CN101119485A - A Progressive Transmission Method of 3D Model Based on Feature Preservation - Google Patents

Abstract

A progressive transmission method for 3D models based on feature preservation, which improves the feature preservation on the basis of the original 3D model simplification algorithm. While simplifying the geometric model, it maintains the topological properties and attribute characteristics of the model, and obtains a relatively large amount of data. A small base grid; and on this basis, the compression coding based on octree division is implemented to construct a progressive grid file. When transmitting, first use the breadth of the octree to first traverse the decoding base grid, and then use the progressive grid transmission method to transmit a series of detail recovery information, continuously improve the model accuracy, and restore the original model, thus solving the problem of response time of traditional transmission methods long question. In addition, an object-oriented viewpoint-related transmission strategy is designed for the 3D scene, which accelerates the real-time rendering of the scene, and puts the object visibility judgment on the client to reduce the server load.

Description

A Progressive Transmission Method of 3D Model Based on Feature Preservation

technical field

The present invention relates to the technical field of 3D model simplification, geometric compression coding, and progressive transmission, especially a feature-preserving 3D model progressive transmission method, which can be applied to the processing of various 3D models with complex texture and structural features and web publishing. In addition, it is also applied to the network publishing of precious cultural relics in digital museums.

Background technique

In computer graphics, triangular meshes are generally used to describe complex three-dimensional shapes. With the continuous improvement of 3D data capture technology, the amount of grid data will be very large, which brings many difficulties to the storage, transmission and rendering of 3D models. The digital museum contains a large number of 3D models that need to be displayed online. The most basic ones, such as the transmission of a single model exhibit, and the complex ones, such as real-time roaming of large scenes, all need to solve the bottleneck limitation of network transmission.

The traditional network transmission of a single-resolution model adopts the Download-and-Display mode, that is, all relevant data of the 3D model is downloaded to the client and then displayed. The disadvantage of this method is that the user wait time is too long. The improvement of this is the progressive grid method proposed by Hoppe (see Hoppe H. Progressive Meshes. ACM Computer Graphics, 1996, 30 (1): 99-108), that is, firstly use vertex deletion (see SchroederW. Decimation of Triangle Meshes .Computer Graphics, 1992, 26(2): 65-70), quadratic error (see Garland M.Surface Simplification using Quadric Error Metrics.Computer Graphics, 1997, 31(3): 209-216) and other model simplification algorithms to get Multi-resolution model, and record each step of operation during the simplification process, so as to decompose the original mesh model into a rough base mesh and a series of detail optimization information that can reverse restore the base mesh; during the transmission process, First transmit the base grid to the client and display it, and then gradually transmit the detailed optimization information to refine the initial rough model; such a lossless model reconstruction and encoding method can give users a very short time Response, and the transmission can be stopped at any time, which is a progressive transmission method in the Download-while-Display mode.

For the progressive grid method, there have been many methods to improve it. However, most of these methods are for relatively regular geometric models without attributes, which do not involve attribute information such as texture, and do not support 3D scenes. In the field of practical application, whether it is a single model or a 3D scene, it not only contains geometric and topological information, but also often includes attributes such as color and texture, and there are inevitably boundaries and holes. Therefore, it is necessary to adopt a suitable method to simplify while retaining these characteristic information. In addition, many methods only use model simplification to obtain progressive grid files, and do not further compress the model data. If the progressive grid and compression coding can be combined to reconstruct the model file, it will be more It is beneficial to reduce the amount of model data and improve the real-time response of model transmission and browsing.

Contents of the invention

The technical problem of the present invention is to overcome the deficiencies of the existing technology, and provide a progressive transmission method of 3D models based on feature retention, which is based on feature retention, that is, to maintain the topological properties and texture or color attribute characteristics of the original model during processing , and maintain high model realism, low requirements on network bandwidth and client machine performance, and support progressive transmission of 3D scenes.

