Point cloud registration for bridge defect tracking in as-built models

In the task of assessing the rate of change of bridge defects over time, photographs have the potential to provide us with detailed time series data of defect propagation. However, in order to make an accurate comparison between two photographs, these must ideally be taken using the same camera with the same pose and conditions, which is not necessarily practical in a sequence of bridge inspections. One possible solution to these challenges lies in point cloud registration, particularly where those point clouds are generated through Structure from Motion. Structure from Motion is an established photogrammetric range imaging technique for 3D reconstruction from 2D images. The output consists of a 3D point cloud together with camera poses for the associated images. It is also possible to obtain similar outcomes with LiDAR scanning, which can give better results where uniform surfaces preclude successful keypoint matching for the Structure from Motion method. Recent advances in computer vision allow faster and more effective point cloud registration by better discarding outlier correspondences. Whence, given two point clouds of the same defect, we are able to align them to have the same origin, orientation and scale. Since the model comprises camera poses, these are re-aligned also, thus solving the problem of different camera poses and facilitating quantitative computational comparison of defects between inspections.