DE10331231B4 - Method for transferring depth information into the depth memory of a graphics card for displaying 3D panoramic images - Google Patents

Abstract

method for transmission depth information into the depth memory of a video card for presentation of 3D panoramic images, characterized in that the transmission the depth information for each pixel of the 3D panoramic image to be displayed on the graphics chip the graphics card using a program running on the graphics chip takes place, wherein as input data for the program at least one Texture that encodes the depth information to be displayed Contains 3D panoramic image, to disposal where the depth information of the 3D panoramic image to be displayed is in the at least one texture are coded such that the depth information of a pixel by a quotient of two in the texture used existing values is formed.

Description

The The invention relates to a method for transmitting depth information in the depth memory of a graphics card for the display of 3D panoramic images.

The Invention is particularly in rapid representation methods a landscape picture with simultaneous representation of one or several flying objects applicable. Because the flying objects in the room Before the viewer can move relatively freely, they must from the landscape partially covered can be - depending on their distance to the viewer. The landscape is in shape a two-dimensional photograph, so does not contain any information about the removal of objects visible to the viewer. The information about the Distance to the viewer ("depth information") must therefore a second way into the system. For the user The easiest way to do this is by means of a kind of "depth image" in which colors are used encodes the depth information individually for each pixel. With this type of data generation, the user can very easily to integrate a new landscape with a 360 ° -photography only produce a suitable depth image.

application find such methods u. a. in visualization systems of military Training devices. The exercise space ROLAND 2 e.g. has a single-channel optical Visor. The visor image shows a fixed eye point 360 ° -Landscape with a photo-realistic terrain representation. The viewer can turn around a virtual axis, so that the visor image represents any part of the landscape. Within this visor image can moving flight destinations. By bringing information about the Distance to the viewer can the flight destinations partly covered by the terrain become. Other boundary conditions for the visor image are one Animation rate of at least 60 frames per second, a resolution of at least 768 × 768 Pixels and the possibility that the viewer is up to 90 ° per Second turn around its axis.

Out cost reasons is for the representation of the landscape and the flying objects standard PCs (so-called COTS products: Commercial Often The Shelf). Due to the rapid development of computer technology in recent years Years became the efficiency of these systems - in particular in the graphics area - so greatly increased, that a realization seems possible in principle. Indeed Such COTS products are not for this particular task Graphic generation for the generation of 3D images by means of a photograph and a "depth image", so that it usually lacks the necessary speed becomes.

In the DE 196 20 858 A1 there is shown a method of color rendering in computer graphics systems, in particular for producing a color representation involving the distance of the viewer to the displayed subject. This distance is approximated by the depth information of the object.

The WO98 / 40848 describes a method for storing pixel data for one three-dimensional representation of an image scene, after which the same Memory area as depth memory as well as image memory can be used. Because the depth information in the course of the procedure will be overwritten by the display information (color values) the presentation information marked as such, to override again to prevent.

task The invention is to provide a method using from standard consumer-grade hardware to a quick representation the 3D landscape at a resolution of at least 768 × 768 Pixels with at least 60 frames per second. The viewer should be very fast (i.e., up to 90 ° / s) around his Axis can rotate.

These Task is solved by the method according to claim 1. A advantageous embodiment of Method is the subject of another claim.

According to the invention, within the 3D visualization process, the transmission of the depth information of the 3D panoramic image to be displayed into the depth memory (eg the Z-buffer, but there are also other forms of depth memory such as the W-buffer) of the graphics card on per-pixel -Basis directly on the graphics chip of the graphics card by means of a program running on the graphics chip. For this purpose who the program as input data at least one texture that contains in coded form the depth information of the 3D panorama image to be displayed provided.

There the presentation process in the representation of the landscape none Overwrites image areas several times, can the landscape in almost constant speed incl. the depth information is displayed. This will be guarantees for the Display speed possible. The presentation speed of the landscape including the depth values is only on the processing speed of the program and the necessary time for the transfer of the data from the main memory of the PC to the graphics memory the graphics card (in current PC systems, the AGP bus comes to Use).

