Evaluating the potential of the Time-SIFT approach using Pléiades satellite imagery for 3D change detection

Optical satellite remote sensing images allow the analysis of topographic change detection based on digital elevation model differencing. Accurate change detection requires co-registration of 3D models derived from multitemporal datasets, so far, usually achieved with external reference data or manual intervention. Based on structure from motion photogrammetric algorithms, co-registering images from different epochs together by extracting multitemporal tie-points, such as the time–scale-invariant feature transform (SIFT) approach, has been proven to be efficient for aerial images. We propose a first test of the time-SIFT approach on Pléiades tri-stereoscopic satellite imagery, covering the coastal plain in northeastern Tunisia. The multitemporal tie-points, coupled with the rational polynomial coefficient models of all the images, are refined in a joint bundle adjustment routine. The accuracy of the derived digital elevation models is assessed in absolute (using external ground truth data) and relative (between epochs). The time-SIFT approach reduces the normalized median absolute deviation from 1.80 to 1.14 m when compared with the classical approach. Our results demonstrate that the time-SIFT approach can be extended for change detection analysis using Pléiades images, without ground control points or a posteriori co-registration, which paves the way to automated investigation of time series of past and future 3D satellite datasets. We highlight the application of the time-SIFT approach to Pléiades images leveraging their high spatial resolution for change detection.