CN102982575B - Hair rendering method based on ray tracking - Google Patents

Abstract

The invention discloses a hair rendering method based on ray tracking. Through the method, the sampling rate required for hair rendering is remarkably reduced through collecting a sample in a pixel into a light cone. The method has the beneficial effects that in order to reduce the cost generated by the crossing of the light cone and hairlines, a surrounding body accelerating structure is established to quickly obtain the hairlines possibly crossed with the light cone, and hairlines projected onto an imaging plane are approximated by using a series of connected belt shapes; and in addition, the expenditures of synthesis and filter transparent sampling are greatly reduced through ignoring light cone internal coloring, non-transparency and shading change. Compared with the performance of other methods, the performance of the hair rendering method is improved by an order of magnitude under the situation that images with the same quality are generated.

Description

A kind of fur rendering method based on ray trace

Technical field

Design configuration rendering technique field of the present invention, particularly relates to a kind of ray tracing method for fur rendering.

Background technology

Hair is most important characteristic in virtual image, and most film and game in virtual role in all apply hair.

Traditional high-quality for hair is played up and is depended on ray trace.Hair is subdivided into micro-polygon by the people such as Hou, and the ray trace then by optimizing is played up, specifically can with reference to HOU, Q., QIN, H., LI, W., GUO, B., AND ZHOU, K.2010.Micropolygon ray tracing with defocus andmotion blur.ACM Trans.Graph.29 (July), 64:1 – 64:10.Hair is directly carried out ray trace as a kind of special geometric expression by the people such as Nakamaru, see NAKAMARU, K., AND OHNO, and Y.2002.Ray tracing for curves primitive.In WSCG, 311 – 316.Without exception, the fine structure of hair is for sampling and anti-aliasingly define very large difficulty, and these methods all need the very high undistorted high-quality of sampling rate guarantee to play up.

Beam ray tracing technology expands for the one of ray tracking technology, and that namely carries out following the tracks of is being not a line but a light beam.This kind of technology is intended to solve the undersampling problem that traditional ray trace faces.The shape of concrete light beam can be divided into prism, circular cone and hypercube, concrete make can respectively with reference to HECKBERT, P.S., AND HANRAHAN, P.1984.Beam tracing polygonal objects.SIGGRAPHComput.Graph.18 (January), 119-127, AMANATIDES, J.1984.Ray tracing withcones.SIGGRAPH Comput.Graph.18 (January), 129-135, ARVO, J., AND KIRK, D.1987.Fast ray tracing by ray classification.SIGGRAPH Comput.Graph.21 (August), 55-64.Igehy proposes the interaction that a kind of method that reach by light beam is expressed as derivative is come between processing beam and scene, and the definition of light derivative can with reference to IGEHY, H.1999.Tracing raydifferentials.In ACM SIGGRAPH, 179-186.The all possible light in beam range is effectively considered, so they can alleviate the problem of traditional ray trace undersampling due to these class methods.But construct light beam and consider that the computation complexity acted between light beam and object scene is often higher than only considering single ray.So usually need specific pencil tracing method for object different classes of in scene.

The people such as Wand propose a kind of method point set with anisotropic light cone and filtering in advance being asked friendship, and apply a hierarchical structure for accelerating, concrete visible WAND, M., AND STRASSER, W.2003.Multi-resolution point-sample raytracing.In Graphics Interface, 139-148.But their method be not suitable for fur rendering, in order to avoid very thin hair is together fuzzy, its method requires that the distance between point must be over the distance between hair, this degree just making a denseness of set exceed reality can to bear.The method is used for the shade calculating aggregation (as bushes or hair) by the people such as Lacewell, and first the shielding rate of precomputation storage direction on an individual accelerating structure, then operationally calculates soft shadow and ambient light occlusion effectively.But because the light of the method and hair ask friendship very approximate, it can not be used for the sight line process of hair.

Beam ray tracing technology for hair mainly faces two unsolved challenges of forefathers.First light cone and hair ask the computational complexity of friendship.Accurately calculating these asks friendship to require high sampling rate, and this disagrees with the original intention reducing sampling rate.Second challenge is the interior visibility problem with transparent effect of process light cone.This visibility of accurate Calculation needs a large amount of circular cone samplings, brings high lucidification disposal cost.

Summary of the invention

The object of the invention is to for the deficiencies in the prior art, a kind of fur rendering method based on ray trace is provided.The high-quality that the present invention is used for carrying out hair and camera lens special efficacy thereof (refraction, reflection and profound) is played up.

