CN106485699A - A kind of reference mark localization method based on Point matching - Google Patents

Abstract

The present invention provides a kind of positioning precision height, real-time good and the reference mark localization method based on Point matching have higher robustness to illumination, belongs to technical field of image processing.Comprise the steps：Step one：The offline crucial summit obtaining sign Mark point shape；Step 2：Obtain the picture of Mark point to be positioned online, carry out angle point or central point detection, obtain angle point or the central point of Mark point；Step 3：The angle point that step 2 is detected, is polymerized, and extracts sub-pix angle point；Step 4：Set up the sub-pix angle point extracting from step 3 or the central point that obtains of step 2 is converted into the spatial alternation function on the accordingly crucial summit that step one obtains, using method for registering, ask for the parameter of spatial alternation function set up, that is,：Obtain the Mark point location information of the picture of Mark point to be positioned.The present invention is used for the positioning of the Mark point to pcb board for the SMT.

Description

A kind of reference mark localization method based on Point matching

Technical field

The present invention relates to a kind of localization method, particularly to a kind of reference mark localization method based on Point matching, belong to Technical field of image processing.

Background technology

The extensive application of SMT (surface mounting technology, Surface Mount Technology), is greatly improved electronics The automaticity of industry, reduces cost of labor, increases production efficiency.The index of maturation of SMT technology the competing of electronic industry Strive power.And placement accuracy and speed, as two important indicators, determine the advanced degree of this technology.In the positioning in SMT It is mostly based on Mark point, the vision localization accuracy and speed of therefore Mark point has decisive role in SMT technology.

Existing Mark independent positioning method be concentrated mainly on using based on bianry image and based on edge.But it is existing The versatility of this Mark independent positioning method is not high, and positions real-time, positioning precision and all undesirable to the robustness of illumination.

Content of the invention

For above-mentioned deficiency, the present invention provides a kind of positioning precision height, real-time is good and has higher robustness to illumination The reference mark localization method based on Point matching.

A kind of reference mark localization method based on Point matching of the present invention, comprises the steps：

Step one：The offline crucial summit obtaining sign Mark point shape；

Step 2：Obtain the picture of Mark point to be positioned online, carry out angle point or central point detection, obtain the angle of Mark point Point or central point；

Step 3：The angle point that step 2 is detected, is polymerized, and extracts sub-pix angle point；

Step 4：Set up the sub-pix angle point extracting from step 3 or the central point that obtains of step 2 is converted into step one The spatial alternation function on the accordingly crucial summit obtaining, using method for registering, asks for the parameter of the spatial alternation function of foundation, I.e.：Obtain the Mark point location information of the picture of Mark point to be positioned.

Preferably, described step one includes：

Step is one by one：Set the Mark point template of SMT, determine the crucial summit of Mark point template；

Step one two：Compare the Mark point template setting, obtain information and the corresponding pass key top of corresponding Mark point offline The information of point.

Preferably, described step is one by one：

The Mark point template setting SMT includes circle, triangle, rectangle, rhombus, butterfly, double square, cross and well shape Mark point template；

The searching mode on crucial summit is：For Circular Mark point template, crucial summit is the center of Circular Mark point；Right In other Mark point templates, crucial summit is the summit of Mark point.

Preferably, described step 2 includes：

Step 2 one：When the picture of Mark point to be positioned is Circular Mark point, the picture of Mark point to be positioned is carried out Edge extracting, then the center of the Mark point of picture of Mark point to be positioned is obtained using Hough circle detection method；

When the picture of Mark point to be positioned is non-circular Mark point, the information of the offline Mark point extracting is incorporated into angle In point detective operators, then Corner Detection is entered using corner detection operator, obtain the angle of the Mark point of picture of Mark point to be positioned Point；

Step 2 two：The center of the Mark point of acquisition or angle point are changed to template coordinate system from image coordinate system；

Described image coordinate system be the picture of Mark point to be positioned left upper apex be initial point, level be to the right x-axis, vertical It is downwards the coordinate system of y-axis；

Described template coordinate system be with picture centre as initial point, level to the right as x-axis, the coordinate system as y-axis vertically downward.

