KR100743391B1 - An improved method for selecting level-of-details using viewpoint adjustment in stereoscopic terrain visualization - Google Patents

Abstract

The present invention is an improved detail level selection method in three-dimensional terrain visualization using a photo tree, and more specifically, instead of the conventional method of calculating the time-dependent error twice for each of the two eyes, the eyes are viewed at the midpoint of both eyes. An improved detail level selection method which greatly increases the rendering speed of stereoscopic terrain visualization by calculating the dependency error only once and performing the detail level selection using the calculated error. The present invention is a method of detail level selection in three-dimensional terrain visualization, comprising: calculating view dependent geometric errors for midpoints of both eyes for all leaf nodes along altitude field data And transmitting the error calculated for all leaf nodes to its parent nodes, and applying an accuracy level determination function to each node using the transmitted error so that the current node can have child nodes. Determining whether there is, thereby determining the visibility of each node, and generating a triangle mesh using the determined information.

According to the present invention, three-dimensional terrain visualization is generated because only one polygon mesh is generated by calculating the visual-dependent error only once using a single viewing specification at the midpoint of both eyes and performing the detail level selection using the calculated error. It can greatly improve the rendering speed of, and also eliminate the inconsistency of the polygon mesh used for both eyes, which occurred in the existing level of detail selection method.

Photo Tree, Stereoscopic Terrain Visualization, Level of Detail Selection (LOD), Visual-Dependent Error, Midpoint of Both Eyes, Render Speed, Mesh, Mesh Mismatch, Visual Frustum Screening

Description

Improved Level of Detail Selection Using Viewpoint Adjustment in Stereo Terrain Visualization

1 is a diagram for explaining a relationship between a size of a node and a distance from a viewer.

2 is a diagram for explaining a method of calculating an upper bound for an approximation error.

3 is a diagram illustrating a comparison between a conventional method and a method proposed by the present invention in a method of calculating a time-dependent geometric error.

4 is a diagram illustrating an example in which different meshes are generated for the same screen, which may occur when using an existing detail level selection method.

FIG. 5 is a diagram illustrating a comparison of rendering time between an existing method and a method proposed by the present invention. FIG.

FIG. 6 is a diagram illustrating a comparison of a resultant image by a conventional rendering method and a rendering method proposed by the present invention. FIG.

<Explanation of symbols for the main parts of the drawings>

d: the size of the node

l: distance from viewer

dh ₁ , dh ₂ , dh ₃ . dh ₄ : Altitude difference computed at the four edges of the node

dh ₅ , dh ₆ : computed altitude difference along two diagonals of the node

P _l : position of left eye

P _r : Right eye position

P _m : the position of the midpoint of both eyes

d _eyes : the distance between both eyes

l _l : distance from left eye to target area

_{r r} : distance from the right eye to the target area

l _m : the distance from the midpoint of both eyes to the target area

The present invention is an improved detail level selection method in three-dimensional terrain visualization using a quadtree, and more specifically, the midpoint of both eyes instead of the conventional method of calculating the time-dependent error twice for each eye. The present invention relates to an improved detail level selection method that greatly improves the rendering speed of stereoscopic terrain visualization by calculating the visual-dependent error only once and performing the detail level selection using the calculated error.

Terrain visualization is widely used in applications such as Geographic Information System (GIS), computer games and flight simulation. Photo-level continuous level-of-detail (CLOD) is one of the techniques for visualizing terrain data. The photo tree data structure enables fast processing in detail step determination and visual frustum screening, so that high quality images can be generated relatively quickly. Since the height field, the input data for this terrain rendering system, has a huge amount of sample points, an enormous amount of polygons can be generated during rendering. Therefore, various methods for reducing the number of polygons in consideration of viewing conditions have been proposed using hierarchical data structures.

