Automated registration of SFM-MVS multitemporal datasets using terrestrial and oblique aerial images

Accurate alignment of 3D models is critical for valid change-detection analysis from multitemporal photogrammetric datasets. This paper assesses an automated registration strategy which uses the scale-invariant feature transform (SIFT) algorithm implemented in modern photogrammetric software. This registration solution, also known as ‘Time-SIFT’, was tested at two study sites featuring vertical surfaces, including a sea cliff (~500 m2) and a quarry face (~50 000 m2). Tests demonstrated that the investigated registration strategy can achieve accurate alignments between multitemporal point clouds even when using multisource and multi-perspective data, captured across widely varying spatial and temporal scales and under a range of weather and illumination conditions. The combination of the Time-SIFT approach with an ICP algorithm produced moderate improvements in the alignment. Furthermore, the use of an innovative direct georeferencing technique, that used the tracking feature of a robotic total station, allowed for accurate georectification of 3D models.