3D model-based image registration for change detection in historical structures via unmanned aerial vehicle

Abstract

Visual inspection is a common technique to detect and examine the state of health of the structural system. Periodic inspection is carried out to determine if anomalies, such as cracks and surface paint, found in previous visits have changed in appearance over time. The image-based change detection techniques require accurate geometrical and photometrical corrections in pre-processing steps to minimize errors. Although several techniques have been proposed to remove geometrical errors, they still fail to align image correctly, which often results in inaccuracy in a change detection system. In this paper, a change detection system is proposed to tackle this problem. The system acquired images via an unmanned aerial vehicle. Then, the acquired images were manually processed to identify damages, such as cracks, which were used to guide a drone to obtain more images of damage areas for monitoring purpose. The images were then used to obtain a 3D surface model and camera calibration through Structure from Motion (SfM), which were used in the image synthesis technique to obtain an image with identical camera parameters as a queried image for accurate geometrical adjustment. The synthesized images were used to compare with the queried image to see if there were changes between them. In this research project, it was showed that the drone can be used to monitor problematic areas and the image synthetic technique via 3D modeling can be used in geometrical registration to improve a change detection system.