CN106204603A - Three-dimensional camera solid matching method - Google Patents

Abstract

Three-dimensional camera solid matching method, fixes two three-dimensional camera；The public viewing area of two three-dimensional camera is placed object of reference；Two three-dimensional camera gather three dimensional point cloud respectively, are filtered respectively processing to cloud data according to the color of object of reference, it is thus achieved that only comprise the three dimensional point cloud of object of reference, be called cloud data collection X={x_iAnd point set Y={y_i}；On two cloud data collection, obtain the mean place of color region in each point converges and the boundary position of different colours two-by-two according to the new cloud data collection X ' of three somes compositions of equal interval sampling and Y ' according to six kinds of different colours point clouds on object of reference respectively；Spin matrix T is calculated according to new cloud data collection₀, translation matrix R₀；According to spin matrix T₀, translation matrix R₀And the point set X={x of the regions of different colours of object of reference_iAnd point set Y={y_i, calculate accurate spin matrix T, translation matrix R.Present invention greatly reduces and need cloud data amount to be processed, improve speed and the matching precision of coupling.

Description

Three-dimensional camera solid matching method

Technical field

The present invention relates to coupling, technical field of measurement and test, particularly relate to the three-dimensional camera Stereo matching side towards three-dimensional coupling Method.

Background technology

3 D stereo coupling refers to imitate the visual performance of human eye, utilizes two three-dimensional cameras from different perspectives to tested Object shooting image, is analyzed image mating, and calculated the three-dimensional geometric information of this object by principle of triangulation Method.3 D stereo matching system applies to each big necks such as reverse-engineering, quality testing, vehicle guidance the most more and more widely Territory.

The stereo-picture that 3 D stereo matching system is shot by different angles obtains object dimensional geological information.Matching system The twin camera image that obtains of shooting between coupling and the three-dimensionalreconstruction problem that is closely related therewith be 3 D stereo Mating most important is also a most difficult step.

Current matching process mainly uses ICP algorithm, directly according to two cloud data collection to be matched by repeatedly In generation, progressively mates, and obtains rotation and translation change that one of them cloud data collection is done relative to another when coupling Change.Owing to two cloud data collection public territorys to be matched are less, Iterative matching process is by the beginning of spin matrix and translation matrix The impact of beginning value, the convergence of iteration is difficult to ensure that and iterations is the biggest, adds not common region point cloud data set Impact on Iterative matching, matching precision is not the highest.

Document is had to describe quick, the matching algorithm in high precision of some practicalities, before the iteration by artificial treatment to point Cloud data set selects, and only utilizes the cloud data of those artificial selections to mate, thus obtains spin matrix and translation Matrix.But artificial selection's cloud data collection is the most inconvenient, particularly relate to workload during multiple-camera coupling bigger.

It practice, after obtaining three-dimensional colour point clouds data, the band color object of rule can be used completely to provide color Information, it is simple to computer carries out choosing of cloud data collection automatically, utilizes the cloud data collection after this selection to mate, thus Obtain relative transform matrix, build 3 D stereo scene.

Summary of the invention

In order to solve the technical problem of existing existence, the invention provides three-dimensional camera solid matching method, greatly subtract Lack and needed cloud data amount to be processed, improve speed and the matching precision of coupling.

The invention provides three-dimensional camera solid matching method, specifically include:

S1: fix two three-dimensional camera；

S2: place object of reference on the public viewing area of described two three-dimensional camera；

S3: described two three-dimensional camera gather three dimensional point cloud respectively, divide cloud data according to the color of object of reference It is not filtered processing, it is thus achieved that only comprise the three dimensional point cloud of object of reference, be called cloud data collection X={x_iAnd point set Y={y_i}；

S4: on said two cloud data collection, obtains respectively according to six kinds of different colours point clouds on object of reference respectively The mean place (totally 6 points) of the color region in point converges and the boundary position of different colours two-by-two are according to equal interval sampling Three points (totally 18 points) form new cloud data collection X ' and Y ', respectively comprise 24 points, and are matched；According to described newly Cloud data collection calculates spin matrix T₀, translation matrix R₀；

S41: the position of centre of gravity of calculating cloud data collection X ' and Y ' respectively:

$\stackrel{\→}{{\mathrm{\μ}}_x}=\frac{1}{24}\underset{i=1}{\overset{24}{\Σ}}\stackrel{\→}{x_i}$