The running on the graphics chip Program may be one in modern 3D visualization technology used so-called "pixel Shader "(in the Literature often also called texture shader, fragment shader or just as Shader). "Pixel Shader "is common as a synonym for a present on the graphics card assembler program, which calculates a point of a polygon to be displayed. The name derives is made up of two parts. "Pixel" refers to the calculation of a point. "Shader" refers to the actual purpose of a pixel shader - the calculation shades or lighting and color information of a Point.

With introduction of high-end graphics cards of the Geforce4 family has the manufacturer nVidia a method called "Z-Correct Bump mapping "presented, that specifically fix the problem of water surface penetration should. Details can be found on the website the manufacturer www.nvidia.com. This technique was first a possibility created on a per pixel basis influence the depth value of a To take pixels inside a polygon. The present invention based on that for the "Z-Correct Bump Mapping procedure Pixel shader commands one at a time other purpose, namely for the fast transmission the depth values in the depth memory of the graphics card to feed.

This succeeds by a suitable encoding of the input data. Because the used texture contains depth information, it will also below referred to as depth texture. This is to delineate textures, which the actual image data of the 3D panorama image to be displayed contain as they usually do from pixel shaders are processed.

by virtue of the given functionality The pixel shader instructions according to the "Z-Correct Bump Mapping" method must have the values contained in the depth texture are encoded in such a way that those generated by the pixel shader and transferred to the depth memory Values the actual Depth values of each pixel. Lying for example those generated by the pixel shader and transferred to the depth memory Depth values in the form of a quotient of two in depth texture contained values before, so must the two values of the deep texture are preset so that you Quotient just the actual Depth value of a pixel.

According to the invention the generation of the depth information for the individual pixels does not occur Interpolation defined only in the vertices of a polygon Depth information (the latter corresponds to the standard rendering pipeline according to the state the technique in the 3D visualization technique, as explained in more detail below), but by accessing a pre-made depth texture that the Depth values on a per-pixel basis and in one processing by the pixel shader contains customized form. The processing of the depth information by means of pixel shaders takes place on the graphics chip of the graphics card in a specially optimized for this Processor area and is therefore very fast.

Testing at the required resolution from 1024 × 768 Points with a landscape area of 768 × 768 points at one color depth of 16 bit and a Geforce4 Ti4600 graphics card have shown that the desired Picture display frequency of 60 Hz can be achieved. This could even with a resolution of the Landscape area of 1024 × 1024 Points are shown.

The inventive method especially under both major graphics standards (Open GL, DirectX) can be used on different operating systems.

It can u. a. when using a graphics processor from nVidia as well as the company ATI.

The Invention will be described with reference to specific embodiments and with reference closer to figures explained. They show:

1 the rendering pipeline of DirectX 8.1,

2 the prescribed instruction order of a pixel shader program.

To the better understanding will be first in general, the modern 3D visualization on which the inventive method builds, closer explained. The points at which the inventive method of the standard method deviates become detailed explained.

• – Eine Menge von 3D-Koordinaten: Sie beschreiben alle Eckpunkte (Vertices) der verwendeten Dreiecke.
• – Koordinaten-Systeme: Sie beschreiben den Zusammenhang von Mengen von 3D-Koordinaten. Jedes bewegliche darzustellende Objekt verwendet sein eigenes lokales Koordinatensystem. Dieses wird mittels einer Transformationsvorschrift in das Welt-Koordinatensystem eingebracht. Dieser Vorgang ist rekursiv, so dass ein bewegliches Objekt bewegliche Teile haben kann.
• – Dreiecks-Listen: Eine Beschreibung, welche 3D-Koordinaten zu einem Dreieck zusammengefasst werden.
• – Oberflächen-Beschreibungen: Jedes Dreieck bildet eine Fläche im Raum. Die Oberfläche wird beschrieben durch Farbeigenschaften, Beleuchtungseigenschaften und Texturierungseigenschaften. Für die Beleuchtungseigenschaften kann die Angabe von Normalen-Vektoren in den Eckpunkten der Dreiecke notwendig sein. Diese können nicht nur relativ zur Fläche des Dreiecks ausgerichtet sein, sondern auch relativ zur durch das Dreieck nachgebildeten gebogenen Oberfläche. Damit lässt sich über die Fläche des Dreiecks die Beleuchtung der gebogenen Fläche berechnen und so den Eindruck der Krümmung vermitteln. Alle verwendeten Informationen bezüglich der Oberflächen-Beschreibung werden in den Eckpunkten des Dreiecks abgelegt. Es ist nicht möglich, einem beliebigen Punkt innerhalb des Dreiecks eine individuelle Beschreibung zuzuordnen.
• – Umweltparameter: Lichtquellen, Nebel etc.

