CN102222361A - Method and system for capturing and reconstructing 3D model - Google Patents

Abstract

The invention provides a method for capturing and reconstructing a dynamic 3D model. The method comprises the following steps of obtaining a static 3D model; converting a surface model of the static 3D model into a volume model to be used as a default scene representation of motion tracking; obtaining an initial 3D motion of top points of the model at the next moment; selecting an accurate top point from the obtained top points according to a preset space-time constraint condition and using the accurate top point as the position constraint of body deformation; and driving a Laplace body deformation framework to update to a dynamic 3D model according to the position constraint. In addition, the invention also provides a system for capturing and reconstructing a 3D model.

Description

The seizure of three-dimensional model and method for reconstructing and system

Technical field

The present invention relates to the computer video processing technology field, particularly a kind of seizure of three-dimensional model and method for reconstructing and system.

Background technology

For the three-dimensional reconstruction problem of dynamic scene, a lot of work are regarded it as static scene and are rebuild problem simply adding up on time dimension, promptly do not utilize temporal information to assist scene rebuilding, and each frame carries out static three-dimensional modeling separately.But this method complexity height, memory space are big, can not guarantee the topological consistance of model between frame and the frame, are easy to generate jitter phenomenon.Adopting said method to carry out three-dimensional modeling in addition can't effectively analyze the motion conditions of non-rigid model, also can't obtain the model of any time by the interpolation on the time domain.By this class problem is studied, prior art has proposed to unite the method for reconstructing of finding the solution 3D scene flows and geometric model.Further also proposed to use variational method to unify the reconstruction of how much of dynamic scenes and motion, yet because geometrical reconstruction and restructure from motion are that iteration is carried out, promptly by certain geometrical reconstruction constantly as the initial value of restructure from motion next Model Reconstruction constantly of deriving, therefore the space-time unite of this method reconstruction efficient is still not high, and actual effect is also also unsatisfactory.

Therefore, rebuild big, the fair problem of difficulty for fear of space-time unite, another kind of dynamic three-dimensional reconstruction method based on video is then represented as scene with the static three-dimensional reconstruction result of initial frame, applying three-dimensional motion tracking algorithm is found the solution the motion of this three-dimensional body then, and use suitable deformation algorithm driving static model motion, thereby obtain the dynamic three-dimensional reconstruction result.At present, the three-dimensional motion tracking based on video can be divided into two classes: the three-dimensional motion of tape label is followed the tracks of and unmarked three-dimensional motion is followed the tracks of.Wherein, the three-dimensional motion tracking of tape label is accurate, but the actor that need be captured is put on the fitted garment that has mark, thereby has limited the seizure to shape and texture.And unmarked three-dimensional motion track side rule has overcome above defective.A kind of unmarked three-dimensional motion tracking is the motion of catching the human body of the general more clothes of dress by Union Movement model and garment form, but this method can't capture the precise geometrical structure of motion object.Another kind of unmarked three-dimensional motion track side rule is the motion of captured object skeleton and shape simultaneously, yet because some local surfaces are not made due change in time, so this method still can't be carried out the three-dimensional motion tracking effectively.In addition, because this method only relies on profile information, so very responsive to profile errors.Though this unmarked method dirigibility has improved, and is difficult to the precision that reaches identical with the tape label method.In addition, most of three-dimensional motion trackings all need to help catch motion by extracting the kinematics skeleton, and the kinematics skeleton can only be followed the tracks of rigid motion, so the shape that this method became when often needing other scanning techniques to come auxiliary catch.At last, all methods all can not be followed the tracks of informal attire dress ornament people's motion more than.

In recent years, in computer graphics, new animation seizure and design, animate during editing and distortion transmission method continue to bring out.These methods are no longer dependent on kinematics skeleton and kinematic parameter, and are based on surface model and general warpage method, thereby can catch the distortion of rigid body and non-rigid body.But, all this motion-captured and restoration methods based on multi-angle video, the static three-dimensional reconstruction of initial frame all needs to adopt laser scanner to carry out.Though laser scanner can obtain the high accuracy three-dimensional reconstructed results, the laser scanner costliness, to waste time and energy, the scan period people must be in complete transfixion state.And for the convenience of follow-up work, the people normally two holds that fist stands, and captured multi-angle video also is that both hands are clenched fist and done action.In addition, the method so that the reconstructed results of laser scanner is represented as initial scene is all keeping the surface characteristics of some on the model when scanning in the whole dynamic 3 D sequence of being recovered always, as the fold of clothes etc.

