CN105261062A - Character segmented modeling method - Google Patents

Abstract

The invention relates to a character model reconstruction method, and discloses a character segmented modeling method. The method includes: building a character head model by employing a three-dimensional scanner to acquire cloud data of character head points; pre-processing the built character head model, and obtaining the pre-processed character head model; simplifying the pre-processed character head model by employing an edge collapse simplification method based on edge curvature and area error, and maintaining detail features of the character head model by employing an interactive method; and performing matching and fusion of the obtained character head model and a human torso model in a human torso database, and finally obtaining a complete human body model. According to the method, the modeling efficiency is improved, the diversity of models is enhanced, and the application value in crowd simulation is high.

Description

A kind of personage's segmentation modeling method

Technical field

The present invention relates to a kind of modeling method, particularly relate to a kind of personage's segmentation modeling method.

Background technology

Computing machine group animation, because have feature that is accurate, intelligent and strong operability, achieves good effect in animation animation, network game making, traffic administration, crowd evacuation and disaster escape etc.How real human body being carried out to three-dimensional modeling and is applied in group animation is the significant challenge that current group Animation Simulating faces.

Traditional Stereo face recognition uses structured light or laser scanner usually, although high-precision personalized three-dimensional manikin can be obtained, and the high and complicated operation of cost.The Kinect device that Microsoft releases utilizes infrared technique, can realize the fast acquiring of three-dimensional information under low cost.This breakthrough, has greatly promoted the application that some use 3-D technology, as human action identification, bone modeling, recognition of face, scene three-dimensional reconstruction etc. based on Kinect, has all become the study hotspot of association area.Kinect depth camera, with its cost feature low and simple to operate, also is often used for building personalized three-dimensional (3 D) manikin real-time by as scanner.

But because the model point cloud density of Kinect scanning is too large, the three-dimensional (3 D) manikin built by it is difficult to be widely used.Such as, in crowd simulation scene, because crowd size is huge, require that human body real-time rendering speed is fast.If directly the three-dimensional (3 D) manikin that Kinect scans is applied in colony's emulation, system overhead can be increased undoubtedly, reduce the efficiency of crowd simulation.Therefore the minutia of how simplified model and how reserving model, and obtain one simple and be a problem highly significant with the model of master mould high similarity.

Summary of the invention

In order to solve the shortcoming of prior art, the invention provides a kind of personage's segmentation modeling method.The method obtains the head model of the realistic personalization of head high precision based on 3D equipment, adopt the edge contraction short-cut method based on limit curvature and area error, and adopt interactive mode to retain the minutia of head model, then select suitable body model and head model automatically to merge, rebuild a complete person model.

For achieving the above object, the present invention is by the following technical solutions:

A kind of personage's segmentation modeling method, comprising:

Step (1): use spatial digitizer to gather the cloud data of personage's head point, build personage's head model;

Step (2): pre-service is carried out to the personage's head model built, obtains pretreated personage's head model;

Step (3): adopt the edge contraction short-cut method based on limit curvature and area error to simplify pretreated personage's head model, and adopt interactive approach to carry out retaining the minutia of personage's head model;

Step (4): the personage's head model obtained step (3) mates with the Human torso in trunk database and merges, and finally obtains complete manikin.

In described step (2), pretreated detailed process is:

Step (2.1): will build the topology map of personage's head model on figure, adopts the fragment in the method removal personage head model of graph theory, obtains the personage's head model removing fragment;

Step (2.2): judge whether there is leak in personage's head model of removal fragment, if exist, the Hierarchical Approach supporting radial basis function is then adopted to repair personage's head model, the smoothing process of personage's head model after adopting Laplace method to repair leak;

If there is not leak, then adopt Laplace method to the smoothing process of personage's head model without leak.

The detailed process of described step (2.1) is:

Step (2.1.1): will the topology map of personage's head model be built on figure, the topological structure of initialization personage head model;

Step (2.1.2): the extreme saturation method traversing graph G adopting figure, retains maximum connected graph G _max;

Step (2.1.3): pass through G _maxreconstruct personage head model, realizes the removal fragment process of personage's head model.