Stereoscopy, on the other hand, is a well-known visualization method for immersive display of virtual or real screens. Among the factors that cause the three-dimensional effect, the binocular disparity that appears because our eyes are about 65mm apart in the horizontal direction is the most important factor of the three-dimensional effect. In other words, both eyes see different two-dimensional images, and when these two images are transmitted to the brain through the retina, the brain accurately fuses each other to reproduce the depth and reality of the original three-dimensional image. Is called stereoscopic. Most virtual reality applications generate stereoscopic images as their final output. They help us understand the spatial relationship between two objects in the scene and also experience a depth cue from video or computer generated animation. The brute force approach, the easiest approach to apply two-dimensional rendering techniques in three dimensions, renders the scene twice under different viewing conditions, which results in at least twice the rendering time and additional storage space. in need. Several optimization methods have been proposed to reduce the rendering time of stereoscopic visualization.

However, while the proposed methods reduce the number of polygons to be transferred to the graphics hardware, they are all the same in that they have to create two sets of polygons in a stereoscopic application. One for the left eye and one for the right eye. In the level of detail selection, the mesh data had to be transmitted twice since the polygon mesh for the left eye was different from the polygon mesh for the right eye. Therefore, much more time was required to render three-dimensional terrain.

The present invention proposes a method that can fundamentally improve the level of detail selection in stereoscopic visualization, and instead of the conventional method of calculating the time-dependent error twice for each of the eyes, it is possible to visualize at the midpoint of both eyes. It is an object of the present invention to provide an improved detail level selection method that greatly increases the rendering speed of stereoscopic terrain visualization by calculating the dependency error only once and performing the detail level selection using the calculated error.

In addition, the present invention, by generating a single polygon mesh using a single viewing specification at the center of both eyes in the view dependent error calculation step performed for detail level selection, the existing detail level selection method It is a further object to provide an improved level of detail selection method that can eliminate inconsistencies in the polygon mesh used for both eyes which occurred in.

According to a feature of the present invention for achieving the above object, a detailed level selection method in three-dimensional terrain visualization,

Calculating view dependent geometric errors for the midpoints of both eyes for all leaf nodes following altitude field data;

Transmitting the error calculated for all leaf nodes to its parent nodes;

Determining whether the current node can have child nodes by applying an accuracy level determination function to each node using the error transmitted, thereby determining the visibility of each node; And

Generating a triangular mesh using the determined information

It characterized by including the.

Hereinafter, with reference to the accompanying drawings, it will be described in detail an embodiment of the present invention.

First, the continuous detail level technique based on a photo tree for terrain rendering and a method of applying the technique to stereoscopic terrain rendering will be briefly described.

In the continuous detail level technique based on the photo tree, the rendering process is performed by the following steps. (1) In the first step, the error metric is repeatedly calculated for all leaf nodes while following the altitude field data, and then the calculated error metric is transmitted to the parent nodes of the leaf nodes. (2) In the second step, it is determined whether the current node can have child nodes by applying an accuracy level determination function to each node using the set of error metrics calculated in the first step. (3) In the third step, after generating the triangle mesh while repeatedly following the photo tree generated from the altitude field data, visual frustum screening is performed to reduce the number of generated triangles. (4) In the final fourth step, render the triangle mesh.

In the course of the second step of determining whether subdivision is required at the leaf nodes, the distance between the viewer and a particular point on the terrain should be taken into account. That is, the size of the node in the photo tree should increase as the distance between the viewer and a specific point on the terrain increases. 1 is a view for explaining the relationship between the size of a node and the distance to the viewer. In FIG. 1, the size of each node of the picture tree is indicated by d, and the distance to the viewer is indicated by l. If C is the minimum desired resolution for the image, then the size of the node can be controlled with respect to the distance satisfying l / d < The number of triangles is determined by the value of C. For example, if a large value is set for C, a larger number of triangles will be generated since each node can be subdivided into smaller nodes.

