REALISTIC

RAIN RENDERING

Yoshiki Mizukami, Katsuhiro Sasaki and Katsumi Tadamura

Graduate School of Science and Engineering, Yamaguchi University, 2-16-1 Tokiwa-Dai, Ube 755-8611, Japan

Keywords:

Animation, Rendering natural phenomena, Particle systems.

Abstract:

We propose a method for rendering realistic three-dimensional raining scene. The proposed method gives

two main contributions to rendering raining scene. The ﬁrst one is to propose a simple model that deals with

temporal-spatial localities such as wind effect, density and intensity of rainfall, and raindrop brightness, so

as to represent environmental conditions that differs on a scene-to-scene basis. The other one is to propose

a raindrop trajectory computing method whose computation load immunes to the number of raindrop, wind

effect and complicated camera movement. Due to the above-mentioned contributions, the proposed method

can represent essential aspects in raining scene, such as spatial changes of rain in urban canyon or temporal

changes due to the moving wisp of rain.

1 INTRODUCTION

Although various factors are concerned with the re-

ality of landscape rendered with Computer Graph-

ics (CG), the representation about weather or mete-

orological condition is one of the essential factors.

This paper proposes a method for realistically render-

ing raining scenes, which can represent the temporal-

spatial localities such as wind effect, density and in-

tensity of rainfall, and raindrop brightness.

For the clariﬁcation of background, the previous

studies related to rainfall is summarized below, es-

pecially focusing on the relationship with our study.

Kaneda et al. proposed a representation method of

water droplet on the windshield for driving simulator

(Kaneda et al., 1999). Sato et al. proposed a faster

method by using the capability of graphics hardware

(Sato et al., 2002). Many researchers have studied

rendering methods of the trajectory of raindrops. In

early CG animation, the falling motion of translucent

white spheres was synthesized into the background

scene on the basis of the particle system (Reeves,

1983; Sims, 1990). Starik et al. proposed a tech-

nique for combining raindrop trajectory with video

images, where rain intensity and raindrop size can be

adjusted by controlling parameters (Starik and Wer-

man, 2003). Feng et al. proposed a fast GPU-based

method of rendering the trajectory of windblown rain-

drops and raindrop splashes (Feng et al., 2005a; Feng

et al., 2005b). Garg and Nayar proposed a method

for improving appearance of rain trajectory that uti-

lizes a database of rain streaks under various illumina-

tion conditions (Garg and Nayar, 2006). Several other

studies have been done for creating real-time rain an-

imation taking into account environmental light con-

dition. For example, handwritten rain strokes were

colored based on the environmental light condition

(Wang et al., 2006). Multiple layers of rain tex-

ture were used particularly for expressing rain drop

motion parallax (Tatarchuk, 2006). A texture dis-

torted according to optical properties of raindrops was

mapped onto a raindrop (Rousseau et al., 2006).

These previous contributions are very useful for

rendering raining scenes with feasible computation

requirements. Now it should be noted that parti-

cle system was employed in all study except Starik’s

method. These particle-based methods generated

raindrops randomly and uniformly. In addition, the

raindrop movement was not inﬂuenced by the wind.

Even so, these wind conditions were uniform. How-

ever it is clear that the density and the falling trajec-

tory of real-life raindrops reﬂect the temporal-spatial

changes of the wind conditions. These characteristic

localities in raining scene are considered as essential

factors for providing raining reality, and the problem

about these factors in the previous studies should be

improved for more broad applications. In addition,

apparent movements of raindrops caused by the cam-

era movement were not considered or if so, they were

considered partially.

273

Mizukami Y., Sasaki K. and Tadamura K. (2008).

REALISTIC RAIN RENDERING.