In described step (2.2), the Hierarchical Approach of employing support radial basis function to the process that personage's head model is repaired is:

Random several points of acquisition in personage's head model, and obtain the normal vector corresponding to each point simultaneously, the surface conversion of personage's head model is become implicit surfaces;

The surface vertices set of personage's head model is fitted to a parallelepipedon, then the grouping of surface vertices set and surface vertices set is recursively sub-divided into the large quadrants such as 8, constructs stratification point set;

Adopt different Interpolation-Radix-Functions to carry out multi-level interpolation to the stratification point set different levels built, finally realize the recovery to personage's head model.

The process adopting Laplace method to carry out smoothing processing personage head model in described step (2.2) is:

Adopt Laplace method to be moved by the centre of gravity place on the three-dimensional position on summit in personage's head model summit towards periphery, summit and surrounding vertex gap are minimized;

For each point on personage's head model, according to the positional information of surrounding vertex, the locus recalculating respective point on personage's head model is come personage's head model smoothing.

The process adopting interactive approach to retain the minutia of personage's head model in described step (3) is:

Step (3.1): according to the set on the reserve area limit in personage's head model, calculates the collapse cost on non-reserved limit and the mean value of non-reserved edge contraction cost;

Step (3.2): need the collapse cost on the limit retained to equal the product of the mean value of non-reserved edge contraction cost and the random weight of this mean value, the collapse cost sequence on the limit retained as required, carries out edge contraction operation from small to large.

Described collapse cost equals the area error sum after deleting this non-reserved limit in the limit curvature on the non-reserved limit of the reserve area in personage's head model and reserve area.

The process that described step (4) personage's head model and Human torso merge is:

Step (4.1): the boundary edge according to personage's head model and Human torso carries out determining respective integration region respectively, determines personage's head model integration region F ¹with Human torso integration region F ², and be mapped to two-dimensional space H respectively ¹, H ²;

Step (4.2): by two-dimensional space H ¹, H ²merge, obtain two-dimensional space H ^c; Adopt FCF method to two-dimensional space H ^cin each point be reconstructed blend surface F ^c, obtain the curved surface that personage's head model and Human torso merge;

Step (4.3): automatically merge personage's head model and Human torso according to the curved surface that personage's head model and Human torso merge, construct level manikin.

In described step (1), 3D scanner is Kinect3D scanner.

Adopt FCF method to two-dimensional space H in described step (4.2) ^cin each point be reconstructed blend surface F ^cprocess be:

First, blend surface F is calculated ^cvertex v ^crespectively at F ¹on coordinate v ¹and at F ²on coordinate v ²;

Then, according to v _c=f (s) v ¹+ (1-f (s)) v ², obtain blend surface F ^cvertex v ^ccoordinate, wherein, f (s) represents non-homogeneous three b spline curve interpolation algorithms.

Beneficial effect of the present invention is:

(1) the present invention adopts 3D equipment to obtain the head model of the realistic personalization of head high precision, adopts the edge contraction short-cut method based on limit curvature and area error, and adopts interactive mode to protect the minutia effectively remaining model;

(2) maximum between person model difference is the minutia on head, especially in colony's emulation, detailed information on trunk seems it is not so important, and the most obvious position distinguished colony personage is at head, so this method adopts based on the head of Kinect and the Method Modeling of trunk segmentation, suitable body model and head model is selected to merge, rebuild a complete person model, adopt FCF fusion method, automatic fusion between implementation model, not only can simplify the operation, but also the efficiency of modeling can be improved, increase the diversity of model, in colony's emulation, there is very high using value.

