Evaluating effects of glare on the monitoring of building facade health condition by analyzing the infrared thermal images collected under different weather conditions

Abstract

Infrared thermal images are widely adopted to monitor the health condition of building facades. Image segmentation can usually segment IRT images by grouping those pixels with similar surface temperatures so that the segmented regions, which correspond to different surface temperatures, can be used for defect detection. Recently, researchers have proposed that intensity inhomogeneity can be approximated, implying that extra information (like glares, shadows, etc.) is included in the pixels of IRT images. The approximated intensity inhomogeneity can be used to enhance or smooth the given IRT images so that the images can be easily interpreted. We propose an innovative image model incorporating intensity inhomogeneity. Assuming that intensity inhomogeneity can be linearly interpreted, it can be presented using Taylor’s expansion. For simplicity, only the first and second terms are included in the image model. The optical-radiative properties of façade materials are usually unknown, while the IRT image containing different façade materials is processed. The entropy of the IRT image reflects these properties. The entropy of a given image can be high if the areas with high-intensity inhomogeneity need more pixels to be included in the image model. By contrast, the areas with low-intensity inhomogeneity need fewer pixels to be included in the image model. Hence, the entropy of the IRT image is used to determine the window sizes. The proposed image model involving Taylor’s expansion and multiple window sizes can be used to determine intensity inhomogeneity and segmented regions through the introduction of level-set functions and an iterative scheme. The processed results demonstrate that while the glare effects dominate the intensity inhomogeneity, the segmented results are affected, and the corresponding image regions cannot be used for defect detection. IRT images are collected on sunny days. In this study, the proposed approach is used to evaluate three sets of IRT images collected on rainy days, and the processed results indicate that intensity inhomogeneity exists in those collected images, but the thermal patterns of defects are not fully developed. The proposed image model incorporates image segmentation. The given images can be segmented, and their intensity inhomogeneity can be computed by introducing level-set functions and an iterative scheme. When the approximated intensity inhomogeneity is less than 1.0 in an IRT image, the areas are enhanced. Conversely, if the approximated intensity inhomogeneity is larger than 1.0, the areas are smoothed. The segmented results offer an important clue for defect detection. Furthermore, incorrect segmentation because of intensity inhomogeneity can be minimized.