Registering georeferenced photos to a building information model to extract structures of interest

Abstract

Vision-based techniques are being used to inspect structures such as buildings and infrastructure. Due to various backgrounds in the acquired images, conventional vision-based techniques rely heavily on manual processing to extract relevant structures of interest for subsequent analysis in many applications, such as distress detection. This practice is laborious, time-consuming, and error-prone. To address the challenge, this study proposes a new method that automatically matches a georeferenced real-life photo with a building information model-rendered synthetic image to allow the extraction of relevant structure of interest. Field experiments were conducted to validate and evaluate the proposed method. The average accuracy of this method is 79.21% and the processing speed is 140 s per image. The proposed method has the potential to reduce the workload of image processing for vision-based structural inspection.