The modern 3D visualization is based on the representation of triangles (generally: polygons). All bodies that are to be represented are split into triangles. Curved or circular surfaces are approximated by a mesh of triangles and "rounded" by the lighting techniques used, so a scene to be displayed is described by relatively few elements:

• - A set of 3D coordinates: They describe all the vertices of the triangles used.
• - Coordinate systems: They describe the relationship between sets of 3D coordinates. Each moving object to be displayed uses its own local coordinate system. This is introduced by means of a transformation rule in the world coordinate system. This process is recursive, so that a moving object can have moving parts.
• - Triangle Lists: A description of which 3D coordinates are combined into a triangle.
• - Surface Descriptions: Each triangle forms one surface in space. The surface is described by color properties, lighting properties and texturing properties. For the lighting properties, the specification of normal vectors in the vertices of the triangles may be necessary. These may be aligned not only relative to the surface of the triangle, but also relative to the curved surface simulated by the triangle. Thus, the illumination of the curved surface can be calculated over the area of the triangle and thus convey the impression of curvature. All information related to the surface description is stored in the vertices of the triangle. It is not possible to assign an individual description to any point within the triangle.
• - Environmental parameters: light sources, fog etc.

Such a scene is processed by the graphics hardware in several steps. First, all coordinates are transformed so that their positions relative to the current camera are known. Since the screen is only two-dimensional, the coordinates must then be projected onto the visible two-dimensional area. As a rule, one uses a central projection for this, since it corresponds to the natural visual behavior of humans. In a second step, the resulting two-dimensional images of the triangles are "cropped" to fit on the screen, ie, do not protrude beyond the edges, and then the visible triangles are drawn, and the problem arises when drawing the triangles If, for example, a house is in a landscape, then the landscape behind the house is present, but in the resulting image the house must be presented in front of these landscape areas.It is very elaborate or even impossible, the triangles of their distance to the viewer Instead, almost all modern graphics systems use what is known as a Z-buffer or depth memory: each pixel on the screen is assigned a depth value in the Z-buffer, and all triangles are easily calculated the depth value for each calculated pixel is compared to the one for the sen pixel already existing value compared. If the value is smaller, the calculated pixel is closer to the viewer than the previously calculated pixel and is therefore stored including the depth value. In this way, the coverage of the triangles can be calculated very easily. However, this usually also leads to large image parts that are drawn several times. In order to be able to draw a triangle, it is "rasterized", ie, starting from the vertices, the triangle is split into horizontal and vertical lines in screen resolution in order to determine the pixels belonging to the triangle In particular, the depth information is only interpolated between the vertices, ie that a triangle always yields a flat surface in space, based on the problem underlying the invention of the 3D representation by means of a two-dimensional Texture (photography) and a depth image, in which the depth information is coded pixel by pixel in colors, this means that you have the depth image not easy to take over.

In In recent years, two major standards have become 3D visualization enforced: OpenGL and DirectX. Both work on very similar Way and differ primarily on the supported hardware platforms, the driver support by the manufacturers of graphics cards and the type of control.

The following describes how DirectX works. The processing of the graphic data, ie the 3D coordinates, triangular lists, surface descriptions, etc., takes place in the so-called rendering pipeline. In DirectX 8.1, it is divided into three main sections, each of which is divided into different sections. These are processed in parallel by the graphics hardware in the form of a pipeline, ie that different areas of the graphics hardware simultaneously work on different tasks in order to achieve the best possible overall performance. In 1 the three main areas are shown in simplified form and the subareas particularly relevant to the invention are marked. Depending on the hardware implementation of DirectX, the sequence of subsections shown may well differ from the sequence ultimately used.