The technical solution of the present invention: a method for progressive transmission of a three-dimensional model based on feature retention, characterized in that it comprises the following steps:

(1) Simplify the original grid model with retained features to form a rough base grid;

(2) Implement octree compression coding on the base grid to form a progressive grid file;

(3) Adopt the progressive transmission mode of octree decoding-progressive grid transmission under the Browser/Server structure;

(4) On the basis of the above steps, an object-oriented viewpoint-dependent transmission strategy is used to transmit the 3D scene.

In the simplification, the triangle folding method that preserves the topological boundaries and texture or color attributes is adopted. The main operation steps are:

(1) Classify the original triangular mesh: boundary triangles, corner triangles, internal triangles, and feature triangles;

(2) According to the classification of triangles, different error measurement methods are adopted:

a. For the inner triangle, equip each triangle T _i with a 4×4 error matrix Q _i , and assuming its new point after folding is v _i , define the folding error of the triangle as $\mathrm{\ϵ}\left(T_i\right)=v_i^T{Q}_i{v}_i,$ Get the position of v _i by minimizing the error;

b. For the corner triangle, limit its folded new point v _i to be a vertex on the original boundary, and use the same method as step a to calculate the folding error;

c. For boundary triangles and feature triangles, set the folding error to the maximum value.

(3) After obtaining the folding errors of all triangles, select the triangle with the smallest error to perform the folding operation.

In the described step (3), the method of geometric compression decoding and progressive grid combination is adopted in the process of transmission, and the steps are:

(1) At the initial stage of transmission, the decoding method of octree breadth-first traversal is adopted until the base grid data transmission is completed;

(2) The progressive grid transmission method is used to maintain the continuous improvement of model accuracy under high similarity.

The step of adopting the object-oriented viewpoint-related transmission strategy for the three-dimensional scene in the described step (4) is:

(1) Construct a separate data stream for each object in the scene, and adopt the aforementioned progressive transmission mode of octree decoding-progressive grid transmission for each data stream;

(2) Set download, analysis, and drawing threads on the client side, and perform synchronous control to ensure parallel transmission of objects;

(3) Calculate the viewpoint information on the client side, and judge the current object to be transmitted according to the visibility of the cone;

(4) The server receives the desired object identifier sent by the client, and sends corresponding optimization information.

Compared with the existing publishing method, the present invention has the following advantages:

(1) The topological properties and texture/color properties of the original model are preserved

Most of the traditional methods deal with relatively regular geometric models without attributes; in practical applications, 3D data not only contains geometric and topological information, but also often includes attributes such as color and texture, and there are inevitably boundaries and holes, this information is important to maintain the look and feel of the model and must be preserved during the simplification process.

Based on the triangle folding method, the present invention provides a non-closed mesh model simplification method combined with attributes, which not only effectively controls the geometric error between the simplified mesh and the original mesh, but also better preserves the three-dimensional model Compared with traditional methods, the topological boundaries and attributes such as color, texture, and normal vector have a greater scope of application and practicability.

In addition, the model simplification algorithm of the present invention can specify the accuracy of model simplification according to the user's need to retain different proportions of geometry/attributes. For an initial model of hundreds of thousands or even millions of polygons, when it simplifies 99% After that, the high geometric similarity and boundary topology and attribute features can still be maintained.

(2) Flexible requirements for network bandwidth and client machine performance

The traditional 3D transmission mode of "download first and then display" needs to download all relevant data of the 3D model to the client before displaying it. Therefore, it has high requirements for network bandwidth. For 3D models with a scale of more than one million polygons, this mode cannot be applied at all; in addition, the graphics processing performance of different client machines is different. The same model, for some users It may already be difficult to deal with, but other users may also need more high-precision effects.

The remote rendering system proposed in recent years is a Client/Server structure. The client is responsible for accepting user input and sending a corresponding rendering request to the server. The server with high graphics performance is responsible for rendering and returning the obtained image to the server that sent the request. Although such a system solves the problem of insufficient graphics performance of the client machine, it limits the maximum number of supported users because all rendering work is placed on the server.