Accompanying drawing explanation

Fig. 1 is personage's segmentation modeling method flow diagram of the present invention;

Fig. 2 is a) remove the personage's head model before fragment;

Fig. 2 b) be remove the personage's head model after fragment;

Fig. 3 is a) original personage's head model;

Fig. 3 b) be repair the personage's head model after leak;

Fig. 4 is a) level and smooth front personage's head model;

Fig. 4 b) be level and smooth after personage's head model;

Fig. 5 is a) limit curvature estimation schematic diagram;

Fig. 5 b) be that area error calculates schematic diagram;

Fig. 6 is a) 20921 points, original personage's head model of 41261 tri patchs;

Fig. 6 b) be simplification rate be 80% personage's head model;

Fig. 6 c) be simplification rate be 90% personage's head model;

Fig. 6 d) be simplification rate be 95% personage's head model;

Fig. 7 is a) the fusion process flow diagram that personage's head model and Human torso merge;

Fig. 7 b) be the FCF function improved;

Fig. 8 is the syncretizing effect figure that personage's head model and Human torso merge.

Embodiment

Below in conjunction with accompanying drawing and embodiment, the present invention will be further described:

As shown in Figure 1, a kind of personage's segmentation modeling method, comprising:

Step (1): use spatial digitizer to gather the cloud data of personage's head point, build personage's head model;

Step (2): pre-service is carried out to the personage's head model built, obtains pretreated personage's head model;

Step (3): adopt the edge contraction short-cut method based on limit curvature and area error to simplify pretreated personage's head model, and adopt interactive approach to carry out retaining the minutia of personage's head model;

Step (4): the personage's head model obtained step (3) mates with the Human torso in trunk database and merges, and finally obtains complete manikin.

Obtain the manikin of different accuracy like this, use these manikins can build stratification human body model data storehouse.Wherein, in step (1), 3D scanner can select Kinect3D scanner or other scanners.Adopt Kinect as depth camera, have simple to operate, easily realize, use the advantages such as flexible, can be good at rebuilding high-precision personage's head model.

For Kinect3D scanner:

Use Kinect3D scanner in step (1), MicrosoftVisualStudio platform adopts C++ to programme, gather head cloud data.

Step (2) is pretreatment operation, and fundamental purpose removes the fragment on personage's head model, and the leak likely occurred when repairing data acquisition, to the smoothing process of personage's head model.

The detailed process of step (2) comprising:

Step (2.1): remove the fragment on personage's head model, by the topology map of personage's head model on figure, adopt the method for graph theory to be removed by fragment unnecessary in model, its process is:

Step (2.1.1): use chart to let others have a look at the topological structure of thing head model, the topological structure G={V of initialization personage head model, F, L}, V represents the point set of personage's head model, and F represents the set in personage's head model face, and L represents the set on personage's head model limit;

Step (2.1.2): the extreme saturation method traversing graph G adopting figure, maximum connected graph is the topological structure needing to retain, and makes maximum connected graph be G _max;

Step (2.1.3): pass through G _maxreconstruct personage head model, realizes the operation of removing fragment, as Fig. 2 a) be remove fragment before personage's head model, Fig. 2 b) be personage's head model after removing fragment.

Step (2.2): judge whether there is leak in personage's head model of removal fragment, if exist, the Hierarchical Approach supporting radial basis function is then adopted to recover personage's head model, revert to implicit surface, reconstruction model, the smoothing process of personage's head model after adopting Laplace method to repair leak;

If there is not leak, then adopt Laplace method to the smoothing process of personage's head model without leak.

In step (2.2), the Hierarchical Approach of use Compactly supported radial basis function to the specific implementation process that model recovers is:

Suppose P={p _ithe N number of scattered points obtained at random in a model, and obtain the normal vector n corresponding to each point simultaneously _i; The Implicitly function that three-dimensional model surface conversion becomes is defined as F (x), wherein zero level integrates the surface of F (x)=0 as grid model, F (x) <0 region representation model is inner, and F (x) >0 region is that model is outside;

To model surface summit p _idetermine a local orthogonal coordinate system (u _i, v _i, w _i), wherein w _ipositive dirction and p _ithe direction of normal vector is identical.Definition Implicitly function g _i(x)=w _i-h (u _i, v _i) approach summit p _ithe basic configuration of small neighbourhood around.