Summary of the invention

Purpose of the present invention is intended to solve above-mentioned technological deficiency at least, the present invention proposes seizure and the method for reconstructing and the system of a kind of static state and dynamic 3 D model.

For achieving the above object, one aspect of the present invention has proposed a kind of seizure and method for reconstructing of static three-dimensional model, may further comprise the steps: image acquisition is carried out in the moving object in the toroidal field; Obtain the visual shell model; Obtain the depth point cloud at each visual angle according to each visual angle image, described visual shell model and preset restriction condition; The depth point cloud at described each visual angle of obtaining merged obtain static three-dimensional model.

The present invention has also proposed a kind of seizure and reconstructing system of static three-dimensional model on the other hand, comprising: a plurality of video cameras around toroidal field are used for image acquisition is carried out in the moving object of toroidal field; The static three-dimensional model reconstructing device, be used to obtain the visual shell model, and obtain the depth point cloud at each visual angle, and the depth point cloud at described each visual angle of obtaining merged obtain static three-dimensional model according to each visual angle image, described visual shell model and preset restriction condition.

Further aspect of the present invention has also proposed a kind of seizure and method for reconstructing of dynamic 3 D model, may further comprise the steps: obtain static three-dimensional model; The surface model of static three-dimensional model is converted to the phantom type, and its acquiescence scene as motion tracking is represented; Obtain model vertices in next initial three-dimensional motion constantly; According to predetermined when empty constraint condition from the summit that obtains, select the position constraint of accurate summit as body deformability; Drive Laplce's body deformability framework according to described position constraint and upgrade the dynamic 3 D model.

Further aspect of the present invention has also proposed a kind of seizure and reconstructing system of dynamic 3 D model, comprising: a plurality of video cameras around toroidal field are used for image acquisition is carried out in the moving object of toroidal field; The static three-dimensional model deriving means is used to obtain static three-dimensional model; Dynamic 3 D Model Reconstruction device, be used for the surface model of static three-dimensional model is converted to the phantom type, and its acquiescence scene as motion tracking represented, and obtain model vertices in next initial three-dimensional motion constantly, and according to predetermined when empty constraint condition from the summit that obtains, selects the position constraint of accurate summit as body deformability, drive Laplce's body deformability framework renewal dynamic 3 D model according to described position constraint.

Can guarantee the accuracy and the integrality of static three-dimensional model reconstruction shape by the present invention, the new three-dimensional motion method of estimation that the present invention is based on the rarefaction representation Design Theory in addition, and optimize framework based on the distortion of phantom type, therefore can access the dynamic reconstructed results of high-quality.In addition, the present invention can not rely on spatial digitizer and optical markings, so cost is not high, and can follow the tracks of informal attire dress ornament people's motion.

Aspect that the present invention adds and advantage part in the following description provide, and part will become obviously from the following description, or recognize by practice of the present invention.

Description of drawings

Above-mentioned and/or additional aspect of the present invention and advantage are from obviously and easily understanding becoming the description of embodiment below in conjunction with accompanying drawing, wherein:

Fig. 1 is the seizure and the method for reconstructing process flow diagram of the static three-dimensional model of the embodiment of the invention;

Fig. 2 be 20 video cameras of the embodiment of the invention ringwise distribution rings around scene to be collected;

Fig. 3 is the seizure and the method for reconstructing process flow diagram of the dynamic 3 D model of the embodiment of the invention;

Fig. 4 is the schematic block diagram of the whole dynamic three-dimensional reconstruction method of the embodiment of the invention; And

The dynamic 3 D model result of Fig. 5 for adopting the inventive method to obtain to two long time seriess.

Embodiment

Describe embodiments of the invention below in detail, the example of described embodiment is shown in the drawings, and wherein identical from start to finish or similar label is represented identical or similar elements or the element with identical or similar functions.Below by the embodiment that is described with reference to the drawings is exemplary, only is used to explain the present invention, and can not be interpreted as limitation of the present invention.

