CN106384386A - Grid processing method for LOD model generation and grid processing system thereof and 3D reconstruction method and system - Google Patents

Abstract

The invention discloses a grid processing method for LOD model generation and a grid processing system thereof and a 3D reconstruction method and system. The grid processing method for the LOD model generation process comprises the steps that an initial grid is acquired; the initial grid is cut so that multiple LOD partitions are obtained; the LOD partitions of the current level are sutured through grid merging so that a merged grid is obtained; grid simplification is applied to the merged grid so that the LOD partitions of the next level are obtained; and grid merging is iteratively performed on the LOD partitions of the next level until the final merged grid is obtained. The grid processing method for LOD model generation is robust and extensible for the ever increasing demand caused by the fast growing 3D reconstruction technology. Three main grid operations can be optimized, and the experiment on large data sets proves that the grid processing method for LOD model generation has robustness and extensibility.

Description

Grid processing method, system and 3D method for reconstructing and system in the generation of LOD model

Technical field

Embodiment of the present invention is related to grid process side during LOD (Level-of-detail, level of detail) model generates Method, system and 3D (three-dimensional) method for reconstructing and system.

Background technology

In past 10 years, by the universalness of improved computing hardware and unmanned plane photography, large-scale 3D reconstruction being become can Energy.These rebuild 3D model have from Buildings Modeling, urban planning, measurement, traces protection typically used as to such as machine The extensive application of the application of relatively frontier of the obstacle avoidance in people's technology and 3D printing etc..Coml unmanned plane is photographed especially Fast-developing in recent years.Certainly, future will may require that even more large-scale and higher-quality 3D reconstruction model.

The possible mode processing this large data is divide and conquer (divide-and-conquer).Identical technology is It is applied to various visual angles stereo reconstruction (multi view stereo, MVS) step, it is to produce the object of capture or the height of scene The required step of density points cloud and be the prior step that large-scale LOD model generates.Hereafter, extract net from each cluster of a cloud Lattice face.This means that produced grid surface is typically fragment and discontinuous.And, the generally chi in these initial mesh faces Very little too big so that it cannot pass through the Internet transmission, and limit its usability.Preferably solution is initial according to these Grid surface generates multi-resolution models or level of detail (LOD) grid, to meet the accuracy requirements of the different levels of model.Example As, transmission application requirement model the most detailed, and the model simplifying is best suited for on-line normalization and rapid preview scene.Only pass through The grid producing LOD form can make large-scale 3D reconstruction model constantly develop and more practical.

However, most of grid operations technology of exploitation are conceived to the single object of process, almost it bel not applied to process this Plant large data.When for this large data, some in these technology start to expose some problems, and these are asked Topic does not almost cause concern in the past.And, fragment and highdensity initial mesh is converted into continuous LOD form grid Need to mix several kinds of grid operations, which again increases one layer of difficulty.

Content of the invention

In computer graphicss, level of detail (LOD) is to choice (trade between model fidelity and rendering speed Off research).Basic thought is using the model representation simplifying is inessential or object of viewpoint further away from each other reduces and renders Load.For many years many researcheres have pointed out different methods and generate LOD model and put into application.Works " 3D figure Level of detail (Level of Detail For 3D Graphics) " be LOD aspect extraordinary general survey and its intermediary The classical technology continuing is still related so far.Outside all LOD frameworks, the present invention makes on discrete LOD.Discrete LOD Refer to different level of detail same model be discrete and each level produce offline first and be based on observer and Mo The distance between type operationally selects suitably to render level.Advantage is that each level of model is static and thus Impact can be produced to increase its rendering speed to graphic hardware.With respect to other frameworks, for example continuous LOD or regard related LOD, from Scattered LOD be more prone to program because its by lattice simplified with render decoupling.This decoupling also makes lattice simplified parallelization offline And it is thus more expansible.Fig. 5 shows the exemplary of the discrete LOD of single object.

Different from simple, the single object LOD example of as described before, the problem that the present invention faces is that scale is much bigger simultaneously And constantly increase, must also be using technology of dividing and ruling in LOD flow process.Thus, the invention provides a kind of LOD model generate in net Lattice processing method, system and 3D method for reconstructing and system.

On the one hand, permissible according to the method processing grid in a kind of model generating process in LOD of embodiment of the present invention Including：Obtain initial mesh；Shear described initial mesh to obtain multiple LOD subregions；By mesh update by current level The suture of LOD subregion is to obtain the grid merging；The lattice simplified LOD to obtain next level of grid application of described merging is divided Area；Described mesh update is executed until obtaining the grid of final merging to the LOD subregion iteration of next level described.

In another embodiment of the present invention, the hole that methods described also includes the grid merging is existed carries out hole filling So that adjacent mesh surface is adapted.

Alternatively, the filling of described hole may include the first filling and processes：Build the grid of the topological structure with known hole Patch；The border vertices in the hole on target gridding are mapped to the border vertices of this grid patch；Position using this border vertices Put and as constraints, internal vertex is smoothed.Additionally, the filling of described hole may also include the second filling processing：Grid is mended Fourth carries out Initial Triangulation Approach；By increasing internal vertex, trigonometric ratio is refined；The internal vertex of increase is placed on makes benefit The position that fourth surface is adapted with adjacently situated surfaces.

Wherein, carry out Initial Triangulation Approach to grid patch can include：Define two error functions of triangle：W_AreaWith W_{Dihedral angle-angle}, wherein, W_AreaRefer to the area of triangle, W_{Dihedral angle-angle}Refer to the maximum dihedral angle angle with adjacent triangle；Definition Grid patch error function W is W_AreaThe minimum error of total or W_{Dihedral angle-angle}Maximum minimum error；By making W minimum Change and to complete Initial Triangulation Approach.

Additionally, or alternatively, described second filling process also include：The surface in the hole of smooth new filling, minimizes side Normal variations between boundary summit and the internal vertex of increase.

In an alternative embodiment of the invention, described first filling processes for the hole more than or equal to predetermined threshold, Described second filling is processed for the hole less than described predetermined threshold.

Alternatively, described shearing can include：Define secant plane；All and this cut-out is searched in described initial mesh The triangle of level-crossing；The triangle of described intersection is separated into little triangle along shearing line；By using described little by three Angular geometric center as test point by all of little triangular marker in the either side of described secant plane, and remove this Unwanted triangle in side.

Alternatively, described mesh update can include：Combined side along grid to be combined cuts away predetermined offset to build Stand the merging space between grid to be combined；Combined side along grid to be combined is passed through to mate between two merging grids Closely border vertices set up the bridge between grid to be combined；Hole filling is carried out to complete to close to hole between described net to be combined And.Additionally, or alternatively, described mesh update also includes：Along bridge described in Boundary Traversal, and record the order of bridge, check Discontinuous order, the bridge being intersected by the sequential search of bridge, and abandon the bridge of described intersection.

Alternatively, the various different embodiments of the present invention carry out institute based on QSlim algorithm using following Quadratic Surface Equations State lattice simplified：

Wherein, Q_vIt is expressed as quadratic surface, v represents summit, p represents the support plane on summit, planes (v) represents v's Support plane set, K_pRepresent that in grid, summit is to the squared-distance matrix supporting plane, α ∈ [0,1] determines and detected The significance level of plane domain, β ∈ [0,1] determines the simplification priority of plane domain, K_{planar region plane(v)}Represent net The squared-distance matrix of the extremely affiliated characteristic plane in summit in lattice.

