Hybrid MBlur A Systematic Approach to Augment Rasterization with Ray Tracing for Rendering Motion Blur in Games Yu Wei Tan

2025-04-27 0 0 8.5MB 11 页 10玖币

侵权投诉

Hybrid MBlur: A Systematic Approach to Augment Rasterization with

Ray Tracing for Rendering Motion Blur in Games

Yu Wei Tan a, Xiaohan Cui band Anand Bhojan c

School of Computing, National University of Singapore, Singapore

{yuwei, cuixiaohan}@u.nus.edu, banand@comp.nus.edu.sg

Keywords: Real-time, Motion Blur, Ray Tracing, Post-processing, Hybrid Rendering, Games.

Abstract: Motion blur is commonly used in game cinematics to achieve photorealism by modelling the behaviour of

the camera shutter and simulating its effect associated with the relative motion of scene objects. A common

real-time post-process approach is spatial sampling, where the directional blur of a moving object is rendered

by integrating its colour based on velocity information within a single frame. However, such screen space

approaches typically cannot produce accurate partial occlusion semi-transparencies. Our real-time hybrid ren-

dering technique leverages hardware-accelerated ray tracing to correct post-process partial occlusion artifacts

by advancing rays recursively into the scene to retrieve background information for motion-blurred regions,

with reasonable additional performance cost for rendering game contents. We extend our previous work with

details on the design, implementation, and future work of the technique as well as performance comparisons

with post-processing.

1 INTRODUCTION

We showcase a novel real-time hybrid rendering tech-

nique for the Motion Blur (MBlur) effect that com-

bines ray tracing and post-processing, exploiting ray-

traced information within a selection mask to reduce

partial occlusion artifacts in post-processed MBlur.

We provide a thorough illustration and evaluation of

hybrid MBlur, extending our published work (Tan

et al., 2020b) with the following additions:

• Extensive discussion of the design and implemen-

tation details of hybrid MBlur.

• Visual quality and performance evaluations of our

technique in comparison to post-processing.

• Explanation of presented extensions and future

work to the approach.

1.1 Background Information

1.1.1 Hybrid Rendering

Ray tracing is a common approach to produce realis-

tic effects like glossy reﬂections, depth of ﬁeld, and

ahttps://orcid.org/0000-0002-7972-2828

bhttps://orcid.org/0000-0002-6152-1665

chttps://orcid.org/0000-0001-8105-1739

motion blur. However, its high computation time has

limited its use mostly to ofﬂine rendering. On the

other hand, many post-process techniques have been

devised to approximate these effects on rasterized im-

ages to meet the constraint of interactive frame rates

for games, albeit with limited visual quality.

Given recent developments in hardware accelera-

tion, it is now the time to marry ray tracing with ras-

terization for games. Ray tracing can be employed

to correct certain visual ﬂaws originating from pure

post-process approaches, while still adhering to the

performance budgets of real-time rendering. This re-

sults in more realistic and convincing graphics while

keeping the gaming experience interactive, enhancing

the overall immersion of the player in the game.

1.1.2 MBlur

MBlur is the streaking or smearing effect of objects

in the direction of relative motion with respect to the

camera. It is employed in cinematics to emphasize the

relative speed of moving objects, producing a blurring

effect as shown in Figure 1. The MBlur effect has

been used to express the speed of rapid movement, the

disoriented state of mind of a character, or the dream-

like quality of a scene.

In cameras, MBlur is produced with the choice of

exposure time or shutter speed, which is the duration

arXiv:2210.06159v1 [cs.GR] 11 Oct 2022

Figure 1: MBlur effect. Image from authors.

for which the camera shutter is open per second. This

duration directly correlates to the amount of time the

camera sensor is exposed to light. Hence, when the

exposure time is long, every resulting image does not

represent an instantaneous moment but rather, an in-

terval consisting of successive moments in time. An

object moving in relation to the camera will therefore

appear on multiple areas of the sensor over time, re-

sulting in a directional blur effect of the object corre-

sponding to its motion.

As described in Jimenez (2014), moving objects

blur both outwards and inwards at their silhouettes,

making the region around their silhouettes appear

semi-transparent. Outer blur represents an object’s

blur into its neighbouring background while inner

blur applies to the blur produced within the silhouette

of the object itself. According to Cook et al. (1984),

accurate motion blurring takes into account areas of

background geometry occluded by any blurred fore-

ground. Hence, realistic MBlur should have accu-

rate partial occlusion by recovering true background

colour in inner blur regions instead of approximat-

ing this colour like many post-process approaches.