The object of the invention is to be achieved through the following technical solutions: a kind of fur rendering method based on ray trace, the method includes following steps:

(1) hair is subdivided into band shape, and enclosure body accelerating structure is set up to all band shapes;

(2) generate light cone according to the information of scene light line following, traversal enclosure body accelerating structure, obtain may with the band shape of conical intersection;

(3) calculate band shape crossing of circular cone and expression hair, and synthesize sampled result;

(4) result of being sampled by hair and the result of scene light line following are synthesized, and generate final image.

The invention has the beneficial effects as follows, method of the present invention is a kind of based on circular cone, and specially for the pencil tracing method of hair.Circular cone is selected to be because circular cone is the most close to the shape of image filter as the shape of light beam.Hair is approximately some end to end band shapes by the method, and is projected to light cone normal plane, circular cone and hair is asked friendship problem to be converted into circularly to ask friendship problem with a series of quadrilateral.In addition, the present invention have ignored the change that circular cone is inner painted and block, and carries out the synthesis of transparent sampling in units of light cone.In order to make such synthesis mode more accurate, introduce the concept of " effective opacity ", using each circular cone and the factor of crossing area as its opacity being with shape.Therefore, the present invention succinctly effectively can process the hair high-quality comprising lens effect and play up problem.Apply method of the present invention and generate more than the image of an equal quality time order of magnitude fewer than existing additive method used, and be similar to the error caused and can be increased a little over-extraction sample and effectively control.

Accompanying drawing explanation

Fig. 1 is that light cone is formed and parameterized schematic diagram;

Fig. 2 is that the band shape in light cone coordinate system is similar to schematic diagram;

Fig. 3 is that quadrilateral and circle ask the schematic diagram handing over the summit, limit of quadrilateral in fast method to be all positioned at circle;

Fig. 4 quadrilateral and circle are asked and are handed over the summit, limit of quadrilateral in fast method to lay respectively at the schematic diagram in circle and outside circle;

Fig. 5 quadrilateral and circle ask hand over quadrilateral in fast method summit, limit all outer the and Bian Yuyuan of circle without the schematic diagram of intersection point;

Fig. 6 quadrilateral and circle ask summit, the limit all schematic diagram that intersect of the outer and Bian Yuyuan of circle handing over quadrilateral in fast method.

Embodiment

Light cone tracking of the present invention, includes following steps:

(1) hair is subdivided into band shape, and enclosure body accelerating structure is set up to all band shapes

First all hairs are subdivided into according to the size of screen prjection and are no more than the long line segment of 3 pixels, then on each line segment end points, carry out painted calculating, the single photon diffusion models that painted computing application Marschner proposes, circular can with reference to MARSCHNER, S.R., JENSEN, H.W., CAMMARANO, M., WORLEY, S., AND HANRAHAN, P.2003.Light scatteringfrom human hair fibers.ACM Trans.Graph.22,3,780 – 791.The width data of hair is stored on each line segment end points together with colouring value.Then according to surface area, an enclosure body accelerating structure is built with the line segment of width to this series: for each line segment, and two implicit expression spheroids that the width at its end points place is formed, build the bounding box of an axle alignment, and using this bounding box as the base unit building enclosure body accelerating structure.Concrete construction method can with reference to GOLDSMITH, J., AND SALMON, J.1987.Automatic creation of object hierarchies for ray tracing.IEEE CG & A7,5,14 – 20.

(2) generate light cone according to the information of scene light line following, traversal enclosure body accelerating structure, and obtain may with the band shape of conical intersection.

Be the image of m for one containing n pixel and hair sampling rate, need n × m circular cone, n and m is natural number, if m=1, and each circular cone correspondence pixel, if m>1, each circular cone correspondence sub-pixel; Each circular cone is represented by two vectorial Rx and Ry on focus o, light cone direction v and normal plane.Rx and Ry represents two axles of the ellipse of light cone formation crossing with normal plane.Normal plane is defined as plane along radiation direction advance unit distance and vertical with radiation direction.The light that the present invention plays up generation for direct light, the light reflecting in scene or produce after reflection and the corresponding Deep Canvas sent by viewpoint generates circular cone.

Playing up for the main line of sight, using the focus of viewpoint as circular cone, and connects on imaging plane and covers the circle of a pixel to form the circular cone required for tracking, as shown in Figure 1.For the effect that the depth of field and refraction reflect, the present invention adopts and WAND, M., AND STRASSER, W.2003.Multi-resolutionpoint-sample raytracing.In Graphics Interface, and the method that 139 – 148 are identical generates circular cone.