In Proceedings of the Third International Conference on Computer Graphics Theory and Applications, pages 273-280

DOI: 10.5220/0001095702730280

 SciTePress

This paper proposes a rendering method of the

falling trajectory of raindrops that can deal with both

the temporal-spatial localities of raining and arbi-

trary camera movements. At ﬁrst, section 2 gives

a brief overview of the proposed method and sec-

tion 3 describes a simulation model of rainfall and 3-

D wrapping-around of raindrops that was introduced

for permitting arbitrary camera movements. Sec-

tion 4 explains the raindrop-generating model for the

temporal-spatial localities of raining, that is, the den-

sity and intensity of raining. This section also de-

scribes the method for rendering raindrops consider-

ing with locality of occlusion of the sky. Section 5

shows usefulness of the method using CG animations

rendered with the proposed method. Finally, section

6 describes the conclusion and the future work.

2 OVERVIEW

2.1 Prior Conditions

In real-life raining, all the raindrops have differences

in location, velocity, its size and shape. In addi-

tion, there are several factors to be considered in ren-

dering, that is, the changes in direction and velocity

of raindrops due to non-steady wind effect, and the

changes in light intensity coming through a raindrop

depending on illumination condition due to the sky-

light. The above-mentioned factors are referred as

temporal-spatial locality of raining in our study. It

is desired to reﬂect these localities for improving the

rendering reality, but practical real-time applications

such as walk-through systems or driving simulators

require as simple processing as possible. In order to

solve this trade-off, we introduce the following prior-

conditions:

(1) The viewpoint is located near the ground level.

Therefore the vertical velocity of every raindrop

is assumed to be equivalent of terminal velocity.

On the other hand, only horizontal velocity is con-

trolled by temporal-spatial changes in wind.

(2) Trajectory of a raindrop is drawn with a polyline

connecting the positions of the raindrop at neigh-

boring two frames.

(3) The shape of a raindrop is represented with

sphere and its diameter is determined by the rain-

ing intensity.

2.2 Fundamental Idea

From two viewpoints, the following description

illustrates the problems in conventional methods and

the characteristics of the proposed method.

(1) Generation and Tracking of Raindrops. Prob-

lems in the previous methods which employ particle

system for generating and vanishing raindrops, are

summarized as :

• Stationary camera; Some researches discussed the

dynamic camera, but they synthesized stationary-

camera rendering of raindrops with the dynamic-

camera background.

• Temporal-spatial uniform generation of raindrops.

• Due to the consistency of the vertical axis of the

camera-coordinate system with that of the world-

coordinate system, raindrops are generated above

the top face of the view volume and fall down only

in the direction of gravitational force.

Conventional methods focused on rendering the rain-

drop trajectory and some of them were very realis-

tic for off-line processing or very fast for real-time

applications. On the contrary, the above-mentioned

constraints narrow the range of their applications. To

solve these problems, we provide a new raindrop

movement model with the following characteristics:

• A raindrop generating model that can deal with

localities on density and intensity of raining.

• A raindrop movement model that can express ap-

parent movements of raindrops caused by the ro-

tation and/or translation of camera.

• A raindrop descent tracking model that invokes

no rapid ﬂuctuation in the number of raindrops

with satisfying two above-mentioned points.

(2) Rendering of Raindrops. As suggested by Wang

et al. (Wang et al., 2006), the light intensity com-

ing through a raindrop is determined on basis of the

reﬂection of the surrounding environment. Wang et

al. utilized a environment mapping for obtaining the

reﬂection-based light intensity. The luminance of the

object surface, however, distinctly differs from that

of the sky during daytime even in the rain. For in-

stance, in the light-raining scene shown in Figure 1

(captured by a digital camera with automatic exposure

mode), the actual measurement of luminance gives

6,886cd/m

, 562cd/m

and 53cd/m

at point A, B

and C, respectively. On the other hand, the pixel

value on the digital camera picture gives (R,G,B) =

(254,254,254),(182, 171, 181) and (51,50,46). As

you can see, 8 bit-colored image-based method is en-

tirely inadequate for computing the optical energy re-

ceived from the surrounding environment due to lack

of the dynamic range, therefore resulting in inaccu-

rate light transport. In order to solve this problem,

we assume that the light intensity coming through a

GRAPP 2008 - International Conference on Computer Graphics Theory and Applications

274

Figure 1: A photograph of a light-rain day.