The present invention adopts a progressive transmission mode based on the Browser/Server structure. It only needs to transmit a rough base grid with a small amount of data to the client, and the client can display the base grid data immediately after receiving the base grid data, and then The server continues to send optimization information for the base grid, and the display effect of the client is gradually refined until the transmission is completed or the user's requirements for the fineness of the model are met. In this way, users can choose models of corresponding precision to view according to their own needs and machine performance, and can stop or continue transmission at any time; since the rendering task is distributed to each client, it can support more users to the maximum extent .

(3) Short response time and high precision

The biggest problem with the "download before display" approach is the long response time, requiring the entire model to download before it can start displaying.

As a fine-grained model multi-resolution division scheme, progressive grid transfer can immediately start the rendering and refinement process of the rough model after obtaining the base grid data, and can provide higher accuracy than other methods model; but its disadvantage is that compared with the non-progressive method, it takes a long time to recover the complete highest resolution model; in addition, although the traditional progressive transmission method is "display while downloading" in the overall process ", but at the beginning stage, it is still necessary to wait for the basic grid data to be downloaded before displaying and progressively updating. For 3D models with high precision and attributes such as texture/color, the basic The amount of grid data is still large. For example, the size of the base grid file after 99% simplification of a 3D model with 1,000,000 polygons is still as much as 3MB. In the case of low bandwidth, the download time it requires is also intolerable of.

The geometric progressive compression based on the tree structure is a coarse-grained model multi-resolution partitioning scheme, which can display the model from "zero", and update the geometric data of the model in batches as the recovery information is transmitted and decoded. Therefore, the response time is minimized; due to the effective compression in batches, the data volume of the generated model file is small and the total transmission time is much shorter than that of the progressive grid method; but its disadvantage is that the improvement of the model resolution is not continuous, and there are jumps change, and the accuracy of the model in the recovery process is lower than that of the progressive grid method.

However, the present invention combines the advantages of both progressive grid and geometric compression, further compresses and encodes its base grid on the basis of grid simplification, and adopts octree breadth-first traversal decoding and progressive grid The recovery scheme combining transmission adopts the compression decoding transmission method at the initial stage of transmission when the response time is high but the model accuracy is not high, until the basic grid data transmission is completed; after that, the progressive grid transmission method is adopted, Response times are further improved by maintaining a high and continuous model resolution increase so that the user can abort at any time.

(4) Support 3D scene

A 3D model includes an object model and a 3D scene. A 3D scene is composed of different objects, each of which is an independent shape or structure. Most of the traditional methods only deal with the object model, without considering the 3D scene.

However, the present invention regards the scene model as a combination of several object models, and performs grid simplification with fully preserved topological boundaries for each object, so as to achieve the purpose of simplifying the entire model. In the transmission recovery phase, multiple data streams are used to form each object into a separate data stream, and a viewpoint-related transmission strategy is adopted to control the transmission of currently visible objects and ensure that all objects are in a unified world coordinate system. The complete original model can be restored.

Description of drawings

Fig. 1 is the system implementation process of the present invention;

Fig. 2 is the experimental result that the present invention carries out retaining feature simplification to three-dimensional model;

Fig. 3 is the viewpoint-related transmission framework of the three-dimensional scene of the present invention;

Fig. 4 is the process of progressive transmission of the Sun Moon Avalokitesvara model in the present invention;

FIG. 5 is a process of view-related decoding and transmission of a three-dimensional scene containing 25 terracotta warriors and horses according to the present invention.

Detailed ways

The present invention includes three aspects: model simplification, compression coding, and progressive transmission, and the realization process is shown in FIG. 1 .

1. Simplification of 3D model

The invention adopts the triangle folding algorithm based on the secondary error to simplify the three-dimensional model, and preserves the topological boundary and attribute characteristics of the model while simplifying. The basic operation of the algorithm is triangle folding and shrinking, that is, by performing folding operation on triangle T _i , its three vertices are merged into one vertex v _i , thus simplifying the original mesh.

The algorithm first classifies all triangles in the original mesh, and the division principles are as follows:

For each triangle in the mesh model, if one of its edges is only owned by one triangle (that is, the triangle), then the edge is a boundary edge, and the triangle is a boundary triangle; for non-boundary triangles, if there is at least one If the vertex is on the boundary edge, then the vertex is the boundary point, and the triangle is a corner triangle; if none of the three vertices is on the boundary edge, the triangle is an interior triangle.