Wherein, h (u, v) ≡ Au ²+ 2Buv+CV ²expression quadric surface approaches summit p _icircumferential shape, the coefficient A in h (u, v), B and C use least square method to obtain, and are obtained by solution formula (1).

$\underset{(u_j,v_j,w_j)=p_j\&Element;P}{\&Sigma;}{\mathrm{\&phi;}}_{\mathrm{\&sigma;}}\left(\right||p_j-p_i|\left|\right){(w_j-h(u_j,v_j\left)\right)}^2\&RightArrow;\min ---\left(1\right)$

In formula (1), φ _σ(r)=(1-r) ⁴+ (4r+1) is Wendaland Compact support RBF, and r represents variable, and σ is for supporting radius.

According to three-dimensional model surface mesh vertex set P, basis function RBFs is utilized to be implicit surfaces F (x) by three-dimensional model surface conversion according to formula (2), wherein, λ _ifor control coefrficient.

$F\left(x\right)=\underset{p_i\&Element;P}{\&Sigma;}{\mathrm{\&psi;}}_i\left(x\right)=\underset{p_i\&Element;P}{\&Sigma;}\&lsqb;g_i\left(x\right)+{\mathrm{\&lambda;}}_i\&rsqb;{\mathrm{\&phi;}}_{\mathrm{\&sigma;}}\left(\right||x-p_i|\left|\right)---\left(2\right)$

Calculate control coefrficient λ _i.By each summit p _ibring formula (2) into and can obtain formula (3)

$F\left(p_j\right)=0=\underset{p_i\&Element;P}{\&Sigma;}\&lsqb;g_i\left(p_j\right)+{\mathrm{\&lambda;}}_i\&rsqb;{\mathrm{\&phi;}}_{\mathrm{\&sigma;}}\left(\right||p_j-p_i|\left|\right)---\left(3\right)$

Can obtain according to formula (3)

$\underset{p_i\&Element;p}{\&Sigma;}{\mathrm{\&lambda;}}_i{\mathrm{\&Phi;}}_{ij}=-\underset{p_i\&Element;p}{\&Sigma;}{g}_i\left(p_j\right){\mathrm{\&Phi;}}_{ij},{\mathrm{\&Phi;}}_{ij}={\mathrm{\&phi;}}_{\mathrm{\&sigma;}}\left(\right||p_j-p_i|\left|\right)---\left(4\right)$

In formula (4), right side is known.Therefore one can be obtained about λ _isparse vectors, utilize process of iteration can solve this sparse vectors, unknown number λ can be obtained _ivalue;

The λ i calculated is brought in formula (2), its implicit surface function can be obtained according to the three-dimensional model surface vertices collection P of input;

Build stratification point set { P ⁿ=P ..., P ², P ¹.First P is fitted to a parallelepipedon, then it and it grouping is recursively sub-divided into large quadrant such as 8 grades.Point set P is polymerized by the parallelepipedon based on Octree; Point inside P is existed for each cell, and is calculated by equalization and normalization, P unit normal is distributed to barycenter;

Multi-level interpolation.Build stratification point set { P ⁿ=P ..., P ², P ¹after, first define Interpolation-Radix-Function, as formula (5) by recursively mode:

f ^k(x)＝-1,k＝0

(5)

f ^k(x)＝f ^k-1(x)+o ^k(x),k＝1,2,...,N

Wherein work as f ^kinterpolation P during (x)=0 ^k, o ^kx () is penalty function, computing formula is (6):

$o^k\left(x\right)=\underset{p_i^k\&Element;P^k}{\&Sigma;}\&lsqb;g_i^k\left(x\right)+c_i^k\&rsqb;{\mathrm{\&phi;}}_{{\mathrm{\&sigma;}}^k}\left(\right||x-p_i^k|\left|\right)---\left(6\right)$

Partial approximation toroidal function g _ix () is by P ^kapplication least square fitting; Weights c _ipass through solve linear equations obtain, use pre-service bi-Conjugate Gradient Method to solve c _i.Wherein support radius sigma ^k+1=σ ^k/ 2, σ ¹=aL, L are the length of border diagonal of a parallelogram, and parameter a makes the octant of bounding box always by a radius sigma ¹ball cover, when finding a=0.75 in repeatedly practice, effect is best.Usual segmentation level n passes through σ ⁰and σ ¹determine, and work as time can obtain reasonable result.If Fig. 3 is a) master pattern, Fig. 3 b) be through the design sketch after Model Reconstruction, clearly repair the leak of model.

