Visionix实时渲染引擎架构：浙江大学CAD实验室研发

需积分: 12 169 浏览量更新于2024-07-27 收藏 898KB PDF 举报

"该资源是浙江大学CAD实验室自主研发的实时图形渲染引擎架构设计的详细资料，主要探讨了如何构建一个能够处理大规模场景模型的系统，并在多个虚拟现实应用中得到实际运用，如城市规划、船舶驾驶模拟及交互式游戏。Visionix系统作为案例，其优势在于支持插件式渲染模块和强大的场景模型管理系统，能适应不同应用需求并处理大量3D数据。" 正文: 实时图形渲染引擎是计算机图形学中的核心组成部分，主要用于快速生成逼真的3D图像。在这个领域，浙江大学CAD实验室研发的Visionix系统以其独特的优势脱颖而出，不仅能够处理大规模场景模型，还具备高度的灵活性和可配置性。本文将深入解析实时渲染引擎的架构及其关键特性。首先，实时渲染引擎的架构通常包括以下几个关键部分： 1. **输入与数据管理**：这部分负责处理来自各种来源的数据，如3D模型、纹理、光照信息等，并进行有效的存储和管理。在Visionix系统中，其强大场景模型管理系统能够高效地处理大容量的3D场景数据，确保在内存和性能之间取得平衡。 2. **场景图与对象组织**：为了优化渲染效率，场景通常被组织成一棵场景图。这种数据结构使得遍历和操作场景中的对象变得高效，同时支持动态加载和卸载，以适应不同的应用程序需求。 3. **渲染管线**：渲染管线是渲染引擎的心脏，它将3D几何数据转化为2D屏幕上的像素。通常包括几何处理、光栅化和像素处理等阶段。Visionix系统支持插件式渲染模块，这意味着用户可以根据特定需求定制渲染流程，比如引入新的光照模型或后处理效果。 4. **光照与材质**：实时渲染引擎需要快速计算光照对物体的影响，包括漫射、镜面反射、折射等。Visionix系统可能包含多种光照模型，以实现逼真的视觉效果。 5. **纹理与贴图**：纹理是赋予物体表面细节的关键，Visionix系统很可能支持多种纹理技术，如环境映射、法线贴图等，以提升图像的真实感。 6. **阴影与深度缓冲**：阴影的处理是增加场景真实感的重要一环。Visionix可能采用多种阴影算法，如阴影贴图、VSM（体积阴影映射）等。深度缓冲则用于确定像素的正确顺序，避免绘制顺序导致的错误遮挡。 7. **性能优化**：为了在实时环境下保持流畅的帧率，引擎需要进行大量的性能优化，如多线程、GPU计算、LOD（级别细节）技术等。 8. **用户界面与交互**：实时渲染引擎不仅需要生成高质量的图像，还要提供友好的用户交互界面，如相机控制、场景编辑工具等。 Visionix系统因其在多个实际应用中的成功实施，证明了其架构的有效性和灵活性。通过学习和理解Visionix的设计，开发者可以借鉴这些经验来构建自己的实时渲染引擎，以应对不断发展的图形技术挑战。无论是用于游戏开发、虚拟现实应用还是其他领域，这样的引擎都将是不可或缺的工具。

definition of Billboard in X3D.

CSuperShape

In many situations there are objects of huge volumes in 3D scenes. These objects impact the

efficiency of view culling —— even if only a small part of the object is visible, we have to

render the entire object. This would severely reduce the rendering speed. The normal way to

solve this problem is to divide these objects into a set of sub-objects. Such division is performed

according to the space occupied by the object, so the sub-objects themselves do not have any

logical or semantic meanings. If we want to give attention to both rendering efficiency and

object’s semantic meaning, a data structure is needed to keep the original object and divided

sub-objects simultaneously.

So in Visionix we introduce CSuperShape to solve this problem. The declaration of data

member of the class CSuperShape is listed as follows.

class CSuperShape : public CShape

{

public:

enum Enum_OptMask{

PARTITIONED = 1,

SUBDIVIDED =2

};

protected:

TMF<TAddress<CShape> > m_Subshapes;

SFUInt32 m_Option;

};

On the basis of CShape, we add a container to store divided sub-shapes inside CSuperShape.

By this means, we achieve the goal to store both the original object and subdivided objects.