$\stackrel{\→}{{\mathrm{\μ}}_y}=\frac{1}{24}\underset{i=1}{\overset{24}{\Σ}}\stackrel{\→}{y_i}$

S42: utilize position of centre of gravityWithCalculate the Cross-covariance ∑ of two data sets_xy:

${\mathrm{\Σ}}_{xy}=\frac{1}{24}\underset{i=1}{\overset{24}{\Σ}}\[(\stackrel{\→}{x_i}-\stackrel{\→}{{\mathrm{\μ}}_x}){(\stackrel{\→}{y_i}-\stackrel{\→}{{\mathrm{\μ}}_y})}^T\]=\frac{1}{24}\underset{i=1}{\overset{24}{\Σ}}\[\stackrel{\→}{x_i}{\stackrel{\→}{y_i}}^T\]-\stackrel{\→}{{\mathrm{\μ}}_x}\stackrel{\→}{{\mathrm{\μ}}_y}$

S43: utilize Cross-covariance ∑_xyAntisymmetric matrix structure column vector A_ij=(∑_xy-∑_xy ^T)_ijStructure row are vowed Amount Δ=[A₂₃A₃₁A₁₂]^T, obtain symmetrical matrix Q (∑ according to this column vector_xy)；

$Q\left({\mathrm{\Σ}}_{xy}\right)=\left[\begin{array}{cc}tr\left({\mathrm{\Σ}}_{xy}\right)& {\mathrm{\Δ}}^T\\ \mathrm{\Δ}& {\mathrm{\Σ}}_{xy}+{{\mathrm{\Σ}}_{xy}}^T-tr\left({\mathrm{\Σ}}_{xy}\right)I_3\end{array}\right]$

S44: solve symmetrical matrix Q (∑_xy) eigenvalue of maximum corresponding unit character vector

S45: by unit character vectorObtain spin matrix R₀；

$R_0=\left[\begin{array}{ccc}{q}_0^2+q_1^2-q_2^2-q_3^2& 2(q_1{q}_2-q_0{q}_3)& 2(q_1{q}_3+q_0{q}_2)\\ 2(q_1{q}_2+q_0{q}_3)& q_0^2-q_1^2+q_2^2-q_3^2& 2(q_2{q}_3-q_0{q}_1)\\ 2(q_1{q}_3-q_0{q}_2)& 2(q_2{q}_3+q_0{q}_1)& q_0^2-q_1^2-q_2^2+q_3^2\end{array}\right]$

S46: obtained translation matrix T by spin matrix₀。

$T_0=\stackrel{\→}{{\mathrm{\μ}}_x}-R_0\stackrel{\→}{{\mathrm{\μ}}_y}\stackrel{\mathrm{\Δ}}{=}\[\begin{array}{ccc}{q}_4& q_5& q_6\end{array}\]$

S5: according to described spin matrix T₀, translation matrix R₀And the point set X=of the regions of different colours of described object of reference {x_iAnd point set Y={y_i, calculate accurate spin matrix T, translation matrix R.

S51: accurately before coupling, making initial point set is data set X₀=X ',K=0.

S52: according to currentEuclidean distance is used to calculate point set X_kNearest point set Y with point set Y_k=C (P_k, Y), and According to point set X_kWith point set Y_kCalculateApply new registration vectorObtain sampled point X_kNew match point Y_k+1, will join The square distance of the point after to and D_kEvaluation criterion as precision.

$D_k=\frac{1}{N}\underset{i=1}{\overset{N}{\Σ}}\left|\right|(\stackrel{\→}{x}-\stackrel{\→}{y})\left|\right|$

S53: if k > k_maxOr | D_k-D_k-1|_k< τ, then iteration terminates, otherwise k=k+1 return the 2nd step.

Preferably, described object of reference is heptahedron, and its bottom surface is, has six kinds of regions of different colours.

Preferably, in step s 4, described spin matrix T is calculated by method of least square₀, translation matrix R₀。

Accompanying drawing explanation

Fig. 1 is the object of reference floor map of the present invention；

Fig. 2. the object of reference schematic perspective view of the present invention；

Fig. 3 is the three-dimensional camera stero realizing three-dimensional camera solid matching method of the present invention.