On the other hand, can include according to a kind of 3D method for reconstructing of embodiment of the present invention：By the three-dimensional weight of various visual angles Build the point cloud that MVS obtains object；Obtain surface mesh from described cloud；Based on the surface mesh of described acquisition, using above-mentioned each Plant the grid processing method described in different embodiments and generate LOD grid；Complete the 3D of described object using the LOD grid generating Rebuild.

Another further aspect, can according to the system processing grid in a kind of model generating process in LOD of embodiment of the present invention To include：

Obtain module, for obtaining initial mesh；

Shear module, for shearing described initial mesh to obtain multiple LOD subregions；

Merge module, for the grid obtaining the LOD subregion suture of current level merging by mesh update；

Simplify module, for lattice simplified to the grid application of described merging to obtain the LOD subregion of next level；

Wherein, described mesh update is executed until obtaining final merging to the LOD subregion iteration of next level described Grid.

Preferably, the hole that described system also includes that the grid merging is existed carries out hole filling so that adjacent mesh surface phase The filling module adapting to.Alternatively, described filling module can be used for executing the first filling process：Build opening up with known hole Flutter the grid patch of structure；The border vertices in the hole on target gridding are mapped to the border vertices of this grid patch；Using this The position of border vertices smooths to internal vertex as constraints.Additionally, described filling module can be additionally used in second and fills out Fill process：Initial Triangulation Approach is carried out to grid patch；By increasing internal vertex, trigonometric ratio is refined；The inside that will increase Summit is placed on the position making patch surface be adapted with adjacently situated surfaces.Wherein, carry out Initial Triangulation Approach to grid patch can wrap Include：Define two error functions of triangle：W_AreaAnd W_{Dihedral angle-angle}, wherein, W_AreaRefer to the area of triangle, W_{Dihedral angle-angle}Refer to Maximum dihedral angle angle with adjacent triangle；Definition grid patch error function W is W_AreaThe minimum error of total or W_{Dihedral angle-angle}Maximum minimum error；To complete Initial Triangulation Approach by making W minimize.Optional embodiment party in the present invention In formula, described second filling processes and may also include：The surface in the hole of smooth new filling, minimizes border vertices and increased inside Normal variations between summit.

In the selectivity embodiment of the present invention, described first filling is processed for more than or equal to predetermined threshold Hole, described second filling is processed for the hole less than described predetermined threshold.

Preferably, described shearing may include：Define secant plane；Described initial mesh is searched all flat with this cut-out The triangle that face intersects；The triangle of described intersection is separated into little triangle along shearing line；By using described little triangle The geometric center of shape as test point by all of little triangular marker in the either side of described secant plane, and remove this side In unwanted triangle.

Preferably, described mesh update may include：Combined side along grid to be combined cuts away predetermined offset to set up Merging space between grid to be combined；Nearest between two merging grids by mating along the combined side of grid to be combined Set up the bridge between grid to be combined apart from border vertices；Hole filling is carried out to complete to merge to hole between described net to be combined. Alternatively, described mesh update may also include：Along bridge described in Boundary Traversal, and record the order of bridge, check discontinuous Sequentially, the bridge being intersected by the sequential search of bridge, and abandon the bridge of described intersection.

Preferably, the system of embodiment of the present invention be based on QSlim algorithm carried out using following Quadratic Surface Equations described Lattice simplified：

Wherein, Q_vIt is expressed as quadratic surface, v represents summit, p represents the support plane on summit, planes (v) represents v's Support plane set, K_pRepresent that in grid, summit is to the squared-distance matrix supporting plane, α ∈ [0,1] determines and detected The significance level of plane domain, β ∈ [0,1] determines the simplification priority of plane domain, K_{planar region plane(v)}Represent net The squared-distance matrix of the extremely affiliated characteristic plane in summit in lattice.

Another aspect, a kind of 3D reconstructing system according to embodiment of the present invention can include：

First acquisition unit, obtains the point cloud of object by various visual angles stereo reconstruction MVS；

Second acquisition unit, obtains surface mesh from described cloud；

Signal generating unit, based on the surface mesh of described acquisition, at the grid described in above-mentioned various different embodiments Reason method generates LOD grid；

Reconstruction unit, is rebuild using the 3D that the LOD grid generating completes described object.

According to the various embodiment of the present invention, it is for the growing need to be carried by the 3D reconstruction technique of rapid growth It is robust and extendible for asking.It optimizes three main grid operations, described grid operations be mainly used in flow process, Challenging grid flaw that fixation is produced by aforementioned reconstruction procedures, merge grid and decline without overlay requirements or do not produce Subtract triangle, be to be suitable to City scenarios grid by general simplification algorithm optimization.Experiment on large data collection shows that it has Robustness and extensibility.

Brief description

Fig. 1 shows the example flow of the grid processing method according to embodiment of the present invention.

Fig. 2 shows the example flow of the 3D method for reconstructing according to embodiment of the present invention.

Fig. 3 shows the exemplary architecture of the lattice processing system according to embodiment of the present invention.

Fig. 4 shows the exemplary architecture of the 3D reconstructing system according to embodiment of the present invention.

Fig. 5 shows the exemplary of the discrete LOD of single object.

Fig. 6 shows a kind of original LOD design.

Fig. 7 a shows the 3D reconstruction image example obtaining using original LOD design.

Fig. 7 b shows that restrained boundary produces undesirable star network during simplifying.

Fig. 8 shows a kind of LOD design according to embodiment of the present invention.

Fig. 9 shows the definition in hole.

Figure 10 shows across the detection of the hole of subregion.

Figure 11 a shows typical case's input of hole filling algorithm.

Figure 11 b shows and does not carry out refining the hole filling algorithm with shaping.

Figure 11 c shows the hole filling algorithm having carried out refinement and shaping.

Figure 11 d shows the comparison of hole filling algorithm and true ground.

Figure 12 shows typical input data scale.

Figure 13 shows the effect of α in archioporus filling algorithm.

Figure 14 shows new hole fill method.

Figure 15 shows the sparse linear system adopting in new method.

Figure 16 shows the visual comparison (line architecture (wireframe)) of archioporus filling algorithm and new method.

Figure 17 shows the visual comparison (surface) of archioporus filling algorithm and new method.

Figure 18 shows the restriction of new method.

Figure 19 shows strange Self-crossover after simplifying the region of new method filling.

Figure 20 shows a case that possible in shearing.

Figure 21 shows the input merging algorithm.

Figure 22 a shows the step 1 merging algorithm.

Figure 22 b shows the step 2 merging algorithm.

Figure 22 c shows the step 3 merging algorithm.

Figure 23 show and the algorithm that improves final result.

Figure 24 shows process bridge cross-cutting issue.

Figure 25 shows the purpose of QSlim geometric error company.

Figure 26 shows the comparison of the method for simplifying of QSlim and embodiment of the present invention, and wherein, in input, black region is The plane domain detecting.

Figure 27 shows the comparison of the method for simplifying of QSlim and embodiment of the present invention.

Figure 28 shows the large data collection for experiment.

Figure 29 shows the 5th grade of LOD grid of Toulouse.

Figure 30 shows the 4th grade of LOD grid of Toulouse.

Figure 31 shows the 3rd level LOD grid of Toulouse.

Figure 32 shows the 2nd grade of LOD grid of Toulouse.

Figure 33 shows the 1st grade of LOD grid of Toulouse.

Figure 34 shows the 0th grade of LOD grid of Toulouse.

Specific embodiment

For the ease of understanding various aspects, feature and the advantage of technical solution of the present invention, below in conjunction with the accompanying drawings to this Bright it is specifically described.It should be appreciated that following various embodiments are served only for illustrating, not for the restriction present invention's Protection domain.