Preferably, described step 3 includes：

Step 3 one：By the use of the offline Mark point obtaining parameter as threshold value, the angle point that step 2 obtains is gathered Close：If the distance existing between angle point is less than threshold value, these angle points are polymerized, ask for the angle point after being polymerized using the method for average Position；

Step 3 two：According to the parameter setting detection window of the offline Mark point obtaining, using detection window to step 3 Corner location after one polymerization is detected, obtains the image of each angle point region；

Step 3 three：The image zooming-out edge image of each angle point region that step 3 two is obtained；

Step 3 four：The edge image that step 3 three is extracted carries out fitting a straight line, obtains two straight lines through screening, two The intersection point of straight line is sub-pix angle point.

Preferably, described step 4, sets up the sub-pix angle point extracting from step 3 and is converted into what step one obtained Accordingly the spatial alternation function on crucial summit is：

F (X)=XR^T+T

Wherein, X is the sub-pix angle point collection extracting in step 3, and R is the spin moment with anglec of rotation θ as parameter for D × D dimension Battle array, meets R^TR=I and | R |=1, T=1_Mt^T, t is translation vector.

Preferably, described step 4, using method for registering, the parameter asking for the spatial alternation function of foundation includes：

Using registering energy function E, derivation is carried out to parameter θ and t：

Wherein, X=[x₁,…,x_i,…,x_M]^T, x_iRepresent the element that sub-pix angle point is concentrated, i=1,2 ..., M, M represent The quantity of sub-pix angle point concentration element, 1_DIt is 1 D dimensional vector for all elements, 1_MThe M being 1 for all elements tie up to Amount,Represent dot product；Registering energy function E is written as M × D dimension matrix form G=[G to the derivative of conversion point set U₁,…,G_i,…, G_M]；Conversion point set U=[u₁,…,u_i,…,u_M]^T, u_i=f (x_i)；

Try to achieve parameter θ and t.

Preferably, described registration energy function E：

Wherein, E () is registering energy function, Σ be sub-pix angle point collection with key vertex set mate matrix, N () table Show with u_iFor average, with σ²I is the Gaussian function of covariance, and I is unit battle array, σ²For positive number, y_jExpression is accordingly in crucial vertex set Element, j=1,2 ..., N, N represent the quantity of element in corresponding crucial vertex set.

Preferably, described G_iFor：

The beneficial effects of the present invention is, the present invention using the angle point to illumination with higher robustness, using angle point Carry out Mark point location, with respect to based on region and the amount of calculation based on edge significantly decrease.The present invention is achieving While the vision localization function of Mark point of pcb board, a comparison is even more reached on the speed of service and repeatable accuracy high Level.Shown according to retest result, the present invention is to selected required time average out to 20ms of example calculation, identical outer Under the conditions of boundary, the repeatable accuracy of 100 retests also reaches in ± 0.001mm it is seen that a kind of Mark point of this view-based access control model is fixed Method for position no matter precision, real-time, stability all very high-performance.

Brief description

Fig. 1 is that a kind of principle of the reference mark localization method based on Point matching described in the specific embodiment of the invention is shown It is intended to；

Fig. 2 is the schematic diagram of Circular Mark point template, and in figure numeral represents number and the position on crucial summit；

Fig. 3 is the schematic diagram of triangle Mark point template, and in figure numeral represents number and the position on crucial summit；

Fig. 4 is the schematic diagram of Rectangle Mark point template, and in figure numeral represents number and the position on crucial summit；

Fig. 5 is the schematic diagram of rhombus Mark point template, and in figure numeral represents number and the position on crucial summit；

Fig. 6 is the schematic diagram of butterfly Mark point template, and in figure numeral represents number and the position on crucial summit；

Fig. 7 is the schematic diagram of double square Mark point template, and in figure numeral represents number and the position on crucial summit；

Fig. 8 is the schematic diagram of cross Mark point template, and in figure numeral represents number and the position on crucial summit；

Fig. 9 is the schematic diagram of well shape Mark point template, and in figure numeral represents number and the position on crucial summit.

