real-time animation of realistic virtual humans

Real-Time Animation of Realistic Real-Time Animation of Realistic Virtual HumansVirtual Humans

Real-Time Animation of Realistic Virtual HumansReal-Time Animation of Realistic Virtual HumansReal-Time Animation of Realistic Virtual HumansReal-Time Animation of Realistic Virtual Humans

1. The 3D virtual player is controlled by the real 1. The 3D virtual player is controlled by the real

people who has a HMD and many sensors people who has a HMD and many sensors

2. Unlike in video games, animation is not predefined.2. Unlike in video games, animation is not predefined.

3. Unlike in rending films, animation is in real-time.3. Unlike in rending films, animation is in real-time.


To achieve a real-time virtual humans application we To achieve a real-time virtual humans application we

need to consider:need to consider:

1. Modeling people.1. Modeling people.

2.2. How to present deformation of human body. How to present deformation of human body.

3.3. Motion control.Motion control.

Real-Time Animation of Realistic Real-Time Animation of Realistic Virtual HumansVirtual Humans

1.Body Creation and Skeleton Animation 1.Body Creation and Skeleton Animation -Zhou Bin-Zhou Bin

2.Facial Deformation 2.Facial Deformation -Hu Yi-Hu Yi

3. Body Deformation and Animation Framework3. Body Deformation and Animation Framework

--Yang YufeiYang Yufei


Body Creation:Body Creation:

First layer: Skeleton hierarchy.First layer: Skeleton hierarchy.

Second layer: Metaballs attached on the skeleton.Second layer: Metaballs attached on the skeleton.

Third layer: Convert metaballs to mesh surface (skin mesh).Third layer: Convert metaballs to mesh surface (skin mesh).

Fourth layer: Texture.Fourth layer: Texture.


First layer: Skeleton hierarchyFirst layer: Skeleton hierarchy

1. Skeleton articulated just similar to real human.1. Skeleton articulated just similar to real human.

2. We can define all the human postures using this skeleton.2. We can define all the human postures using this skeleton.

3. Relative limb sizes of the skeleton decides the figure of the 3. Relative limb sizes of the skeleton decides the figure of the

virtual people. virtual people.


Second layer: Metaballs attached on the skeletonSecond layer: Metaballs attached on the skeleton

Construct human body by column.Construct human body by column.

Can’t avoid gap on human body.Can’t avoid gap on human body.

Construct human body by column Construct human body by column

and sphere joint. Can avoid gap butand sphere joint. Can avoid gap but

can’t present human body’s smoothcan’t present human body’s smooth

and gradual shape.and gradual shape.

Construct human body by metaballs.Construct human body by metaballs.

Metaballs includes spheres and ellipMetaballs includes spheres and ellipsoids. Left figure is constructed by tsoids. Left figure is constructed by two ellipsoids. wo ellipsoids.


Second layer: Metaballs attached on the skeletonSecond layer: Metaballs attached on the skeleton

1.These two human models are assembled1.These two human models are assembled

by Metaballs (ellipsoid components). by Metaballs (ellipsoid components).

2. Metaballs are smooth and gradual, so 2. Metaballs are smooth and gradual, so

they are suitable for modeling human bodiesthey are suitable for modeling human bodies

3. Different color depends on the attributes of 3. Different color depends on the attributes of

that metaball: blendable or unblendable,that metaball: blendable or unblendable,

deformable or nondeformable. deformable or nondeformable.


Third layer: Convert metaballs to mesh surfaces (skin mesh).Third layer: Convert metaballs to mesh surfaces (skin mesh).

1. Convert metaballs to a number of meshes.1. Convert metaballs to a number of meshes.

We can do this job by doing ray tracing, and We can do this job by doing ray tracing, and

sampling the intersection points. sampling the intersection points.

2. These meshes can decide the lighting effect2. These meshes can decide the lighting effect

of human bodies.of human bodies.

3. Texture mapping corresponds to these 3. Texture mapping corresponds to these

mesh surfaces.mesh surfaces.


Fourth layer: Mapping textureFourth layer: Mapping texture

Texturing has low cost of computation but can Texturing has low cost of computation but can

observably improves the quality of the virtualobservably improves the quality of the virtual

object.object.

Real human hand textureReal human hand texture

Textured 3D modelTextured 3D model

3D Bernard Boxton3D Bernard Boxton 3D Bernard Boxton3D Bernard Boxton

without texturewithout texture


Body Animation:Body Animation:

First layer: Skeleton motion.First layer: Skeleton motion. --Let avatars make postures and actions.--Let avatars make postures and actions.

Second layer: Mesh surface (skin mesh) deformationSecond layer: Mesh surface (skin mesh) deformation ..

--Present deformation of humans’ skin.--Present deformation of humans’ skin.


Skeleton:Skeleton:

1. The hierarchy of skeleton is defined by a set of joints.1. The hierarchy of skeleton is defined by a set of joints.

This set of joints corresponds to the main joints of real humansThis set of joints corresponds to the main joints of real humans

2. Each joint consists of a set of degrees of freedom (DOF). 2. Each joint consists of a set of degrees of freedom (DOF).

DOFs decide ranges of the joint can translate and rotate. DOFs decide ranges of the joint can translate and rotate.


Three methods of skeleton motion control:Three methods of skeleton motion control:

1. The skeleton motion is captured in real time and drives the 1. The skeleton motion is captured in real time and drives the avatar.avatar.

2. The skeleton motion is predefined. It will be activated from a2. The skeleton motion is predefined. It will be activated from a

database as a response to human’s input. E.g. Sony’s Eye Toy.database as a response to human’s input. E.g. Sony’s Eye Toy.