Example block diagram below with reference to method, system, unit and programming and computer program to describe this Invention.It should be appreciated that each of example block diagram frame and combinations thereof can be realized by programming instruction, described programming refers to Order includes computer program instructions.These computer program instructions can be loaded to computer or other programmable datas process dress Put, the instruction of execution so on computer or other programmable data processing units produces and is used for realizing in block diagram or flow chart The instruction of the function of specifying.These computer program instructions are also storable in computer-readable memory, may indicate that computer Or other programmable data processing units run in a specific way, so, the instruction being stored in computer-readable memory is produced The manufacture of the raw instruction of function including realizing specifying in block diagram or flow chart.Computer program instructions also can load to calculating Machine or other programmable data processing units, to produce a series of execution in described computer or other programmable devices to produce The operating procedure of raw executive program, so, the instruction of execution on described computer or other programmable devices provides The step of the function that realization is specified in block diagram or flow chart.Programming instruction also can be stored in electronic circuit and/or by electricity Sub-line road is implemented to realize each function, the step of the present invention.

Grid processing method

Fig. 1 shows the method according to processing grid in a kind of model generating process in LOD of embodiment of the present invention.As Shown in Fig. 1, described can include, but are not limited to：

110. acquisition initial mesh；

120. shear described initial mesh to obtain multiple LOD subregions；

130. pass through the grid that the LOD subregion suture of current level is obtained merging by mesh update；

The grid application of 140. pairs of described merging lattice simplified to obtain the LOD subregion of next level；

The LOD subregion iteration of next level described in 150. pairs executes described mesh update until obtaining the net of final merging Lattice.

In the selectivity embodiment of the present invention, the hole that methods described can also include the grid merging is existed is carried out Hole filling is so that adjacent mesh surface is adapted.Alternatively, the filling of described hole may include the first filling and processes, as follows：Build band There is the grid patch of the topological structure in known hole；The border vertices in the hole on target gridding are mapped to the side of this grid patch Boundary summit；The position that this border vertices are used smooths to internal vertex as constraints.Additionally, the filling of described hole also may be used Process including the second filling, as follows：Initial Triangulation Approach is carried out to grid patch；By increasing internal vertex, trigonometric ratio is carried out carefully Change；The internal vertex of increase is placed on the position making patch surface be adapted with adjacently situated surfaces.

Wherein, carry out Initial Triangulation Approach to grid patch can include：Define two error functions of triangle：W_AreaWith W_{Dihedral angle-angle}, wherein, W_AreaRefer to the area of triangle, W_{Dihedral angle-angle}Refer to the maximum dihedral angle angle with adjacent triangle；Definition Grid patch error function W is W_AreaThe minimum error of total or W_{Dihedral angle-angle}Maximum minimum error；By making W minimum Change and to complete Initial Triangulation Approach.

Additionally, or alternatively, described second filling process can also include：The surface in the hole of smooth new filling, minimum Change the Normal variations between border vertices and the internal vertex of increase.

In an alternative embodiment of the invention, described first filling processes for the hole more than or equal to predetermined threshold, I.e. macropore, described second filling is processed for the hole less than described predetermined threshold, i.e. aperture.Described predetermined threshold can be according to reality Border needs to be determined, for example, 600 summits.

Alternatively, in one embodiment of the invention, described shearing can include, but are not limited to：Definition cut-out is flat Face；All trianglees intersecting with this secant plane are searched in described initial mesh；By the triangle of described intersection along cutting Tangent line is separated into little triangle；By using described little triangle geometric center as test point by all of little triangle mark Note is in the either side of described secant plane, and removes unwanted triangle in this side.

Alternatively, in one embodiment of the invention, described mesh update can include, but are not limited to：Along treating The combined side merging grid cuts away predetermined offset to set up the merging space between grid to be combined；Along grid to be combined Combined side sets up the bridge between grid to be combined by mating the minimum distance border vertices between two merging grids；To described Between net to be combined, hole carries out hole filling to complete to merge.Additionally, described mesh update can also include：Along Boundary Traversal institute State bridge, and record the order of bridge, check discontinuous order, the bridge being intersected by the sequential search of bridge, and abandon described friendship The bridge of fork.

Alternatively, in the various different embodiments of the present invention, entered using following Quadratic Surface Equations based on QSlim algorithm Row is described lattice simplified：

Wherein, Q_vIt is expressed as quadratic surface, v represents summit, p represents the support plane on summit, planes (v) represents v's Support plane set, K_pRepresent that in grid, summit is to the squared-distance matrix supporting plane, α ∈ [0,1] determines and detected The significance level of plane domain, β ∈ [0,1] determines the simplification priority of plane domain, K_{planar region plane(v)}Represent net The squared-distance matrix of the extremely affiliated characteristic plane in summit in lattice.

According to the various embodiment of the present invention, it is for the growing need to be carried by the 3D reconstruction technique of rapid growth It is robust and extendible for asking.It optimizes three main grid operations, described grid operations be mainly used in flow process, Challenging grid flaw that fixation is produced by aforementioned reconstruction procedures, merge grid and decline without overlay requirements or do not produce Subtract triangle, be to be suitable to City scenarios grid by general simplification algorithm optimization.Experiment on large data collection shows that it has Robustness and extensibility.

3D method for reconstructing

Fig. 2 shows a kind of 3D method for reconstructing according to embodiment of the present invention.As shown in Fig. 2 described method for reconstructing can To include, but are not limited to：

210. obtain the point cloud of object by various visual angles stereo reconstruction MVS；

220. obtain surface mesh from described cloud；

230. surface meshes based on described acquisition, using the grid processing method described in above-mentioned various different embodiments Generate LOD grid；

240. are rebuild using the 3D that the LOD grid generating completes described object.

According to the various embodiment of the present invention, it is for the growing need to be carried by the 3D reconstruction technique of rapid growth It is robust and extendible for asking.It optimizes three main grid operations, described grid operations be mainly used in flow process, Challenging grid flaw that fixation is produced by aforementioned reconstruction procedures, merge grid and decline without overlay requirements or do not produce Subtract triangle, be to be suitable to City scenarios grid by general simplification algorithm optimization.Experiment on large data collection shows that it has Robustness and extensibility.

Lattice processing system

Fig. 3 shows according to the system processing grid in a kind of model generating process in LOD of embodiment of the present invention.As Shown in Fig. 3, lattice processing system can include, but are not limited to obtain module 310, shear module 320, merge module 330 and Simplify module 330.Wherein, obtain module 310 to can be used for obtaining initial mesh, shear module 320 can be used for shearing by described , to obtain multiple LOD subregions, merging module 330 can be used for ought by mesh update for the initial mesh obtaining module 310 acquisition The LOD subregion of front level sutures to obtain the grid merging, and simplification module 340 can be used for the grid application to described merging Lattice simplified to obtain the LOD subregion of next level, wherein, described grid is executed to the LOD subregion iteration of next level described Merge until obtaining the grid of final merging.

