As-Built BIM Updating Based on Image Processing and Artificial Intelligence

Image-based technology is effective for recording on-site data geospatially and chronologically. It has gained increasing attention in the construction field for progress monitoring. However, a notable downside of image processing is the light condition, particularly for noisy environments such as construction sites. Poor or undesirable ambient light conditions produce low quality images which significantly affect the accuracy of data extracted from related images. This paper proposes an innovative approach based on thermal image analysis to mitigate problems related to the image quality. Construction materials with various emissivity resulting in measurable temperature differences can be a crucial evidence to identify different materials. In addition, a convolutional neural network was used to segment different components semantically. Together with a camera-view image extracted from BIM model, a cross matching can be applied to confirm the presence of different building elements with specific materials, which is the key step to estimate the construction process qualitatively.