Background colour recovery in MBlur also helps to

prevent inaccuracies between real and approximated

backgrounds for sharp and blurred regions respec-

tively. Hence, we employ ray tracing in our approach

to retrieve the exact colour of occluded background.

2 RELATED WORK

2.1 Hybrid Rendering

The use of ray tracing to complement rasterization

techniques to achieve high graphics quality while

maintaining interactive frame rates has been at-

tempted for several visual effects over the years. In

Beck et al. (1981) and Hertel et al. (2009), shadow

mapping is employed to mark shadow boundaries

and reduce the actual number of shadow rays to be

traced, of which only rays from regions deemed po-

tentially inaccurate by shadow mapping are traced

in Lauterbach and Manocha (2009) to generate pre-

cise and alias-free shadows. For realistic reﬂections,

Macedo et al. (2018) invokes ray tracing only on pix-

els with reﬂections that cannot be solved with just

screen space information. In Marrs et al. (2018), tra-

ditional temporal anti-aliasing (TAA) is also extended

with ray-traced supersampling for regions with a high

chance of TAA failure.

In Cabeleira (2010), diffuse illumination is com-

puted via rasterization while reﬂections and refrac-

tions are handled by ray tracing. In more general

pipeline approaches, Chen and Liu (2007) substitutes

the generation of primary rays in recursive ray tracing

(Whitted, 1979) with rasterization for improved per-

formance. This approach is extended in Andrade et al.

(2014) which adheres to a render time limit by tracing

rays only for objects of the highest importance.

2.2 MBlur

Current implementations of MBlur include the use of

an accumulation buffer. Within the exposure time

when the camera shutter is set to be open, a series

of discrete snapshots rendered at multiple time offsets

from the actual time value of the image is integrated to

produce MBlur such as in Korein and Badler (1983)

and Haeberli and Akeley (1990). However, as dis-

cussed in McGuire et al. (2012), this approach is com-

putationally expensive in terms of shading time. Fur-

thermore, in the case of undersampling, this method

can lead to a series of disjoint ghosting artifacts in the

direction of motion instead of a continuous blur.

Stochastic sampling (Cook, 1986) for MBlur is

a Monte Carlo technique based on Halton (1970)

in which sampling is performed in a randomized

and nonuniform manner along the time dimension.

McGuire et al. (2010) incorporates stochastic sam-

pling with ray tracing by conservatively estimating

the total convex hull bound for the area of each

moving triangle during the exposure time, leverag-

ing hardware rasterization. Stochastically distributed

rays are then shot into the scene within this bound for

shading. This approach makes use of distributed ray

tracing (Cook et al., 1984), where rays are distributed

in time and the colour at each hit point is averaged to

produce the ﬁnal image.

Cook et al. (1987) translates micropolygons for

each sample according to a jittered time. On the

other hand, Fatahalian et al. (2009) and Sattlecker and

Steinberger (2015) achieve efﬁciency in performance

by extending the bounding box of each micropolygon

to consider its pixels and their associated velocities in-

stead. Hou et al. (2010) is an approach that makes use

of 4D hyper-trapezoids to perform micropolygon ray

tracing. Methods of approximating the visibility func-

tion to sample have also been devised such as in Sung

et al. (2002). However, as explained in Sattlecker and

Steinberger (2015), at low sample rates, these meth-

ods will also exhibit ghosting artifacts whereas in-

creasing the number of samples would lead to noise.

To produce the right amount of blur, some real-

time approaches like Rosado (2007), Ritchie et al.

(2010) and Sousa (2013) make use of per-pixel ve-

locity information by accumulating samples along the

magnitude and direction of velocities in the colour

buffer. Other techniques in Korein and Badler (1983),

Catmull (1984) and Choi and Oh (2017) accumulate

the colours of visible passing geometry or pixels with

respect to a particular screen space position while

Gribel et al. (2011) makes use of screen space line

samples instead. Potmesil and Chakravarty (1983)

represents the relationship between objects and their

corresponding image points as point-spread functions

(PSFs), which are then used to convolve points in

motion. Leimk¨

uhler et al. (2018) splats the PSF of

every pixel in an accelerated fashion using sparse

representations of their Laplacians instead. Time-

dependent edge equations, as explained in Akenine-

M¨

oller et al. (2007) and Gribel et al. (2010), and 4D

polyhedra primitives (Grant, 1985) have also been

used for MBlur geometry processing. Recently, a

shading rate-based approach involving content and

motion-adaptive shading in Yang et al. (2019) has also

been developed for the generation of MBlur.