It is worthy to note that since we adopt TAddress pointers to make inter-reference of features

or entities, there is only one copy of the same data in divided subshapes and the original shape.

Therefore, there is no storage space wasted.

CSuperShape can not only be used to represent the subshapes of huge-volume objects, but

also the objects that need subdivision out of other reasons. One common situation is the object

with multiple materials. Take a triangular mesh as example, if some triangles on this mesh’s

surface have red diffuse material, while others have blue specular material, we can not use one

CShape instance to represent such mesh. The traditional method is to adopt multiple CShape

instances. This would lead to the loss of the semantic meaning of the original model.

CSuperShape can solve this problem satisfyingly by dividing the object according to its material

to make each subshape has only one material.

If we start from the definition of the class, any subshape of a CSuperShape instance can also

be a CSuperShape instance. Only if there are two different divisions needed to be represented

simultaneously, we would need the above complex representation. For example, if the object

with multiple materials also needs spatial division, two-level hierarchical structure of

CSuperShape is required.

m_Option describes some Boolean values in CSuperShape. It is used together with the

enumeration types defined by Enum_OptMask.

m_Option& PARTITIONED: If it is 1, this means subshapes come from the cut of the

geometry in CSuperShape. New vertices are generated in this situation. If it is 0, this means

subshapes come from the classification of the geometry in CSuperShape. No new vertices would

be produced.

m_Option&SUBDIVIDED: If it is 1, this means the geometry of subshape comes from the

subdivision of the geometry in SuperShape. New vertices and topology are produced. Otherwise,

there is no subdivision.

We give some examples to explain the meanings of these two values:

Example 1: The object M is composed of two triangles A and B. If these two triangles are

subdivided to get two triangular mesh M(A) and M(B), M(A) and M(B) form two subshapes.

Under this situation, m_Option& PARTITIONED is set to be 0 and m_Option&SUBDIVIDED is

set to be 1.

Example 2: The object M is composed of two triangles A and B. If we cut triangle A and

triangulate two sub-regions of A, we get three triangles A1, A2, and A3. Triangle B is not cut.

Triangles A1 and A2 form a subshape. Triangles A3 and B form the other subshape. Under this

situation, m_Option& PARTITIONED is set to be 1 and m_Option&SUBDIVIDED is set to be

Example 3: The object M is composed of two triangles A and B. If we cut triangle A and

triangulate two sub-regions of A, we get three triangles A1, A2, and A3. Triangle B is not cut.

Then triangles A1, A2, A3 and B are subdivided. Under this situation, m_Option&

PARTITIONED is set to be 1 and m_Option&SUBDIVIDED is set to be 1. If we perform

subdivision before cut, the values of these two properties are also set to be 1.

CComposedShape

The declaration of data member of the class CComposedShape is listed as follows.

class CComposedShape : public IShapeFeature

{

protected:

TMF<TAddress<IGroupingFeature> > m_ShapeGraph;

};

The data member m_ShapeGraph in CComposedShape contains an array of TAddress

pointers pointing to IGroupingFeature instances. Its name shows that m_ShapeGraph is a

hierarchical structure——a tree structure composed of subclasses of IGroupingFeature and

IChildFeature (if we consider the inter-references among these subclasses, the tree is actually a

non-cyclic digraph). The type of the middle node of the tree is the subclass of IGroupingFeature.

The type of the leaf node is the subclass of IChildFeature.

The concrete composition of CComposedShape is flexible. However, such flexibility would

bring complexity to the applications. In order to reduce the complexity of data types, we need to

limit CComposedShape properly according to the application. Such limitation is called Scene

Model Scheme which will be introduced later. In most situations, the type of the leaf node is

CShape. However, if we start from class definition, we do not exclude the possibility to adopt

other subclasses of IChildFeature as the type of leaf nodes in some applications.

Note that CComposedShape is derived from IShapeFeature instead of CShape, while

CSuperShape is derived from CShape. Hence, CSuperShape actually gives two representative

models for one object——one is the non-divided model and the other is the model after object

division. CComposedShape only gives one representative model. It describes how this model is

composed of multiple sub-models in detail.