Geometry pipeline

The Geometry Pipeline is present on all modern graphics cards. It is often referred to as the "T & L" unit (Transform & Lightning). Received as input data this pipeline stage the untransformed 3D coordinates (vertices) and lighting parameters of a scene. From this she generates the Position and illumination of these coordinates on the screen View of the camera used.

In The latest version of DirectX will be called vertex shaders used. These are small programs in a special assembler dialect (Vertex shader assembler), which are applied to every 3D coordinate. The instruction set of Assembler language includes commands for coordinate transformation, illumination calculations and some others for 3D coordinates useful Things.

The The result of the Geometry Pipeline are the transformed and illuminated ones Coordinates or vertices incl. The corresponding to the vertices Texturing information from which the polygons are generated.

Pixel and Texture blending

The The result of the Geometry Pipeline are the transformed and illuminated ones Coordinates or vertices incl. The corresponding to the vertices Texturing information from which the polygons are generated. With this information, the polygons to be displayed are rasterized. This process is done in several steps. In the first step become the vertices of the polygon with all the corresponding ones Interpolate data along the joint edges of the vertices, so for every line to be rendered creates a pair of edge information - the state data at the beginning of a line and the status data at the end of a line. These state data include information about the distance from the viewer, texturing information and the Space coordinate of the corresponding point on the edge of the polygon. In the second step, each line to be displayed is between the pair calculated from edge information. For this purpose, the edge information interpolated along the line. This process corresponds to the interpolation along the edges. This creates the information to calculate each point to be displayed inside the polygon. The last step is the calculation of the representation of a point required Information. Which includes the color information of the point, its alpha value and the distance to the viewer.

With DirectX version 8 will be for these steps already mentioned Pixel shaders used. Similar to in the vertex shaders come here small written in pixel shader assembler Programs are used, primarily the calculation of the color value and the alpha value of a point. With pixel shaders Is it possible, complex calculations and joins to realize textures. In particular, can be so realistic Lighting effects are achieved.

rasterization

In the final stage of the rendering pipeline, the calculated information for a pixel is written to the designated memory areas. Before this happens, some tests can be done to determine whether the calculated information should even get into the final picture. The tests can be activated or deactivated individually by the application at any time.

• • Alpha-Test: Sofern der Alpha-Test aktiviert ist, wird der Alphawert des Punktes mit einem Referenzwert verglichen. Ist die Bedingung des Vergleiches erfüllt, so gilt der Test als Bestanden. Andernfalls wird der Punkt nicht dargestellt und verworfen. Dieser Test wird häufig verwendet, um Punkte die sehr bzw. vollständig durchsichtig sind, möglichst frühzeitig von der Darstellung und weiteren Tests bzw. Berechnungen auszuschließen.
• • Stencil-Test: Der Stencil-Test dient dem Maskieren von Bildschirmbereichen. Ist dieser Test aktiviert, wird überprüft, ob die Zielposition des Punktes in der resultierenden Grafik gültig ist. Es wird ein der Zielposition korrespondierender Wert aus dem so genannten Stencil-Buffer mit einem Referenzwert mittels eines Operators verglichen. Ist die Bedingung erfüllt, kann der Punkt dargestellt werden, andernfalls wird er verworfen.
• • Depth-Test: Der abschließende Test überprüft, ob der darzustellende Punkt weiter vom Betrachter entfernt ist, als ein an der entsprechenden Position bereits dargestellter Punkt. Ist dies der Fall, wird der Punkt verworfen und nicht dargestellt. Mittels dieses Tests wird die Überdeckung von Bildbereichen bestimmt. Die Tiefeninformationen für diesen Test werden im Z-Buffer gespeichert.