Detailed description of the invention

Technical scheme is further described below in conjunction with detailed description of the invention.Should be appreciated that and be described herein as Detailed description of the invention only in order to explain the present invention, be not intended to limit the present invention.

Fig. 1 is the object of reference floor map of the present invention.As it is shown in figure 1, the three-dimensional camera Stereo matching that the present invention provides Method a, it is desirable to provide known form and the object of reference 1 of color.Fig. 2. the object of reference schematic perspective view of the present invention.Such as Fig. 2 institute Show, this object of reference 1 be ground be regular hexagon, side is the positive seven face cones (regions of different colours in Fig. 2 with 6 trianglees The most specifically draw).These positive seven face cones have six different color regions 11,12,13,14,15,16.

The size changing object of reference 1 about can be placed in the square area of 1 square metre, and its height is not above reference Thing 1 is to the half of the distance of three-dimensional camera.

According to different demands and design, the size of object of reference can arbitrarily change.

Fig. 3 is the three-dimensional camera stero realizing three-dimensional camera solid matching method of the present invention.As it is shown on figure 3, this three Dimension camera stero includes two three-dimensional camera, object of reference 1 in the viewing area that is placed in two three-dimensional camera and two three Information process unit that dimension camera connects and receive the image information of information process unit the display shown.

The three-dimensional camera solid matching method that the present invention provides is specific as follows: first as in figure 2 it is shown, fix in the left and right sides Two three-dimensional camera；Then on the public viewing area of these two three-dimensional camera, place object of reference 1；Secondly described two three-dimensionals Camera gathers three dimensional point cloud respectively, is filtered respectively processing to cloud data according to the color of object of reference, it is thus achieved that only bag Three dimensional point cloud containing object of reference, is called cloud data collection X={x_iAnd point set Y={y_i}；Then at said two point On cloud data set, obtain putting down at the color region each put in converging according to six kinds of different colours point clouds on object of reference respectively The boundary position of equal position (totally 6 points) and two-by-two different colours forms new according to three points of equal interval sampling (totally 18 points) Cloud data collection X ' and Y ', respectively comprises 24 points, and is matched；Spin moment is calculated then according to described new cloud data collection Battle array T₀, translation matrix R₀；

The position of centre of gravity of calculating cloud data collection X ' and Y ' the most respectively:

$\stackrel{\&RightArrow;}{{\mathrm{\&mu;}}_x}=\frac{1}{24}\underset{i=1}{\overset{24}{\&Sigma;}}\stackrel{\&RightArrow;}{x_i}$

$\stackrel{\&RightArrow;}{{\mathrm{\&mu;}}_y}=\frac{1}{24}\underset{i=1}{\overset{24}{\&Sigma;}}\stackrel{\&RightArrow;}{y_i}$

2. utilize position of centre of gravityWithCalculate the Cross-covariance ∑ of two data sets_xy:

${\mathrm{\&Sigma;}}_{xy}=\frac{1}{24}\underset{i=1}{\overset{24}{\&Sigma;}}\&lsqb;(\stackrel{\&RightArrow;}{x_i}-\stackrel{\&RightArrow;}{{\mathrm{\&mu;}}_x}){(\stackrel{\&RightArrow;}{y_i}-\stackrel{\&RightArrow;}{{\mathrm{\&mu;}}_y})}^T\&rsqb;=\frac{1}{24}\underset{i=1}{\overset{24}{\&Sigma;}}\&lsqb;\stackrel{\&RightArrow;}{x_i}{\stackrel{\&RightArrow;}{y_i}}^T\&rsqb;-\stackrel{\&RightArrow;}{{\mathrm{\&mu;}}_x}\stackrel{\&RightArrow;}{{\mathrm{\&mu;}}_y}$

3. utilize Cross-covariance ∑_xyAntisymmetric matrix structure column vector A_ij=(∑_xy-∑_xy ^T)_ijStructure column vector Δ=[A₂₃A₃₁A₁₂]^T, obtain symmetrical matrix Q (∑ according to this column vector_xy)；

$Q\left({\mathrm{\&Sigma;}}_{xy}\right)=\left[\begin{array}{cc}tr\left({\mathrm{\&Sigma;}}_{xy}\right)& {\mathrm{\&Delta;}}^T\\ \mathrm{\&Delta;}& {\mathrm{\&Sigma;}}_{xy}+{{\mathrm{\&Sigma;}}_{xy}}^T-tr\left({\mathrm{\&Sigma;}}_{xy}\right)I_3\end{array}\right]$

