CN101840566A - Real-time shadow generating method based on GPU parallel calculation and system thereof - Google Patents

Abstract

The invention discloses a real-time shadow generating method based on GPU parallel calculation and a system thereof. The real-time shadow generating method comprises the following steps: storing object depth information by using a light source viewpoint to render a scene; carrying out function conversion for once on the depth information; storing the converted depth information; carrying out parallel filtering on the depth information by using the GPU; obtaining the distance between an object and the light source by using the viewpoint to render the scene; carrying out function conversion on a distance value; multiplying a depth value and the distance value which are converted to obtain the shadow value information of the current render point; and using the shadow value information to render the scene and finally obtaining smooth shadow effect. The shadow generated by using the method and the system has natural and smooth shadow edges, and the invention has high-efficiency rendering speed in a complex scene.

Description

A real-time shadow generation method and system based on GPU parallel computing

technical field

The invention belongs to the field of graphics, and in particular relates to a real-time shadow production method and system based on GPU parallel computing.

Background technique

In recent years, computer graphics has developed greatly, and has been widely used in games, virtual reality, film special effects production, etc. With the continuous increase of application requirements, people's demand for realistic graphics is also increasing day by day . There are many aspects to achieving photorealistic graphics generation, and shadows are one of the more difficult aspects. With the continuous enhancement of GPU and CPU capabilities, especially the continuous improvement of graphics cards, most of the data originally processed by the CPU has been processed by the GPU, and some general functions have been realized on the hardware, such as changes and lighting calculations, which greatly greatly The processing power has been improved, and the photorealistic rendering method has gradually been widely used.

As an extremely important part of the realistic rendering field, the shadow method mainly includes scan line method, detail polygon method (twice blanking method), shadow volume polygon method (shadow volumes), shadow mapping (shadow mapping), Z -buffer method, ray tracing method, radiosity method, etc. However, shadow mapping and its improved methods are most used in interactive 3D applications. This approach and its improvements are increasingly used in games, virtual reality, medical imaging, filmmaking, and more.

In the process of continuous improvement of computer hardware performance in the future, photorealistic rendering methods are bound to develop towards high-speed, real-time, and realistic directions. Among them, there must be higher requirements for the shadow generation method. Therefore, the research in this direction has extremely high application value.

Contents of the invention

One of the objectives of the present invention is to overcome the deficiencies of the prior art, and propose a real-time shadow generation method based on GPU parallel computing. The second object of the present invention is to propose a real-time shadow generation system based on GPU parallel computing according to the method of the first object.

In order to realize the purpose of the invention one, the technical scheme adopted is as follows:

Render the scene with the viewpoint of the light source to save the depth information of the object; perform a function transformation on the depth information; store the transformed depth information; use the GPU to filter the depth information in parallel; render the scene with the viewpoint to obtain the distance between the object and the light source; do a function on the distance value Transformation; multiply the transformed depth value and distance value to obtain the shadow value information of the current rendering point; use this shadow value information to render the scene, and finally obtain a smooth shadow effect.

In order to realize the purpose of the invention two, the technical scheme adopted is:

A real-time shadow generation system based on GPU parallel computing, including a new shadow map texture generation module, a GPU parallel processing shadow map texture module, and a rendering module for generating real-time shadows from shadow maps.

The new shadow map generation module includes a depth acquisition submodule and a continuous function shadow map generation submodule.

The GPU parallel processing shadow map texture module includes a GPU shadow map loading sub-module and a GPU parallel filtering sub-module.

The rendering module for generating real-time shadows by shadow maps includes a continuous function shadow map shadow map map reduction sub-module and a shadow generation sub-module.

The new shadow map generation module, after collecting the depth information from the scene object to the light source, uses a continuous function to map the depth information once, and uses the mapped value to generate a shadow map.

Send the shadow map to the GPU shadow map loading submodule, including the following three steps:

Divide the original image into small blocks of the same size; read each small block into the memory of the processing unit of the GPU; expand the edge part of the read data.

If it is the outermost border of the image, use 0 to expand, and the number of extended rows depends on the filter order; if it is the inner block boundary of the image, use adjacent row or column data to copy and expand, and the number of extended rows depends on the filter order Certainly.

