CN101246595A - Multi-view point data splitting method of optical three-dimensional scanning system - Google Patents

Abstract

The present invention provides a multiple-view cloud data assembling method in the optical three dimensional scanning system. The method is realized under the supporting of the optical scanning image system, including following steps: (1) adopting patch which center is white solid circle and the surrounding is black as reference unit, (2) sticking at least three reference units on the scanning object, (3) directly extracting information, acquiring data of reference unit center; (4) converting the acquired reference unit center data to the centroid position information of 3D reference unit, (5) obtaining the distance matrix which the number is identical to the digital camera by the help of the analysis method, (6) determining corresponding common reference point of the two points cloud set, (7) adopting at least three pairs corresponding coordinate parameter of the corresponding common reference point to find rotary matrix R and translation matrix T of the two coordinates conversion, according to the conversion formula: P=R*P'+T directly completing the changes of the point-cloud coordinate data, completing the assemble of the two point-cloud data. The invention has a simple and accurate data assemble.

Description

Multi-view point data splitting method in the optical three-dimensional scanning system

Technical field

This method belongs to the computer digital image treatment technology, multi-view point data splitting method in a kind of specifically optical three-dimensional scanning system.

Background technology

3 D scanning system has obtained updating and developing in actual applications, can be by obtain to reproduce the intactly cloud data collection of object images by means of the direct shooting for sweep object.But be difficult to just can obtain by one-shot measurement the cloud data of objects intact when large piece is surveyed and drawn, a camera often can't scan panorama, often needs to change different angles and measures; Or adopt plural at least digital camera to carry out the cloud data collection that can reflect the complete object looks that scanning survey just can obtain simultaneously.Each or can only obtain the related data of the object in certain visual angle at every turn, only obtain complete cloud data, need carry out amalgamation to the cloud data of different visual angles.

At present existing combination method mainly contains: be that this prerequisite of curvature of curved surface is carried out amalgamation based on the curved surface topological characteristic 1), promptly search for the higher point of cloud data mean curvature, utilize these characteristic informations to come amalgamation partial points cloud as reference point.This just makes that those curved surfaces are level and smooth, and surface characteristics information is not obvious, does not have the cloud data of higher curvature to be difficult to obtain amalgamation effect preferably.And the curvature of finding the solution each point in the cloud data is one, and neighborhood is counted chooses the direct curvature result of calculation that influences this point than time-consuming job, and uncertainty is a lot; 2. the method for man-machine interaction.More typical program step is at first artificial selected some to corresponding point, finishes whole amalgamation earlier, and just thick amalgamation work utilizes the ICP algorithm to carry out interative computation to being total to each concentrated element of public cloud data then, makes the amalgamation result reach best.The shortcoming of this method is to the having relatively high expectations of operating personnel, and requires selected corresponding point accurate, and the ICP algorithm is responsive to initial amalgamation result, if thick amalgamation deviation is bigger, can bring bigger calculated amount for accurate amalgamation.

Summary of the invention

The objective of the invention is at the difference that occurs in many viewpoints scanning imaging system cloud data independently, find a method that realizes the data amalgamation quickly and accurately, realization is carried out standard to the three dimensional point cloud that obtains under the different visual angles, seek a precision height, fast, the widely applicable method of speed coaches, to realize that the cloud data of different visual angles is carried out amalgamation.Design basis of the present invention is carried out based on the 3 D scanning system at image in the process of cloud data extraction, and what drawn is the coordinate data of being established by online reference system coordinate information.Carry out the process of point data splitting, be actually the process of two reference system coordinate transforms.Therefore, method of the present invention is exactly to find out the operating process of the undetermined coefficient of two coordinate system transformation formula.

This method is at computing machine, has the supporting storer of scanning imaging optical system management software and have under the support of at least two digital cameras and realize finishing, and key is that this method may further comprise the steps:

1. the paster that adopts middle white filled circles and circle outer black closed region formation is as the reference unit,

2. in the scope that the correlated digital video camera that is in different points of view all can scan simultaneously, on institute's scanned objects, paste at least three reference units,

3. the system by means of on-line operation carries out the direct information extraction to the reference unit of pasting on the scanned objects, obtain the data message in the reference unit center of circle, after adopting of black, the white transition boundary extraction of area identification method again to reference point, the method of utilizing least square to combine is obtained the home position data of reference unit

4. with the position of form center data of the reference unit of above acquisition, utilize theory on computer vision, carry out the solid coupling, the reference unit center of circle data-switching that is about to the plane becomes the position of form center information of three-dimensional reference unit,

5. according to the position of form center information of reference unit, the data message of each point during the scanning mapping of reference number video camera draws and a little converges, obtain the same number of distance matrix of digital camera that the distance parameter of reference point selected in a little converging and the position of form center of reference unit forms and adopted by analytical method