After obtaining the light cone needing to follow the tracks of, light cone and enclosure body accelerating structure is first taken to carry out asking friendship.Each light cone is guarded approximate with two pyramids, two pyramids all send from light cone focus, respectively along the positive negative direction of light.Four faces of pyramid are respectively by four line segments vertical respectively with the positive negative direction of oval two axles on normal plane, and light cone focus is formed.Utilize segmentation theorem of principal axes (concrete visible http://en.wikipedia.org/wiki/Separating_axis) afterwards, judge whether the bounding box in light cone and enclosure body structure asks friendship fast.Light cone and enclosure body structure ask intersection to obtain a series of band shape that may be crossing with it, it should be noted that, in order to realize the correct synthesis of hair and scene light line following result in following step 4, ask the depth capacity needing during friendship to consider other geometric objects in light cone and scene, and abandon the band shape that the degree of depth exceedes this depth capacity.Next concrete calculate that light cone is with shape with these ask knot really.

(3) calculate band shape crossing of circular cone and expression hair, and synthesize sampled result.

First illustrate that a band shape and light cone ask the calculating of friendship.First two of line segment end points and corresponding width thereof are projected to the normal plane of light cone.The width information that end points after projection stores can as radius with this point for the center of circle generates a circle, an optional diameter can produce two points.For the point that hair end does not have other line segment to share like this, directly get with project after the vertical diameter of line segment; And for the point having two line segments to share in the middle part of hair, get the center line of two line segments to generate a pair new point.So, a quadrilateral will be formed after each band shape projection.And light cone is formed on normal plane is an ellipse determined by Rx and Ry.According to two axle Rx and Ry of ellipse, conversion T=[Rx, Ry] can be defined.By its inverse transformation, can transform in the coordinate system of light cone by oval and quadrilateral, elliptic transformation is circular, and quadrilateral is still transformed into quadrilateral.As shown in Figure 2.In the coordinate system, the problem intersecting area transforms in order to quadrilateral and circular crossing area for cutting.

The invention provides the area that an efficient method calculates any one quadrilateral and circular intersection.The center of circle on every bar limit and circle that the present invention gets quadrilateral forms a triangle.Obtain with each triangle intersect the area that has positive negativity afterwards by circular.The absolute value that finally all signed areas are added its result is exactly the area of quadrilateral and circular intersection.Triangle and the signed area of circular intersection gained are calculated respectively by the different situations of triangle and circular intersection, summarizes four classes below and intersect situation, as shown in figures 3 to 6.Without loss of generality, the quadrilateral of current consideration is called while AB by the present invention, is called leg-of-mutton two other while AO and BO.If the straight line at AB place, limit and circle have two intersection points and all outside AB, then this is first kind situation (Fig. 3).Crossing area is now exactly leg-of-mutton area.If limit AB place straight line and circle have two intersection points and only have an intersection point outside, then this is Equations of The Second Kind situation (Fig. 4)).Crossing area be now one of them little triangle and one fan-shaped and.If two intersection points are all within AB, then this is the 3rd class situation (Fig. 5)).Area be now one of them little triangle and two fan-shaped and.If limit AB is completely outside circle, this is the 4th class situation (Fig. 6) and its area is a fan-shaped area.Be worth noting that, the inventive method only consumes the interim storage of fixed qty, thus realizes easily through GPU register.

The sampled point synthesis that next each band shape will be produced is to generate final sampled result.In the inventive method, assuming that the opacity of hair in light cone, painted and to block be all constant, the result of sampling can be approximately thus:

In this formula, Ω is the set of band shape that possible crossing with light cone, and Υ is light sampling all in light cone, Υ _jthe light sampling sending to band shape j, α _jthe opacity of respective belt shape j, φ _jthe on average painted of band shape j, and Ω _jall set of blocking the band shape of band shape j.It should be noted that || Υ _j||/|| Υ || be exactly the ratio that band shape covers light cone area, this ratio be can be regarded as a part for opacity by the present invention, thus obtains effective opacity α ' _k=α _k|| Υ _k||/|| Υ ||.

In other words, the band shape that the present invention may be crossing with light cone to each, calculates an average colouring value and effective opacity, and generates a sampling for synthesis below.