3D CG

software

background

image

object data

solar position

camera

motion

depth (from

the view point)

main-process

final image

rain parameter

sky-illumination

calculation

illuminance

table

Figure 2: Rain rendering procedure.

raindrop is proportional to the optical energy received

from the sky, and determines the light intensity based

on the illuminance due to skylight.

2.3 Procedure of Generating Raining

Scene

As depicted in Figure 2, the proposed method for gen-

erating raining scene consists of three steps. As a

preprocessing, the illuminance due to skylight is cal-

culated. In the second step, background image and

depth information for each animation frame is gener-

ated and stored. In the third step, the raindrops are

generated based on both the camera movement and

the local density of raindrop occurrence. The velocity

and direction of wind ﬁeld are computed and applied

to the updating of raindrops to be tracked. Then, the

trajectories of raindrop are rendered based on both the

illuminance due to skylight at the view point and the

raindrop movement in the camera-coordinate system.

Finally, the background image and the corresponding

raindrop-rendering image are synthesized.

3 RAINDROP MOVEMENT

In this section, the computing and drawing method

for the trajectory of raindrops is described. The pro-

posed method regards only raindrops in the rectangu-

lar parallelepiped (RP, see Figure 3) located in front

of the viewpoint as a target for computing and draw-

ing. At ﬁrst, let us explain how to compute the veloc-

ity vector of a raindrop including its apparent move-

ment due to camera movement. Then the effective

wrapping-around reuse of raindrops running over the

RP is shown.

3.1 Calculation of Velocity Vector

(1) Stationary Camera Model. The RP is deﬁned as

a bounding box that including a part of the view vol-

ume, which is illustrated with bold lines in Figure 3.

We consider that every raindrop which is tracked its

trajectory exists only in this RP as described before.

According to the prior condition (1), the factors

affecting the raindrops’ movement are the gravity and

wind. Consequently, the velocity vector of a raindrop

in Figure 4) can be computed by synthesizing

the following two factors. The ﬁrst one is the

vertical falling velocity which reaches to the terminal

velocity, V

. The second one is the wind velocity, V

that is calculated by solving Navier-Stokes equation

(We discuss more in Section 4.2).

(2) Expansion to Dynamic-camera Model. In ap-

plications such as walk-through, the camera move-

ment can be classiﬁed into dolly, pan and tilt. The

camera movement causes an apparent movement of

raindrops on the screen. The proposed method trans-

lates this apparent movement into the velocity. Let

us explain with the case of dolly as shown in Fig-

ure 4. The dolly movement V

results in an apparent

movement with the opposite-direction velocity V

all raindrops. In the case of pan and tilt movements

that arise from the rotation of camera axis, they give

apparent movement with the opposite-rotation to all

raindrops. The proposed method regards the appar-

ent movement of raindrop caused by pan and tilt as a

set of linear movement on each frame. Consequently,

the apparent movement caused by camera movement

can be represented as a velocity vector. Then, ﬁnal

raindrop’s movement, V

, is determined by synthesiz-

ing V

calculated by physics and V

given by apparent

movement as shown in Figure 4.

3.2 Three-dimensional

Wrapping-around of Raindrops

(1) Basic Idea. In the proposed method, only

raindrops that exist in the RP are regarded as a target

for computing and drawing. In a windless scene,

as shown in Figure 5(a), all raindrops arising from

REALISTIC RAIN RENDERING

275

right

bottom

far

top

left

near

view volume

RP*

camera coordinate system

* RP: rectangular parallelepiped for drawing

trajectory of raindrop

Figure 3: Geometry of the rectangular parallelepiped (RP)

for drawing a trajectory of a raindrop.

V’

raindrop

Figure 4: Determination of velocity vector of a raindrop.

the top face of the RP, T , fall down and reach the

bottom face, B. However, in case of windy scene as

depicted in Figure 5(b), some of raindrops run over

the side face before arriving at B and run into region

P, while region Q has no raindrop arising from the

top face. One of the solutions for this problem is to

extend the top face horizontally as a function of wind

speed. However it will bring redundant computation

of raindrops. The proposed method employs more

efﬁcient method. Since we may consider that the

amount of raindrops in region P and Q are almost

the same with the geometrical property of the RP,

the raindrops running into region P can be reused as

raindrops going into region Q, which is a basic idea

of 3-D wrapping-around of raindrops. The details

will be described below.