There are three types of tests:

• • Alpha Test: If the alpha test is enabled, the alpha value of the point is compared to a reference value. If the condition of the comparison is fulfilled, then the test is passed. Otherwise the point will not be displayed and discarded. This test is often used to exclude points that are very or completely transparent, as early as possible from the presentation and further tests or calculations.
• • Stencil test: The stencil test is used to mask screen areas. If this test is activated, it is checked whether the target position of the point in the resulting graphic is valid. A value corresponding to the target position from the stencil buffer is compared with a reference value by means of an operator. If the condition is met, the point can be displayed, otherwise it is discarded.
• • Depth Test: The final test verifies that the point to be displayed is farther from the viewer than a point already shown at the corresponding position. If this is the case, the point is rejected and not shown. This test determines the coverage of image areas. The depth information for this test is stored in the Z-buffer.

are all activated tests passed successfully, the point can be displayed become. The color information of the dot is put into the frame buffer written. The depth information of the point is put in the Z-buffer written and the alpha value is no longer needed and discarded.

As previously described, must be in the standard rendering pipeline the generation of the depth information about the polygons is done. To assign a specific depth value to each pixel individually, should one for one define each pixel its own polygon. At a resolution of 768 × 768 = 589,824 pixels would have so at least 35,389,440 polygons per second are displayed to achieve the underlying task (animation rate of 60Hz). In view of the fact that for the calculation of these polygons also their basic data, i. Position information of the corners and Texturing needed is not available today Graphics card capable of the depth information for the 2D photo represent this base.

Around not for To use each pixel a polygon, was the method of the invention for transmission the depth information into the depth memory develops. It realized in terms of gaining the depth information an alternative Rendering pipeline, which relies favorably on pixel shaders, the to a coded depth texture of the 3D panorama image to be displayed access. The depth information obtained from the polygonal calculation will be overwritten.

Of the Pixel Shader Assembler is a programming language that is based on the needs texture and lighting calculations in a rendering pipeline is tailored. The programs become DirectX in textual form passed as ASCII text and there in the binary code of the graphics processor. The generated binary code can subsequently using an identifier associated with this binary code be used.

The Pixel Shader Assembler language consists of a set of instructions, which operate on dot data kept in registers. Each operation consists of an instruction and one or more Operands. The instructions use registers to store data with the ALU (arithmetic logic unit) to process, and to temporary results to save between. All processing is done on the graphics chip the graphics card in a specially optimized processor area. Execution Pixel shader programs are therefore very performant. The length of a Pixel shader program, however, is strictly limited to one for one Limit the maximum processing time to be expended. at Version 1.3 pixel shader programs are e.g. only max. 12 instructions allowed.

The Technique of pixel shaders is still very young and learns one fast development. Under DirectX 8.1, this already has five different ones Versions led: 1.0, 1.1, 1.2, 1.3 and 1.4. Versions 1.0 to 1.3 build on each other up and use a very similar one Architecture. Version 1.4 is based on a new architecture, the also the basis for to make the upcoming version 2.0.

The pixel shader program developed according to the invention is based on a "misappropriation" of a command for simulating "rough" surfaces by means of "Z-correct bump mapping" a method in which - similar to a texture with image information - a kind of "deflection texture" is applied to the polygon.

The basic idea of the invention is now that specifically for "Z-Correct Bump mapping "developed and per se known pixel shader commands for fast transfer of depth information for that to be displayed landscape image. This succeeds by appropriate choice of input data and a suitable data format for the textures used, as shown below using the example of an nVidia graphics chip Under the pixel shader version 1.3 is explained in more detail.

In 2 is the prescribed instruction order of a Version 1.3 pixel shader program. Each program must adhere to this specification and generates output from the inputs in the input registers by executing the instructions using temporary registers in the output registers.

• • c_n: Konstanten-Register. Es stehen min. 8 Konstanten-Register (c₀-c₇) zur Verfügung. Sie können vom Pixel Shader nicht verändert werden und müssen von der Applikation vor Anwendung des Pixel Shaders gesetzt werden. Jedes Register kann vier Festkommawerte mit Wertebereich [–1,1] enthalten.
• • t_n: Textur-Register. Es stehen min. 4 Textur-Register (t₀-t₃) zur Verfügung. Jedes Register besteht aus vier Festkommawerten (r, g, b, a bzw. u, v, w, q oder x, y, z, w geschrieben als t_n.rgba bzw. t_n.uvwq oder t_n.xyzw) mit einem minimalen Wertebereich von jeweils [–1,1]. Jedes Textur-Register korrespondiert zu einer so genannten Texture-Stage. Jede Texture-Stage umfasst die Beschreibung einer verwendeten Textur und der für die Verwendung benötigten Attribute. Ein Textur-Register beinhaltet zunächst Textur-Koordinaten, die vom System automatisch interpoliert werden. Möchte man den korrespondierenden Farbwert der Textur haben, so muss man die entsprechende Textur „sampeln". Dies geschieht unter Berücksichtigung der Attribute der entsprechenden Texture-Stage.
• • v_n: Farbregister. Es stehen min. 2 Farbregister (v₀-v₁) zur Verfügung. Sie werden automatisch mit interpolierten Werten im Bereich [0,1] gefüllt und sind für unsere Betrachtung nicht weiter von Interesse.