Wherein, use Laplace method to the smoothing process of model.

The centre of gravity place on the three-dimensional position on summit summit is towards periphery moved, summit and surrounding vertex gap are minimized.For each point on model, according to the positional information of surrounding vertex, recalculate the locus of this point.Laplce's smoothing formula is as shown in formula (7):

$\stackrel{\&OverBar;}{x_i}=\frac{1}{N}\underset{j=1}{\overset{N}{\&Sigma;}}{x}_j---\left(7\right)$

N is the number of vertices around current point, it is the new coordinate on i-th summit.If Fig. 4 is a) model before smoothing processing, Fig. 4 b) be design sketch after process, clearly achieve the smoothing processing to model by the method.

Wherein, in step (3) limit curvature and the edge contraction short-cut method limit curvature of area error and the computing method of area error as follows:

As Fig. 5 a) shown in, the account form of limit curvature is:

$E_c=\mathrm{\&beta;}(1-cos\mathrm{\&alpha;})=(\frac{l}{h_1}+\frac{l}{h_2})(1-n_1{n}_2)---\left(8\right)$

Wherein l=||q-p||, represents the distance between some q and some p, h _i=d{v _i, e}, represents some v _ito the distance of limit e, β represents a coefficient, n ₁, n ₂presentation surface (v respectively ₁, q, p) and face (v ₂, p, q) normal vector, α represents the angle of two normal vectors.

As Fig. 5 b) shown in, the account form of area error is:

If the angle of triangle rotating is θ, then the error Q produced by deletion face t _tfor:

Q _t＝l _t×θ(9)

Wherein l _t=(A ₁+ A ₂)/2, A ₁for triangle t=(v ₀, v ₁, v ₂) area, A ₂for triangle t'=(v, v ₁, v ₂) area.Make θ=1-n _tn _t', wherein n _tthe normal vector of presentation surface t, n _{t '}the normal vector of presentation surface t '.Then delete the area error E that limit e produces _afor:

$E_s=\underset{t\&Element;P}{\&Sigma;}{Q}_t---\left(10\right)$

Wherein P is and v ₀the set of all tri patchs be connected.

In sum, the computing formula of edge contraction cost is formula (11):

E＝E _c+E _s(11)

Wherein, the process adopting interactive approach to retain the minutia of personage's head model in step (3) is:

Step (3.1): the set making us the reserve area limit in thing head model is P={e _i', wherein i=0,1,2 ..., c, c represent the total number retaining limit; Calculate the collapse cost E on non-reserved limit _jand the mean value of non-reserved edge contraction cost wherein j represents jth bar non-reserved limit; N represents the total number on non-reserved limit;

Step (3.2): order wherein represent the collapse cost needing the limit retained, (rand ()+0.3) represents the random weight of E; According to collapse cost sequence, carry out edge contraction operation from small to large.

If Fig. 6 is a) master pattern, there are 20921 points, 41261 tri patchs.Fig. 6 b), Fig. 6 c), Fig. 6 d) and adopt the short-cut method of the reserving model detail characteristic proposed in the present invention, remain the minutia of eye and nose.Fig. 6 b) there are 3640 points, 7986 tri patchs, simplification rate is 80%; Fig. 6 c) there are 1916 points, 3799 tri patchs, simplification rate is 90%; Fig. 6 d) there are 1014 points, 1995 tri patchs, simplification rate is 95%.By Fig. 6 b) ~ Fig. 6 d) can find out that the method for the reserving model detail characteristic in the present invention effectively remains the minutia of model.