For models containing color attributes, the average value of the color values of the three vertices of each triangle is used as the equivalent color value of the triangle, if the difference between the color values of the current triangle and its surrounding triangles exceeds the pre-specified threshold τ, that is

$\underset{\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; p</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; +</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; g</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; g</span>}}_{\mathrm{<span\; class="notranslate">\; p</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; +</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; p</span>}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; ></span>\mathrm{<span\; class="notranslate">\; \&tau;</span>}$

Then mark the triangle as a characteristic triangle, where [r _i g _i b _i ] ^T , [r _p g _p b _p ] ^T are the color values of triangle T _i and T _p respectively, and the threshold τ can be determined according to the user's geometry and It is selected according to different requirements for the retention of attribute features. The higher the requirement for attribute details, the smaller the value of τ, and vice versa.

In addition, for textures, a similar method is also adopted for processing; the difference is that the color information of the corresponding vertices needs to be obtained from the texture image through the texture coordinates, and then the texture equivalent color value of the triangle is calculated.

Afterwards, compute the geometric folding error for each triangle:

(1) For each internal triangle T _i in the original grid, equip an error matrix Q _i , if the triangle T _i is folded into a point v _i =[xi _{y i} z _i 1] ^T , then define the fold of the triangle The error is the sum of the squares of the distances from the new point to all adjacent triangle planes of the original triangle:

$\mathrm{<span\; class="notranslate">\; \&epsiv;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; p</span>}}^{\mathrm{<span\; class="notranslate">\; T</span>}}{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}$

Here p=[a b c d] ^T represents the plane equation ax+by+cz+d=0 of the plane of each triangle in the triangle set adjacent to triangle T _i (where a ² +b ² +c ² =1). Then the error standard given by the above formula can be converted into the following quadratic form:

$\mathrm{<span\; class="notranslate">\; \&epsiv;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; T</span>}}\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; p</span>}}^{\mathrm{<span\; class="notranslate">\; T</span>}}{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; pp</span>}}^{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; p</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; v</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}$

Where M _p is a 4×4 symmetric matrix, which can represent the quadratic distance from any point in the triangular mesh to the plane:

${\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; p</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; pp</span>}}^{\mathrm{<span\; class="notranslate">\; T</span>}}<span\; class="notranslate">\; =</span>\left[\begin{array}{cccc}{\mathrm{<span\; class="notranslate">\; a</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}& \mathrm{<span\; class="notranslate">\; ab</span>}& \mathrm{<span\; class="notranslate">\; ac</span>}& \mathrm{<span\; class="notranslate">\; ad</span>}\\ \mathrm{<span\; class="notranslate">\; ab</span>}& {\mathrm{<span\; class="notranslate">\; b</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}& \mathrm{<span\; class="notranslate">\; bc</span>}& \mathrm{<span\; class="notranslate">\; bd</span>}\\ \mathrm{<span\; class="notranslate">\; ac</span>}& \mathrm{<span\; class="notranslate">\; bc</span>}& {\mathrm{<span\; class="notranslate">\; c</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}& \mathrm{<span\; class="notranslate">\; cd</span>}\\ \mathrm{<span\; class="notranslate">\; ad</span>}& \mathrm{<span\; class="notranslate">\; bd</span>}& \mathrm{<span\; class="notranslate">\; cd</span>}& {\mathrm{<span\; class="notranslate">\; d</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}\end{array}\right]$

And the error matrix Q _i is the sum of the plane quadratic matrices of all adjacent triangles $Q_i=\underset{p\&Element;C_i}{\mathrm{\&Sigma;}}{m}_p;$ The selection of the position of v _i is obtained by calculating to minimize the value of ε(T _i ).

(2) For each corner triangle, define its folded new point v _i as a vertex on the original boundary, and pass $\mathrm{\&epsiv;}\left(T_i\right)=v_i^T{Q}_i{v}_i$ Calculate the folding error;

(3) For boundary triangles and characteristic triangles, set the folding error to the maximum value.

