Photomosaicing and automatic topography generation from stereo aerial photography

Abstract

The Autonomous Vehicle Systems Lab specializes in using autonomous planes for remote sensing applications. By developing an inexpensive image acquisition platform and the algorithms to post process the data, remote sensing can be performed at a lower monetary cost with shorter lead times. This thesis presents one algorithm that has shown to be an effective alternative to the traditional Bundle Adjustment (BA) algorithm used for making composite images from many individual overlapping images. BA simultaneously estimates camera poses and visible feature locations from blocks of overlapping imagery, but is computationally expensive. The alternate algorithm (ABA) uses a cost function that does not explicitly include the feature locations. For photographic sets covering large areas, but having overlap only between adjacent photos, the search space and consequently the computational cost is significantly reduced when compared to typical BA. The usefulness of the algorithm is demonstrated by comparing a digital elevation model created through the ABA with LIDAR data.