Specific embodiment

Specific embodiment one：In conjunction with Fig. 1, present embodiment is described, a kind of described in present embodiment is based on Point matching Reference mark localization method, including：

Step one：The offline crucial summit obtaining sign Mark point shape；

Step 2：Obtain the picture of Mark point to be positioned online, carry out angle point or central point detection, obtain the angle of Mark point Point or central point；

Step 3：The angle point that step 2 is detected, is polymerized, and extracts sub-pix angle point；

Step 4：Set up the sub-pix angle point extracting from step 3 or the central point that obtains of step 2 is converted into step one The spatial alternation function on the accordingly crucial summit obtaining, using method for registering, asks for the parameter of the spatial alternation function of foundation, I.e.：Obtain the Mark point location information of the picture of Mark point to be positioned.

Present embodiment is carried out with the crucial summit of corresponding template using the sub-pix angle point detecting or circular central point Registration, obtains positional information, realizes the vision localization function of pcb board Mark point, and versatility is high, and the requirement to illumination is low, Shandong Rod is good.

Step one is the offline crucial summit obtaining and characterizing Mark point shape, and in preferred embodiment, step one includes：

Step is one by one：Set the Mark point template of SMT, determine the crucial summit of Mark point template；

The crucial summit of present embodiment is the summit that can characterize Mark point shape；

Step one two：Compare the Mark point template setting, obtain information and the corresponding pass key top of corresponding Mark point offline The information of point；

Offline obtain Mark point information when, available image processing method or manually adjust, obtain each of Mark point Plant parameter, for example：The radius of Circular Mark point, each length of side of triangle Mark point etc..

Present embodiment is by way of the Mark point template of SMT, finds the crucial summit of Mark point template, for follow-up Step uses.The Mark point template of SMT is conventional, so versatility is fine.

In determining the preferred embodiment on crucial summit of Mark point template, as shown in Fig. 2 to Fig. 9, step one two is：

The Mark point template setting SMT includes circle, triangle, rectangle, rhombus, butterfly, double square, cross and well shape Mark point template；The initial point of template coordinate system is element central, and for level to the right, y-axis is vertically downward to x-axis.

The searching mode on crucial summit is：For Circular Mark point template, crucial summit is the center of Circular Mark point；

For other Mark point templates, crucial summit is the summit of Mark point.

After determining crucial summit, write down the position on crucial summit.

Step 2 is that the picture to Mark point to be positioned carries out angle point or central point detection, in preferred embodiment, step 2 Including：

Step 2 one：When the picture of Mark point to be positioned is Circular Mark point, the picture of Mark point to be positioned is carried out Edge extracting, then the center of the Mark point of picture of Mark point to be positioned is obtained using Hough circle detection method；

The edge extracting of present embodiment can adopt Canny edge extracting method；

When the picture of Mark point to be positioned is non-circular Mark point, in order to reduce the introducing of noise spot, will extract offline The information of Mark point be incorporated in corner detection operator, then Corner Detection is entered using corner detection operator, obtains to be positioned The angle point of the Mark point of the picture of Mark point；

The corner detection operator of present embodiment can be Harris operator or Shi-tomasi operator etc.；

Step 2 two：The center of the Mark point of acquisition or angle point are changed to template coordinate system from image coordinate system；

Described image coordinate system be the picture of Mark point to be positioned left upper apex be initial point, level be to the right x-axis, vertical It is downwards the coordinate system of y-axis；

Step 3 purpose is for non-circular Mark point, the angle point that step 2 is detected, and extracts sub-pix angle point, preferably In embodiment, described step 3 includes：

Step 3 one：By the use of the offline Mark point obtaining parameter as threshold value, the angle point that step 2 obtains is gathered Close：If the distance existing between angle point is less than threshold value, these angle points are polymerized, ask for the angle point after being polymerized using the method for average Position；

The threshold value of present embodiment refers to, during for Circular Mark point, threshold value is then the offline Circular Mark point obtaining Radius；During for triangle Mark point, threshold value is then each length of side of triangle Mark point.