3. The skeleton motion is dynamically calculated. Doesn’t require3. The skeleton motion is dynamically calculated. Doesn’t require

a user’s continual intervention. E.g. complex games, and AIa user’s continual intervention. E.g. complex games, and AI

application. application.

Human Head Modeling Facial Animation


• To simulate humans requires real-time visualization and animation.

Human Head Modeling


• Scanning

scan the surface of the head and construct a head model

• Use a sculpturing model----Software Sculptor

Use a sculpturing model----Software Sculptor


Facial Animation


• Facial deformation model

• Facial motion control

Facial Deformation Model


1. Consider human face as a polygonal mesh

2. Define regions on the mesh 3. Define a control lattice on the region of in

terest 4. Muscle actions are simulated by changing

the control points’ weight 5. A stiffness factor allows the amount of def

ormation for each point

Facial Deformation Model


Facial Motion Control


MuscularAction

An AnimationSequence Real-time

Animation

Coding Composer

The Facial Action Coding System

1. Define basic motion parameters as minimum perceptible actions (MPAs)

open_mouth, close_upper_eyelids, or raise_corner_lip



1. Each MPA has a corresponding set of visible features

e.g. movement of eyebrows, jaw, or mouth



2. Real-time facial animation module uses three different input methods

• Video• Audio or Speech• Predefined Actions


I. Video Input

• Recognition and tracking of the facial features is based on color sample identification, edge detection, and other image processing operations. The feature capture and tracking rate is about 20 frames per second (fps)


Facial FeatureExtraction And

Tracking

Video InputMagnenat-Thalmann

et al. Array of MPAs

II. Audio Input segment audio into phonemes with their durationText input to phonemes phonemes decomposes into MPAs


III. Predefined Action an action that has intensity, duration, and a start time e.g. emotion (surprise, anger)

head gestures (nodding, turning) decomposes into MPAs

predefined action of speechinputaudio input

video input

interpolation


IV. Synchronization

go to the composer

V. Composition


defined a sinusoid

weight function

considered 10 percent of

the action’s total duration

no jump effect

with the same MPA overlap

Body deformations• The ways we do the representing humans

– Polygonal representation– Visual accuracy representation– The combination of previous two

• Implement results


Animation framework

Two case studies


The first method

• The skin wrapped around the skeleton is represented with a fixed mesh divided at important joints where deformations occur.

Polygonal representation

• Pros– Simple and easy to

implement

• Cons– The virtual human

appears “rigid” and lacks realism

– Visually distracting artifacts may arise at joints

• The combined method

• Visual accuracy representation

Body deformation

Image from the internet


The second method

• The application compute the skin from implicit primitives and use a physical model to deform the body’s envelope.

• Pros– Stress visual accuracy and

yields very satisfactory results in terms of realism

• Cons– So computationally

demanding – Unsuitable for real-time

applications

• Polygonal representation

• The combined method

Visual accuracy representation

Body deformation


The third method

• It’s a combination of previous two methods, allowing a good trade-off between realism and rendering speed.

• Steps– Constructing a body mesh– Deforming by manipulating skin contours

• Polygonal representation

The combined method

• Visual accuracy representation

Body deformation


The combined method--- step one

• Output the body data as cross-sectional contours.

• Convert contours data to triangle meshes.– Easy to render, performs better– Construct a triangle strip by connecting

the points from two adjacent cross-section

• Connecting two different body parts proves a bit more complicated.

The combined method Constructing the body mesh

• Manipulating skin contours

Body deformation

Image from: Daniel Thalmann, Jianhua Shen, Eric Chauvineau, 1996


The combined method--- step two

• Transform a complicated 3D operation into a 2D operation by manipulate the cross-section contours.

• By setting the orientation and position of the plane in which every contour lies, we can achieve a smooth deformation of the skin.

The combined method

• Constructing the body mesh Manipulating skin contours

Body deformation

Image from: Daniel Thalmann, Jianhua Shen, Eric Chauvineau, 1996


The combined method--- step two (cont.)

• Very joint lies in the plane of its contour when the skeleton is in the at-rest posture.

• Two segments of the arm whose directions are L1 and L2.• Nu, N0 and Nl are the normal vectors of the cross-section planes.• Oi and Ni are the center and normal respectively of the ith cross-section plane.• Since we know Ni by

interpolation, we can

compute each vertex

belonging to the ith contour.

The combined method


Body deformation

Image from:Prem Kalra, Nadia Magnenat-Thalmann, Laurent Moccozet, and Gael Sannier, 1998


The combined method --- step two (conclusion)

• The contours run parallel at the region above the elbow.• This is because our eyes naturally go to the areas

surrounding major joints like the elbows or knees.• Practically we determined the number of upper and lower

contours to deform in a heuristic fashion.• It saves the rendering time and has little degradation.

The combined method


Body deformation



Results• Using real-time-oriented 3D graphics

toolkit called Performer.

• In right image: the virtual human on the left is made up of 14,000 vertices and containing 13,500 textured triangles using deformation; on the right uses rigid meshes with 17,000 triangles.

Image and figure from:Prem Kalra, Nadia Magnenat-Thalmann, Laurent Moccozet, and Gael Sannier, 1998


Animation framework

• The close link between modeling and animation.• The system separates into three units: modeling,

deformation and motion control.– Modeling provides geometrical models for the body, hands and

face.– Deformations are performed separately on different entities

based on the model used for each part.– Motion control generates and controls the movements for

different entities.


Animation framework

Figure from:Prem Kalra, Nadia Magnenat-Thalmann, Laurent Moccozet, and Gael Sannier, 1998


Case studies--- CyberTennis



Case studies--- CyberDance


real-time animation of realistic virtual humans

Documents

virtual people

skeleton animation zhou

d virtual player

deformation of human

human postures

skeletonconstruct human

human models

body deformation