4. solve symmetrical matrix Q (∑_xy) eigenvalue of maximum corresponding unit character vector

5. by unit character vectorObtain spin matrix R₀；

$R_0=\left[\begin{array}{ccc}{q}_0^2+q_1^2-q_2^2-q_3^2& 2(q_1{q}_2-q_0{q}_3)& 2(q_1{q}_3+q_0{q}_2)\\ 2(q_1{q}_2+q_0{q}_3)& q_0^2-q_1^2+q_2^2-q_3^2& 2(q_2{q}_3-q_0{q}_1)\\ 2(q_1{q}_3-q_0{q}_2)& 2(q_2{q}_3+q_0{q}_1)& q_0^2-q_1^2-q_2^2+q_3^2\end{array}\right]$

6. obtained translation matrix T by spin matrix₀。

$T_0=\stackrel{\&RightArrow;}{{\mathrm{\&mu;}}_x}-R_0\stackrel{\&RightArrow;}{{\mathrm{\&mu;}}_y}\stackrel{\mathrm{\&Delta;}}{=}\&lsqb;\begin{array}{ccc}{q}_4& q_5& q_6\end{array}\&rsqb;$

Then according to described spin matrix T₀, translation matrix R₀And the point set X=of the regions of different colours of described object of reference {x_iAnd point set Y={y_i, calculate accurate spin matrix T, translation matrix R.

1., before accurately mating, making initial point set is data set X₀=X ',K=0.

2. according to currentEuclidean distance and same color region is used to go to calculate point set X_kClosest approach with point set Y Collection Y_k=C (P_k, Y), and according to point set X_kWith point set Y_kCalculateApply new registration vectorObtain sampled point X_kNew Match point Y_k+1, by square distance and the D of the point after pairing_kEvaluation criterion as precision.

$D_k=\frac{1}{N}\underset{i=1}{\overset{N}{\&Sigma;}}\left|\right|(\stackrel{\&RightArrow;}{x}-\stackrel{\&RightArrow;}{y})\left|\right|$

If 3. k > k_maxOr | D_k-D_k-1|_k< τ, then iteration terminates, otherwise k=k+1 return the 2nd step.

Calculating spin matrix T₀, translation matrix R₀Step in, can calculate with method of least square.

In sum, the invention provides three-dimensional camera solid matching method, considerably reduce and need to be processed some cloud Data volume, improves speed and the matching precision of coupling.

Above embodiment is the preferred embodiment of the present invention, not thereby limits the patent protection model of the present invention Enclose.Those skilled in the art belonging to any present invention, in the premise without departing from spirit and scope disclosed in this invention Under, the equivalent structure being done present disclosure each falls within claimed the scope of the claims with the conversion of equivalent step Within.

Claims (3)

1. three-dimensional camera solid matching method, it is characterised in that

S1: fix two three-dimensional camera；

S2: place object of reference on the public viewing area of described two three-dimensional camera；

S3: described two three-dimensional camera gather three dimensional point cloud respectively, enter cloud data respectively according to the color of object of reference Row Filtering Processing, it is thus achieved that only comprise the three dimensional point cloud of object of reference, is called cloud data collection X={x_iAnd point set Y= {y_i}；

S4: on said two cloud data collection, obtains at respective point according to six kinds of different colours point clouds on object of reference respectively The boundary position of the mean place (totally 6 points) of the color region in converging and two-by-two different colours is according to equal interval sampling three Point (totally 18 points) forms new cloud data collection X ' and Y ', respectively comprises 24 points, and is matched；According to described new some cloud Data set calculates spin matrix T₀, translation matrix R₀, concrete, S41: the position of centre of gravity of calculating cloud data collection X ' and Y ' respectively:

$\stackrel{\&RightArrow;}{{\mathrm{\&mu;}}_x}=\frac{1}{24}\underset{i=1}{\overset{24}{\&Sigma;}}\stackrel{\&RightArrow;}{x_i}$

$\stackrel{\&RightArrow;}{{\mathrm{\&mu;}}_y}=\frac{1}{24}\underset{i=1}{\overset{24}{\&Sigma;}}\stackrel{\&RightArrow;}{y_i}$