Send the expanded data to the GPU parallel filtering sub-module, including the following three steps:

Each processing unit simultaneously reads in the supplemented and extended data; each processing unit uses filtering to check the data for multiplication and addition filtering; writes the result after the operation into the common memory area, and returns the end signal after waiting for thread synchronization.

Then enter the continuous function shadow map shadow map map restoration sub-module and shadow generation sub-module.

The map texture reduction sub-module performs functional transformation on the distance value; multiplies the transformed depth value and the distance value to obtain the shadow value information of the current rendering point. Send it to the shadow generation sub-module, use this shadow value information to render the scene, and finally get a smooth shadow effect.

The invention has the advantages of fast shadow generation speed and smooth edges. In the case of multiple light sources in the scene, it can effectively increase the shadow generation speed by more than 50% compared with the general method. It has great advantages in the rendering application requirements of complex scenes.

Description of drawings

Fig. 1 is the framework schematic diagram of the real-time shadow generation system based on GPU parallel computing of the present invention;

Fig. 2 is the flowchart of the real-time shadow generation method based on GPU parallel computing of the present invention;

Figure 3 is a flow chart of the shadow map loading sub-module;

FIG. 4 is a flow chart of GPU parallel processing of shadow map textures.

Detailed ways

In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present invention. For those skilled in the art, other drawings can also be obtained according to these drawings without any creative effort.

Figure 1 shows the framework of real-time shadow generation system based on GPU parallel computing. First, the rendering system obtains the depth information of all objects to be rendered in the scene from the scene, and there are no special requirements for this step.

After obtaining the depth information, pass the depth information to the continuous function shadow map module. This module completes the following tasks:

First, this module obtains a discrete shadow test function based on the depth information. This shadow test function is defined as if the depth information value from the light source to the object is smaller than the depth information value from the light source to the object in the viewpoint space, it is considered that the point is in the shadow. The shadow test function is a 0-1 discrete function. Secondly, a continuous function is selected to approximate the discrete function. There are many choices of functions, and developers can try to select different types of continuous functions. The selection purpose of all functions is to approximate and simulate the 0-1 discrete function, and finally store the shadow test function transformed by the continuous function into a shadow map.

Next, the stored shadow map is sent to the GPU shadow map loading sub-module. This module is responsible for pre-processing the shadow map to prepare for GPU parallel filtering. The specific work is shown in Figure 3, including the following three steps:

1. Divide the original shadow map into small pieces of the same size. The size of the blocks in this step will vary according to the size of the filter kernel and the size of the image itself, and should meet the general principles: the number of blocks should meet the GPU data alignment principle, that is, the number of blocks should be the power of 2.

2. Read each small block into the memory of the processing unit of the GPU. This step is to provide raw data for the GPU parallel computing unit.

3. Expand the edge part of the read data. This step is to calculate the value of the edge of the block. According to the definition of convolution filtering, this step cannot be missing, otherwise the image will shrink. The extension methods are described below:

If it is the outermost border of the image, use 0 to expand, and the number of extended rows depends on the filter order; if it is the inner block boundary of the image, use adjacent row or column data to copy and expand, and the number of extended rows depends on the filter order Certainly.

Next, as shown in Figure 4, the expanded data is sent to the GPU parallel filtering sub-module, which performs parallel filtering on the shadow map, including the following three steps:

1. Each processing unit simultaneously reads in the completed and expanded data;

2. Each processing unit uses filtering to check the data to perform multiplication and addition filtering operations; the filtering method here can be different, and this system uses Gaussian filtering.

3. Write the calculated result into the public storage area, and wait for the thread synchronization to return the end signal. So far, the parallel filtering of the image has been completed, and the result is stored in the common storage area.

Then enter the continuous function shadow map shadow map map restoration sub-module and shadow generation sub-module.

The shadow map restoration sub-module performs functional transformation on the distance value; multiplies the transformed depth value and the distance value to obtain the shadow value information of the current rendering point. Send it to the shadow generation sub-module, use this shadow value information to render the scene, and finally get a smooth shadow effect naturally.