(2) Face Attributes of RP. As illustrated in Fig-

ure 5(a), the top face, T , is usually given as an

attribute of raindrop-generating face, G, and the

bottom face, B, is given an attribute of raindrop-

vanishing face, V . Other four faces have the same

attribute of raindrop wrapping-around face. However,

in the case of large camera tilt and strong wind as

shown in Figure 5(c), there will be a problem that

all raindrops may ﬂy out from the RP just after

being generated at the top face, T. Therefore, the

proposed method selects the face whose inward-

looking normal vector gives maximum inner product

with raindrop’s velocity vector (V

in Fig. 4) as the

raindrop-generating face, G, and its opposite face as

T (G)

B (V)

L R

(a)

L R

(b)

wind

(c)

(d)

T: To p

B: Bottom

G: Generating face

V: Vanishing face

L: Left

R: Right

N: Near

F: Far

velocity

vector

view

direction

Figure 5: 3-D Wrapping-around of the falling trajectory of

the raindrop.

the raindrop-vanishing face, V .

(3) 3-D Wrapping-around of Raindrops. Fig-

ure 5(d) illustrates 3-D wrapping-around for tracking

the raindrop trajectory. If a raindrop hits a wrapping-

around face (in this case S

), the raindrop is auto-

matically moved to the same position on the opposite

wrapping-around face (in this case S

) and continues

the same movement. In this example, only x coordi-

nate of the raindrop is changed from x

to x

as the

result of the 3-D wrapping-around processing. Rain-

drops can go under this processing more than once

until reaching the vanishing face.

4 RAINING LOCALITY

This section describes the way of representing raining

localities of intensity, density, movement and appear-

ance, which is one of the main contributions of the

proposed method.

4.1 Raindrop Generation with Local

High-density Distribution

In the situation of shower or heavy rain, we can see

cluster of raindrops with high intensity and high den-

sity moving. This paper develops a simple model for

rendering these non-uniform characteristics of rain-

ing scene. It is assumed that there are two kinds of

area in a raining scene. The ﬁrst area has temporally-

and-spatially uniform characteristics, while the other

GRAPP 2008 - International Conference on Computer Graphics Theory and Applications

276

area does not. We employ a raindrop-generating

model that deals with these different areas separately.

To be more precise, this proposed model generates

raindrops uniformly in the RP discussed in section

3 and synthesizes local high-density distributions of

raindrops. This synthesis procedure provides blocky

placements of raindrop in the rendered space. The lo-

cal high-density distributions of raindrops should be

moved over the course of time due to its nature prop-

erty. On the other hand, as described in section 3, this

proposed method only tracks the trajectories of rain-

drop in the rectangle parallelepiped and renders them.

Consequently, in the local high-density distributions

of raindrops, one needs to consider only raindrops

whose initial positions are included in the generating

face of the RP. However, for example, in producing an

animation sequence where the view direction gradu-

ally moves upward from initial horizontal direction,

the raindrop-generating face changes as the viewing

line tilts. Therefore, we deﬁne local high-density dis-

tribution on another plane that is independent of the

raindrop-generating face on the RP.

Figure 6 illustrates the raindrop-generating pro-

cedure of the proposed method. As shown in Fig-

ure 6(a), we deﬁne a plane S that is parallel to the

xz-plane of the world-coordinate system. As shown

in Figure 6(b), the local high-density distribution is

given as a form of Gauss distribution centered at the

reference point Q on the plane S. Since all rain-

drops have to be generated on the raindrop-generating

face of the RP in the proposed method, the raindrops

generated in the local high-density distribution are

projected onto the raindrop-generating face and are

merged with the raindrops normally generated in the

RP. The reference point Q is regarded as a particle

that is inﬂuenced by the wind effect and is moved like

a particle in the wind ﬁeld. The computing method

of the wind vector inﬂuencing the reference Q is the

same with the computing method of the wind vector

inﬂuencing the raindrops that is described in the next

subsection. Since local high-density distributions can

be processed independently, it is possible to move

several distributions simultaneously in a scene.