In an alternative embodiment of the invention, the hole that described system may also include to the grid presence merging carries out hole and fills out It is filled with the filling module making adjacent mesh surface be adapted.Alternatively, described filling module can be used for executing the first filling process： Build the grid patch of the topological structure with known hole；The border vertices in the hole on target gridding are mapped to this grid mend The border vertices of fourth；The position that this border vertices are used smooths to internal vertex as constraints.Additionally, described filling Module can be additionally used in the second filling and processes：Initial Triangulation Approach is carried out to grid patch；By increasing internal vertex, trigonometric ratio is entered Row refinement；The internal vertex of increase is placed on the position making patch surface be adapted with adjacently situated surfaces.Wherein, to grid patch Carry out Initial Triangulation Approach may include：Define two error functions of triangle：W_AreaAnd W_{Dihedral angle-angle}, wherein, W_AreaRefer to triangle Area, W_{Dihedral angle-angle}Refer to the maximum dihedral angle angle with adjacent triangle；Definition grid patch error function W is W_AreaTotal The minimum error of meter or W_{Dihedral angle-angle}Maximum minimum error；To complete Initial Triangulation Approach by making W minimize.At this In bright optional embodiment, described second filling processes and may also include：The surface in the hole of smooth new filling, minimizes border top Normal variations between point and increased internal vertex.

In the selectivity embodiment of the present invention, described first filling is processed for more than or equal to predetermined threshold Hole, described second filling is processed for the hole less than described predetermined threshold.

In the various embodiments of the present invention, described shearing may include：Define secant plane；In described initial mesh Search all trianglees intersecting with this secant plane；The triangle of described intersection is separated into little triangle along shearing line； By using described little triangle geometric center as test point by all of little triangular marker in described secant plane Either side, and remove unwanted triangle in this side.Preferably, described mesh update may include：Along grid to be combined Combined side cut away predetermined offset to set up the merging space between grid to be combined；Combined side along grid to be combined is led to Two minimum distance border vertices merging between grid of overmatching set up the bridge between grid to be combined；To described net to be combined Between hole carry out hole filling with complete merge.Alternatively, described mesh update may also include：Along bridge described in Boundary Traversal, and And the order of record bridge, check discontinuous order, the bridge being intersected by the sequential search of bridge, and abandon the bridge of described intersection.

In the various embodiments of the present invention, described lattice processing system is based on QSlim algorithm and adopts following secondary songs Face equation carries out described lattice simplified：

Wherein, Q_vIt is expressed as quadratic surface, v represents summit, p represents the support plane on summit, planes (v) represents v's Support plane set, K_pRepresent that in grid, summit is to the squared-distance matrix supporting plane, α ∈ [0,1] determines and detected The significance level of plane domain, β ∈ [0,1] determines the simplification priority of plane domain, K_{planar region plane(v)}Represent net The squared-distance matrix of the extremely affiliated characteristic plane in summit in lattice.

According to the various embodiment of the present invention, it is for the growing need to be carried by the 3D reconstruction technique of rapid growth It is robust and extendible for asking.It optimizes three main grid operations, described grid operations be mainly used in flow process, Challenging grid flaw that fixation is produced by aforementioned reconstruction procedures, merge grid and decline without overlay requirements or do not produce Subtract triangle, be to be suitable to City scenarios grid by general simplification algorithm optimization.Experiment on large data collection shows that it has Robustness and extensibility.

3D reconstructing system

Fig. 4 shows a kind of 3D reconstructing system according to embodiment of the present invention.In embodiments of the present invention, described 3D Reconstructing system can include：

First acquisition unit 410, obtains the point cloud of object by various visual angles stereo reconstruction MVS；

Second acquisition unit 420, obtains surface mesh from described cloud；

Signal generating unit 430, based on the surface mesh of described acquisition, using the grid described in any of the above different embodiments Processing method generates LOD grid；

Reconstruction unit 440, is rebuild using the 3D that the LOD grid generating completes described object.

According to the various embodiment of the present invention, it is for the growing need to be carried by the 3D reconstruction technique of rapid growth It is robust and extendible for asking.It optimizes three main grid operations, described grid operations be mainly used in flow process, Challenging grid flaw that fixation is produced by aforementioned reconstruction procedures, merge grid and decline without overlay requirements or do not produce Subtract triangle, be to be suitable to City scenarios grid by general simplification algorithm optimization.Experiment on large data collection shows that it has Robustness and extensibility.

Based on the various different embodiments of method, system, the present invention is described in detail above, below with concrete Application the present invention is illustrated.

Application example

As it was previously stated, different from simple, the single object LOD example of as described before, the problem that the present invention faces is rule Mould is much bigger and constantly increases, must also be using technology of dividing and ruling in LOD flow process.Step before LOD flow process is MVS, its Produce the discontinuous surface mesh of high density.These initial mesh are as the input of LOD flow process.It should be noted that at the beginning of these Beginning grid is in total cooperative system, and therefore, they overlap around in its borderline region.

A kind of original design is each initial mesh to be processed as single object and independently produces LOD grid, referring to Fig. 6.Although this design has extraordinary autgmentability, major defect is there is boundary difference between adjacent grid Different.A kind of means to save the situation is to constrain or limit these borders during lattice simplified, but this will be lattice simplified for big havoc Effectiveness because the ratio regular meeting on fixing summit quick increase and produce star over-simplification grid, referring to Fig. 7 a and 7b.

Therefore, embodiment of the present invention adopts quaternary tree (quad-tree) to design.Referring to Fig. 8.First by all input nets Lattice shearing (trim) are overall 2D grid (global 2D grid), and less rectangular mesh is referred to as subregion (tiles).Each side The number of subregion is 2 upwardsⁿ, wherein, n is natural number.This input level is referred to as n-th grade of LOD.Producing each next stage During LOD, first pass through mesh update and the LOD subregion of 2 × 2 current levels is sutured.Subsequently, by all grid flaws (mesh artifacts) is fixed in the grid of each merging.Finally, to merge grid application lattice simplified to obtain under The LOD of one-level.It is sequentially repeated this three step process until obtaining the final grid (that is, the 0th grade of LOD) merging of only one. The number of level is subsequently n+1, and, at n-th grade, the sum of subregion is 2ⁿ×2ⁿ=2²ⁿ.Therefore, by moving to next stage, The number of subregion is reduced to a quarter of current level.

Quaternary tree design can solve the problems, such as the border-differential in grid and over-simplification problem.This design also reduces literary composition Number of packages mesh and network I/O expense (overhead), this is because reduce total grid number.For example, if obtaining 2⁴×2⁴=16 × 16=256 subregion is as input, then altogether will produce 2⁸+2⁶+2⁴+2²+2⁰=341 discrete grids, comparatively speaking, In having the original design of LOD of this level quantity, grid sum is 2⁸× 5=1280.It can be seen that, total grid number is able to significantly Reduce.This design of embodiment of the present invention is extendible, because simply use local when producing next stage LOD adjacent Grid, it can be carried out in a distributed manner.Only need a data aggregate in each level.The granularity of subregion, the number of LOD level and Simplifying rate depends on hardware limitation and application purpose.In general, the granularity of subregion is more fine and LOD level deeper become In taking longer time, however it is necessary that less strength machine (powerful machines) and higher parallelization (parallelization).

This design can be implemented using three kinds of main grid operations, i.e. hole filling (hole filling), shearing and conjunction And (trimming and merging), simplification (simplification).The purpose of " hole filling " is fixing initial mesh The flaw (artifacts) existing, described flaw can flowed due to removing non-manifold (non-manifold) grid Journey internal dynamic produces." shearing " is applied to initial mesh to remove overlapping region and " merging " is used for removing border-differential simultaneously Move up quaternary tree." simplification " is the core that LOD generates, and wherein, simplification can subtract while maintaining grid geometric fidelity Few size of mesh opening.Thoroughly discuss each operation below.

【Hole is filled】

In theory, it is related to two kinds of hole in grid, i.e. border hole and topological hole (topological hole), ginseng See Fig. 9.Referring briefly to border hole as hole for the sake of here discussion border hole and simplicity.Assume the hole filling algorithm of reality first And discuss that it limits and problem.Subsequently introduce and process the described method limiting with problem.