The systems and methods described above are provided to enable those skilled in the art to make and use the present invention. Those skilled in the art can make corresponding modifications to the specific implementation, but the maximum scope corresponding to it is consistent with the principles and novelties disclosed here.

Claims (3)

1. the real-time shadow generation method based on the GPU parallel computation is characterized in that, comprises the steps: that playing up scene with light source viewpoint preserves object depth information; Depth information is done the linear function conversion; Depth information after the memory mapping; With GPU to the depth information parallel filtering; Play up scene with viewpoint, obtain the distance of object and light source; The value of adjusting the distance is done functional transformation; Depth value after the conversion and distance value are multiplied each other, obtain the current shade value information of playing up a little; Utilize this shade value information to play up scene, finally obtain the smooth shade effect.

2. the real-time shadow generation system based on the GPU parallel computation is characterized in that, this system comprises that new shade mapping pinup picture generation module, GPU parallel processing shade mapping pinup picture module and shadow map generate the rendering module of real-time shadow;

Described new shade mapping pinup picture generation module comprises degree of depth collection submodule, generates continuous function shadow map submodule;

Described GPU parallel processing shade mapping pinup picture module comprises that the GPU echo is written into submodule, GPU parallel filtering submodule;

The rendering module that described shadow map generates real-time shadow comprises that continuous function shadow map shade mapping pinup picture reduction submodule, shade generate submodule;

New shade mapping pinup picture generation module, collect object scene behind the depth information of light source, with continuous function depth information is done once mapping, utilize the value after shining upon to generate shadow map, shadow map is sent into the GPU echo to be written into submodule the data marginal portion is expanded, data after the expansion are sent into GPU parallel filtering submodule, enter continuous function shadow map shade mapping pinup picture reduction submodule and shade afterwards and generate submodule, the mapping pinup picture reduction submodule value of adjusting the distance is done functional transformation, depth value after the conversion and distance value are multiplied each other, obtain the current shade value information of playing up a little, send into shade and generate submodule, utilize this shade value information to play up scene, finally obtain the smooth shade effect.

3. the real-time shadow generation system based on the GPU parallel computation according to claim 2, it is characterized in that, the continuous function shadow map module shadow testing function that obtains dispersing according to depth information at first, if this shadow testing function definition promptly is considered as this point in shade less than light source in the view space to the depth information value of object to the depth information value of object for light source, the shadow testing function that obtains like this is the 0-1 discrete function; Next selects for use a continuous function to approach this discrete function, and it is multiple that the choosing of function can have, and the developer can attempt choosing different types of continuous function; All functions choose purpose all approaching simulation 0-1 discrete function, at last the shadow testing function after the continuous function conversion is stored in the shadow map;

Next the shadow map of storage is sent into the GPU echo and be written into submodule, this module is responsible for that shadow map is done early stage and is handled, and purpose is for the GPU parallel filtering provides preparation, comprises the processing of following 3 steps:

A1: original shadow map is divided into the identical fritter of size, this step operation is according to the size of filtering core, with the size of image itself divide block size have the institute different, should satisfy rule: the piecemeal number should satisfy GPU alignment of data principle, promptly piecemeal should be number formulary 2 times;

B1: each fritter is read in the internal memory of the processing unit of GPU, and this step operation is to provide raw data for GPU parallel computation unit;

C1: the data edges after will reading in is partly expanded, and this step operation is in order to calculate the value of block margin, can not lack according to this step of definition of convolutional filtering, otherwise image will dwindle; Extended method is as described below:

If image outermost layer border is expanded with 0, the expansion line number is decided according to the filtering exponent number; If the inner block boundary that divides of image duplicates the polishing expansion with adjacent lines or column data, the expansion line number is decided according to the filtering exponent number;

Next the data after will expanding are sent into GPU parallel filtering submodule, and this module is carried out parallel filtering with shadow map, comprise the processing of following 3 steps:

A2: each processing unit reads in the data after the polishing expansion simultaneously;

B2: each processing unit is taken advantage of data with filtering core and is added filtering operation; The method of filtering herein can be different, and what native system adopted is gaussian filtering;

C2: calculated result is write common storage area, and return end signal after waiting for thread synchronization; So far finished the parallel filtering to image, the result is stored in common storage area.

Images