This paper classiﬁes raining types into three cat-

egories, that is, gentle rain, hard rain and heavy rain.

Each type has fundamental properties described in Ta-

ble 1. The proposed method assigns one of the fun-

damental properties of raining to the raindrops gen-

erated in the RP and assigns a heavier property to

the raindrops generated in the local high-density dis-

tribution. Consequently, two groups of raindrops in

the scene have distinctive differences in the terminal

velocity and their size, in addition to the raindrop-

generating density.

camera coordinate

system

world coordinate

system

(a)

camera coordinate

system

distribution

of raindrops

(b)

Figure 6: Local high-density distribution of raining.

4.2 Local Wind Inﬂuence

We assume that the raindrops are inﬂuenced just hor-

izontally by the wind ﬁeld and recognizes only two

factors as external forces on a raindrop as described

with the prior condition (1) in section 2.1, that is, the

vertical gravity and the horizontal wind inﬂuence. As

shown in Table 1, the vertical downward velocity is

assumed to be equivalent of a constant terminal ve-

locity depending on the degree of raining intensity.

On the other hands, the horizontal velocity is deter-

mined based on the two-dimensional ﬂuid simulation.

These vertical-and-horizontal results are utilized for

computing the trajectories of raindrop. It should be

noticed that this simpliﬁed procedure is very efﬁcient

for computing the unsteady wind ﬁeld and its inﬂu-

ence to the raindrop movement.

We employ a solving technique of Navier-Stokes

equations suggested by Stam (Stam, 1999) for simu-

lating two-dimensional wind ﬁeld. A simulation ﬁeld

is conﬁgured on the scene with enough area to deal

with changes in the ﬁeld of view and the calculus of

ﬁnite differences is applied to its mesh representation.

At the mesh points, the wind velocity vectors are com-

puted at the frame rate. The wind velocity vector at

the sub-mesh points can be interpolated linearly based

on four values at the nearest surrounding mesh points

and it is used as the wind velocity vector, V

in Fig-

ure 4.

REALISTIC RAIN RENDERING

277

Table 1: Fundamental properties of three types of raining

intensity.

hard

gentle

heavy

type

4.0

rainfall (mm/h)

diamiter (mm)

terminal

velocity (m/s)

15.0

40.0

1.0

4.0

1.5

5.0

2.0

6.0

4.3 Raindrop Rendering Considering

Illumination Locality

It is indispensable to reﬂect the occluding condition

of the sky and cloudiness on the trajectories of rain-

drop with a simpliﬁed method. Since the image-based

method using the environment mapping technique for

representing the reﬂection to the surface of raindrops

leads to the shortage problem of dynamic range

as discussed in section 2, the illuminance due to

skylight is utilized for computing the reﬂection on

the raindrop surface. The following parts describe

a simpliﬁed rendering model for representing the

inﬂuence of the lighting environment on a raindrop,

the computation of the illuminance due to skylight,

and the procedure of producing an illuminance map.

(1) Raindrop Rendering Model. The intensity of light

arriving at the viewpoint comes through a raindrop,

, is expressed as the following equation based on the

intensity of light reﬂected at a point on the raindrop,

, and the intensity of light transmitted through the

rain drop, I

∑

+ (1 − r

)ω

, (1)

where r

is a reﬂection ratio t a point x

on the rain-

drop, ωi is a contribution of x

(

∑

= 1.). It is well

known that the computation cost to obtain both I

and

is heavy. Then we set the following assumptions in

order to reduce computation cost for rendering rain-

drop considering locality of illumination:

• Since the major part of the I

consists of energy

which comes from the sky, the intensity of I

proportional to the illuminance due to skylight at

the location of the raindrop.

• Only the light comes from background of a rain-

drop whose direction is the same as I

, I

in Fig-

ure 7, is considered as a transmitted component.