Hole is topological characteristic, and it easily can detect by using half of (Half-Edge) data structure, described Half- Edge data structure is the modal mode representing grid.Each non-boundary edge can and border adjacent with two gores Edge is only adjacent with a face, so that opposite side is opened, and thus becomes the part in hole.Subsequently by through all of border Half edge, can detect all holes and each hole is expressed as the ordered list at border half edge (or summit).Real in the present invention Apply in mode, many using CGAL (Computational Geometry Algorithms Library, computational geometry algorithms library) Face volume data structure, it is substantially the half of data structure representing grid.

In one embodiment of the invention, LOD design all input grids be cut into 2D lattice structure and each Rectangular mesh is referred to as subregion.How to detect the so big hole existing across subregion, as one of illustration in Figure 10 (left) Individual hole.

The detection in hole can not be completely covered all of scope by identifying first and selecting to have when being projected in X/Y plane Subregion.The coverage of subregion may be defined as its projected area with respect to rectangular area (that is, the area of complete rectangular sub-area) Ratio.Its projected area can be by using the half of delaunay trigonometric ratio amounting to its 2D restriction as the edge limiting in border hole Triangle in (delaunay triangulation) and find.Another quick estimation to projected area is to calculate its projection The area of the 2D convex surface shell on summit.For example, CGAL provides the software kit (packages) easily calculating two values.Subsequently Selection subregion not being completely covered for all that expands outwardly the subregion of an adjacent ring.Finally, merging is selected Subregion and carry out hole detection.Referring to Figure 10.

In one embodiment of the invention, described hole filling algorithm can be analyzed to three key steps：

1. Initial Triangulation Approach (Initial triangulation).Define two error functions of triangle, W first_Area And W_{Dihedral angle-angle}, W_AreaRefer to the area of triangle, W_{Dihedral angle-angle}Refer to the maximum dihedral angle angle with adjacent triangle.Subsequently define Grid patch (mesh patch) error function W is W_AreaThe minimum error of total or its trigonometric ratio W_{Dihedral angle-angle}Maximum Minimum error.Therefore, the Initial Triangulation Approach of grid patch be make Initial Triangulation Approach that W minimizes and this can be compiled by dynamic Journey is searched.

2. refine (Refinement).Initial Triangulation Approach has the hole of base closed, but trigonometric ratio too coarse so that The patch (patch) of nature complexion cannot be shown, because it lacks internal vertex.This step is based on some heuristic rules (heuristic criteria) iteratively increases more internal vertexs on trigonometric ratio.This standard is by the parameter control of referred to as α System.In short, higher α leads to degree of refinement higher and takes longer time operation.

3. shaping (Fairing).After inserting new summit in refinement step, shaping will place them in so that mending Fourth surface and adjacently situated surfaces seem natural position.This is by repeating the surface in the hole of smooth new filling, minimizing border top Point and in refinement step the Normal variations between the new summit of insertion completing.

These three steps, in Figure 11 a, are illustrated in 11b, 11c, 11d.Wherein, there is no refinement step, Initial Triangulation Approach Strange shape can be produced.After refinement and shaping step, the shape of grid patch of filling is adapted with adjacently situated surfaces.

Because the thorough dynamic programming retrieval of the step 1 in algorithm, the step 1 in algorithm is in O (n³) time operation.In step Run time complexity in rapid 2 is more difficult to accurately define, because the number on summit of insertion and the surface area of patch and parameter alpha Proportional rising, but surface area cannot be defined on hole in the 3 d space well.In practice, good-looking net can be produced Lattice patch, is adapted from different adjacent shapes but for fine setting parameter with for using given input prediction total run time is It is highly desirable to skill.

In another embodiment of the invention, there is provided the filling algorithm of improved hole.In embodiments of the present invention, The appearance in hole comes from three sources.

1) they are the flaws of the initial mesh that MVS step produces.The reflecting surface of the known picture water surface, the window in building etc. Cannot rebuild and leave macropore in surface mesh.

2) in following shearings and merging process, in order to be filled with deliberately to produce hole as the purpose merging.

3) hole filling algorithm produces non-manifold grid sometimes.By removing non-manifold surface, it produces new hole.

This algorithm effective process is passed through because its size is limited in latter two source in hole.However, in initial mesh Flaw brings very big challenge to algorithm, because they have very big input size and cover very big area.Greatly Input size will slow down step 1, due to its O (n³) run time essence, and larger area also will slow down step 2 and Step 3.

The present inventor is tested on the machine with intel i7-4770K CPU, and this CPU has four Kernel operation device simultaneously supports 8 threads and 32GB internal memory.Result of the test is as shown in Figure 12 and Biao 1.

Table 1：The assessment of hole filling algorithm

Form from above can be seen that the run time when inputting size and increasing, and exponentially form gradually rises.In order to Algorithm, in another embodiment of the present invention, can be expanded to by the dynamic programming in decomposition step 1 and prop up by solve problem Hold multithreading.Although run time can successfully be improved several times (several folds) by this, this acceleration nevertheless suffers from machine The restriction of device core number and scale cannot be expanded, and the size of macropore is practically without the upper limit.

And, the inventors found that the default value of parameter alpha (i.e.,) (that defining the degree of refinement) Cannot produce and seem gratifying grid patch.Described inventor passes through repeatedly experiment adjusting parameter and finally sets α's Default value about 1.8.Referring to Figure 13.

Finally, the present inventor has processed the macropore of these types by another diverse method.The method Inspired by the description below：These macropores most of are substantially flat.Initial Triangulation Approach step is typically unnecessary, because It is it is contemplated that it is only plane for minimum error result.Actually nor is it necessary that will iteratively tessellated mesh patch To adapt to neighbouring grid.In another embodiment of the invention, the basic thought of this new method is to build with first The grid patch knowing topological structure and the border vertices that the border vertices on target gridding are mapped to this grid patch.Subsequently It is used border vertices position as constraints, internal vertex to be smoothed.Referring to Figure 14.

Smoothing problasm is represented by sparse linear system (sparse linear system) Ax=B.For example, with reference to figure Example shown in 15.

All of internal vertex is located at the mean place of its adjacent ring and all border vertices are fixed.Due to matrix Sparse essence, linear system can use as Biconjugate gradient method linear solution device (linear solver) big Cause O (n²) rapid solving in the time.The grid patch obtaining is the smooth surface with rough isodensity summit.In order to further Using the shape on macropore border generally simple the fact, sampled by reducing, extract each k-th summit come as input approximate Input border.Set first and input threshold value, such as 500 or 1000, subsequently reduce sampling input vertex, so its equivalent inpnt chi Very little less than threshold value.By this way, the run time of filling macropore can finally be limited.

When hole is larger, almost plane and during convex surface, the method is run optimal.The grid that the speed of service is exceedingly fast and obtains The quality of patch can be matched in excellence or beauty with the quality of the grid patch from primal algorithm, referring to Figure 16,17.

In addition, because the method does not consider any aesthetics as minimizing the angle with the dihedral angle of adjacent triangle Standard, the grid patch that it obtains seems unnatural.The method also creates Self-crossover.Referring to Figure 18.

Fortunately, these macropores generally fall into plane domain, and as the lake in natural phenomena or river, and grid patch can It is thoroughly mixed with adjacently situated surfaces.Therefore, the LOD of embodiment of the present invention is designed to aperture and macropore are processed respectively, i.e. aperture Using primal algorithm, to macropore using the method based on new linear system.In one embodiment of the invention, by macropore Threshold value is set to 600 summits, and achieves balance result well.

