×
![]( "Buchcover ISBN 9783906467917")
Texture mapping and true orthophoto generation of 3D objects
von Thomas HanuschPhotogrammetry provides, in contrast to other surveying methods, a complete coverage of the object or region by image data. For a long time, the main purpose of surveys was to extract geometrical information such as elevation models, distances, or point coordinates from images. Since the first computers became readily available, more prospects have opened to use this data and the photorealistic visualization of the acquired side came into clearer focus. This application was unique and proved a big advantage for photogrammetry in comparison with other surveying techniques. Over many years, the representation of 2.5D data, such as digital elevation models derived from aerial images, was the development goal. With the improvement of the computer systems and the available algorithms, photogrammetry paved the way to acquire 3D objects in aerial photogrammetry and close range applications.
The change from 2.5D to 3D lead to an enormous change in the requirements concerning the algorithms used. Due to the much higher complexity in handling 3D data as compared to 2.5D data, a large number of existing approaches and algorithms were no longer usable. This work was motivated by the need to develop new algorithms to visualize 3D data in combination with the acquired image data. The procedure to do this is called texture mapping. The basic idea behind texture mapping is to attach the image information onto the geometrical data. The result is a photorealistic visualization of the acquired object. Even at the time of the start of this work, a wide range of algorithms were available to perform texture mapping. The problem was that the existing algorithms were developed with the focus on a different data configuration and requirement. Most of the existing texture mapping algorithms, that could handle 3D data were developed for computer vision applications, e. g. real time visualizations. In those applications, the available data is quite different than the data we are focused to handle in this work. For example, one difference is the number of available images. In our applications, an object should be covered with a useful number of images in a high quality. In contrast, most of the existing approaches use data with low resolution images, acquired using video devices. This work is aimed at analyzing existing approaches concerning their usability to process photogrammetric datasets. Even in the best cases, most of the algorithms had to be adapted to the special needs of the photogrammetric specifications and a number of new algorithms had to be developed for necessary processing steps.
This thesis presents algorithms and software modules covering the workflow for texture mapping from the point of provided images and surface models to the photo-realistic textured 3D model. The completed work results in a wide variety of algorithms of which there are two types of algorithms distinguished:
The second set of algorithms deals with colour and brightness corrections, and combines different images acquired under different conditions to achieve a seamless textured 3D model. Those algorithms are:
The author wishes to thank the Swiss National Science Foundation for their financial support over 3 years of this work.
The change from 2.5D to 3D lead to an enormous change in the requirements concerning the algorithms used. Due to the much higher complexity in handling 3D data as compared to 2.5D data, a large number of existing approaches and algorithms were no longer usable. This work was motivated by the need to develop new algorithms to visualize 3D data in combination with the acquired image data. The procedure to do this is called texture mapping. The basic idea behind texture mapping is to attach the image information onto the geometrical data. The result is a photorealistic visualization of the acquired object. Even at the time of the start of this work, a wide range of algorithms were available to perform texture mapping. The problem was that the existing algorithms were developed with the focus on a different data configuration and requirement. Most of the existing texture mapping algorithms, that could handle 3D data were developed for computer vision applications, e. g. real time visualizations. In those applications, the available data is quite different than the data we are focused to handle in this work. For example, one difference is the number of available images. In our applications, an object should be covered with a useful number of images in a high quality. In contrast, most of the existing approaches use data with low resolution images, acquired using video devices. This work is aimed at analyzing existing approaches concerning their usability to process photogrammetric datasets. Even in the best cases, most of the algorithms had to be adapted to the special needs of the photogrammetric specifications and a number of new algorithms had to be developed for necessary processing steps.
This thesis presents algorithms and software modules covering the workflow for texture mapping from the point of provided images and surface models to the photo-realistic textured 3D model. The completed work results in a wide variety of algorithms of which there are two types of algorithms distinguished:
The second set of algorithms deals with colour and brightness corrections, and combines different images acquired under different conditions to achieve a seamless textured 3D model. Those algorithms are:
- Vignetting detection and correction to remove natural vignetting.
- Global adjustment of colour or brightness between images.
- An algorithm to remove local colour differences caused by various factors, e. g. flash lights.
- Procedures to detect and eliminate shadows.
The author wishes to thank the Swiss National Science Foundation for their financial support over 3 years of this work.