Error propagation for DEM-based georegistration of motion imagery

A new method for georegistering motion imagery has recently been introduced. It requires a digital elevation model (DEM) from which it generates and registers predicted images to actual images. For aerial imagery, including wide area motion imagery (WAMI) and full motion video (FMV), the method fits a multi-parameter camera model composed of exterior and interior orientation parameters including radial distortion. In this paper we describe an algorithm for estimating the geospatial accuracy of DEM-based georegistration algorithms. It employs statistical methods to calculate the error distributions of the georegistered camera model parameters and propagate them to the ground. Monte Carlo methods are employed to achieve accurate results for oblique imagery with occluded features. For FMV imagery georegistered to a high-resolution LIDAR DEM, we obtained a horizontal error of 2.3 m CE90 and vertical error of 0.66 m LE90. These values increased in regions of the images where the obliquity angles and ranges increased. Comparison with GPS data demonstrated that the error estimates were reliable. For WAMI data georegistered to a low-resolution USGS DEM, we obtained an accuracy of 16.6 m CE90 and 8.2 m LE90. In this case poor DEM accuracy limited the georegistration accuracy.