6. find out their maximum isomorphism subclass by contrasting in two matrixes element value, thereby find out reference point, be defined as the corresponding common reference point of two points in converging with identical topological characteristic,

7. adopt the respective coordinates parameter of at least three pairs of corresponding common reference point to find out rotation matrix R and the translation matrix T of changing about two coordinate systems, again according to conversion formula: P=R*P '+T directly finishes a transformation of cloud coordinate data, finishes two some cloud numbers

This method is actually by means of the reference paster that designs, the large-sized object that is scanned mapping artificially is provided with boundary and sign, by the paster information processing that the cloud data that twice different visual angles collected is concentrated, enough public corresponding point have been found with reference significance.All corresponding point are at same position but have diverse coordinate information, find out that the undetermined coefficient in the conversion formula provides necessary condition between two coordinate system correspondence position points thereby give.Each point that different coordinates is put in the cloud can be finished conversion easily according to formula, just easily finished data amalgamation task exactly.Under the assistance of existing system and management software, become so simple and accurately in the prior art under the support of the complicated and difficult guidance that is operated in this method and computer system, this good effect of the present invention just.

Description of drawings

Fig. 1 is to be the paster of reference unit with the circle.

Fig. 2 is to be the paster of reference unit with the rectangle.

Fig. 3 is the mark merger process flow diagram of equal value in the processing procedure.

Embodiment

Multi-view point data splitting method in the optical three-dimensional scanning system, this method is at computing machine, has the supporting storer of scanning imaging optical system management software and have under the support of at least two digital cameras and realize finishing, and this method may further comprise the steps:

1. the paster that adopts middle white filled circles and the outer black of circle closed region formation is as the reference unit;

2. in the scope that the correlated digital video camera that is in different points of view all can scan simultaneously, on institute's scanned objects, paste at least three reference units;

3. the system by means of on-line operation carries out the direct information extraction to the reference unit of pasting on the scanned objects, obtain the data message in the reference unit center of circle, after adopting black, the white transition boundary extraction of area identification method to reference point again, the method for utilizing least square to combine is obtained the home position data of reference unit;

4. with the position of form center data of the reference unit of above acquisition, utilize theory on computer vision, carry out the solid coupling, the reference unit center of circle data-switching that is about to the plane becomes the position of form center information of three-dimensional reference unit;

5. according to the position of form center information of reference unit, the data message of each point during the scanning of reference number video camera mapping draws and a little converges is obtained the distance parameter formation of position of form center of the reference point selected in a little converging and reference unit and the same number of distance matrix of digital camera that is adopted by analytical method;

6. find out their maximum isomorphism subclass by contrasting in two matrixes element value, thereby find out reference point, be defined as the corresponding common reference point of two points in converging with identical topological characteristic;

7. adopt the respective coordinates parameter of at least three pairs of corresponding common reference point to find out rotation matrix R and the translation matrix T of changing about two coordinate systems, again according to conversion formula: P=R*P '+T directly finishes a transformation of cloud coordinate data, finishes the amalgamation to two cloud datas.

The reference unit that step is used in 1. during concrete enforcement is to be the black paster of white filled circles in the middle of circle or the rectangle.

More than said step 2. in said reference unit evenly be attached to two digital cameras and can scan the public domain.

3. above step carries out the direct information extraction by means of the system of on-line operation to the reference unit of pasting on the scanned objects, it is the position of form center that the method that adopts area identification to combine with least square is obtained reference point, at first adopt the SEQUENTIAL ALGORITHM scan image, the object under test image is done area identification and of equal value to dividing, to belong to of a sort gauge point merger, and give identical mark of equal value; Utilize the frontier point recognition methods to obtain the border of reference point then, promptly for the point of inside, border, then the point on 8 directions all is a black around this point; Otherwise borderline point does not just have such character; Or the judgment task of doing frontier point according to 4 neighborhoods of point goes out the position of form center and the relevant data message of reference point by least square fitting, deposits in the distributor above relevant information stand-by.

