Vergleich der Algorithmen automatischer Bildzuordnungen zur Erstellung digitaler Oberflächenmodelle in Stadtgebieten mittels sehr hoch auflösender optischer Satellitenbilder (2012)

Dieser Artikel liegt nur in englischer Sprache vor!!

Abstract:

The extraction of the third dimension from stereoscopic image pairs is a well known technique. Since in a number of countries aerial images and laser scanner data are unavailable, expensive or classified, high resolution optical satellite images provide a viable alternative to generate digital surface and digital terrain models. Especially the automatic extraction of highly accurate 3D surface models in urban areas is still a very complicated task due to occlusions, large differences in height and the variety of objects and surface types.

Problem:

In built-up area, due to density of buildings, many parts of buildings can be partially occluded by other buildings, this causes that many pixels don’t have a correspondence in the other image. Also sudden changes in height at building boundaries by classic area based matching algorithms lead to incorrect matches and incorrect height models as well as to blurring building forms. due to using a window with constant shape, the area based matching algorithm is not able to track the building outlines, leading to smoothed building forms because some pixels in the template may be located on top of the building and others on the ground or a wall.
 The modeled profile (red line) does not follow the true profile of the building and lead to blurring building forms.
 

Project Goals:

• The goal of this research is an Empirical comparison of three known matching algorithms for generating urban DSMs which can extract boundaries of building or building blocks with sufficient accuracy and reliability for practical applications in built-up area.
• Development of an optimal matching procedure, which can extract DSMs with satisfied accuracy and reliability. [Precise DSM]

Motivation:

• Using very high resolution satellites images instead of other sources such as aerial images or laser scanner measurements. [Easy access and relatively cheap]
 
• Increase of stereo matching accuracy, especially at sharp buildings boundaries by using appropriate matching algorithms. [High accurate]

Background:

We examine three effective methods in order to find the optimal method to generate a digital surface model from very high resolution stereo satellite images with efficient and high accuracy in Urban Regions.

• Least squares matching (LSM; Förstner 1982)
 
• Dynamic programming (DP) according to Birchfield, Tomasi (1999)
 
• Semiglobal matching (SGM; Hirschmüller, 2008)