ABSTRACT
To date, numerous research papers on urban change detection have been published. Close-to-nadir satellite images are commonly used in the publications to avoid the misregistration caused by relief displacements in the images. This severely limits the data sources for change detection, because the majority of high-resolution satellite images are taken with an off-nadir angle and most portions of aerial images have off-nadir views. This causes misregistration between two images used for change detection and therefore significantly reduces the accuracy of change detection. To solve this problem, a change detection framework is designed that uses the patch-wise co-registration (PWCR) method to overcome the misregistration problem and integrates other methods to enable accurate change detection using images taken from different sensors onboard different platforms. In the PWCR method, a digital surface model and the exterior orientation parameters of the images are used to guide the detection of the corresponding points in the bi-temporal images. Then, from the corresponding points, corresponding patches are generated. Finally, the multivariate alteration detection method is employed to detect the changed patches based on the spectral properties of the corresponding patches. The approach was tested on different combinations of off-nadir satellite and airborne images. The accuracy achieved was from 89% to 92% in co-registration and more than 90% in change detection. This approach demonstrates the potential to open up the remote sensing data sources for urban change detection.