The embodiment of the invention has proposed the seizure and the method for reconstructing of static three-dimensional model and dynamic 3 D model respectively; can be but need to prove the seizure of dynamic 3 D model and rebuild based on the static three-dimensional model that obtains by the present invention; also can be based on the static three-dimensional model that obtains by other means; existing spatial digitizer etc. for example, these all should be included within protection scope of the present invention.

As shown in Figure 1, seizure and method for reconstructing process flow diagram for the static three-dimensional model of the embodiment of the invention may further comprise the steps:

Step S101 carries out image acquisition to the moving object in the toroidal field.For example, be provided with 20 video cameras in toroidal field, the frame per second of each video camera was 30 frame/seconds, and control is respectively organized video camera the moving object in the toroidal field is gathered.Certainly those skilled in the art also can select more video camera to obtain more visual angle image, also can reduce the quantity of video camera certainly, and these all should be included within protection scope of the present invention.An example of the present invention, as shown in Figure 2, for 20 video cameras of the embodiment of the invention ringwise distribution rings around scene to be collected.Wherein, Ci represents the i video camera.The resolution of camera acquisition image is 1024 * 768.The personage that gathers stands in annular center.

Step S102 obtains the visual shell model (visual hull) of initial time.

Step S103 obtains the depth point cloud at each visual angle according to each visual angle image, described visual shell model and preset restriction condition.Specifically can comprise:

Step S201 asks friendship to obtain the visible point cloud at each visual angle on each visual angle image and the described visual shell model that obtains.

Step S202 projects to this visual angle image with the visible point cloud at each visual angle, obtains the estimation of initial depth point cloud, and promptly d=(a, b, 1) is along the skew to the polar curve direction.

Step S203 estimates and described preset restriction condition obtains accurate depth point cloud according to initial depth point cloud, and wherein, described preset restriction condition comprises intrafascicular approximately one or more of utmost point geometrical constraint, brightness constraint, gradient constraint and flatness.In a preferred embodiment of the invention, can comprise above-mentioned four kinds of constraints simultaneously, obtain accurate depth point cloud by following formula:

$E(a,b)={\∫}_{\mathrm{\Ω}}\mathrm{\β}\left(x\right)\mathrm{\Ψ}({|I(x_b+d)-I\left(x\right)|}^2+\mathrm{\γ}{|\▿I(x_b+d)-\▿I\left(x\right)|}^2)\mathrm{dx}+\mathrm{\α}{\∫}_{\mathrm{\Ω}}\mathrm{\Ψ}({|\▿a|}^2+{|\▿b|}^2)\mathrm{dx},$

Wherein, x:=(x, y, c) defined a location of pixels in the reference viewing angle c image (x, y), its brightness is defined as I (x); x _b:=(x _b, y _b, c) being antipodal points on the c+1 of visual angle, w is the skew of the corresponding point of x on the c+1 of visual angle;

Be the spatial gradient operator; β (x) is an Occlusion Map, is 1 for the pixel in unshielding zone, otherwise is 0.Consider the influence of wild value in the model hypothesis, we adopt the penalty of robust

Produce a total variation regularization, wherein ε is that this formula of a very little value (being made as 0.001 in experiment) has comprised four constraints: to utmost point geometrical constraint (x _b+ d=x+w), brightness constraint (I (x _b+ d)=and I (x)), gradient constraint

Retrain with flatness

Step S104 merges the depth point cloud at described each visual angle of obtaining and to obtain static three-dimensional model.Specifically can may further comprise the steps:

Step S301 removes some wild values with the depth point cloud fusion at each visual angle and by the profile constraint.

Step S302 rebuilds the full surface model by mobile cube method, obtains static three-dimensional model.

Can guarantee that by the present invention static three-dimensional model rebuilds the accuracy and the integrality of shape, the accuracy of static three-dimensional model and integrality are the bases of dynamic 3 D Model Reconstruction.

As shown in Figure 3, seizure and method for reconstructing process flow diagram for the dynamic 3 D model of the embodiment of the invention may further comprise the steps:

Step S401 is converted to the phantom type with the surface model of static three-dimensional model, and its acquiescence scene as motion tracking is represented.

Step S402 obtains model vertices in next initial three-dimensional motion constantly.Particularly, can may further comprise the steps:

Step S501 calculates next light stream of each visual angle image constantly.

Step S502 asks for the scene flows of visible point from each visual angle light stream and the light stream of adjacent visual angle, and composing for the scene flows of invisible point is a big relatively value, and for example 10000.

Step S503, in the hope of each visual angle scene flows for row, structural matrix M ∈ ^{M * n}, wherein m is the surface vertices number.

Step S504 based on the rarefaction representation theory, recovers problem by finding the solution following low-rank matrix, obtains new matrix X.

minimize ‖X‖ _*

subject?to

Wherein, X is a known variables, Ω be [m] * [n] complete element set a subclass ([n] be defined as ordered series of numbers 1 ..., n}),

For sampling operation, be defined as

Step S505 is with the motion of each mean value of going among the matrix X as this row institute corresponding vertex

Thereby obtain next vertex position constantly

Step S403, according to predetermined when empty constraint condition from the summit that obtains, select the position constraint of accurate summit as body deformability.In embodiments of the present invention, described predetermined when empty constraint condition comprise:

$C_{\mathrm{sp}}=\frac{1}{N}\underset{n=0}{\overset{N-1}{\mathrm{\Σ}}}(1-P_{\mathrm{sil}}^n\left(v_i^{\′}\right)),$

$C_{\mathrm{tmp}}=\frac{1}{N_v}\underset{n\∈v\left(i\right)}{\mathrm{\Σ}}(1-P_z^n(p\left(v_i\right),p\left(v_i^{\′}\right))),$

$C_{\mathrm{smth}}=\left|\right|\stackrel{\→}{f}\left(v_i\right)-\frac{1}{N_s}\underset{j\∈N}{\mathrm{\Σ}}\stackrel{\→}{f}\left(v_j\right)\left|\right|.$

Wherein,

Be the profile errors of estimated value, if v ' _iProject to next constantly the pixel on camera n image within profile then this functional value be 1, otherwise be 0; V (i) is v _iThe set of visible camera; N _vIt is visible camera number;

Calculate v _iAnd v ' _iIn the ZNCC correlativity between the projected position on the camera n image; N _sBe vertex v _iThe immediate neighbor number.

Step S404 drives Laplce's body deformability framework according to position constraint and upgrades the dynamic 3 D model.Particularly, comprising:

Step S601 sets up following Laplce's body deformability linear system, for each v ' _i, have

$\underset{j\∈N\left(i\right)}{\mathrm{\Σ}}{\mathrm{\ω}}_{\mathrm{ij}}({v_i}^{\′}-{v_j}^{\′})=\underset{j\∈N\left(i\right)}{\mathrm{\Σ}}\frac{{\mathrm{\ω}}_{\mathrm{ij}}}{2}(R_i+R_j)(v_i-v_j),$

Wherein, R _iAnd R _jBe rotation matrix, and be initialized as unit matrix.

Step S602, the definition covariance matrix

$C_i=\underset{j\∈N\left(i\right)}{\mathrm{\Σ}}{\mathrm{\ω}}_{\mathrm{ij}}(v_i-v_j){(v_i^{\′}-v_j^{\′})}^T=V_i{D}_i{V}_i^{\′T},$

To C _iCarrying out svd has

So

If det is (R _i)≤0 then changes U _iIn corresponding to the symbol of the row of minimum singular value;

Step S603, if profile errors less than given threshold value, new model more then, otherwise return step S601.

As the preferred embodiments of the present invention, the seizure and the method for reconstructing of static three-dimensional model that the present invention is above-mentioned and dynamic 3 D model can use simultaneously, as shown in Figure 4, are the schematic block diagram of the whole dynamic three-dimensional reconstruction method of the embodiment of the invention.

As shown in Figure 5, the dynamic 3 D model result for adopting the proposed invention method to obtain to two long time seriess.Wherein, first width of cloth figure of each sequence results is total figure that each moment model is put together, and the picture of back is respectively each modeling result constantly.

The embodiment of the invention has also proposed a kind of seizure and reconstructing system of static three-dimensional model, and this system comprises: around a plurality of video cameras and the static three-dimensional model reconstructing device of toroidal field.Wherein, a plurality of video cameras around toroidal field are used for image acquisition is carried out in the moving object of toroidal field; The static three-dimensional model reconstructing device is used to obtain the visual shell model, and obtain the depth point cloud at each visual angle, and the depth point cloud at described each visual angle of obtaining merged obtain static three-dimensional model according to each visual angle image, described visual shell model and preset restriction condition.Wherein, the concrete course of work of static three-dimensional model reconstructing device can not repeat them here with reference to the seizure of above static three-dimensional model and the embodiment of method for reconstructing.

In addition, the embodiment of the invention has also proposed a kind of seizure and reconstructing system of dynamic 3 D model, comprising: around a plurality of video cameras, static three-dimensional model deriving means and the dynamic 3 D Model Reconstruction device of toroidal field.Wherein, a plurality of video cameras around toroidal field are used for image acquisition is carried out in the moving object of toroidal field; The static three-dimensional model deriving means is used to obtain static three-dimensional model; Dynamic 3 D Model Reconstruction device, be used for the surface model of static three-dimensional model is converted to the phantom type, and its acquiescence scene as motion tracking represented, and obtain model vertices in next initial three-dimensional motion constantly, and according to predetermined when empty constraint condition from the summit that obtains, selects the position constraint of accurate summit as body deformability, drive Laplce's body deformability framework renewal dynamic 3 D model according to described position constraint.Wherein, the concrete course of work static and dynamic 3 D Model Reconstruction device can not repeat them here with reference to the seizure of above static state and dynamic 3 D model and the embodiment of method for reconstructing.

Can guarantee the accuracy and the integrality of static three-dimensional model reconstruction shape by the present invention, the new three-dimensional motion method of estimation that the present invention is based on the rarefaction representation Design Theory in addition, and optimize framework based on the distortion of phantom type, therefore can access the dynamic reconstructed results of high-quality.In addition, the present invention can not rely on spatial digitizer and optical markings, so cost is not high, and can follow the tracks of informal attire dress ornament people's motion.

Although illustrated and described embodiments of the invention, for the ordinary skill in the art, be appreciated that without departing from the principles and spirit of the present invention and can carry out multiple variation, modification, replacement and modification that scope of the present invention is by claims and be equal to and limit to these embodiment.

Claims (15)

1. the seizure of a dynamic 3 D model and method for reconstructing is characterized in that, may further comprise the steps:

Obtain static three-dimensional model;

The surface model of static three-dimensional model is converted to the phantom type, and its acquiescence scene as motion tracking is represented;

Obtain model vertices in next initial three-dimensional motion constantly;

According to predetermined when empty constraint condition from the summit that obtains, select the position constraint of accurate summit as body deformability;

Drive Laplce's body deformability framework according to described position constraint and upgrade the dynamic 3 D model.

2. the seizure of dynamic 3 D model as claimed in claim 1 and method for reconstructing is characterized in that, described static three-dimensional model is measured and preserved, and perhaps described static three-dimensional model is that the scene obtains,

Wherein, described scene obtains static three-dimensional model and comprises:

Image acquisition is carried out in moving object in the toroidal field;

Obtain the visual shell model;

Obtain the depth point cloud at each visual angle according to each visual angle image, described visual shell model and preset restriction condition;

The depth point cloud at described each visual angle of obtaining merged obtain static three-dimensional model.

3. the seizure of dynamic 3 D model as claimed in claim 2 and method for reconstructing is characterized in that, the described depth point cloud that obtains each visual angle according to visual shell model and preset restriction condition comprises:

Ask friendship to obtain the visible point cloud at each visual angle on each visual angle image and the described visual shell model that obtains;

The visible point cloud at each visual angle is projected to this visual angle image, obtain initial depth point cloud and estimate;

Obtain accurate depth point cloud according to described initial depth point cloud estimation and described preset restriction condition.

4. the seizure of dynamic 3 D model as claimed in claim 3 and method for reconstructing is characterized in that, described preset restriction condition comprises intrafascicular approximately one or more of utmost point geometrical constraint, brightness constraint, gradient constraint and flatness.

5. the seizure of dynamic 3 D model as claimed in claim 4 and method for reconstructing is characterized in that, obtain accurate depth point cloud by following formula:

$E(a,b)={\∫}_{\mathrm{\Ω}}\mathrm{\β}\left(x\right)\mathrm{\Ψ}({|I(x_b+d)-I\left(x\right)|}^2+\mathrm{\γ}{|\▿I(x_b+d)-\▿I\left(x\right)|}^2)\mathrm{dx}+\mathrm{\α}{\∫}_{\mathrm{\Ω}}\mathrm{\Ψ}({|\▿a|}^2+{|\▿b|}^2)\mathrm{dx},$

Wherein, d=(a, b, 1) is that initial depth point cloud estimates, x:=(x, y, c) be in the reference viewing angle c image a location of pixels (x, y), I (x) is the brightness of this location of pixels; x _b:=(x _b, y _b, c) being antipodal points on the c+1 of visual angle, w is the skew of the corresponding point of x on the c+1 of visual angle;

Be the spatial gradient operator; β (x) is an Occlusion Map,

Penalty for robust.

6. the seizure of dynamic 3 D model as claimed in claim 2 and method for reconstructing is characterized in that, described depth point cloud to described each visual angle of obtaining merges and obtains static three-dimensional model and comprise:

Remove wild value with the depth point cloud fusion at described each visual angle and by the profile constraint;

Rebuild the full surface model by mobile cube method, obtain static three-dimensional model.

7. the seizure of dynamic 3 D model as claimed in claim 1 and method for reconstructing is characterized in that, the described model vertices of obtaining comprises in next initial three-dimensional motion constantly:

Calculate next light stream of each visual angle image constantly;

Ask for the scene flows of visible point from each visual angle light stream and the light stream of adjacent visual angle, composing for the scene flows of invisible point is a big relatively value;

In the hope of each visual angle scene flows for row, structural matrix M ∈ ^{M * n}, wherein m is the surface vertices number;

Based on the theoretical matrix X that obtains of rarefaction representation;

With the motion of each mean value of going among the matrix X as this row institute corresponding vertex Thereby obtain next vertex position constantly

8. the seizure of dynamic 3 D model as claimed in claim 7 and method for reconstructing is characterized in that, describedly comprise based on the theoretical matrix X that obtains of rarefaction representation:

Obtain new matrix X by finding the solution following low-rank matrix recovery problem,

minimize ‖X‖ _*

subject?to

Wherein, X is a known variables, Ω be [m] * [n] complete element set a subclass ([n] be defined as ordered series of numbers 1 ..., n}),

For sampling operation, be defined as

9. the seizure of dynamic 3 D model as claimed in claim 1 and method for reconstructing is characterized in that, described predetermined when empty constraint condition comprise:

$C_{\mathrm{sp}}=\frac{1}{N}\underset{n=0}{\overset{N-1}{\mathrm{\Σ}}}(1-P_{\mathrm{sil}}^n\left(v_i^{\′}\right)),$

$C_{\mathrm{tmp}}=\frac{1}{N_v}\underset{n\∈v\left(i\right)}{\mathrm{\Σ}}(1-P_z^n(p\left(v_i\right),p\left(v_i^{\′}\right))),$

$C_{\mathrm{smth}}=\left|\right|\stackrel{\→}{f}\left(v_i\right)-\frac{1}{N_s}\underset{j\∈N}{\mathrm{\Σ}}\stackrel{\→}{f}\left(v_j\right)\left|\right|.$

Wherein,

Be the profile errors of estimated value, if v ' _iProject to next constantly the pixel on camera n image within profile then this functional value be 1, otherwise be 0; V (i) is v _iThe set of visible camera; N _vIt is visible camera number;

Calculate v _iAnd v ' _iIn the ZNCC correlativity between the projected position on the camera n image; N _sBe vertex v _iThe immediate neighbor number.

10. the seizure of dynamic 3 D model as claimed in claim 1 and method for reconstructing is characterized in that, describedly drive Laplce's body deformability framework according to position constraint and upgrade the dynamic 3 D model and comprise:

The initialization rotation matrix is a unit matrix, R _i=R _j=I;

Utilize Laplce's body deformability to be optimized;

Obtain new rotation matrix R _iAnd R _j

Whether judge profile errors less than predetermined value, if less than described predetermined value then upgrade the dynamic 3 D model, if be not less than described predetermined value then continue to utilize Laplce's body deformability to be optimized.

11. the seizure of a dynamic 3 D model and reconstructing system is characterized in that, comprising:

A plurality of video cameras around toroidal field are used for image acquisition is carried out in the moving object of toroidal field;

The static three-dimensional model deriving means is used to obtain static three-dimensional model;

Dynamic 3 D Model Reconstruction device, be used for the surface model of static three-dimensional model is converted to the phantom type, and its acquiescence scene as motion tracking represented, and obtain model vertices in next initial three-dimensional motion constantly, and according to predetermined when empty constraint condition from the summit that obtains, selects the position constraint of accurate summit as body deformability, drive Laplce's body deformability framework renewal dynamic 3 D model according to described position constraint.

12. the seizure of dynamic 3 D model as claimed in claim 11 and reconstructing system is characterized in that, described static three-dimensional model deriving means obtains static three-dimensional model and comprises:

Image acquisition is carried out in moving object in the toroidal field;

Obtain the visual shell model;

Obtain the depth point cloud at each visual angle according to each visual angle image, described visual shell model and preset restriction condition;

The depth point cloud at described each visual angle of obtaining merged obtain static three-dimensional model.

13. the seizure of dynamic 3 D model as claimed in claim 12 and reconstructing system is characterized in that, described preset restriction condition comprises intrafascicular approximately one or more of utmost point geometrical constraint, brightness constraint, gradient constraint and flatness.

14. the seizure of dynamic 3 D model as claimed in claim 13 and reconstructing system is characterized in that, obtain accurate depth point cloud by following formula:

$E(a,b)={\∫}_{\mathrm{\Ω}}\mathrm{\β}\left(x\right)\mathrm{\Ψ}({|I(x_b+d)-I\left(x\right)|}^2+\mathrm{\γ}{|\▿I(x_b+d)-\▿I\left(x\right)|}^2)\mathrm{dx}+\mathrm{\α}{\∫}_{\mathrm{\Ω}}\mathrm{\Ψ}({|\▿a|}^2+{|\▿b|}^2)\mathrm{dx},$

Wherein, d=(a, b, 1) is that initial depth point cloud estimates, x:=(x, y, c) be in the reference viewing angle c image a location of pixels (x, y), I (x) is the brightness of this location of pixels; x _b:=(x _b, y _b, c) being antipodal points on the c+1 of visual angle, w is the skew of the corresponding point of x on the c+1 of visual angle;

Be the spatial gradient operator; β (x) is an Occlusion Map,

Penalty for robust.

15. the seizure of dynamic 3 D model as claimed in claim 11 and reconstructing system is characterized in that, described predetermined when empty constraint condition comprise:

$C_{\mathrm{sp}}=\frac{1}{N}\underset{n=0}{\overset{N-1}{\mathrm{\Σ}}}(1-P_{\mathrm{sil}}^n\left(v_i^{\′}\right)),$

$C_{\mathrm{tmp}}=\frac{1}{N_v}\underset{n\∈v\left(i\right)}{\mathrm{\Σ}}(1-P_z^n(p\left(v_i\right),p\left(v_i^{\′}\right))),$

$C_{\mathrm{smth}}=\left|\right|\stackrel{\→}{f}\left(v_i\right)-\frac{1}{N_s}\underset{j\∈N}{\mathrm{\Σ}}\stackrel{\→}{f}\left(v_j\right)\left|\right|.$

Wherein,

Be the profile errors of estimated value, if v ' _iProject to next constantly the pixel on camera n image within profile then this functional value be 1, otherwise be 0; V (i) is v _iThe set of visible camera; N _vIt is visible camera number;

Calculate v _iAnd v ' _iIn the ZNCC correlativity between the projected position on the camera n image; N _sBe vertex v _iThe immediate neighbor number.

Images