After the folding error of each triangle is obtained, the triangles are arranged according to the final error from small to large, and the triangle with the smallest folding error is selected from the triangle sequence to perform the folding operation. In this way, the boundary triangles and Feature triangles do not participate in the folding, thus preserving the boundary and attribute details of the model.

Using the above simplification method, the model is still very close to the original model after simplification of 70%. When the simplification is more than 90%, the obtained result can not only accept the geometric similarity, but also better retain the boundary and outline of the original model. and attributes, as shown in Figure 2.

为顶点对应的标量属性(例如对于颜色属性来说， ${\stackrel{\&RightArrow;}{A}}_i=r_i,g_i,b_i$ 代表RGB颜色值；对于纹理属性， ${\stackrel{\text{\&RightArrow;}}{A}}_i=u_i,v_i$ 表纹理坐标)，经过以上处理，就可以将属性信息记录在递进网格文件中，从而在重建原始几何模型的同时提供属性的恢复。While the model is being simplified, the coordinates of vertices of the triangles involved in folding and disappearing and the information of their related triangles are recorded to construct a progressive mesh file, thus supporting effective incremental transmission. For models containing attributes, the expression of vertex coordinates can be extended, and the vertex information can be expressed as v _i (x _i , y _i , z _i , ), where x _i , y _i , z _i are the vertex coordinates,

is the scalar attribute corresponding to the vertex (for example, for the color attribute, ${\stackrel{\&Right\; Arrow;}{A}}_i=r_i,g_i,b_i$ Represents RGB color values; for texture properties, ${\stackrel{\text{\&Right Arrow;}}{A}}_i=u_i,v_i$ Table texture coordinates), after the above processing, the attribute information can be recorded in the progressive mesh file, so as to restore the attribute while reconstructing the original geometric model.

2. Base grid compression coding

The present invention carries out compression coding based on octree division to the simplified base grid, quantifies the geometric coordinates of vertices with 12 bits, and mainly divides into three steps:

(1) Perform multi-resolution hierarchical organization of octrees on the vertices of the geometric model until the most refined tree nodes can express the quantization accuracy: first, by setting a minimum bounding box for the model, and according to the octree The division method performs recursive subdivision, in which each tree node records the number of vertices contained in its subtree, and the tree nodes without vertices will no longer be subdivided, while the tree nodes containing vertices will be iteratively subdivided until The node is empty or contains only one vertex and reaches the maximum precision level; therefore, the center of the leaf node in the generated octree implicitly represents the spatial position of the vertex it contains.

(2) Traverse each node of the octree in a breadth-first manner from coarse to fine, and output the data stream describing the position of the vertex: according to the number of vertices contained in the node, for each non-empty node, use 1bit to identify Whether it is a single node; for a single node, use 3bits to identify which child node in the single node contains a unique vertex; and for a non-empty non-single node, an 8bits identifier is generated to indicate whether the child node of the node is Non-empty; loop like this until the node at the finest level is visited.

(3) Rearrange and compress the vertex indexes of the topology data: rearrange the vertex indexes in the order of the vertices contained in the leaf nodes from left to right, and update the corresponding index values in the original topology data, and compress the topology data .

(4) Arrange and compress the attribute information in advance according to the order of the leaf vertices of the geometric node flow: arrange the attribute information in advance according to the order of the leaf vertices of the geometric node flow, and synchronize it with the geometric information during the encoding process.

(5) Perform arithmetic coding on all data streams.

After adopting compression coding, the base grid file size has been further reduced.

1. Progressive transmission

The invention adopts the progressive transmission mode of octree decoding-progressive grid transmission under the Browser/Server structure. In the transmission and decoding stage, the base grid is first restored by layer-by-layer refinement based on the breadth-first traversal of the octree; after the base grid information is transmitted, the progressive grid transmission is continued to transmit a series of detailed recovery information , to optimize and restore the model.