Then, equation 1 can be re-written as follows,

= r

sky

+ (1 − r

T c

, (2)

where α

sky

is the coefﬁcient to transform from

illuminance due to skylight, E

sky

to intensity of light,

T c

is intensity of light which arrives at the viewpoint

raindrop

Figure 7: Raindrop rendering model.

passing through the center of the raindrop, r

is the

reﬂection ratio of the light corresponding to I

T c

(2) Computation of Illuminance due to Skylight. The

proposed method calculates the illuminance due to

skylight, E

sky

considering with sky occlusion in the

same manner of (Tadamura et al., 1993) with the fol-

lowing equation.

sky

∑

i=0

∑

j=0

τ(i, j)L(i, j)H(i, j)∆S

i j

, (3)

where τ(i, j) is an attenuation ratio due to raining

related to the sky-element (i, j). Since it is assumed

that there is no directional difference of attenuation,

τ(i, j) can be a constant. L(i, j) is a luminance at

the sky-element (i, j), H(i, j) is a blocking function

that if the sky-element (i, j) is blocked (black pixels

in Figure 8(b) and (c)) it gives 0 otherwise (gray

pixels) 1. ∆S

i j

is the area onto which the sky-element

(i, j) is projected. In the proposed method, the

sky luminance L(i, j) is computed by ALL SKY

MODEL(Igawa et al., 2001) that can consider the

location of the sun even in the overcast-sky (see the

details in (Igawa et al., 2001)). Figure 8(a) and (b)

show the typical occluding condition of the sky, at a

place surrounded by tall buildings and a place where

there is no tall building, respectively. It should be

noted that the surrounding buildings determine the

occluding conditions. Therefore, the light intensity

coming through a raindrop at the two places differs

greatly.

(3) Illuminance Map. Computing the illuminance at

all positions of raindrops with the above procedure re-

sults in a huge load. Therefore, we employ a scheme

for reducing the computation load without keeping

the locality of the light intensity coming through

a raindrop based on the occluding condition of the

sky. In this scheme, at the mesh points used for

two-dimensional wind simulation, the illuminance is

computed in advance and the obtained illuminance is

stored in the data table. This data table is referred as

an illuminance map.

GRAPP 2008 - International Conference on Computer Graphics Theory and Applications

278

(a) (b)

Figure 8: Examples of the sky occlusion; (a)surrounded by

tall buildings; (b)no tall building.

5 RESULTS AND DISCUSSIONS

We implemented the proposed method described

above on a PC with Intel Pentium 4 2.8GHz pro-

cessor, nVIDIA GeForce 6600GT and 2GB RAM.

In order to obtain a raining animation, we prepared

sequence images which were generated by a 3D-CG

package software as back-ground images for rain. At

ﬁrst, we will show an example for dealing with local-

ity of wind. Figure 9(a) is a scene obtained by apply-

ing conventional method, that is, a stable wind blows

with uniform direction and velocity over the scene.

Figure 9(b) is a frame of an animation created by the

proposed method treating with the temporal-spatial

locality of unsteady wind by solving Navier-Stokes

equation. Two portions of the image surrounded by

white broken line ellipses, P and Q, in Figure 9(b)

show the difference of the trajectory of raindrops.

The direction of the trajectory of the raindrops in Q

are quite different from the same potion of the Fig-

ure 9(a). Figure 10 shows a frame of an animation

for demonstrating the usefulness of non-uniform den-

sity and intensity of the rain distribution. We can see

the high density raindrops in the vicinity of the cen-

ter of the image. The animation allows us to inves-

tigate temporal movement of the high density rain-

drops. Figure 11 demonstrates the advantage of ren-

dering a raindrop with illuminance due to skylight.

The skylight around the sun is the brightest in the

whole sky even in the overcast sky. Figure 11(a) is

an example for a case that the high intensity portion

of the sky is not occluded by the buildings because of

high solar altitude, while bright potion is almost oc-

cluded by the buildings because of low solar altitude

in (b). These ﬁgures are noticed that illumination due

to skylight deeply affects on appearance of raindrop.

It takes about one second per frame to add rain tra-

jectories to the background image with our proposed

method. Since the proposed method draws the rain

streak with CPU, more than 70% of computation time

is consumed for scan-conversion of the rain streaks.