In order to the internal memory consumption of limit algorithm, present invention employs self-adaptation Transparency Methodology (with reference to SALVI, M., MONTGOMERY, J., AND LEFOHN, A.E.2011.Adaptive transparency.InProceedings of HPG, 119 – 126), the method allows the transparent sampled point of Dynamic Maintenance in units of light cone.The method is adopted to completely avoid data contention and control internal memory consumption completely.

(4) result of being sampled by hair and the result of scene light line following are synthesized, and generate final image.

Carry out fur rendering because method of the present invention is specifically designed, so the complete image of band scene will be generated, need the inventive method and common Ray Tracing Algorithm to carry out integrated.Because the light cone for hair ray trace and the light for scene light line following are not relations one to one, therefore need by therefore needing, the ray trace sampled point of other scene geometric object on the ray trace sampled point of hair corresponding in each light cone and corresponding light to be synthesized, and consider mutually to block.For this reason, The present invention gives the technical scheme of following synthesis two parts ray trace result: first ray trace carried out to scene and on every bar light, obtain sampled point (ray tracing method can with reference to TIMO AILA and SAMULI LAINE, 2009, Understanding the Efficiency of Ray Traversal on GPUs, Proc.High-PerformanceGraphics2009,145-149); Owing to screening band shape by the degree of depth of the maximum scene aggregation sampler body in light cone in step 2, eliminate completely by hair that scene is blocked, therefore only have in the sampling of self-adaptation transparency and do not sampled by the hair that scene is blocked completely, sampling scene light line following on these samplings and corresponding light produced, according to the synthesis of degree of depth tandem, just obtains the color value needing to export.When calculating the degree of depth of band shape sampling, adopt band shape and the depth capacity of light cone intersection and the mean value of minimum-depth.Can with reference to LEVOY by the method that tandem carries out synthesizing by transparent sample, M.1990.Efficient raytracing of volume data.ACM Trans.Graph.9,3,245 – 261.

Inventor is equipped with CPU(4GB internal memory at one) and NVIDIA GTX570GPU(1280MB video memory) machine on achieve some embodiments of the present invention.

Embodiment one, inventor depict three typical hair scenes by the inventive method, comprise two squirrels to stand in complex scene and to stand on sphere of reflection with the scene (scene one) of Deep Canvas and refraction reflecting effect, squirrel and with Deep Canvas scene (scene two), one dynamically with the scene (scene three) of the hair let it flow of Deep Canvas.Table 1 is the performance statistics table of the inventive method in all scenes, and method of the present invention is compared with traditional Ray Tracing Algorithm (generating same quality to need).

Table 1:

Easily find out from table 1, owing to reducing the sampling rate required for traditional ray tracing method, method of the present invention accelerates playing up of the main line of sight greatly; And reduce few due to the sampling rate reflecting reflection ray, thus reflect reflecting effect play up accelerate little.

Embodiment two, inventor test the control action of light cone track algorithm over-extraction sample for its approximate error in above-mentioned scene one.The inventive method mainly have employed two and is similar to accelerate high-quality fur rendering.First enough little have ignored, light cone inside is painted for the band shape generated by hypothesis segmentation, opacity and the change of blocking.Secondly, the hair after projection is approximated to a series of quadrilateral by the present invention.But these two the approximate errors introduced can by reducing the size of light cone, and the sampling rate that namely raising light cone is followed the tracks of effectively controls.Compared with the result produced under different sampling rate and legitimate reading (classic method 31x31 sample gained), calculate the difference value between two results, use RMS misrepresentation.As shown in table 2, along with the lifting of light cone tracking sampling rate, approximate error is in remarkable minimizing.

Table 2:

This result clearly shows method of the present invention can carry out control algolithm error by improving sampling rate.It should be noted that the inventive method has seemed just the same with reference picture when 7x7 samples, and still can near an order of magnitude on the time.

Claims (1)

1., based on a fur rendering method for ray trace, it is characterized in that, the method includes following steps:

(1) hair is subdivided into band shape, and enclosure body accelerating structure is set up to all band shapes; Be specially: be first subdivided into according to the size of screen prjection by all hairs and be no more than the long line segment of 3 pixels, then on each line segment end points, carry out painted calculating, the width data of hair is stored on each line segment end points together with colouring value; Then according to surface area, an enclosure body accelerating structure is built with the line segment of width to this series: for each line segment, and two implicit expression spheroids that the width at its end points place is formed, build the bounding box of an axle alignment, and using this bounding box as the base unit building enclosure body accelerating structure;