(1) Browser-server

The present invention uses the Java Applet embedded in the Web browser on the browser side, so that the three-dimensional model can be rendered without the need for the user to install an independent application program; the user can perform interactive operations such as rotation, translation, and zooming on the model, It is also possible to roam in the scene, view the area of interest, etc.; in addition, due to the use of progressive compression and transmission strategies, the response time is greatly shortened, and it is more suitable for ordinary users to browse the Internet.

In the process of 3D data transmission, when the user needs to view and operate the model, he can control and pause the transmission at any time. At this time, the browser sends a waiting request to the server, which contains the current data stream position and status, and the server suspends the transmission. Data; when the user stops operating, the browser sends a download request to the server, and the server continues to send data to the browser according to the saved state information, thereby continuing the refinement process of the 3D model.

(2) Communication thread and its management

In order to achieve the purpose of progressive transmission, it is necessary to download the file data while performing model rendering, so a multi-threaded implementation is adopted.

The server transmits the model files to the client sequentially, and the client analyzes the data content while receiving the data, and builds different internal model data representations for different content, so as to start drawing; at the same time, when each new information arrives, it needs to be updated internal data content, and re-updates the image on the screen. Three threads are needed here, namely: data download thread, data analysis thread and graphics drawing thread (main thread).

The download thread is responsible for downloading file data from the server. Whenever the client has data, it will be handed over to the analysis thread for semantic analysis of the data to form an internal data description format. The drawing thread is responsible for displaying these internal data to the screen. In the subsequent process, whenever a new message is received, it will be converted into an internal data description, and then displayed; so repeated.

In this process, the synchronization problem between threads needs to be solved: first, for the download thread and the analysis thread, it is meaningless to analyze before the data is downloaded, which requires synchronization between these two threads; Second, when the analysis thread analyzes the downloaded file data, the download thread may download new data and require the file data to be updated, which requires mutually exclusive access to data between these two threads; similarly, There is also such a problem between the analysis thread and the drawing thread. The drawing thread can only display after the analysis thread is partially completed (the analysis of several pieces of optimization information is completed), and there is also a read-write mutual exclusion between them.

Here, the critical section is used to realize the mutually exclusive access of these three threads to resources. When the download thread has not downloaded the data to the local client, the analysis thread can only wait and work only after getting the data. Depending on the speed of the two, it can be divided into two situations:

(1) The download speed is slower than the analysis speed: After analyzing the current data segment, the analysis thread needs to determine its own running state according to the current download state. If the current file has not been downloaded, continue to wait for the arrival of the next piece of data; if the current file has been downloaded, the analysis ends.

(2) The download speed is faster than the analysis speed: at this time, for the analysis thread, the data to be analyzed always exists, and it only needs to wait for the arrival of the first data at the beginning.

For the analysis thread and the drawing thread, the synchronization between them is relatively simple, as long as the drawing thread can draw after the analysis of new data ends.

1. 3D scene transmission

A 3D scene is composed of different objects, each of which is an independent shape or structure. Therefore, in the simplification and encoding stage, the 3D scene is regarded as a combination of several object models, and the mesh simplification with fully preserved topological boundaries is performed on each object in it, while in the transmission restoration stage, the multi-stream method, and adopts an object-oriented viewpoint-related transmission strategy, and selects different model objects for optimization according to different viewpoints of users.

First, a separate data stream is established for each object in the scene, so for each data stream, the aforementioned progressive transmission mode of octree decoding-progressive grid transmission can be used. Different from simple object model transmission, the present invention designs an object-oriented viewpoint-related transmission framework, as shown in Figure 3, the server side provides model data and saves the transmission state of the current model object, and the client downloads the data and reconstructs, Render the model. In addition to the download, analysis, and drawing threads, add a visibility judgment thread for the client to calculate the viewpoint information and judge the objects to be transmitted according to the visibility of the cone; the server only needs to receive the required data sent by each client. Object flag, you can send the corresponding optimization information. In this way, the calculation load is distributed to each client, thereby improving the efficiency of the server; and for the client, since this visibility judgment is coarse-grained and object-oriented, it will not cause a large amount of calculation. Affects visual effects.