As another criterion, an increase in resolution is required for areas with high surface roughness. If you drop the hierarchy one level, new errors are created at exactly five points. Here, five points are one center point of the picture tree node and four midpoints between respective edges. By adopting the maximum of the absolute value for the altitude difference computed along the four edges and the two diagonals of the nodes, an upper bound to the approximation error can be given. 2 is a diagram for explaining this. The altitude differences calculated at the four edges in FIG. 2 are dh ₁ , dh ₂ , dh ₃ , respectively. The difference in altitude calculated along the two diagonal lines is represented by dh ₄ and dh ₅ , dh _{6. The} upper limit of the approximate error of the node can be obtained by calculating the maximum of the absolute values for these six values. . If the maximum value of absolute values for altitude differences is defined as d ₂ , d ₂ may be calculated as in Equation 1 below.

The subdivision criterion for determining whether to subdivide is a form including a d ₂ value for processing the surface roughness calculated above, and may be modified by Equation 2 as an expression of the selection variable f.

Here, the newly introduced constant c specifies the desired global resolution, which directly affects the number of polygons to be rendered per frame.

Next, a method of applying the aforementioned continuous detail level technique to stereoscopic rendering will be described. When applying the continuous detail level technique to stereoscopic rendering by the brute force approach that is the easiest to think of, the detail level selection step is performed twice. That is, the first detail level selection is performed for the viewing specification of the left (right) eye, and the second detail level selection is performed for the opposite eye. Level of detail selection is known to be the most time consuming step in the overall terrain visualization procedure because it has to calculate the geometric error value for each node and pass it over to its parent nodes repeatedly.

In the method proposed by the present invention, unlike the conventional method of performing the detail level selection twice for each eye once, the detail level selection is performed only once. Hereinafter, a detailed level selection method proposed by the present invention will be described in detail with reference to FIG. 3.

FIG. 3 illustrates a comparison of a conventional method and a method proposed by the present invention in a method of calculating a time-dependent geometric error. 3 (a) is a conventional method of calculating the time-dependent geometric error, each of which is visual-dependent geometric error under two different viewing conditions of the position of the left eye (P _l ) and the position of the right eye (P _r ). Is being calculated. In contrast, FIG. 3 (b) is a method proposed by the present invention and calculates the time-dependent geometric error only once at the midpoint P _m of both eyes. As shown in FIG. 3, when the distance from the left eye to the target area is l _l , and the distance from the right eye to the target area is l _r , even if the two distances l _l and l _r are not exactly the same, Since the distance d _eyes between both _eyes is small, the difference between the two values is also small. Therefore, time-The method proposed by the present invention for calculating a one-time-dependent geometrical errors in both eyes position P _l to an intermediate point P _m P a _r becomes feasible. When the vector V _m is represented using the vectors V _l and V _r shown in FIG. 3, Equation 3 is obtained.

여기서, V _l, V _r, V _m은 각각 왼쪽 눈으로부터 타깃 영역까지의 벡터, 오른쪽 눈으로부터 타깃 영역까지의 벡터, 양쪽 눈의 중간점으로부터 타깃 영역까지의 벡터를 나타낸다.

Here, V _l , V _r , V _m represent a vector from the left eye to the target area, a vector from the right eye to the target area, and a vector from the midpoint to the target area of both eyes.

The picture tree based method calculates the time-dependent geometric error, determines the visibility of each node, and then generates a triangular mesh using the determined information. Since both eyes have different viewing conditions, the resulting polygon mesh may not be the same. 4 illustrates an example in which different meshes are generated for the same screen, which may occur when using an existing detail level selection method. 4 (a) and 4 (b) show a result mesh for the left eye and a result mesh for the right eye, respectively, for the same screen, in which a rectangular area shown in red is displayed as different geometric resolutions for both eyes. As can be seen in FIG. 4, according to the existing level of detail selection method, the same object can be represented with different geometric resolutions for each eye, which generates visual artifacts for stereoscopic images. As a result, it becomes difficult for the viewer to experience an immersive screen.

The method proposed in the present invention provides a solution to this problem. In other words, since only one polygon mesh is generated using a single viewing specification at the center point of both eyes in the visual-dependent error calculation step, the inconsistency of the polygon mesh generated for both eyes can be eliminated. In addition, the mesh data is only sent once to the GPU, which greatly reduces the overall rendering time.