Model Fusion process as Fig. 7 a) shown in, the process that step (4) personage's head model and Human torso merge is:

Step (4.1): the boundary edge according to personage's head model and Human torso carries out determining respective integration region respectively, determines integration region F ¹, F ², and be mapped to two-dimensional space H ¹, H ²;

Step (4.2): by H ¹, H ²merge, obtain H ^c; Adopt FCF method to H ^cin each point be reconstructed blend surface F ^c, obtain the curved surface that personage's head model and Human torso merge;

Step (4.3): automatically merge personage's head model and Human torso according to the curved surface that personage's head model and Human torso merge, construct level manikin.

Wherein, H ^ccomprise two class summits: a class is directly by F ¹, F ²inherit the original summit of coming, another kind of is by the newly-generated point of calculated crosswise.As Fig. 7 b), this method is improved, for the fusion of Controlling model the s in non-homogeneous three b spline curve interpolation algorithm f (s).

F ^cvertex v ^ccoordinate be:

v _c＝f(s)v ¹+(1-f(s))v ²

Wherein, v ¹represent v ^cat F ¹on coordinate, v ²represent v ^cat F ²on coordinate, s=1-l/L, l are for point is to H ^cthe distance of lower boundary, L is the distance of up-and-down boundary.

As shown in Figure 8, the fusion method in use the present invention achieves the automatic fusion between model, can not lose the minutia of model after fusion, and very level and smooth after former boundary edge area merges.Based on the fusion method of boundary edge, the size of the integration region of model is controlled, and the complexity effect of new summit to model produced because of merging is little, negligible, as there being 3740 points before Fig. 8 Model Fusion, merging rear 3934 points, only generating less than 200 points.Fusion method of the present invention can not be too much the complexity of change model, the Size estimation before model still can be made to keep it to merge.

Finally, construct stratification human body model data storehouse and be applied in colony's emulation.The scene of crowd's emergency evacuation in teaching building of 600 people can be simulated.Use LOD method, meticulous manikin (having 43261 tri patchs) is used when view distance is less than 40, when view distance be greater than 40 be less than 100 time use the model (having 3216 tri patchs) of medium accuracy, use coarse model (having 1046 tri patchs) when view distance is greater than 100.The most complicated model is used in scene.The average frame per second wherein using LOD method is 22FPS, and uses the frame per second of most complex model to only have 7.5FPS.The visible hierarchical model storehouse that builds in colony's emulation can the performance of effective elevator system.

The personage's modeling method based on the realistic personalization of Kinect that the present invention proposes can rebuild the individual face object model of high realism, can improve modeling speed again, increases model diversity.Not only have the very high sense of reality by the person model of the method reconstruct in the present invention, and model complexity is low, the LOD person model database using the method in the present invention to build effectively can improve the efficiency of crowd evacuation.

By reference to the accompanying drawings the specific embodiment of the present invention is described although above-mentioned; but not limiting the scope of the invention; one of ordinary skill in the art should be understood that; on the basis of technical scheme of the present invention, those skilled in the art do not need to pay various amendment or distortion that creative work can make still within protection scope of the present invention.

Claims (10)

1. personage's segmentation modeling method, is characterized in that, comprising:

Step (1): use spatial digitizer to gather the cloud data of personage's head point, build personage's head model;

Step (2): pre-service is carried out to the personage's head model built, obtains pretreated personage's head model;

Step (3): adopt the edge contraction short-cut method based on limit curvature and area error to simplify pretreated personage's head model, and adopt interactive approach to carry out retaining the minutia of personage's head model;

Step (4): the personage's head model obtained step (3) mates with the Human torso in trunk database and merges, and finally obtains complete manikin.

2. a kind of personage's segmentation modeling method as claimed in claim 1, it is characterized in that, in described step (2), pretreated detailed process is:

Step (2.1): will build the topology map of personage's head model on figure, adopts the fragment in the method removal personage head model of graph theory, obtains the personage's head model removing fragment;

Step (2.2): judge whether there is leak in personage's head model of removal fragment, if exist, the Hierarchical Approach supporting radial basis function is then adopted to repair personage's head model, the smoothing process of personage's head model after adopting Laplace method to repair leak;

If there is not leak, then adopt Laplace method to the smoothing process of personage's head model without leak.