(a) A scene with uniform wind.

(b) A scene with non-uniform wind.

Figure 9: Effects of dealing with wind locality.

Figure 10: A scene with temporal-spatial localities of rain

intensity and density.

6 CONCLUSIONS

We have proposed a method for rendering the tra-

jectory of raindrops that can deal with both the

temporal-spatial localities of raining and arbitrary

camera movements. These localities include wind

velocity and direction, intensity and density of the

raindrops and light intensity coming through the rain-

drop. We employed 3-D wrapping-around of rain-

drops for efﬁcient calculation of the trajectory of the

raindrops. 2-D wind ﬁeld was simulated by solv-

REALISTIC RAIN RENDERING

279

(a) High solar altitude

(b) Low solar altitude

Figure 11: Comparison of the results with different solar

altitude.

ing Navier-Stokes equations and used for achieving to

represent temporal-spatial localities of wind. We used

a numerical sky luminance distribution model which

can consider the location of the sun for the overcast-

sky.

In order to use the proposed method to the real-

time application such as games and drive simulators,

we need to accelerate the rain streak drawing process

by using GPU. In future work, we will modify the

proposed method for utilizing GPU ability.

ACKNOWLEDGEMENTS

This research work was partially supported by the

Ministry of Education, Science, Sports and Cul-

ture of Japan, Grant-in-Aid scientiﬁc research(C)

19500084.

REFERENCES

Feng, Z., Tang, M., Dong, J., and Chou, S. (2005a). Real-

time rain simulation in cartoon style. In 9th Inter-

national Conference on Computer Aided Design and

Computer Graphics.

Feng, Z., Tang, M., Dong, J., and Chou, S. (2005b).

Real-time rendering of raining animation based on

the graphics hardware acceleration. In 9th Interna-

tional Conference on Computer Supported Coopera-

tive Work in Design, pages 734–739.

Garg, K. and Nayar, S. (2006). Photorealistic rendering of

rain streaks. In SIGGRAPH’06, pages 996–1002.

Igawa, N., Koga, Y., Matsuzawa, T., and Nakamura, H.

(2001). Development of all sky model of sky ra-

diance distribution under various atmospheric con-

ditions. In Istanbul 2001 International Lighting

Congress 2, pages 331–333.

Kaneda, K., Ikeda, S., and Yamashita, H. (1999). Animation

of water droplets moving down a surface. The journal

of visualization & computer animation, 10(1):15–26.

Reeves, T. W. (1983). Particle systems - a technique for

modeling a class of fuzzy objects. ToG, 2(2):359–376.

Rousseau, P., Jolivet, V., and Ghazanfarpour, D. (2006). Re-

alistic real-time rain rendering. Computers & Graph-

ics, 30(4):507–518.

Sato, T., Dobashi, Y., and Yamamoto, T. (2002). A method

for real-time rendering of water droplets taking into

account interactice depth of ﬁeld effects. In IWEC’02,

pages 110–117.

Sims, K. (1990). Particle animation and rendering us-

ing data parallel computation. Computer Graphics,

24(4):405–413.

Stam, J. (1999). Stable ﬂuids. In SIGGRAPH’99, pages

121–128.

Starik, S. and Werman, M. (2003). Efﬁcient simulation of

rain in videos. In Texture 2003 Nice, pages 95–100.

Tadamura, K., Nakamae, E., Kaneda, K., Baba, M., Ya-

mashita, H., and Nishita, T. (1993). Modeling of

skylight and rendering of outdoor scenes. Computer

Graphics forum, 12(3):189–200.

Tatarchuk, N. (2006). Artist-directable real-time rain ren-

dering in city environments. In SIGGRAPH2006

Course Note 26, chapter 3, pages 23–64. ACM.

Wang, L., Lin, Z., Fang, T., Yang, X., and Kang, S.

(2006). Real-time rendering of realistic rain. In SIG-

GRAPH’06 Sketches.

GRAPP 2008 - International Conference on Computer Graphics Theory and Applications

280