CStaticLOD

class CStaticLOD : public class IShapeFeature

{

protected:

TMF< TAddress< IChildFeature> > m_Children;

MFFloat m_Range;

SFVec3f m_Center;

}

The data members in CStaticLOD are the same as that of LOD node in X3D. However, we

let it derived from IShapeFeature instead of IGroupingFeature, which is different from X3D, due

to the reason that the child features contained in CStaticLOD do not have composition relation

semantically. That is to say, all child features in CStaticLOD depict the same object in the scene

with different level of detail.

The CStaticLOD class specifies various levels of detail for a given object, and provides hints

allowing programs choose the appropriate detail level of the object according to the distance from

the viewpoint The m_Children field contains a list of TAddress<IChildFeature>, which represent

the same object at different level of detail, ordered from highest level to the lowest level. The

m_Range field specifies the distances at which to switch between the levels. The m_Center field

specifies the centre of the CStaticLOD feature for distance calculations.

CSkinShape

CSkinShape class represents the dynamic objects which have skeletal animation. We will

introduce this class when we introduced IAnimatedFeature subclass.

3.5.6 IAnimatedFeature

So far, Visionix only provides skeletal animation based dynamic features——CSkinAnim.

consisting of CBone instances. This structure describes the skeleton composition. Besides

CBone::m_Children containing an array of TAddress<CBone> pointing to the children bone node,

there is a matrix m_WorldToBoneSpaceMatrix, which transforms world coordinate frame to the

bone’s local coordinate frame. Such transformation is to make more convenient the

implementation of skinning animation. To compute skinning animation, the first step is to

transform the coordinates of vertices of the skinning mesh to world coordinate frame. Then,

according to the binding relation between the vertices and the bones, these vertices are

transformed to their corresponding bones’ local coordinate frames by multiplying

m_WorldToBoneSpaceMatrix of the bones’. After that, correct animation effect can be realized by

further multiplying the bones’ animation matrixes to the vertices.

The data member m_Motions in CSkeleton is the array of

TAddress<CComposedRigidMotion>. Each CComposedRigidMotion instance describes one set of

skeleton movement, which means that CSkeleton::m_Motions describes multiple sets of skeleton

movements.

CComposedRigidMotion is the subclass of IMotionFeature. IMotionFeature is the abstract

class to describe objects’ movement. The data member

CComposedRigidMotion::m_pTimeSequence points to the time sequence CTimeSequence, which

records the duration time of the whole time sequence (denoted by m_Duration) and the timing of

some key frames during the duration time (denoted by m_Times). The data members

m_RotationSequences and m_TranslationSequences in CComposedRigidMotion record the

rotation sequences and translation sequences respectively. The time sequence in

CComposedRigidMotion is valid for all rotation and translation sequences. This means that for the

i-th key frame at time t

, the i-th element in any rotation sequence is the object’s rotation

quaternion at time t

, and the i-th element in any translation sequence is the object’s translation

vector at time t

. The data member CComposedRigidMotion::m_StartTime records the starting

time of the movement with the default value of 0.

For each CComposedRigidMotion instance in CSkeleton::m_Motions, the rotation sequence

is paired with the translation sequence. Each rotation-translation sequence pair describes one

movement of one bone. Therefore, the number of the bones in a CSkeleton instance should be

equal to the number of rotation-translation sequence pairs in each CComposedRigidMotion

instance in the Skeleton instance.

The data member CSkinAnim::m_pShape points to a CSkinShape instance. As we have

stated before, CSkinShape is the subclass of IShapeFeature.

CSkinShape : public IShapeFeature

{

protected:

TAddress<ICoordinateFeature> m_pCoord;

TAddress<INormalFeature> m_pNormal;

TAddress<ITextureCoordinateFeature> m_pTexCoord;

TAddress<IColorFeature> m_pColor;

TMF< TAddress<IShapeFeature> > m_Parts;

MFVec4f m_BoneWeights;

TMF< SFInt16[4] > m_BoneIndexes;

剩余167页未读，继续阅读

雪落诗痕

粉丝: 3
资源: 8

Visionix实时渲染引擎架构：浙江大学CAD实验室研发

Real-Time 3D Graphics with WebGL 2

Real Time Rendering矩阵操作（演示程序）

Real Time DSP Fundamentals, Implementation and Application 3rd

real-time graphics rendering engine csdn

real-time rendering 4th

real-time rendering 4th edition 中文

《real-time rendering 3rd》 pdf

real-time rendering 4th pdf

real time rendering 4th

《real-time rendering 4th》pdf

最新资源