The input registers are in detail:

• • c _n : constants register. There are min. 8 constant registers (c ₀ -c ₇ ) are available. They can not be changed by the Pixel Shader and must be set by the application before using the Pixel Shaders. Each register can contain four fixed-point values with value range [-1,1].
• • t _n : texture register. There are min. 4 texture registers (t ₀ -t ₃ ) available. Each register consists of four fixed-point values (r, g, b, a or u, v, w, q or x, y, z, w written as t _n .rgba and t _n .uvwq or t _n .xyzw) a minimum value range of [-1,1]. Each texture register corresponds to a so-called texture stage. Each texture stage includes the description of a used texture and the attributes needed for its use. A texture register initially contains texture coordinates that are automatically interpolated by the system. If you want to have the corresponding color value of the texture, then you have to "sample" the corresponding texture, taking into account the attributes of the corresponding texture stage.
• • v _n : color register. There are min. 2 color registers (v ₀ -v ₁ ) available. They are automatically filled with interpolated values in the range [0,1] and are no longer of interest for our consideration.

Two temporary registers r ₀ and r ₁ with a minimum value range of [-1,1] are available for the processing of the instructions. As the output register of the pixel shader program, the temporary register r _{0 is} used. The four fixed-point values contained therein are interpreted as RGBA color values with a value range of [0,1].

The pixel shader program developed according to the invention has the following form. In addition to the transmission of the depth values, this program also adopts the representation of the landscape image in the version chosen here. Other and simpler versions of this program are possible. In particular, other registers may be used. (1) ps.1.3 (2) tex t0 (3) tex t1 (4) texm3x2pad t2, t0 (5) texm3x2detph t3, t0 (6) mov r0, t1

The command (1) indicates the version of the pixel shader program (here 1.3). Like in the 2 As you can see, this must always be the first command of a pixel shader program. The next two commands are used to the texture to "resample". Which is used in a pixel shader program Every texture, is a texture register _n t represents. The index is n while at the same time the number of texture Stage, which corresponds to this register.

First, the texture register contains the texture coordinates in the texture system in t _n .uvwq. The triple (u, v, w) describes the coordinate in a three-dimensional volume texture. A normal two-dimensional texture uses as coordinates only (u, v). By sampling the texture, the color information and the alpha value are generated to the corresponding texture coordinate and stored in t _n .rgba. This overwrites the texture coordinates because t _n .uvwq and t _n .rgba use the same memory area.

With the instruction (2), the texture of the texture stage 0 is thus sampled in the register t ₀ and with command (3) the texture of the texture stage 1 in the register t ₁ . In the two registers are then the Far binations of the two textures for this point. Like in the 2 As can be seen, the processing order of pixel shader programs requires that all textures to be sampled be sampled before the first arithmetic operation.

It is essential for the invention that one of the textures used in this pixel shader program contains information (this does not necessarily have to be color information, other texture formats which are suitable for this task, such as HILO textures), which are used as deep texture ( in register t ₀ ) is used. Another contains information that is used as image information (in register t ₁ ).

• • texm3x2pad t_m, t_n Es wird das Skalarprodukt von t_m und t_n gebildet und in einem internen Register z gespeichert. Für die beiden Indizes m und n muss die Bedingung m > n erfüllt sein.
• • texm3x2depth t_m+1, t_n Es wird das Skalarprodukt von t_m+1, und t_n gebildet und in einem internen Register w gespeichert. Anschließend wird der neue Tiefenwert für den aktuell berechne ten Punkt auf z/w gesetzt, falls w ≠ 0. Andernfalls wird der neue Tiefenwert auf 1 gesetzt.