Operating process is as follows more specifically: at first the image of gathering that comprises reference unit is carried out medium filtering and handle, thereby the noise in the image is removed in realization.Adopt the dynamic threshold dividing method to realize the binaryzation of image then, white is designated as 1, and black is designated as 0.To image from left to right, carry out in the scanning process from top to bottom, if the gray-scale value of current pixel is 1, check its left side, upper left, last, upper right these 4 neighbors.If the gray-scale value of above-mentioned 4 neighbors all is 0, just give new position mark value of current pixel.If having only the gray-scale value of a pixel in 4 neighbors is 1, just this locations of pixels mark value is composed to current pixel.If in 4 neighbors the gray-scale value of the individual pixel of m (1＜m＜=4) being arranged is 1,, determine the position mark value of current pixel then according to left, upper left, last, upper right priority.It is of equal value right then this m the mark value that pixel had note to be done, and it is included into an equivalence in the chained list.The neighborhood point of being correlated with when Fig. 3 represents to calculate O point position number.As seen, the O mark value of ordering is relevant with the mark value of A, B, C, D.If A, B, C, D have 1 value in bianry image, the mark value of O is got the wherein minimum value of mark value, composes to the O point otherwise current maximum mark value is added 1.If all underlined value among A, B, C, the D, then that priority is high mark value mark O point, the mark value that all the other are different are done equivalent processes, so just set up one corresponding to the sequence number matrix M of bianry image and equivalence to chained list EList.Equivalence among the EList to putting in order, is removed duplicate keys wherein, and all pixels of equal value are done same mark.Although it is all of equal value right to have comprised among the EList, but the situation that wherein may have the corresponding different mark value of same mark value as through sequential connection, the situation of 4 and 5 equivalences and 4 and 3 equivalences may occur, need process like this, identify into same mark value 3,4,5.Construct pair of lamina chained list SList then, earlier deposit in SList to the head node among the chained list EList equivalence, all the other each nodes among the EList are done following processing: for two mark value a in the EList present node and b, traversal SList, if find a, the interior chained list pointer record that then will contain a is in pa, otherwise pa is empty; If find b, the interior chained list pointer record that will comprise b is in pb, otherwise pb is empty.According to the value of pa and pb, do processing as shown in Figure 3.Through after the above processing, all mark chained lists of equal value are stored among the SList, each is organized element marking of equal value be changed to minimum mark value in the mark chain of equal value, make all connected regions just all be labeled new same tag.

Carry out Boundary Extraction for the zone with a kind of mark, for the point of inside, border, the point around this point on 8 directions all is a black; Otherwise borderline point does not just have such character.Utilize this method can obtain the closure edge point data of reference point, it is stored in the chained list.The utilization least square fitting method is asked for the centre of form of boundary pixel institute enclosing region:

If given sampling point is p _i(x _i, y _i), i=1...n.

Can obtain:

${(x_i-{\hat{u}}_1)}^2+{(y_i-{\hat{u}}_2)}^2={(x_{i+1}-{\hat{u}}_1)}^2+{(y_{i+1}-{\hat{u}}_2)}^2$

Can not have partially in the hope of the best of geometric centroid by following formula and to be estimated as:

${\hat{u}}_1=\frac{\underset{i=1}{\overset{m-1}{\mathrm{\Σ}}}{b_i}^2\·\underset{i=1}{\overset{m-1}{\mathrm{\Σ}}}{a}_i{c}_i-\underset{i=1}{\overset{m-1}{\mathrm{\Σ}}}{b}_i{c}_i\·\underset{i=1}{\overset{m-1}{\mathrm{\Σ}}}{a}_i{b}_i}{\underset{i=1}{\overset{m-1}{\mathrm{\Σ}}}{a_i}^2\·\underset{i=1}{\overset{m-1}{\mathrm{\Σ}}}{b_i}^2-{\left(\underset{i=1}{\overset{m-1}{\mathrm{\Σ}}}{a}_i{b}_i\right)}^2}$

${\hat{u}}_2=\frac{\underset{i=1}{\overset{m-1}{\mathrm{\Σ}}}{a_i}^2\·\underset{i=1}{\overset{m-1}{\mathrm{\Σ}}}{b}_i{c}_i-\underset{i=1}{\overset{m-1}{\mathrm{\Σ}}}{a}_i{c}_i\·\underset{i=1}{\overset{m-1}{\mathrm{\Σ}}}{a}_i{b}_i}{\underset{i=1}{\overset{m-1}{\mathrm{\Σ}}}{a_i}^2\·\underset{i=1}{\overset{m-1}{\mathrm{\Σ}}}{b_i}^2-{\left(\underset{i=1}{\overset{m-1}{\mathrm{\Σ}}}{a}_i{b}_i\right)}^2}$

Wherein: a _i=2 (x _I+1-x _i), b _i=2 (y _I+1-y _i),

$c_i=x_{i+1}^2+y_{i+1}^2-x_i^2-y_i^2$

Consider in the actual measurement and may produce error, so selected comparison error amount that is provided with when constituting equidistant isomorphism subclass, promptly being defined as the concrete steps that two points converge corresponding common reference point in 6. in above said step is: according to the basic error decision minimum error values δ that this measuring system exists, be condition is found out them by element value from two matrixes maximum isomorphism subclass with the comparison error amount less than δ.It is right further to find out public corresponding point.