One final temporary transient sex chromosome mosaicism is when the grid patch to these types carries out lattice simplified, find another with The relevant problem of this method.Mention, initial mesh is to rebuild from the 3D of Nature condition to obtain, described Nature condition tool before There are natural geometry feature or change, and this method will produce artificial smooth surface.Lattice simplified tend to over-simplification these smooth Region, produces aciculiform and upset triangle.Referring to Figure 19.In order to solve this problem, artificial random noise is introduced final The vertex position being obtained by linear system.Select the amplitude of this noise as the sub-fraction of average edge length to maintain it Impact minimizes and prevents patch over-simplification.

In general, the hole filling algorithm according to embodiment of the present invention, is expanded to support multithreading, is run Time improves several times (several folds).For actually used angle, macropore really occur and primal algorithm property Can be due to its O (n³) essence and start fast decay.Therefore, the present invention implements another algorithm based on linear system with special Process this macropore.Embodiment of the present invention also applies reduction sampling to limit run time.Although this new method is logical With having its limitation in case, but its run in the case of these macropores typically plane and convex surface 3D rebuilds good Good.The mesh quality that run time is improved hundreds times and obtains by the method can be matched in excellence or beauty with primal algorithm.

【Shear and merge】

Due to the essence of dividing and ruling of LOD flow scheme design, grid using shearing and is merged, this is not in other LOD design Common.

<Shearing>

Shearing is quite direct grid operations.In embodiments of the present invention, define secant plane first, then look up All trianglees with this level-crossing.Subsequently, the triangle that those intersect is separated into less triangle along shearing line. Finally, by using their geometric center as test point by all of triangular marker in the either side of plane, and remove Remove the triangle wanting in this side to remove.Exist three kinds of crossing instances, and can according to the number on cross edge and summit to this three Plant crossing instances classification.As follows：

(1) 1 summit 0 side：Need not operate, the +/- side of labelling triangle to plane；

(2) 1 summit 1 side：Triangle is separated into two trianglees, increases a new summit；

(3) 0 summit 2 side：Triangle is separated into three trianglees, increases by two new summits；

Although the triangle of decay is undesirable for many grid operations, it is tolerable in shearing 's.It is fully located at the triangle of the decay of shear plane if there is three summits, can simply directly assign them to arbitrary Side and this does not interfere with result.Otherwise, them are processed as conventional triangle.It is fortunately that shearing manipulation both will not Produce the triangle that non-manifold grid also will not produce decay.Referring to Figure 20.

Wherein, run time is O (n), and wherein, n is face number.

<Merge>

There is the mesh update algorithm of multiple exploitations.One type depends on to surface resampling with from again adopting The surface mesh of merging is extracted in the point cloud of sample.Although they work under general case, they are undesirable, because Resampling can destroy original mesh topological structure and cannot keep enough geometric fidelities, this violates 3D reconstruction Primary objective.Some other methods depend on find merge grid between overlapping region and carefully increase new summit and Face is to connect.However, the method cannot process non-overlapped input grid.

It is true that the LOD design of embodiment of the present invention shows that merging is special circumstances, because merge grid (that is, dividing Area) anticipated rectangular, and the position of known merging and overlapping region need not be detected.Figure 21 shows typical case's input Example.

In embodiments of the present invention, implement new merging algorithm by reseting hole filling algorithm.This merging algorithm has There are three key steps：

1). cut away minor shifts amount (offset) to set up the merging space merging between grid along combined side.

2). pass through to mate two minimum distance border vertices foundation merging merging between grid along their combined side Bridge between grid or connection.

3). applicable holes filling is to complete to merge.

This three step shows in Figure 22 a, 22b, 22c and Figure 23.Wherein, Figure 22 a shows that foundation merges space, figure 22b shows and sets up cross structure, and Figure 22 c shows filling hole, and Figure 23 shows merging algorithm in the form of illustrate only border Final result.

Part the most delicate is step 2.First-selection, needs setpoint distance threshold value to prevent from mating the summit being placed too far from, phase It is typically mistake away from too remote summit.Described threshold value can be by the width of zoom factor and bridge joint space (bridging gap) It is multiplied and set.Secondly, the number across the merging border of two grids may not be 1 to 1 pairing and it may happen that bridge intersects (bridge crossing) problem.This will lead to topological structure or the non-manifold grid of mistake.In fact can detect that bridge intersects to ask Topic, for example, along Boundary Traversal (traverse) and record the order of bridge and check discontinuous order, subsequently can be in step The bridge of those intersections was simply discarded before 3.Referring to Figure 24, left-hand side shows the accurate row from B1 to B5 with from C1 to C5 Sequence；Right-hand side shows bridge cross-cutting issue when in grid Y across border C.It is envisaged that arriving B10 after depositing B9, But replacing it to obtain A4, it is envisaged that arriving A6 after depositing A5, but replacing it to obtain B8.

According to embodiment of the present invention, this merging algorithm has broken away from the requirement of overlap, and overlap is other mesh update algorithms In necessary.Due to establishing merging space in step 1, the final grid merging does not contain the triangle of decay.

【Simplify】

Lattice simplified is core algorithm and many algorithms are suggested in the past used in discrete LOD generates.Some are calculated Method is based on fidelity, and it starts specification error threshold value, subsequently simplifies grid in the range of this limit of error as far as possible. This error is typically measured as the geometric error inputting and simplifying between grid, and it can be global error or local error, and this takes Certainly in practical operation.Some algorithms are based on budget, and it is contrary with the algorithm based on fidelity, replace fixed error Boundary, the method based on budget sets the target numbers of the triangle wanting to realize after simplifying.Subsequently apply a series of offices With simplified model, this minimizes error to portion's grid operations.

Although neither one size matching all of simplification algorithm, QSlim proposes a classical way, and it is by letter Change and be formulated as greedy algorithm (greedy algorithm) to minimize the local geometric in each grid operations step process Error.The contribution of its maximum is that invention uses matrix, thus compactly and easily representing error matrix, described error matrix For calculating the optimal layout on the summit that subsides after opposite vertexes subside.The method has proved to be successful, because it is strong And effectively and for full new class simplification algorithm provide enlightenment.

In embodiments of the present invention, inventor provides a kind of lattice simplified method based on QSlim algorithm, contributes to protecting Deposit and reinforced structure feature.

The general thought of QSlim algorithm is that iteratively to select and destroy opposite vertexes be new summit and by new summit It is placed on optimum position, so, the approximate geometry error minimize of obtained grid.As shown in figure 25, the mesh of each iteration Be make geometric error minimize, described geometric error be defined as from summit to its support plane squared-distance.

Assume the plane that plane p=[a b c d] expression is defined, wherein, a by equation ax+by+cz+d=0²+b²+c²=1 And vertex v=[v₁v₂v₃1] some arbitrfary points in representation space.Subsequently the squared-distance from v to p is expressed as：

squared_dist(v,p)=(p^Tv)²=(v^Tp)(p^TV)=v^TK_pv

Wherein：

$K_p={\mathrm{pp}}^T=\left[\begin{array}{cccc}{a}^2& ab& ac& ad\\ ab& b^2& bc& bd\\ ac& bc& c^2& cd\\ ad& bd& cd& d^2\end{array}\right]$

The squared-distance information of this plane is encoded to 4 × 4 symmetrical matrix data bags by Kp, and it carries 10 uniquenesses altogether Floating numeral.Can be effectively calculated as from the summation of the squared-distance of v to group of plane (planes) of point:

∑_{p∈planes(v)}squared_dist(v,p)=∑_{p∈planes(v)}(v^TK_pV)=v^T(∑_{p∈planes(v)}K_p)^v,

It is the simple summation of matrix K p.In the pre-treatment step of algorithm, by each for each triangle self-corresponding Kp with Each summit Q_vAssociated, Q_vIt is represented simply as quadratic surface, wherein, Q_v=∑_{p∈planes(v)}K_p, wherein, planes (v) is v Support plane set.