3. a kind of personage's segmentation modeling method as claimed in claim 2, it is characterized in that, the detailed process of described step (2.1) is:

Step (2.1.1): will the topology map of personage's head model be built on figure, the topological structure of initialization personage head model;

Step (2.1.2): the extreme saturation method traversing graph G adopting figure, retains maximum connected graph G _max;

Step (2.1.3): pass through G _maxreconstruct personage head model, realizes the removal fragment process of personage's head model.

4. a kind of personage's segmentation modeling method as claimed in claim 2, is characterized in that, in described step (2.2), the Hierarchical Approach of employing support radial basis function to the process that personage's head model is repaired is:

Random several points of acquisition in personage's head model, and obtain the normal vector corresponding to each point simultaneously, the surface conversion of personage's head model is become implicit surfaces;

The surface vertices set of personage's head model is fitted to a parallelepipedon, then the grouping of surface vertices set and surface vertices set is recursively sub-divided into the large quadrants such as 8, constructs stratification point set;

Adopt different Interpolation-Radix-Functions to carry out multi-level interpolation to the stratification point set different levels built, finally realize the recovery to personage's head model.

5. a kind of personage's segmentation modeling method as claimed in claim 2, is characterized in that, the process adopting Laplace method to carry out smoothing processing personage head model in described step (2.2) is:

Adopt Laplace method to be moved by the centre of gravity place on the three-dimensional position on summit in personage's head model summit towards periphery, summit and surrounding vertex gap are minimized;

For each point on personage's head model, according to the positional information of surrounding vertex, the locus recalculating respective point on personage's head model is come personage's head model smoothing.

6. a kind of personage's segmentation modeling method as claimed in claim 1, is characterized in that, the process adopting interactive approach to retain the minutia of personage's head model in described step (3) is:

Step (3.1): according to the set on the reserve area limit in personage's head model, calculates the collapse cost on non-reserved limit and the mean value of non-reserved edge contraction cost;

Step (3.2): need the collapse cost on the limit retained to equal the product of the mean value of non-reserved edge contraction cost and the random weight of this mean value, the collapse cost sequence on the limit retained as required, carries out edge contraction operation from small to large.

7. a kind of personage's segmentation modeling method as claimed in claim 6, is characterized in that, described collapse cost equals the area error sum after deleting this non-reserved limit in the limit curvature on the non-reserved limit of the reserve area in personage's head model and reserve area.

8. a kind of personage's segmentation modeling method as claimed in claim 1, it is characterized in that, the process that described step (4) personage's head model and Human torso merge is:

Step (4.1): the boundary edge according to personage's head model and Human torso carries out determining respective integration region respectively, determines personage's head model integration region F ¹with Human torso integration region F ², and be mapped to two-dimensional space H respectively ¹, H ²;

Step (4.2): by two-dimensional space H ¹, H ²merge, obtain two-dimensional space H ^c; Adopt FCF method to two-dimensional space H ^cin each point be reconstructed blend surface F ^c, obtain the curved surface that personage's head model and Human torso merge;

Step (4.3): automatically merge personage's head model and Human torso according to the curved surface that personage's head model and Human torso merge, construct level manikin.

9. a kind of personage's segmentation modeling method as claimed in claim 1, is characterized in that, in described step (1), 3D scanner is Kinect3D scanner.

10. a kind of personage's segmentation modeling method as claimed in claim 8, is characterized in that, adopts FCF method to two-dimensional space H in described step (4.2) ^cin each point be reconstructed blend surface F ^cprocess be:

First, blend surface F is calculated ^cvertex v ^crespectively at F ¹on coordinate v ¹and at F ²on coordinate v ²;

Then, according to v _c=f (s) v ¹+ (1-f (s)) v ², obtain blend surface F ^cvertex v ^ccoordinate, wherein, f (s) represents non-homogeneous three b spline curve interpolation algorithms.