, adopt svd (SVD) to finish the rotation matrix coordinate transform formula relevant and further finish and change and point data splitting also just has been readily solved (more than 3 pairs) according to the corresponding point that obtained with solving of translation vector.

Further specify as follows:

The pairing reference point set of the cloud A that sets up an office is P={p ₁, p ₂P _n, the pairing reference point set of some cloud B is Q={q ₁, q ₂Q _n, the corresponding relation of these two reference point centrostigmas can be described as q _i=R * p _i+ T i=1,2 ..., n, n＞3

R is one 3 * 3 a rotation matrix in the formula, and T is the translation vector of one 3 dimension, and n is the right number of reference point.Set up following error function

$f(R,T)=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{||q_i-(R\×p_i+T)||}^2$

Adopt svd algorithm to find the solution and place matrix and translation vector, order

$H_{3\×3}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}(p_i-\stackrel{\‾}{p}){(q_i-\stackrel{\‾}{q})}^T$

Wherein: $\stackrel{\‾}{p}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{p}_i,$ $\stackrel{\‾}{q}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{q}_i$ Be respectively the barycenter of P and Q.Matrix H is got as svd: H=UDV ^T, (D=diag (d _i), d ₁〉=d ₂〉=d ₃〉=0),

Rotation matrix: R=UV ^T

Translation vector: T=p-R * q

Solve rotation matrix R and translation matrix T, promptly can realize amalgamation, finally can obtain the complete cloud data of a material object two cloud datas.

Claims (5)

1, multi-view point data splitting method in the optical three-dimensional scanning system, this method is at computing machine, has the supporting storer of scanning imaging optical system management software and have under the support of at least two digital cameras and realize finishing, and it is characterized in that this method may further comprise the steps:

1. the paster that adopts middle white filled circles and circle outer black closed region formation is as the reference unit,

2. in the scope that the correlated digital video camera that is in different points of view all can scan simultaneously, on institute's scanned objects, paste at least three reference units,

3. the system by means of on-line operation carries out the direct information extraction to the reference unit of pasting on the scanned objects, obtain the data message in the reference unit center of circle, after adopting of black, the white transition boundary extraction of area identification method again to reference point, the method of utilizing least square to combine is obtained the home position data of reference unit

4. with the position of form center data of the reference unit of above acquisition, utilize theory on computer vision, carry out the solid coupling, the reference unit center of circle data-switching that is about to the plane becomes the position of form center information of three-dimensional reference unit,

5. according to the position of form center information of reference unit, the data message of each point during the scanning mapping of reference number video camera draws and a little converges, obtain the same number of distance matrix of digital camera that the distance parameter of reference point selected in a little converging and the position of form center of reference unit forms and adopted by analytical method

6. find out their maximum isomorphism subclass by contrasting in two matrixes element value, thereby find out reference point, be defined as the corresponding common reference point of two points in converging with identical topological characteristic,

7. adopt the respective coordinates parameter of at least three pairs of corresponding common reference point to find out rotation matrix R and the translation matrix T of changing about two coordinate systems, again according to conversion formula: P=R*P '+T directly finishes a transformation of cloud coordinate data, finishes the amalgamation to two cloud datas.

2,, it is characterized in that the reference unit of using during step 1. is is the black paster of white filled circles in the middle of circle or the rectangle according to multi-view point data splitting method in the said optical three-dimensional scanning system of claim 1.

3,, it is characterized in that said reference unit evenly is attached to two digital camera retouchs to the public domain during step 2. according to multi-view point data splitting method in the said optical three-dimensional scanning system of claim 1.

4, according to multi-view point data splitting method in the said optical three-dimensional scanning system of claim 1, it is characterized in that 3. step carry out direct information by means of the system of on-line operation to the reference unit of pasting on the scanned objects and extract, it is the position of form center that the method that adopts area identification to combine with least square is obtained reference point, at first adopt the SEQUENTIAL ALGORITHM scan image, the object under test image is done area identification and of equal value to dividing, to belong to of a sort gauge point merger, and give identical mark of equal value; Utilize the frontier point recognition methods to obtain the border of reference point then, promptly for the point of inside, border, then the point on 8 directions all is a black around this point; Otherwise borderline point does not just have such character; Or the judgment task of doing frontier point according to 4 neighborhoods of point goes out the position of form center and the relevant data message of reference point by least square fitting, deposits in the distributor above relevant information stand-by.

5, according to multi-view point data splitting method in the said optical three-dimensional scanning system of claim 1, it is characterized in that being defined as during step 6. the concrete steps that two points converge corresponding common reference point is: according to the basic error decision minimum error values δ that this measuring system exists, be condition is found out them by element value from two matrixes maximum isomorphism subclass with the comparison error amount less than δ.

Images