The entire execution process of the framework is:

(1) The client sends a scene transfer request to the server;

(2) The server receives the client request, establishes a connection with the client, and at the same time checks whether the model data requested by the user has been loaded into the memory, if not, loads it into the memory, and then creates a tree for each object of the scene octree;

(3) The server transmits the base grid data of each object (vertex coordinates, surface indices, and other relevant information) to the client;

(4) After the client receives the base grid information, it displays it;

(5) The client calculates the currently visible objects according to its own viewpoint, and transmits the visibility status to the server;

(6) The server transmits the optimization information of the currently visible object according to the current client state, and suspends the transmission of the optimization information data stream of other objects, and saves the current state;

(7) Each time the client receives a piece of optimization information, it refines the model and redraws it;

(8) If the viewpoint changes, repeat the above (5);

(9) The client exits, and the server cleans up useless state information.

Using the above transmission method, the model is progressively transmitted. Figure 4 shows the transmission of a Guanyin model. First, the octree decoding is performed, and after the basic grid is obtained, the progressive grid transmission and restoration is performed. From the transmission process in Figure 4, it can be seen that the base grid obtained after the octree is started and decoded and transmitted by 20%, is very close to the original model in terms of geometric similarity, and also preserves the boundaries and contours of the original model well. and texture features.

Figure 5 shows the process of progressive transmission of a group of terracotta warrior models in the scene. The initial process is still similar. First, the octree decoding and transmission of the currently visible terracotta warrior models is performed; then the viewpoint is zoomed in, and some objects enter the field of view and participate in refinement ; When the viewpoint continues to rotate, the terracotta warriors and horses model outside the window still appear rough due to the transmission at the beginning.

Claims (4)

1. A three-dimensional model progressive transmission method based on feature retention is characterized by comprising the following steps:

(1) Simplifying the retained features of the original grid model to form a rough base grid;

(2) Carrying out octree compression coding on the formed rough base grid to form a progressive grid file;

(3) Adopting a progressive transmission mode of octree decoding-progressive grid transmission under a Browser/Server structure;

(4) And transmitting the three-dimensional scene by adopting an object-oriented viewpoint related transmission strategy on the basis of the steps.

2. The feature preservation-based three-dimensional model progressive transmission method according to claim 1, characterized in that: in the step (1), the original mesh model is simplified by adopting a method for retaining topological boundary and texture or color attribute based on triangle folding, and the steps are as follows:

(1) Classifying the original triangular mesh: boundary triangles, corner triangles, internal triangles and characteristic triangles;

(2) According to the classification of the triangle, different error measurement methods are adopted:

a. for internal triangles, T is assigned to each triangle _i Provided with a 4 x 4 error matrix Q _i And assume that the new point after it is folded is v _i Defining the folding error of the triangle as

By minimizing the error to obtain v _i Position;

b. for the corner point triangle, defining a new folded point vi as a vertex on the primary boundary, and calculating a folding error by adopting the same method as the step a;

c. the folding error is set to the maximum for the boundary triangle and the feature triangle.

(3) And after the folding errors of all the triangles are obtained, selecting the triangle with the smallest error to perform the folding operation.

3. The feature preservation-based three-dimensional model progressive transmission method according to claim 1, characterized in that: and (4) in the initial transmission stage in the step (3), an octree breadth traversal decoding mode is adopted until the transmission of the base grid data is finished, and then a progressive grid transmission mode is adopted to keep the continuous improvement of the model precision.

4. The progressive transmission method of the three-dimensional model according to claim 1, wherein: the step (4) of adopting an object-oriented viewpoint-related transmission strategy for the three-dimensional scene model comprises the following steps:

(1) Constructing a single data stream for each object in the scene, and adopting the progressive transmission mode of the octree decoding-progressive grid transmission for each data stream;

(2) The method comprises the steps that downloading, analyzing and drawing threads are arranged on a client side, synchronous control is conducted, and parallel transmission of objects is guaranteed;

(3) Calculating viewpoint information at a client, and judging the current object to be transmitted according to the visibility of the view cone;

(4) The server receives the needed object mark transmitted by the client and sends corresponding optimization information.

Images