Subsequently, calculate each feasible summit to V_i,V_j, its be they subside vertex v ' optimum position, so, Error_Vi,Vj=v '^T(Q_Vi+Q_Vj)^v′It is minimized.V' can be by Error_vi,vjPart derivation simultaneously sets it to 0 and obtains Linear system solve and quickly found, by all errors insertion pile structure (heap structure) and in each iteration Middle extraction minimum error summit subsides (collapse) to execute.After subsiding every time, new vertex v ' will be from Qi Muding Inheriting quadratic surface in point is Q_v=(Q_vi+Q_vj), and update accumulation tree, all it is affected for be connected with this new summit Summit pair.If this v' cannot solve, then simply return to be assigned therein as v_iOr v_jOr their intermediate point.Repeat this Process is until reach the triangle of target numbers.

Obtain two kinds of conclusions accordingly.First, the formula of QSlim algorithm shows that minimization problem is local.The top subsiding The optimum position of point is to be determined by the support planar set of formerly vertex set, from described support plane inherit obtain described in subside Summit.This support planar set grows out and the step that stretches out from the vicinity of plane.Lack the consideration to global characteristics, this makes Obtain QSlim sensitive to local grid noise.Secondly, input grid is always subjected to inevitable effect of noise in MVS step. Even in input grid, the completely flat surface in actual scene can comprise noise.

This produces undesirable result, and the plane domain of grid cannot energetically fully simplify and other valuable richnesses The region having feature has of a relatively high priority and is simplified.This comprises substantial amounts of in 3D large-scale reconstruction and in scene Especially prominent during flat surface.

Enlightened by the above, in embodiments of the present invention, inventor improves quadratic surface formula so that its Be conducive to the simplification in plane domain and the summit in secant plane region is to form the plane belonging to summit, thus smoothing that Noise in flat surface a bit.For doing so, the plane domain first in detection grid and its plane equation.The present invention is implemented Mode is optimized to existing method so as to be mated completely with application scenarios of the present invention, and has obtained the two of following improvement Secondary surface equation：

α ∈ [0,1] determines the significance level of detected plane domain and forces corresponding summit to be cut to its inspection The plane measuring.β ∈ [0,1] determines the simplification priority in all feasible these plane domains of summit centering.At this In the experiment of bright embodiment, setting α as 0.9, β is 0.5.This new quadratic surface initializes and other when algorithm starts Step does not change.Figure 26 and 27 shows the difference between standard QSlim and embodiment of the present invention modified version.Input net Lattice have 1.9M summit and 3.9M triangle.Both is reduced to the 1% of input according to the number of triangle.

Above-mentioned experiment confirms newly designing and showing that QSlim leads to summit more really of embodiment of the present invention proposition Plus be uniformly distributed, and the improved method of embodiment of the present invention can more energetically simplify plane domain, is being rich in feature Region (for example, car, tree, elbow and edge) leaves more details.

Experiment

Previously described improved grid operations used below, enterprising by large-scale Grid dataset on a single machine Row experiment designs come the LOD to assess embodiment of the present invention.

<Input data>

Large data collection is named as Toulouse, and it is the city in the French west and south.This data set passes through 3D reconstructing system Produce.It has been cut into 128 × 64 subregion grid in advance, is made up of 8044 available partitions.It has 407M top altogether Put with 798M face and constitute in memory the size of 31GB altogether.They will be supplied to LOD flow process to produce the 7th grade The grid of LOD grid to the 0th grade.It has 4 cross partitions hole and their size is very big.Figure 28 shows the 5th Data set in level LOD, be not filled by cross partitions hole so that reader to understand that what input data set appears to such general Read.6th grade or the 7th grade too greatly so that it cannot be carried in grid browser.

<Start>

The LOD flow process of embodiment of the present invention is implemented in C++ and is integrated into the part in 3D reconstructing system.Flow process Flowing by python Script controlling.When producing next stage LOD, the 3 steps process for each 2 × 2 grid (that is, merge, fill out Fill hole, simplify) parallel running.In order to fill the hole of cross partitions, produce the 7th grade of LOD after with the addition of extra step with Detect its filling.

The purpose of embodiment of the present invention is to produce the LOD grid that mean size is 200KB in all LOD levels.Cause This, set to simplify in each level from the 7th grade to the 0th grade and designed with compensating quaternary tree than for 0.25.Because input file Mean size be 4MB, so the 7th grade of input is set simplifying than being 0.05.

In embodiments of the present invention, the machine of operation test is equipped with i7-470K CPU, with four process cores and Support 8 threads and 32GB RMA.

<Result>

Total run time of flow process is 21,681 seconds or about 6 hours.Experimental summary is presented herein below.

Table 2:The experimental summary of large data collection

Figure 29 shows the 5th grade of LOD grid of Toulouse, and Figure 30 shows the 4th grade of LOD grid of Toulouse, figure The 31 3rd level LOD grids showing Toulouse, Figure 32 shows the 2nd grade of LOD grid of Toulouse, and Figure 33 shows The 1st grade of LOD grid of Toulouse, Figure 34 shows the 0th grade of LOD grid of Toulouse.

Test result indicate that the triangle do not decayed by all of grid that LOD flow process produces, non-manifold edge and top Point.

In sum, the present invention proposes LOD grid and produces flow process, and it to carry for by the 3D reconstruction technique of rapid growth Growing demand for be robust (robust) and extendible.It optimizes three main grid operations, institute State the challenging grid flaw that grid operations are mainly used in flow process, fixation is produced by aforementioned reconstruction procedures, merge grid And without overlay requirements or do not produce decay triangle, by general simplification algorithm optimization be suitable to City scenarios grid.Large-scale Experiment on data set shows that it has robustness and extensibility.

Through the above description of the embodiments, those skilled in the art can be understood that the present invention can be by The mode of software combined with hardware platform is realizing.Based on such understanding, technical scheme makes tribute to background technology That offers can be embodied in whole or in part in the form of software product, and this computer software product can be stored in storage and be situated between In matter, such as ROM/RAM, magnetic disc, CD etc., including some instructions with so that a computer equipment (can be individual calculus Machine, server, or network equipment etc.) execution each embodiment of the present invention or embodiment some partly described methods.

Skilled person should be appreciated that disclosed above be only embodiments of the present invention, certainly can not be with This limiting the interest field of the present invention, the equivalent variations made according to embodiment of the present invention, still belong to the claims in the present invention institute The scope covering.

Claims (24)

1. process the method for grid in a kind of model generating process in LOD it is characterised in that methods described includes：

Obtain initial mesh；

Shear described initial mesh to obtain multiple LOD subregions；

By mesh update, the LOD subregion suture of current level is obtained the grid merging；

Lattice simplified to the grid application of described merging to obtain the LOD subregion of next level；

Described mesh update is executed until obtaining the grid of final merging to the LOD subregion iteration of next level described.

2. the method for claim 1 it is characterised in that also include to merge grid exist hole carry out hole filling with Adjacent mesh surface is made to be adapted.