S42: utilize position of centre of gravityWithCalculate the Cross-covariance ∑ of two data sets_xy:

${\mathrm{\&Sigma;}}_{xy}=\frac{1}{24}\underset{i=1}{\overset{24}{\&Sigma;}}\&lsqb;(\stackrel{\&RightArrow;}{x_i}-\stackrel{\&RightArrow;}{{\mathrm{\&mu;}}_x}){(\stackrel{\&RightArrow;}{y_i}-\stackrel{\&RightArrow;}{{\mathrm{\&mu;}}_y})}^T\&rsqb;=\frac{1}{24}\underset{i=1}{\overset{24}{\&Sigma;}}\&lsqb;\stackrel{\&RightArrow;}{x_i}{\stackrel{\&RightArrow;}{y_i}}^T\&rsqb;-\stackrel{\&RightArrow;}{{\mathrm{\&mu;}}_x}\stackrel{\&RightArrow;}{{\mathrm{\&mu;}}_y}$

S43: utilize Cross-covariance ∑_xyAntisymmetric matrix structure column vector A_ij=(∑_xy-∑_xy ^T)_ijStructure column vector Δ =[A₂₃ A₃₁ A₁₂]^T, obtain symmetrical matrix Q (∑ according to this column vector_xy)；

$Q\left({\mathrm{\&Sigma;}}_{xy}\right)=\left[\begin{array}{cc}tr\left({\mathrm{\&Sigma;}}_{xy}\right)& {\mathrm{\&Delta;}}^T\\ \mathrm{\&Delta;}& {\mathrm{\&Sigma;}}_{xy}+{{\mathrm{\&Sigma;}}_{xy}}^T-tr\left({\mathrm{\&Sigma;}}_{xy}\right)I_3\end{array}\right]$

S44: solve symmetrical matrix Q (∑_xy) eigenvalue of maximum corresponding unit character vector

S45: by unit character vectorObtain spin matrix R₀；

$R_0=\left[\begin{array}{ccc}{q}_0^2+q_1^2-q_2^2-q_3^2& 2(q_1{q}_2-q_0{q}_3)& 2(q_1{q}_3+q_0{q}_2)\\ 2(q_1{q}_2+q_0{q}_3)& q_0^2-q_1^2+q_2^2-q_3^2& 2(q_2{q}_3-q_0{q}_1)\\ 2(q_1{q}_3-q_0{q}_2)& 2(q_2{q}_3+q_0{q}_1)& q_0^2-q_1^2-q_2^2+q_3^2\end{array}\right]$

S46: obtained translation matrix T by spin matrix₀

$T_0=\stackrel{\&RightArrow;}{{\mathrm{\&mu;}}_x}-R_0\stackrel{\&RightArrow;}{{\mathrm{\&mu;}}_y}\stackrel{\mathrm{\&Delta;}}{=}\left[\begin{array}{ccc}{q}_4& q_5& q_6\end{array}\right].$

S5: according to described spin matrix T₀, translation matrix R₀And the point set X={x of the regions of different colours of described object of reference_i} With point set Y={y_i, calculate accurate spin matrix T, translation matrix R, concrete,

S51: accurately before coupling, making initial point set is data set X₀=X ',K=0；

S52: according to currentEuclidean distance is used to calculate point set X_kNearest point set Y with point set Y_k=C (P_k, Y), and according to point Collection X_kWith point set Y_kCalculateApply new registration vectorObtain sampled point X_kNew match point Y_k+1, after pairing The square distance of point and D_kEvaluation criterion as precision；

$D_k=\frac{1}{N}\underset{i=1}{\overset{N}{\&Sigma;}}\left|\right|(\stackrel{\&RightArrow;}{x}-\stackrel{\&RightArrow;}{y})\left|\right|$

S53: if k > k_maxOr | D_k-D_k-1|_k< τ, then iteration terminates, otherwise k=k+1 return the 2nd step.

Three-dimensional camera solid matching method the most according to claim 1, it is characterised in that described object of reference is heptahedron, Its bottom surface is regular hexagon, has six kinds of regions of different colours.

Three-dimensional camera solid matching method the most according to claim 1, it is characterised in that in step s 4, by minimum Square law calculates described spin matrix T₀, translation matrix R₀。