Digital

Wissenschaftliche Publikation

Computer Vision for Large Format Digital Aerial Cameras.

Publikation aus Digital

Roland Perko

PhD thesis, Graz, University of Technology, Austria , 10/2004

Abstract:

Since the end of 2002, the consumer market has seen a massive replacement of analog by digital cameras. However, in photogrammetry this transition is proceeding only slowly. This work emphasizes the  advantages of high-resolution digital image acquisition over the traditional analog photogrammetry and discusses concepts of different digital aerial cameras. All computer vision aspects of using such cameras for a photogrammetric approach, in particular image quality evaluation, image stitching, image reconstruction and methods for digital color sensing are studied. The demand of a quantitative comparison of image quality is satisfied by contributing a novel framework for image characterization. The evaluations performed prove the superior quality of digital over film images. Image stitching, the task of merging smaller images into one large seamless image, is an integral part for area-based digital aerial cameras. This problem is investigated deeply and a highly accurate solution is achieved. A novel fast interpolation method is derived and its functionality and efficiency are mathematically proven. This new development allows a significant acceleration of the calculating time. Finally, the difficulty of digital color sensing is discussed and two standard methods are analyzed in detail, also introducing new error metrics. First, the Bayer pattern technique is presented, a method where only one sensor is needed to capture a full color image. An innovative evaluation technique also proves the geometrical accuracy of this concept. Second, the pansharpening approach is described, where the idea is to capture high-resolution panchromatic images and use them to sharpen low-resolution multi-spectral images. Both methods produce high quality color images. In sum, this thesis makes a crucial contribution to the scientific field of computer vision by introducing innovative algorithms to permit the functionality of large format digital aerial cameras. The proposed workflow is applied within the commercial large format digital aerial camera system UltraCamD by Vexcel Imaging Austria introduced in 2003.

Download (7459 kB)