3. method as claimed in claim 2 is it is characterised in that the filling of described hole includes the first filling process：

Build the grid patch of the topological structure with known hole；

The border vertices in the hole on target gridding are mapped to the border vertices of this grid patch；

The position that this border vertices are used smooths to internal vertex as constraints.

4. method as claimed in claim 3 is it is characterised in that the filling of described hole also includes the second filling process：

Initial Triangulation Approach is carried out to grid patch；

By increasing internal vertex, trigonometric ratio is refined；

The internal vertex of increase is placed on the position making patch surface be adapted with adjacently situated surfaces.

5. method as claimed in claim 4 includes it is characterised in that carrying out Initial Triangulation Approach to grid patch：

Define two error functions of triangle：W_AreaAnd W_{Dihedral angle-angle}, wherein, W_AreaRefer to the area of triangle, W_{Dihedral angle-angle}Refer to Maximum dihedral angle angle with adjacent triangle；

Definition grid patch error function W is W_AreaThe minimum error of total or W_{Dihedral angle-angle}Maximum minimum error；

To complete Initial Triangulation Approach by making W minimize.

6. method as claimed in claim 4 is it is characterised in that described second filling processes and also includes：

The surface in the hole of smooth new filling,

Minimize the Normal variations between border vertices and the internal vertex of increase.

7. method as claimed in claim 4 is it is characterised in that described first filling is processed for more than or equal to predetermined threshold Hole, described second filling process for less than described predetermined threshold hole.

8. the method for claim 1 is it is characterised in that described shearing includes：

Define secant plane；

All trianglees intersecting with this secant plane are searched in described initial mesh；

The triangle of described intersection is separated into little triangle along shearing line；

Geometric center by using described little triangle is flat in described cut-out by all of little triangular marker as test point The either side in face, and remove unwanted triangle in this side.

9. the method for claim 1 is it is characterised in that described mesh update includes：

Combined side along grid to be combined cuts away predetermined offset to set up the merging space between grid to be combined；

Combined side along grid to be combined is set up and is waited to close by mating two and merging the minimum distance border vertices between grids And the bridge between grid；

Hole filling is carried out to complete to merge to hole between described net to be combined.

10. method as claimed in claim 9 is it is characterised in that described mesh update also includes：

Along bridge described in Boundary Traversal, and record the order of bridge, check discontinuous order,

The bridge being intersected by the sequential search of bridge, and abandon the bridge of described intersection.

11. the method for claim 1 are it is characterised in that carried out using following Quadratic Surface Equations based on QSlim algorithm Described lattice simplified：

Wherein, Q_vIt is expressed as quadratic surface, v represents summit, p represents the support plane on summit, planes (v) represents the Zhi Chiping of v Face is gathered, K_pRepresent that in grid, summit is to the squared-distance matrix supporting plane, α ∈ [0,1] determines detected plane area The significance level in domain, β ∈ [0,1] determines the simplification priority of plane domain, K_{planar region plane(v)}Represent in grid and push up The squared-distance matrix of the extremely affiliated characteristic plane of point.

A kind of 12. 3D method for reconstructing are it is characterised in that include：

Obtain the point cloud of object by various visual angles stereo reconstruction MVS；

Obtain surface mesh from described cloud；

Based on the surface mesh of described acquisition, LOD grid is generated using the method described in claim 1 to 11 any one；

Rebuild using the 3D that the LOD grid generating completes described object.

The system of grid is processed it is characterised in that described system includes in a kind of 13. model generating process in LOD：

Obtain module, for obtaining initial mesh；

Shear module, for shearing described initial mesh to obtain multiple LOD subregions；

Merge module, for the grid obtaining the LOD subregion suture of current level merging by mesh update；

Simplify module, for lattice simplified to the grid application of described merging to obtain the LOD subregion of next level；

Wherein, described mesh update is executed until obtaining the grid of final merging to the LOD subregion iteration of next level described.

14. systems as claimed in claim 13 are it is characterised in that the hole that described system also includes the grid merging is existed is entered Row hole filling so that adjacent mesh surface be adapted filling module.

15. systems as claimed in claim 14 are it is characterised in that described filling module is used for execution the first filling process：

Build the grid patch of the topological structure with known hole；

The border vertices in the hole on target gridding are mapped to the border vertices of this grid patch；

The position that this border vertices are used smooths to internal vertex as constraints.

16. systems as claimed in claim 15 are it is characterised in that described filling module is additionally operable to the second filling process：

Initial Triangulation Approach is carried out to grid patch；

By increasing internal vertex, trigonometric ratio is refined；

The internal vertex of increase is placed on the position making patch surface be adapted with adjacently situated surfaces.

17. systems as claimed in claim 16 include it is characterised in that carrying out Initial Triangulation Approach to grid patch：

Define two error functions of triangle：W_AreaAnd W_{Dihedral angle-angle}, wherein, W_AreaRefer to the area of triangle, W_{Dihedral angle-angle}Refer to Maximum dihedral angle angle with adjacent triangle；

Definition grid patch error function W is W_AreaThe minimum error of total or W_{Dihedral angle-angle}Maximum minimum error；

To complete Initial Triangulation Approach by making W minimize.

18. systems as claimed in claim 16 are it is characterised in that described second filling processes and also includes：

The surface in the hole of smooth new filling,

Minimize the Normal variations between border vertices and the internal vertex of increase.

19. systems as claimed in claim 16 are it is characterised in that described first filling is processed for more than or equal to predetermined threshold The hole of value, described second filling is processed for the hole less than described predetermined threshold.

20. systems as claimed in claim 13 are it is characterised in that described shearing includes：

Define secant plane；

All trianglees intersecting with this secant plane are searched in described initial mesh；

The triangle of described intersection is separated into little triangle along shearing line；

Geometric center by using described little triangle is flat in described cut-out by all of little triangular marker as test point The either side in face, and remove unwanted triangle in this side.

21. systems as claimed in claim 13 are it is characterised in that described mesh update includes：

Combined side along grid to be combined cuts away predetermined offset to set up the merging space between grid to be combined；

Combined side along grid to be combined is set up and is waited to close by mating two and merging the minimum distance border vertices between grids And the bridge between grid；

Hole filling is carried out to complete to merge to hole between described net to be combined.

22. systems as claimed in claim 21 are it is characterised in that described mesh update also includes：

Along bridge described in Boundary Traversal, and record the order of bridge, check discontinuous order,

The bridge being intersected by the sequential search of bridge, and abandon the bridge of described intersection.

23. systems as claimed in claim 13 are it is characterised in that entered using following Quadratic Surface Equations based on QSlim algorithm Row is described lattice simplified：

Wherein, Q_vIt is expressed as quadratic surface, v represents summit, p represents the support plane on summit, planes (v) represents the Zhi Chiping of v Face is gathered, K_pRepresent that in grid, summit is to the squared-distance matrix supporting plane, α ∈ [0,1] determines detected plane area The significance level in domain, β ∈ [0,1] determines the simplification priority of plane domain, K_{planar region plane(v)}Represent in grid and push up The squared-distance matrix of the extremely affiliated characteristic plane of point.

A kind of 24. 3D reconstructing systems are it is characterised in that include：

First acquisition unit, obtains the point cloud of object by various visual angles stereo reconstruction MVS；

Second acquisition unit, obtains surface mesh from described cloud；

Signal generating unit, based on the surface mesh of described acquisition, is generated using the method described in claim 1 to 11 any one LOD grid；

Reconstruction unit, is rebuild using the 3D that the LOD grid generating completes described object.