After the resulting mesh is generated, the visual frustum screening step is performed to reduce the number of triangles generated, and finally, the triangle mesh is rendered. In the visual frustum screening step, as in the detail level selection step, an improved visual frustum screening method that performs the visual frustum screening operation only once, unlike the conventional method, can be used with the improved detail level selection method proposed by the present invention. For this, reference may be made to a separate application filed on the same date as the present application.

[Experiment result]

The existing method and the method proposed in the present invention are implemented using a dual Intel Pentium Xeon ^™ 3.0 GHz CPU and 2GB main memory. The graphics accelerator was an NVIDIA GeForce ^TM 6600GT with 128MB video memory and was installed on a PCI Express * 4 BUS. The sample data used for the experiment is Puget altitude field data with a resolution of 256 x 256 and Grand Canyon altitude field data. The viewport size is 512 x 512.

5 shows a comparison of the rendering time between the conventional method and the method proposed by the present invention. FIG. 5 (a) shows the results of experiments on three different viewing conditions view 1, view 2, and view 3 for the Puget altitude field data, and FIG. 5 (b) shows three different types for the Grand Canyon altitude field data. Results of experiments on viewing conditions view 1, view 2, and view 3. As can be seen in Figure 5, the method proposed in the present invention is about 2 times faster than the conventional method. This can be interpreted as the result of the method proposed in the present invention because it performs the detail level selection and the visual frustum screening only once and these two steps consume most of the time for rendering.

FIG. 6 shows a comparison of the resultant image by the conventional rendering method and the rendering method proposed by the present invention. In FIG. 6, the upper drawings are the resultant images generated by the conventional rendering method, and the lower drawings are the resultant images generated by the rendering method proposed by the present invention. Result images, and the images on the right are the result images of the right eye. 6, the detailed level selection operation is performed using the geometric error evaluated at both eye positions according to the conventional method, and the geometric error calculated at the midpoint position of both eyes according to the method proposed by the present invention. In the case of performing the detailed level selection by using the result, it can be seen that there is almost no difference in the image.

Referring to FIGS. 5 and 6, it can be clearly seen that the method proposed in the present invention significantly reduces the total rendering time without degrading the image quality compared to the conventional method.

The present invention described above may be variously modified or applied by those skilled in the art, and the scope of the technical idea according to the present invention should be defined by the following claims.

According to the present invention, instead of the conventional method of calculating the time-dependent error twice for each of the two eyes, the time-dependent error is calculated only once at the midpoint of both eyes and the detailed level selection is performed using the calculated error. By doing so, the rendering speed of the stereoscopic terrain visualization can be greatly improved. In addition, the present invention, by generating a single polygon mesh using a single viewing specification at the center point of both eyes in the step of calculating the time-dependent error performed for detail level selection, both eyes that occurred in the conventional level of detail selection method This can eliminate inconsistencies in the polygon mesh used for. The improved level of detail selection method proposed in the present invention can be applied to any kind of stereoscopic applications.

Claims (2)

As a detail level selection method in three-dimensional terrain visualization, Computing view dependent geometric errors with respect to the midpoints of both eyes for all leaf nodes following altitude field data, where the view-dependent geometric errors occur in dependence on time of day. -Means error between real and real modeled model; Transmitting the error calculated for all leaf nodes to its parent nodes; Determining whether the current node can have child nodes by applying an accuracy level determination function to each node using the error transmitted, thereby determining the visibility of each node; And Generating a triangular mesh using the determined information How to include. The method of claim 1, When the vector from the left eye to the target area is V _l , the vector from the right eye to the target area is V _r , and the vector from the midpoint of the both eyes to the target area is V _m , The step of calculating the time-dependent geometric error with respect to the midpoint of both eyes,

A vector V _m is obtained and then the time-dependent geometric error is calculated using the vector V _m .

Images