Page 1 of 2
Next

Changing Perspective of Aerial Photogrammetry


Christian Heipke
Institute of Photogrammetry and GeoInformation
University of Hannover
Germany


Much has happened in aerial photogrammetry and remote sensing for topographic mapping tasks over the last few years.
In the sensor domain the most noteworthy developments comprise the development and successful commercial use of digital aerial cameras and of laser scanners, the possibility to measure image orientation parameters directly in the air by means of a combination of GPS and inertial measurement techniques, the availability of sub-meter imagery from space on a regular basis. As far as the generation of geospatial information from imagery is concerned, a number of trends can be observed:

• a larger degree of automatic processing,
• a closer link to GIS, including update and refinement of existing GIS data bases,
• the largely increased demand for digital imagery and geospatial data for use in the internet, exemplified by applications such as Goggle Earth and Microsoft’s Virtual Earth.

Without any doubt, the automatic generation of geospatial information in conjunction with a closer link to the GIS world is the most demanding of the mentioned trends, especially if real-time applications such as disaster relief operations are to be handled. Such tasks can only be achieved on the basis of digital images, and sometimes need direct height information. Therefore, the development of digital aerial cameras and of laser scanners can be seen as pre-runners to solve the mentioned tasks. The focus of this article lies on the automatically producing geospatial information and describes the scientific background of the field as well as some applications and future directions.

Background
Photogrammetric image processing is divided into two aspects, i.e. the geometric/radiometric image evaluation and image analysis. Geometric/radiometric image evaluation comprises image orientation, the derivation of geometric surface descriptions and orthoprojection. Image analysis contains the extraction and description of three-dimensional objects. When using digital images, the borders of geometric/radiometric image evaluation and image analysis become blurred, mostly because due to automation the formerly decisive manual measurement effort has lost much of its significance. Therefore, already during the orientation phase a point density can be used, which is sufficient for some digital surface models (DSMs).

Geometric/radiometric image evaluation
Methods for the integrated determination of image orientation, DSMs, and orthophotos are as follows. The components of image orientation are the sensor model, i.e. the mathematical transformation between image space and object space, and the determination of homologous image primitives (mostly image points). As far as the sensor model is concerned, the central projection distinguished from line geometry.

In the context of bundle adjustment the central projection is traditionally described by means of collinearity equations. The determination of homologue points is almost exclusively done by digital image matching. The methods for aerial images and for the satellite sector are almost fully developed and are available for practical purposes under the term “automatic aerial triangulation”. It should be noted that the automatically generated image coordinates of the tie points are often interactively supplemented or corrected.

As an alternative to aerial triangulation the direct and integrated sensor orientation were thoroughly investigated in the last decade. In both cases data from GPS receivers and IMUs (inertial measurement units) are used for determination of the elements of exterior orientation. For direct sensor orientation these data replace tie and (more importantly) also ground control points and thus the entire aerial triangulation. For integrated sensor orientation all information is used in a combined adjustment.

Like image orientation the derivation of geometric surface descriptions from images is based on digital image matching. If a digital terrain model (DTM) is to be derived from the DSM, interfering objects (for the terrain these can be buildings, trees etc.) must be recognized and eliminated. At present this task is solved by comparatively simple image processing operators and statistical methods. As in image orientation, the automatic step is usually followed by a post-editing phase to eliminate blunders and fill in areas in which matching was not successful.