(2) generate light cone according to the information of scene light line following, traversal enclosure body accelerating structure, obtain may with the band shape of conical intersection; Be specially: be the image of m for one containing n pixel and hair sampling rate, need n × m circular cone, n and m is natural number, if m=1, and each circular cone correspondence pixel, if m>1, each circular cone correspondence sub-pixel; Each circular cone is represented by two vectorial Rx and Ry on focus o, light cone direction v and normal plane, Rx and Ry represents two axles of the ellipse of light cone formation crossing with normal plane, normal plane is defined as plane along radiation direction advance unit distance and vertical with radiation direction, and the light playing up generation for direct light, the light reflecting in scene or produce after reflection and the corresponding Deep Canvas sent by viewpoint generates circular cone; After obtaining the light cone needing to follow the tracks of, light cone and enclosure body accelerating structure is first taken to carry out asking friendship; Each light cone is guarded approximate with two pyramids, two pyramids all send from light cone focus, respectively along the positive negative direction of light; Four faces of pyramid are respectively by four line segments vertical respectively with the positive negative direction of oval two axles on normal plane, and light cone focus is formed; Segmentation theorem of principal axes is utilized to judge whether the bounding box in light cone and enclosure body structure asks friendship fast afterwards;

(3) calculate band shape crossing of circular cone and expression hair, and synthesize sampled result; Comprise following sub-step:

(3.1) quadrilateral is formed by being with shape to project on normal plane: the normal plane first two of line segment end points and corresponding width thereof being projected to light cone; The width information that end points after projection stores can as radius with this end points for the center of circle generates a circle, an optional diameter can produce two points; For the point that hair end does not have other line segment to share like this, directly get with project after the vertical diameter of line segment; And for the point having two line segments to share in the middle part of hair, get the center line of two line segments to generate a pair new point; So, a quadrilateral will be formed after each band shape projection; And light cone is formed on normal plane is an ellipse determined by Rx and Ry; According to two axle Rx and Ry of ellipse, conversion T=[Rx, Ry] can be defined; By its inverse transformation, can transform in the coordinate system of light cone by oval and quadrilateral, elliptic transformation is circular, and quadrilateral is still transformed into quadrilateral;

(3.2) area of any one quadrilateral and circular intersection is calculated: every bar limit and the circular center of circle of getting quadrilateral form a triangle, obtain with each triangle intersect the area that has positive negativity afterwards by circular, the absolute value that finally all signed areas are added its result is exactly the area of quadrilateral and circular intersection, triangle and the signed area of circular intersection gained are calculated respectively by the different situations of triangle and circular intersection, the quadrilateral of current consideration is called while AB, leg-of-mutton two other is called while AO and BO, summarize four classes below and intersect situation: if the straight line at AB place, (a) limit and circle have two intersection points and all outside AB, crossing area is now exactly leg-of-mutton area, if b () limit AB place straight line and circle have two intersection points and only have an intersection point outside, crossing area be now one of them little triangle and one fan-shaped and, if c () two intersection points are all within AB, area be now one of them little triangle and two fan-shaped and, if d () limit AB is completely outside circle, area is now a fan-shaped area, then the sampled point synthesis each band shape produced is to generate final sampled result, the opacity of the hair in supposition light cone, painted and to block be all constant, the result of sampling can be approximately:

$\begin{array}{c}L\≈{\mathrm{\Σ}}_{j\∈\mathrm{\Ω}}\left(\right|\left|{\mathrm{\γ}}_j\right|\left|{\mathrm{\α}}_j{\mathrm{\φ}}_j{\mathrm{\Π}}_{k\∈{\mathrm{\Ω}}_j}(1-{\mathrm{\α}}_k^{\′})\right)/\left|\right|\mathrm{\γ}\left|\right|\\ ={\mathrm{\Σ}}_{j\∈\mathrm{\Ω}}\left({\mathrm{\α}}_j^{\′}{\mathrm{\φ}}_j{\mathrm{\Π}}_{k\∈{\mathrm{\Ω}}_j}(1-{\mathrm{\α}}_k^{\′})\right)\end{array};$

In this formula, Ω is the set of band shape that possible crossing with light cone, and γ is light sampling all in light cone, γ _jthe light sampling sending to band shape j, α _jthe opacity of respective belt shape j, φ _jthe on average painted of band shape j, and Ω _jall set of blocking the band shape of band shape j; || γ _j||/|| γ || be the ratio that band shape covers light cone area, this ratio counted effective opacity α ' that opacity obtains being with shape k _k=α _k|| γ _k||/|| γ ||;

(4) result of being sampled by hair and the result of scene light line following are synthesized, and generate final image.