In particular, attempts to simulate nonlinear

MBlur include Gribel et al. (2013) and Woop et al.

(2017). Our hybrid technique only considers linear

inter-frame image space motion for now, but we in-

tend to provide support for higher-order geometry

motion in the future. We also assume mainstream

ray tracing acceleration architecture widely available

in modern gaming workstations. The GA10x RT

Core of the newest NVIDIA Ampere architecture

provides hardware acceleration for ray-traced motion

blur (NVIDIA Corporation, 2021) but is only found in

the premium GeForce RTX 30 Series graphics cards.

3 DESIGN

Our hybrid MBlur approach, as illustrated in Fig-

ure 2, compensates for missing information in post-

processed MBlur with the revealed background pro-

duced by a ray trace-based technique.

A Geometry Buffer (G-Buffer) is ﬁrst gener-

ated under a deferred shading set-up, rendering tex-

tures containing per-pixel information such as cam-

era space depth, screen space velocity and rasterized

colour. The same depth, velocity and colour infor-

mation for background geometry is produced by our

novel ray reveal pass within a ray mask for pixels in

the inner blur of moving foreground objects. A tile-

dilate pass is then applied to these 2 sets of buffers

to determine the sampling range of our gathering ﬁl-

ter in the subsequent post-process pass. Both the ray-

revealed and rasterized output are blurred by this post-

process pass and lastly composited together.

3.1 Post-process

The McGuire et al. (2012) post-process MBlur ren-

ders each pixel by gathering sample contributions

from a heuristic range of nearby pixels. We adapt

this approach to produce a motion-blurred effect sep-

arately for rasterized and ray-revealed information.

As suggested in Rosado (2007), motion vectors

are given by ﬁrst calculating per-pixel world space

positional differences between every frame and its

previous frame, followed by a translation to screen

space. By using the motion vector between the last

frame and the current frame, we simulate the expo-

sure time of one frame. Then, we follow the approach

of McGuire et al. (2012) in calculating the displace-

ment of the pixel within the exposure time by scal-

ing the inter-frame motion vector with the frame rate

of the previous frame as well as the exposure time.

Considering the full exposure as one unit of time,

this displacement can be interpreted as a per-exposure

velocity vector. This approach uses the assumption

that the motion vector of each pixel between the pre-

vious frame and the current frame remains constant

throughout the exposure time.

Jimenez (2014) is a technique that is based on

McGuire et al. (2012). As described in Jimenez

(2014), the major problems to be addressed when

producing a post-processed MBlur are the range of

sampling, the amount of contribution of each sam-

ple as well as the recovery of background geometry

information for inner blur. With our method for cal-

culating per-exposure velocities, we adopt McGuire

et al. (2012)’s approach in determining the magni-

tude of the sampling range and representing different

amounts of sample contribution, as illustrated in the

Appendix for completeness. As shown in Figure 3,

McGuire et al. (2012) centers the sampling area at

the target pixel, creating a blur effect both outwards

and inwards from the edge of the object. Although

this produces a more uniform blur for thin objects and

the specular highlights of curved surfaces, it poses the

problem of having to smoothen the transition between

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

HybridMBlur:ASystematicApproachtoAugmentRasterizationwithRayTracingforRenderingMotionBlurinGamesYuWeiTana,XiaohanCuibandAnandBhojancSchoolofComputing,NationalUniversityofSingapore,Singaporefyuwei,cuixiaohang@u.nus.edu,banand@comp.nus.edu.sgKeywords:Real-time,MotionBlur,RayTracing,Post-processing,Hyb...

展开>> 收起<<

Hybrid MBlur A Systematic Approach to Augment Rasterization with Ray Tracing for Rendering Motion Blur in Games Yu Wei Tan.pdf

共11页,预览3页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

Hybrid MBlur A Systematic Approach to Augment Rasterization with Ray Tracing for Rendering Motion Blur in Games Yu Wei Tan

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: