CN103955939A - Boundary feature point registering method for point cloud splicing in three-dimensional scanning system - Google Patents

Abstract

The invention discloses a boundary feature point registering method for point cloud splicing in a three-dimensional scanning system. The method comprises the following steps that (1) a three-dimensional laser scanner acquires space sampling points on the surfaces of a real object in different view angles; (2) a boundary detecting method of a point cloud gravity center distance feature is used for extracting point cloud boundary feature points in different viewing angles; (3) an improved iterative closest point (ICP) algorithm is used for registering point cloud according to the extracted point cloud boundary feature points; (4) the registering precision is evaluated according to a registering error standard, and whether a registering result meets a registering request or not is verified. According to the boundary feature point registering method for point cloud splicing in the three-dimensional scanning system, boundary feature point extraction is conducted on the point cloud to be registered, the defect that all points in point cloud data need to be traversed to search for the corresponding points in a traditional ICP algorithm is overcome, on the basis that the registering precision is guaranteed, the complexity of the algorithm is effectively lowered, and meanwhile the efficiency of point cloud registering is obviously improved.

Description

3 D scanning system point cloud splicing edge feature point method for registering

Technical field

The invention belongs to the technical field such as reverse-engineering, image processing, be specifically related to a kind of 3 D scanning system point cloud splicing edge feature point method for registering.

Background technology

Cloud data is generally the three-dimensional geometry coordinate that obtains body surface discrete point by surveying instruments such as spatial digitizers, owing to being measured the restriction of the factors such as size, environment and the measuring appliance of object, can only measure a side of object at every turn, therefore, for obtaining the partial data information of testee, need to scan object from multiple different angles.Cloud data unification under different visual angles is to the registration of cloud data to the process in the same coordinate system.Be accompanied by the widespread use of reconstructing three-dimensional model in various fields such as reverse-engineering, industrial detection, Medical Image Processing, historical relic's protections, three dimensional point cloud registration technology also becomes popular and important research topic.

Existing cloud data registration technology is mainly divided into: manual registration, dependence instrument registration and autoregistration, and wherein autoregistration technology has obtained application the most widely, and this technology mainly comprises thick registration and two steps of accuracy registration.Thick registration is the prerequisite that good initial value is provided for accuracy registration, and accuracy registration is the key of commit point cloud registration error, and relevant experts and scholars have done a large amount of correlative studys aspect accuracy registration.Using at present the most widely accuracy registration algorithm is the Proximal Point Algorithm (iterative close point, ICP) of iteration.ICP algorithm is proposed respectively by people such as people and Chen such as Besl the earliest, but its shortcoming mainly contains: the quality of initial point cloud directly has influence on splicing precision, and iteration is consuming time, convergence is slow, registration is subject to noise, be easily absorbed in local optimum etc.For obtaining accuracy registration result, Chinese scholars has been done large quantity research and improvement to classical ICP algorithm.As aspect irregular surface points cloud registration, adopt the improved ICP algorithm of thick registration and rigidity characteristic to carry out registration, realize the accuracy registration of irregular surface points cloud.Although this convergence of algorithm speed and joining quality have all obtained certain raising, do not consider the impact of noise on registration accuracy, robustness is poor in some cases.Can effectively carry out registration to unknown point cloud as the ICP in conjunction with based on curvature unique point improves algorithm, in registration speed, have a clear superiority in.But must be under the guarantee of initial registration, just can make registration be unlikely to tend to the opposite way round.Otherwise, in the time that the rotary shifted and translation dislocation of two point clouds is larger, only use improved ICP accuracy registration can cause registration to be absorbed in local optimum.As give the Revised ICP algorithm of corresponding point weight and introducing M-estimation, on thick registration basis, using Revised ICP algorithm to carry out accuracy registration.This improvement algorithm has solved the efficiency bottle neck of classical ICP algorithm, has effectively rejected the impact of abnormity point on algorithm, has improved accuracy and the reliability of algorithm, but needs on to search efficiency at closest approach further to be improved.

Current, neighbor point (ICP) method of existing iteration is: obtaining by spatial digitizer after the cloud data of different visual angles, use two closest approaches that point is concentrated of the each searching of ICP algorithm, and make the quadratic sum minimum of its Euclidean distance, thereby calculate the rigid body translation between a cloud.This rigid body translation is applied to a cloud and obtains new impact point cloud, if objective function error convergence in given threshold value, termination of iterations, otherwise continue to search closest approach.

As shown in Figure 1, Fig. 1 adopts improvement iterative closest point method to carry out cloud data registration, but this cloud method for registering has the following disadvantages and defect:

(1) to extract the algorithm complex of point cloud boundary characteristics point higher and there is no improved search strategy for this algorithm, causes closest approach not high to search efficiency, thereby affect registration efficiency.

(2) this algorithm must be under the guarantee of initial registration, just can make registration be unlikely to tend to the opposite way round.Otherwise, in the time that the rotary shifted and translation dislocation of two point clouds is larger, only use improved ICP accuracy registration can cause registration to be absorbed in local optimum.

Therefore, expect registration results quickly and accurately, use new point cloud boundary extracting method to obtain edge feature point, in edge feature point, set search by K-D and accelerate to search closest approach pair, can accelerate search speed, further improve registration efficiency.

Summary of the invention

For above shortcomings in prior art, the invention provides a kind of 3 D scanning system point cloud splicing edge feature point method for registering that improves registration speed.

In order to solve the problems of the technologies described above, the present invention has adopted following technical scheme:

Edge feature point method for registering is used in the splicing of 3 D scanning system point cloud, and the method comprises the steps:

(1) profile, with spatial digitizer from different visual angles scanning mock-up, obtains multi-view angle three-dimensional sampling number certificate;

(2), for two three-dimensional sampling number certificates of same mock-up under different visual angles, extract respectively the edge feature point of two three-dimensional sampling number certificates by the boundary detection method of operating point cloud centroidal distance feature;

(3), in the two block boundary unique points of extracting, fixing wherein one be with reference to point set, another piece is target point set, utilize K-D tree method acceleration search to obtain corresponding closest approach pair, and calculate according to unit Quaternion method that to make the rigid body translation of corresponding closest approach to mean distance minimum, this rigid body translation be rotation matrix R and translation matrix T;

(4), utilize rotation matrix R and translation matrix T to carry out coordinate transform to target point set, obtain new target point set, judge new target point set and whether be less than given threshold value with reference to the distance of point set and realize a cloud registration according to stopping criterion for iteration;

(5), carrying out Revised ICP algorithm with this based on a boundary extraction method for cloud centroidal distance feature can carry out a cloud registration, acquisition accuracy registration effect fast.

As a preferred embodiment of the present invention, the boundary detection method of described some cloud centroidal distance feature, concrete steps are as follows:

First the numerical value of representative point cloud elevation information on Z axis is converted into gray-scale value between 0-255 qualitative attribute as a cloud, is designated as H; Secondly find current some P (X by k nearest neighbor search _i, Y _i, Z _i) K arest neighbors, K=16; The barycentric coordinates (X, Y, Z) of the point group of a last calculating K arest neighbors composition, obtain threshold value δ by poor method between maximum kind; Calculation level P, to the Euclidean distance of point group center of gravity, as fruit dot P is greater than threshold value δ to the Euclidean distance of point group center of gravity, thinks that a P is marginal point; Otherwise some P is not marginal point;

Wherein barycentric coordinates (X, Y, the Z) formula of the point group of K arest neighbors composition is as follows:

$X=\underset{i=1}{\overset{i=k}{\mathrm{\Σ}}}{X}_i{H}_i/\underset{i=1}{\overset{i=k}{\mathrm{\Σ}}}{H}_i$

$Y=\underset{i=1}{\overset{i=k}{\mathrm{\Σ}}}{Y}_i{H}_i/\underset{i=1}{\overset{i=k}{\mathrm{\Σ}}}{H}_i$

Z＝H _i

P is as follows to the Euclidean distance formula of point group center of gravity for point:

$\mathrm{Dis}\left(i\right)=\sqrt{{(X_i-X)}^2+{(Y_i-Y)}^2+{(Z_i-Z)}^2}.$

As another kind of preferred version of the present invention, described unit Quaternion method, concrete steps are as follows:

Hypothetical target point set X={x _i| x _i∈ R ³, i=1,2 ... m}, with reference to point set Y={y _j| y _j∈ R ³, j=1,2 ... m}, target point set X is with corresponding one by one with reference to the point in point set Y, and m is cloud data amount;

1) ask respectively target point set X and the center of gravity with reference to point set Y;

${\mathrm{\μ}}_X=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{x}_i,{\mathrm{\μ}}_Y=\frac{1}{m}\underset{j=1}{\overset{m}{\mathrm{\Σ}}}{y}_j$

2) structure covariance matrix;

${\mathrm{\Σ}}_{X,Y}=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\left[(x_i-{\mathrm{\μ}}_X){(y_i-{\mathrm{\μ}}_Y)}^T\right]$

Wherein: (y _i-μ _y) ^tfor matrix (y _i-μ _y) transposition;

3) according to covariance matrix structure 4 × 4 symmetric matrixes;

$\begin{array}{c}Q\left({\mathrm{\Σ}}_{X,Y}\right)\\ =\left(\begin{array}{cc}\mathrm{tr}{\mathrm{\Σ}}_{X,Y}& {\mathrm{\Δ}}^T\\ \mathrm{\Δ}& {\mathrm{\Σ}}_{X,Y}+{{\mathrm{\Σ}}_{X,Y}}^T-\left(\mathrm{tr}{\mathrm{\Σ}}_{X,Y}\right)I_3\end{array}\right)\end{array}$

Wherein: tr Σ _x,Ymatrix Σ _x,Ymark, establish A _i,j=(Σ _x,Y-Σ _x,Y ^t) _i,j, Δ=[A ₂₃, A ₂₁, A ₁₂] ^t, I ₃be 3 × 3 unit matrixs, A _i,jfor (Σ _x,Y-Σ _x,Y ^t) _i,jthe matrix of structure;

4) calculate 4 × 4 symmetric matrix Q (Σ _x,Y) eigenwert and proper vector, eigenvalue of maximum characteristic of correspondence vector is unit quaternion [q ₀, q ₁, q ₂, q ₃] ^t; Rotating vector is q ^r=[q ₀, q ₁, q ₂, q ₃] ^t;

5) calculate rotation matrix

$\begin{array}{c}R\left(q^R\right)=\\ \left(\begin{array}{ccc}{q_0}^2+{q_1}^2-{q_2}^2-{q_3}^2& 2(q_1{q}_2-q_0{q}_3)& 2(q_1{q}_3+q_0{q}_2)\\ 2(q_1{q}_2+q_0{q}_3)& {q_0}^2+{q_2}^2-{q_1}^2-{q_3}^2& 2(q_2{q}_3-q_0{q}_1)\\ 2(q_1{q}_3-q_0{q}_2)& 2(q_2{q}_3+q_0{q}_1)& {q_0}^2+{q_3}^2-{q_1}^2-{q_2}^2\end{array}\right)\end{array};$

6) calculate translation matrix

q ^T＝[q ₄,q ₅,q ₆] ^T＝μ _X-R(q ^R)μ _Y。

As another preferred version of the present invention, it is as follows that described iteration stops Rule of judgment concrete steps:

If new conversion point set be less than given threshold value λ with reference to the European mean distance of point set, stop iteration, otherwise new point set proceeded to iteration as initial value, until meet the requirement of objective function; Wherein objective function is:

$f(R,T)=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\left|\right|m_i-({\mathrm{Rs}}_i+T)\left|\right|}^2$

Wherein: n is cloud data amount, m _ifor reference point clouds subject to registration, R is rotation matrix, and T is translation matrix, S _ifor impact point cloud, i=1,2...n.

The invention has the beneficial effects as follows: need to travel through each for ICP algorithm and put to calculate corresponding point, algorithm calculated amount is larger, the shortcoming that efficiency is very low, the present invention proposes a kind of Revised ICP algorithm of the rim detection based on gravity center characteristics for the accuracy registration of cloud data, the method has not only solved the bottleneck on ICP algorithm search strategy, and has effectively improved registration efficiency on the basis that ensures registration accuracy; Meanwhile, the simplification rate based on Boundary Detection cloud data reaches 2.2431%, has improved more than 25.8% than the registration efficiency of same class methods, shows that the method is more suitable for the cloud data registration that data volume is larger.

Brief description of the drawings

Fig. 1 is the schematic diagram of cloud data registration of the prior art;

Fig. 2 is the schematic diagram of the embodiment mono-of existing boundary detection method and boundary detection method of the present invention contrast;

Fig. 3 is the schematic diagram that classical ICP algorithm, existing improvement ICP method and the present invention improve the embodiment bis-of ICP method contrast;

Fig. 4 is that existing improvement ICP method and the present invention improve the schematic diagram of ICP method for the embodiment tri-of different visual angles point cloud registration contrast;

Fig. 5 is the schematic diagram of existing ICP algorithm and the contrast consuming time of the inventive method registration.

Embodiment

Below in conjunction with the drawings and specific embodiments, the present invention is described in further detail.

Edge feature point method for registering is used in the splicing of 3 D scanning system point cloud, and the method comprises the steps:

(1) profile, with spatial digitizer from different visual angles scanning mock-up, obtains multi-view angle three-dimensional sampling number certificate;

(2), for two three-dimensional sampling number certificates of same mock-up under different visual angles, extract respectively the edge feature point of two three-dimensional sampling number certificates by the boundary detection method of operating point cloud centroidal distance feature;

(3), in the two block boundary unique points of extracting, fixing wherein one be with reference to point set, another piece is target point set, utilize K-D tree method acceleration search to obtain corresponding closest approach pair, and calculate and make the rigid body translation of corresponding closest approach to mean distance minimum according to unit Quaternion method, be rotation matrix R and translation matrix T;

(4), utilize rotation matrix R and translation matrix T to carry out coordinate transform to target point set, obtain new target point set, judge new target point set and whether be less than given threshold value with reference to the distance of point set and realize a cloud registration according to stopping criterion for iteration;

(5), carrying out Revised ICP algorithm with this based on a boundary extraction method for cloud centroidal distance feature can carry out a cloud registration, acquisition accuracy registration effect fast.

Based on a boundary detection method for cloud centroidal distance feature, step is as follows:

1) numerical value of representative point cloud elevation information on Z axis is converted into gray-scale value between 0-255 qualitative attribute as a cloud, is designated as H;

2) find current some P (X by k nearest neighbor search _i, Y _i, Z _i) K arest neighbors, K=16;

3) barycentric coordinates (X, Y, Z) of the point group of a calculating K arest neighbors composition, formula is as follows:

$X=\underset{i=1}{\overset{i=k}{\mathrm{\Σ}}}{X}_i{H}_i/\underset{i=1}{\overset{i=k}{\mathrm{\Σ}}}{H}_i$

$Y=\underset{i=1}{\overset{i=k}{\mathrm{\Σ}}}{Y}_i{H}_i/\underset{i=1}{\overset{i=k}{\mathrm{\Σ}}}{H}_i$

Z＝H _i

Obtain threshold value δ by Otsu (poor method between maximum kind);

4) calculation level P is to the Euclidean distance of point group center of gravity, and formula is as follows:

$\mathrm{Dis}\left(i\right)=\sqrt{{(X_i-X)}^2+{(Y_i-Y)}^2+{(Z_i-Z)}^2}$

5), as fruit dot P is greater than threshold value δ to the Euclidean distance of point group center of gravity, think that a P is marginal point; Otherwise some P is not marginal point.

Application units' Quaternion method carries out rigid body translation, and concrete steps are as follows:

Hypothetical target point set X={x _i| x _i∈ R ³, i=1,2 ... m}, with reference to point set Y={y _j| y _j∈ R ³, j=1,2 ... m}, target point set X is with corresponding one by one with reference to the point in point set Y, and m is cloud data amount;

1) ask respectively target point set X and the center of gravity with reference to point set Y

${\mathrm{\μ}}_X=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{x}_i,{\mathrm{\μ}}_Y=\frac{1}{m}\underset{j=1}{\overset{m}{\mathrm{\Σ}}}{y}_j;$

2) structure covariance matrix;

${\mathrm{\Σ}}_{X,Y}=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\left[(x_i-{\mathrm{\μ}}_X){(y_i-{\mathrm{\μ}}_Y)}^T\right]$

Wherein: (y _i-μ _y) ^tfor matrix (y _i-μ _y) transposition;

3) according to covariance matrix structure 4 × 4 symmetric matrixes;

$\begin{array}{c}Q\left({\mathrm{\Σ}}_{X,Y}\right)\\ =\left(\begin{array}{cc}\mathrm{tr}{\mathrm{\Σ}}_{X,Y}& {\mathrm{\Δ}}^T\\ \mathrm{\Δ}& {\mathrm{\Σ}}_{X,Y}+{{\mathrm{\Σ}}_{X,Y}}^T-\left(\mathrm{tr}{\mathrm{\Σ}}_{X,Y}\right)I_3\end{array}\right)\end{array}$

Wherein tr Σ _x,Ymatrix Σ _x,Ymark, establish A _i,j=(Σ _x,Y-Σ _x,Y ^t) _i,j, Δ=[A ₂₃, A ₂₁, A ₁₂] ^t, I ₃be 3 × 3 unit matrixs, A _i,jfor (Σ _{x, Y-}Σ _x,Y ^t) _i,jthe matrix of structure;

4) calculate 4 × 4 symmetric matrix Q (Σ _x,Y) eigenwert and proper vector, maximum feature characteristic of correspondence vector is unit quaternion [q ₀, q ₁, q ₂, q ₃] ^t; Rotating vector is q ^r=[q ₀, q ₁, q ₂, q ₃] ^t;

5) calculate rotation matrix

$\begin{array}{c}R\left(q^R\right)=\\ \left(\begin{array}{ccc}{q_0}^2+{q_1}^2-{q_2}^2-{q_3}^2& 2(q_1{q}_2-q_0{q}_3)& 2(q_1{q}_3+q_0{q}_2)\\ 2(q_1{q}_2+q_0{q}_3)& {q_0}^2+{q_2}^2-{q_1}^2-{q_3}^2& 2(q_2{q}_3-q_0{q}_1)\\ 2(q_1{q}_3-q_0{q}_2)& 2(q_2{q}_3+q_0{q}_1)& {q_0}^2+{q_3}^2-{q_1}^2-{q_2}^2\end{array}\right)\end{array};$

6) calculate translation matrix

q ^T＝[q ₄,q ₅,q ₆] ^T＝μ _X-R(q ^R)μ _Y。

It is as follows that iteration stops Rule of judgment concrete steps:

If new conversion point set be less than given threshold value λ with reference to the European mean distance of point set, stop iteration, otherwise new point set proceeded to iteration as initial value, until meet the requirement of objective function; Wherein objective function is:

$f(R,T)=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\left|\right|m_i-({\mathrm{Rs}}_i+T)\left|\right|}^2$

Wherein: n is cloud data amount, m _ifor reference point clouds subject to registration, R is rotation matrix, and T is translation matrix, S _ifor impact point cloud, i=1,2...n.

The present invention has adopted new point cloud boundary detection method Revised ICP algorithm.Based on a Boundary Detection of cloud centroidal distance feature as shown in Figure 2, it is as shown in table 1 below that point cloud boundary extracts contrast;

Table 1:

As shown in table 2 below by Fig. 3 being carried out contrast after accuracy registration algorithm;

Table 2:

Method for registering	Registration time/ms	Registration error/mm	Iterations/time
				Classical ICP algorithm	112.1820	0.1544	137
Revised ICP algorithm	80.4597	0.1287	87
				Algorithm herein	51.5570	0.0965	30

As shown in table 3 below by Fig. 4 being carried out to the contrast of different visual angles cloud data registration;

Table 3:

Can find out from Fig. 3, Fig. 4 and Fig. 5 and table 2 and table 3, the present invention carries out cloud data method for registering with edge feature point and has solved the bottleneck on ICP algorithm search strategy, and has effectively improved registration efficiency on the basis that ensures registration accuracy.Meanwhile, the simplification rate based on Boundary Detection cloud data reaches 2.2431%, has improved more than 25.8% than the registration efficiency of same class methods, shows that the improved ICP algorithm of the present invention is more suitable for the cloud data registration that data volume is larger.

Finally explanation is, above embodiment is only unrestricted in order to technical scheme of the present invention to be described, although the present invention is had been described in detail with reference to preferred embodiment, those of ordinary skill in the art is to be understood that, can modify or be equal to replacement technical scheme of the present invention, and not departing from aim and the scope of technical solution of the present invention, it all should be encompassed in the middle of claim scope of the present invention.

Claims (4)

1. edge feature point method for registering is used in the splicing of 3 D scanning system point cloud, it is characterized in that, the method comprises the steps:

(1) profile, with spatial digitizer from different visual angles scanning mock-up, obtains multi-view angle three-dimensional sampling number certificate;

(2), for two three-dimensional sampling number certificates of same mock-up under different visual angles, extract respectively the edge feature point of two three-dimensional sampling number certificates by the boundary detection method of operating point cloud centroidal distance feature;

(3), in the two block boundary unique points of extracting, fixing wherein one be with reference to point set, another piece is target point set, utilize K-D tree method acceleration search to obtain corresponding closest approach pair, and calculate according to unit Quaternion method that to make the rigid body translation of corresponding closest approach to mean distance minimum, this rigid body translation be rotation matrix R and translation matrix T;

(4), utilize rotation matrix R and translation matrix T to carry out coordinate transform to target point set, obtain new target point set, judge new target point set and whether be less than given threshold value with reference to the distance of point set and realize a cloud registration according to stopping criterion for iteration;

(5), carrying out Revised ICP algorithm with this based on a boundary extraction method for cloud centroidal distance feature can carry out a cloud registration, acquisition accuracy registration effect fast.

2. edge feature point method for registering is used in 3 D scanning system point cloud splicing according to claim 1, it is characterized in that, and the boundary detection method of described some cloud centroidal distance feature, concrete steps are as follows:

First the numerical value of representative point cloud elevation information on Z axis is converted into gray-scale value between 0-255 qualitative attribute as a cloud, is designated as H; Secondly find current some P (X by k nearest neighbor search _i, Y _i, Z _i) K arest neighbors, K=16; The barycentric coordinates (X, Y, Z) of the point group of a last calculating K arest neighbors composition, obtain threshold value δ by poor method between maximum kind; Calculation level P, to the Euclidean distance of point group center of gravity, as fruit dot P is greater than threshold value δ to the Euclidean distance of point group center of gravity, thinks that a P is marginal point; Otherwise some P is not marginal point;

Wherein barycentric coordinates (X, Y, the Z) formula of the point group of K arest neighbors composition is as follows:

$X=\underset{i=1}{\overset{i=k}{\mathrm{\Σ}}}{X}_i{H}_i/\underset{i=1}{\overset{i=k}{\mathrm{\Σ}}}{H}_i$

$Y=\underset{i=1}{\overset{i=k}{\mathrm{\Σ}}}{Y}_i{H}_i/\underset{i=1}{\overset{i=k}{\mathrm{\Σ}}}{H}_i$

Z＝H _i

P is as follows to the Euclidean distance formula of point group center of gravity for point:

$\mathrm{Dis}\left(i\right)=\sqrt{{(X_i-X)}^2+{(Y_i-Y)}^2+{(Z_i-Z)}^2}.$

3. edge feature point method for registering is used in 3 D scanning system point cloud splicing according to claim 1, it is characterized in that, and described unit Quaternion method, concrete steps are as follows:

Hypothetical target point set X={x _i| x _i∈ R ³, i=1,2 ... m}, with reference to point set Y={y _j| y _j∈ R ³, j=1,2 ... m}, target point set X is with corresponding one by one with reference to the point in point set Y, and m is cloud data amount;

1) ask respectively target point set X and the center of gravity with reference to point set Y;

${\mathrm{\μ}}_X=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{x}_i,{\mathrm{\μ}}_Y=\frac{1}{m}\underset{j=1}{\overset{m}{\mathrm{\Σ}}}{y}_j$

2) structure covariance matrix;

${\mathrm{\Σ}}_{X,Y}=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}\left[(x_i-{\mathrm{\μ}}_X){(y_i-{\mathrm{\μ}}_Y)}^T\right]$

Wherein: (y _i-μ _y) ^tfor matrix (y _i-μ _y) transposition;

3) according to covariance matrix structure 4 × 4 symmetric matrixes;

$\begin{array}{c}Q\left({\mathrm{\Σ}}_{X,Y}\right)\\ =\left(\begin{array}{cc}\mathrm{tr}{\mathrm{\Σ}}_{X,Y}& {\mathrm{\Δ}}^T\\ \mathrm{\Δ}& {\mathrm{\Σ}}_{X,Y}+{{\mathrm{\Σ}}_{X,Y}}^T-\left(\mathrm{tr}{\mathrm{\Σ}}_{X,Y}\right)I_3\end{array}\right)\end{array}$

Wherein: tr Σ _x,Ymatrix Σ _x,Ymark, establish A _i,j=(Σ _x,Y-Σ _x,Y ^t) _i,j, Δ=[A ₂₃, A ₂₁, A ₁₂] ^t, I ₃be 3 × 3 unit matrixs, A _i,jfor (Σ _x,Y-Σ _x,Y ^t) _i,jthe matrix of structure;

4) calculate 4 × 4 symmetric matrix Q (Σ _x,Y) eigenwert and proper vector, eigenvalue of maximum characteristic of correspondence vector is unit quaternion [q ₀, q ₁, q ₂, q ₃] ^t; Rotating vector is q ^r=[q ₀, q ₁, q ₂, q ₃] ^t;

5) calculate rotation matrix

$\begin{array}{c}R\left(q^R\right)=\\ \left(\begin{array}{ccc}{q_0}^2+{q_1}^2-{q_2}^2-{q_3}^2& 2(q_1{q}_2-q_0{q}_3)& 2(q_1{q}_3+q_0{q}_2)\\ 2(q_1{q}_2+q_0{q}_3)& {q_0}^2+{q_2}^2-{q_1}^2-{q_3}^2& 2(q_2{q}_3-q_0{q}_1)\\ 2(q_1{q}_3-q_0{q}_2)& 2(q_2{q}_3+q_0{q}_1)& {q_0}^2+{q_3}^2-{q_1}^2-{q_2}^2\end{array}\right)\end{array};$

6) calculate translation matrix

q ^T＝[q ₄,q ₅,q ₆] ^T＝μ _X-R(q ^R)μ _Y。

4. edge feature point method for registering is used in 3 D scanning system point cloud splicing according to claim 1, it is characterized in that, it is as follows that described iteration stops Rule of judgment concrete steps:

If new conversion point set be less than given threshold value λ with reference to the European mean distance of point set, stop iteration, otherwise new point set proceeded to iteration as initial value, until meet the requirement of objective function; Wherein objective function is:

$f(R,T)=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\left|\right|m_i-({\mathrm{Rs}}_i+T)\left|\right|}^2$

Wherein: n is cloud data amount, m _ifor reference point clouds subject to registration, R is rotation matrix, and T is translation matrix, S _ifor impact point cloud, i=1,2...n.