• No results found

4.3 Preparation of 3D Content

4.3.1 Common Problems

3D models which are not initially created for interactive rendering have to be adapted. Other-wise they may be too complex for the rendering hardware in terms of memory consumption or interactive frame rates. Also the visual quality can be improved if the 3D model is prepared properly for real time rendering.

4.3.1.1 Conversion Problems

To use existing content in a rendering engine one has to start with choosing a file format. The format has to be exported by the creation tool and imported by the engine. Conversion software like Deep Exploration helps to overcome part of this task but some special features like shader effects may have to be imported separately.

4.3.1.2 Problems related to Construction and Planning Software

Computer-aided design (CAD) tools aim to create an exact representation of a 3D object. The main goal is to define all the details to actually construct the object. This goal may lead to the following problems for real time rendering:

• Unnecessary geometry, curves and labels are included in the 3D content. For example each single construction part is defined in detail, down to the last nut and bolt. Figure4.19shows such an object with unnecessary geometry.

• The geometry gets corrupted by the conversion process. Vertex positions change because of transformations, faces are duplicated. Mirrored faces suddenly appear black because their normal is flipped.

• Freeform surfaces with trim curves etc. may have to be tessellated depending on the render-ing engine. The approximation of curves and surfaces can lead to numerical problems and gaps in the geometry.

• The conversion of freeform surfaces has to be done in an intelligent way. Automatically generated tessellations may lead to a lot of irrelevant polygons filling the rendering pipeline with unnecessary data.

• The material definitions of CAD tools may only consist of references to real materials with-out defining the visual properties. For the construction of an object it is sufficient to know the material. To render the object, either offline or online, a detailed material definition is necessary.

Fig. 4.19 A tiny bolt object is tessellated in high detail. The back side is never visible.

4.3.1.3 Problems related to Marketing and Entertainment Software

Fig. 4.20 The image of the tree looks good from one point of view. Other viewpoints are unsuitable.

Professional creation tools are regularly used for the creation of offline rendered images.

Those images are needed for marketing purposes, for example, to advertise a new product.

And the entertainment industry depends on 3D content for special effects, up to feature films that are more or less completely created in the virtual world. The following aspects may lead to problems when rendered in real time:

• The 3D content is often augmented with 2D elements. Those elements are not recognizable as such in rendered (2D) images and can hardly be combined with 3D content in an inter-active manner if the observer can freely choose the point of view. In Figure4.20the tree is added as a 2D image. The image of the tree looks good from one point of view. For other viewpoints the tree looks like a part of a stage set.

• The scene may be lit by fake lighting effects which are added in a post-production step.

Offline renderings are split into several layers and combined with a depth map. Parts of the image can be relit, repainted and changed to compose the final picture or video frame. Those effects can hardly or practically impossible be added to an interactive visualization. The light setup may have to be changed completely and the workflow for images effects is different.

• 3D objects use special materials which can only be used in the offline renderer. For example the human skin or even metallic car paints have to be used in a different way for real time rendering.

• Decorations are defined as extra geometry that shares the same position with base geometry.

This can lead to so called z-fighting if not rendered in a special way. Figure 4.21shows an arrow made of geometry which is set on top of a surface. With offline rendering (left image) the image looks okay but real time rendering and insufficient z-buffer resolution may produce flickering results (right image).

• The 3D scene contains too much geometry and the textures are often too large for real time rendering with acceptable frame rates.

• There are too many small objects in the scene organized in a scene graph with too many leaves. For example a model of a tree with each leave put into its own graph node produces a lot of overhead and slows down the draw traversal.

Fig. 4.21 The left image is rendered offline. The right image suffers from insufficient z-buffer resolu-tion.

4.3.1.4 Reconstruction and Preservation

Acquisition methods like laser scanning and photogrammetry generate point clouds or polygon meshes. This kind of 3D data is aiming to reconstruct real world objects with the best possible quality. Parts which are well captured get a highly tessellated surface and other parts may be

very coarse or missing. Automatic tools can optimize the resulting content only depending on the captured data. That means it can reduce the amount data according to the structure (e.g. curvature). But it is hardly possible for automatic algorithms to decide, which parts are important in terms of higher level features. For those decisions it is necessary to add semantic information and to define a ranking of importance for the interactive rendering.