Image-Based Change Detection for Bridge Inspection

Abstract

The changes in defects patterns or in element condition index during visual inspection of bridges are primary concerns for inspectors. This paper presents a new approach for change detection of defects in bridges by identifying changes in texture patterns through spectral analysis of digital images. The commonly used method for change detection is image differentiation. This subtraction method requires images to be of same size, scale, and rotation. However, no two images are same in real practices. Thus, image registration is required to align images and to produce change maps. This process is tedious and it is difficult often to achieve a good registered image. But, the change detection task can be readily modeled in frequency domain for texture patterns discrimination and also for quantifying their properties. This paper proposes a novel approach for change detection by transforming digital images into Fourier spectrum. In new coordinate system, 1-D signature functions can be drawn which facilitates easy comparison of textures in different directions. The proposed methodology provides useful tools for comparison of inspection history graphically and quantitatively. In practice, expensive sensors are used to detect subtle change in defect patterns. The proposed method can be used to detect any subtle change in defect patterns using digital images at much lower cost.