Instructions (4) and (5) compute the new depth value for the current point based on depth texture. The command (5) can only be used in combination with command (4) and has the following effect:

• • texm3x2pad t _m , t _n The scalar product of t _m and t _{n is} formed and stored in an internal register z. For the two indices m and n the condition m> n must be fulfilled.
• • texm3x2depth t _{m + 1} , t _n The scalar product of t _{m + 1} , and t _{n is} formed and stored in an internal register w. Then, the new depth value for the currently calculated point is set to z / w if w ≠ 0. Otherwise, the new depth value is set to 1.

So the following operation takes place: z = t 2 .uvw * t 0 .rgb = t 2 .u * t 0 .r + t 2 .v * t 0 .g + t 2 .w * t 0 .b (1) w = t 3 .uvw * t 0 .rgb = t 3 .u * t 0 .r + t 3 .v * t 0 .g + t 3 .w * t 0 .b (2)

• • t₀.rgba: r und g ergeben den Quotienten; b = 0; a ist beliebig
• • t₂.uvwq: u = x; v = 0; w = 0; q ist beliebig
• • t₃.uvwq: u = 0; v = x; w = 0; q ist beliebig

In order to assign a specific depth value to each point in the displayed image, the quotient z / w must therefore give the desired depth value. In order to obtain as simple a relationship as possible between the quotient z / w and the values of the deep texture z / w and to ensure that z / w does not depend on the values of the texture registers t ₂ and t ₃ , the values of the texture become Registers t ₀ , t ₂ and t _{3 are selected as} follows:

• • t ₀ .rgba: r and g give the quotient; b = 0; a is arbitrary
• • t ₂ .uvwq: u = x; v = 0; w = 0; q is arbitrary
• • t ₃ .uvwq: u = 0; v = x; w = 0; q is arbitrary

This leads to the following calculation: z = t 2 .uvw * t 0 .rgb = x · t 0 .r + 0 · t 0 .g + 0 · 0 = x · t 0 (4) w = t 3 .uvw * t 0 .rgb = 0 · t 0 .r + x · t 0 .g + 0 · 0 = x · t 0 .g (5)

Of the Value of the selected x must be in the range (0,1), because 0 does not lead to an admissible result leads and through the mapping rule of texture coordinates mapped a 1 to 0 becomes.

Of the Value of the depth information of a pixel can thus by means of a suitable depth texture can be generated on a per-pixel basis.

With the last command (6) of the Pixel Shader program becomes the color value of the picture to be displayed for transfer the current point to the output register of the Pixel Shader program. Alternatively could one the generation of the depth information and the presentation of the perform the actual image in two separate steps, however would then processed by the graphics processor similar polygons multiple times become. This can negatively affect the overall performance.

To execution of the explained Pixel shader program is the transmission the depth information for completed the individual pixels. This is followed by the further processing steps the standard rendering pipeline (under DirectX especially rasterization) at.

At the Using a graphics card with a graphics processor from ATI (e.g. Radeon 9700 Pro), version 1.3 pixel shaders are not supported. ATI has derived version 1.4 of version 1.2, which makes it possible a data-compatible pixel shader program to the program described above to write.

erzeugt, so dass der Inhalt der Texturregister i₂, i₃ ohne Bedeutung ist.Under this version of the pixel shader, instead of the quotient of two scalar products, the quotient becomes direct

is generated, so that the content of the texture register i ₂ , i _{3 is} meaningless.

Claims (1)

A method for transmitting depth information in the depth memory of a graphics card for the display of 3D panoramic images, characterized in that the transmission of the depth information for each pixel of the 3D panoramic image to be displayed on the graphics chip of the graphics card by means of a program executed on the graphics chip, wherein Input data for the program at least one texture that contains in coded form the depth information of the 3D panorama image to be displayed, are provided, wherein the depth information of the 3D panorama image to be displayed in the at least one texture are coded such that the depth information of a pixel by a Quotient of two values present in the texture used.