based displacement and IR-based displacement are used during the day and night, respectively. At this time, through updating the reference frame, a vision camera sensitive to illumination during the day and an IR camera sensitive to temperature at night enable continuous displacement measurement. The displacement estimation performance of the proposed techniques was experimentally validated. In addition, the effectiveness of all algorithms included in these techniques was investigated, and the application range and limitations of each technique were discussed.