Realistic Visualization of Animated Virtual Cloth

Dissertation, Universität Bonn, Oct. 2007
 

Abstract

Photo-realistic rendering of real-world objects is a broad research area with applications in various different areas, such as computer generated films, entertainment, e-commerce and so on. Within photo-realistic rendering, the rendering of cloth is a subarea which involves many important aspects, ranging from material surface reflection properties and macroscopic self-shadowing to animation sequence generation and compression. In this thesis, besides an introduction to the topic plus a broad overview of related work, different methods to handle major aspects of cloth rendering are described. Material surface reflection properties play an important part to reproduce the look & feel of materials, that is, to identify a material only by looking at it. The BTF (bidirectional texture function), as a function of viewing and illumination direction, is an appropriate representation of reflection properties. It captures effects caused by the mesostructure of a surface, like roughness, self-shadowing, occlusion, inter-reflections, subsurface scattering and color bleeding. Unfortunately a BTF data set of a material consists of hundreds to thousands of images, which exceeds current memory size of personal computers by far. This work describes the first usable method to efficiently compress and decompress a BTF data for rendering at interactive to real-time frame rates. It is based on PCA (principal component analysis) of the BTF data set. While preserving the important visual aspects of the BTF, the achieved compression rates allow the storage of several different data sets in main memory of consumer hardware, while maintaining a high rendering quality. Correct handling of complex illumination conditions plays another key role for the realistic appearance of cloth. Therefore, an upgrade of the BTF compression and rendering algorithm is described, which allows the support of distant direct HDR (high-dynamic-range) illumination stored in environment maps. To further enhance the appearance, macroscopic self-shadowing has to be taken into account. For the visualization of folds and the life-like 3D impression, these kind of shadows are absolutely necessary. This work describes two methods to compute these shadows. The first is seamlessly integrated into the illumination part of the rendering algorithm and optimized for static meshes. Furthermore, another method is proposed, which allows the handling of dynamic objects. It uses hardware-accelerated occlusion queries for the visibility determination. In contrast to other algorithms, the presented algorithm, despite its simplicity, is fast and produces less artifacts than other methods. As a plus, it incorporates changeable distant direct high-dynamic-range illumination. The human perception system is the main target of any computer graphics application and can also be treated as part of the rendering pipeline. Therefore, optimization of the rendering itself can be achieved by analyzing human perception of certain visual aspects in the image. As a part of this thesis, an experiment is introduced that evaluates human shadow perception to speedup shadow rendering and provides optimization approaches. Another subarea of cloth visualization in computer graphics is the animation of the cloth and avatars for presentations. This work also describes two new methods for automatic generation and compression of animation sequences. The first method to generate completely new, customizable animation sequences, is based on the concept of finding similarities in animation frames of a given basis sequence. Identifying these similarities allows jumps within the basis sequence to generate endless new sequences. Transmission of any animated 3D data over bandwidth-limited channels, like extended networks or to less powerful clients requires efficient compression schemes. The second method included in this thesis in the animation field is a geometry data compression scheme. Similar to the BTF compression, it uses PCA in combination with clustering algorithms to segment similar moving parts of the animated objects to achieve high compression rates in combination with a very exact reconstruction quality.

(Thesis submission date: Dec. 2006)

Download: http://nbn-resolving.de/urn:nbn:de:hbz:5N-12418

Bibtex

@PHDTHESIS{sattler-2007-dissertation,
    author = {Sattler, Mirko},
     title = {Realistic Visualization of Animated Virtual Cloth},
      type = {Dissertation},
      year = {2007},
     month = oct,
    school = {Universit{\"a}t Bonn},
  abstract = {Photo-realistic rendering of real-world objects is a broad research area with applications in
              various different areas, such as computer generated films, entertainment, e-commerce and so on.
              Within photo-realistic rendering, the rendering of cloth is a subarea which involves many important
              aspects, ranging from material surface reflection properties and macroscopic self-shadowing to
              animation sequence generation and compression.
              In this thesis, besides an introduction to the topic plus a broad overview of related work,
              different methods to handle major aspects of cloth rendering are described.
              Material surface reflection properties play an important part to reproduce the look {\&} feel of
              materials, that is, to identify a material only by looking at it. The BTF (bidirectional texture
              function), as a function of viewing and illumination direction, is an appropriate representation of
              reflection properties. It captures effects caused by the mesostructure of a surface, like roughness,
              self-shadowing, occlusion, inter-reflections, subsurface scattering and color bleeding.
              Unfortunately a BTF data set of a material consists of hundreds to thousands of images, which
              exceeds current memory size of personal computers by far.
              This work describes the first usable method to efficiently compress and decompress a BTF data for
              rendering at interactive to real-time frame rates. It is based on PCA (principal component analysis)
              of the BTF data set. While preserving the important visual aspects of the BTF, the achieved
              compression rates allow the storage of several different data sets in main memory of consumer
              hardware, while maintaining a high rendering quality.
              Correct handling of complex illumination conditions plays another key role for the realistic
              appearance of cloth. Therefore, an upgrade of the BTF compression and rendering algorithm is
              described, which allows the support of distant direct HDR (high-dynamic-range) illumination stored
              in environment maps.
              To further enhance the appearance, macroscopic self-shadowing has to be taken into account. For the
              visualization of folds and the life-like 3D impression, these kind of shadows are absolutely
              necessary. This work describes two methods to compute these shadows. The first is seamlessly
              integrated into the illumination part of the rendering algorithm and optimized for static meshes.
              Furthermore, another method is proposed, which allows the handling of dynamic objects. It uses
              hardware-accelerated occlusion queries for the visibility determination. In contrast to other
              algorithms, the presented algorithm, despite its simplicity, is fast and produces less artifacts
              than other methods. As a plus, it incorporates changeable distant direct high-dynamic-range
              illumination.
              The human perception system is the main target of any computer graphics application and can also be
              treated as part of the rendering pipeline. Therefore, optimization of the rendering itself can be
              achieved by analyzing human perception of certain visual aspects in the image. As a part of this
              thesis, an experiment is introduced that evaluates human shadow perception to speedup shadow
              rendering and provides optimization approaches.
              Another subarea of cloth visualization in computer graphics is the animation of the cloth and
              avatars for presentations. This work also describes two new methods for automatic generation and
              compression of animation sequences.
              The first method to generate completely new, customizable animation sequences, is based on the
              concept of finding similarities in animation frames of a given basis sequence. Identifying these
              similarities allows jumps within the basis sequence to generate endless new sequences.
              Transmission of any animated 3D data over bandwidth-limited channels, like extended networks or to
              less powerful clients requires efficient compression schemes. The second method included in this
              thesis in the animation field is a geometry data compression scheme. Similar to the BTF compression,
              it uses PCA in combination with clustering algorithms to segment similar moving parts of the
              animated objects to achieve high compression rates in combination with a very exact reconstruction
              quality.},
       url = {http://nbn-resolving.de/urn:nbn:de:hbz:5N-12418}
}