Polymerization makes, if the distance between angle point is less than threshold value then it is assumed that these angle points belong to same summit, can be gathered Close, distance is more than during threshold value then it is assumed that these angle points are not belonging to same summit when between angle point, be then not required to be polymerized；

Step 3 two：According to the parameter setting detection window of the offline Mark point obtaining, using detection window to step 3 Corner location after one polymerization is detected, obtains the image of each angle point region；

Step 3 three：The image zooming-out edge image of each angle point region that step 3 two is obtained；

In present embodiment, edge extracting method can adopt Canny operator or Sobel operator etc.；

Step 3 four：The edge image that step 3 three is extracted carries out fitting a straight line, obtains two straight lines through screening, two The intersection point of straight line is sub-pix angle point.

Step 4 is the Mark point location information of the picture obtaining Mark point to be positioned, in preferred embodiment, set up from The spatial alternation function that the sub-pix angle point extracting in step 3 is converted into the accordingly crucial summit that step one obtains is：

F (X)=XR^T+T

Wherein, X is the sub-pix angle point collection extracting in step 3, and R is the spin moment with anglec of rotation θ as parameter for D × D dimension Battle array, meets R^TR=I and | R |=1, T=1_Mt^T, t is translation vector.

Set up the spatial alternation function that the central point obtaining from step 2 is converted into the accordingly crucial summit that step one obtains For：

F (X)=X+T

Wherein, X is the sub-pix angle point collection extracting in step 3, T=1_Mt^T, t is translation vector.

During for non-circular Mark point location, step 4 utilizes method for registering, asks for the ginseng of the spatial alternation function of foundation Number, in preferred embodiment, including：

Using registering energy function E, derivation is carried out to parameter θ and t：

Wherein, X=[x₁,…,x_i,…,x_M]^T, x_iRepresent the element that sub-pix angle point is concentrated, i=1,2 ..., M, M represent The quantity of sub-pix angle point concentration element, 1_DIt is 1 D dimensional vector for all elements, 1_MThe M being 1 for all elements tie up to Amount,Represent dot product；Registering energy function E is written as M × D dimension matrix form G=[G to the derivative of conversion point set U₁,…,G_i,…, G_M]；Conversion point set U=[u₁,…,u_i,…,u_M]^T, u_i=f (x_i)；

Try to achieve parameter θ and t.

During for Circular Mark point location, above-mentioned process demand of asking for takes t, that is, obtain location information；

The registering energy function E of present embodiment is that L2 distance is minimum to be set up based on making, and the process of foundation includes：

Set up the registering energy function based on L2 distance：

Wherein, E () is registering energy function, Σ be sub-pix angle point collection with key vertex set mate matrix, d_L2(u_i, y_j) represent u_iAnd y_jL2 distance；y_jRepresent the element in crucial vertex set, j=1,2 ..., N, N represent unit in crucial vertex set The quantity of element；

Using Gaussian function property, and ignore constant term, registering energy function is simplified：

In present embodiment, registering energy function E is written as M × D dimension matrix form G=to the derivative of conversion point set U [G₁,…,G_i,…,G_M], wherein

In registration process, also include：

Using pathogenic fire reducing method to δ²It is updated, if pathogenic fire reducing speed is κ, then δ²=κ δ².

The method for registering that the present invention asks for the parameter of spatial alternation function of foundation can also adopt TPS-RPM (The Thin plate spline robust point matching, thin plate spline robust Point matching method) or CPD (Coherent Point Drift, coherent point bleach-out process).

Finally the location information obtaining can be verified.As Circular Mark point needs to verify its radius, to other classes Type Mark point verifies its each side length of side.

Claims (9)

1. a kind of reference mark localization method based on Point matching is it is characterised in that comprise the steps：

Step one：The offline crucial summit obtaining sign Mark point shape；

Step 2：Obtain online the picture of Mark point to be positioned, carry out angle point or central point detection, obtain Mark point angle point or Central point；

Step 3：The angle point that step 2 is detected, is polymerized, and extracts sub-pix angle point；

Step 4：Set up the sub-pix angle point extracting from step 3 or the central point that obtains of step 2 is converted into step one and obtains Accordingly crucial summit spatial alternation function, using method for registering, ask for the parameter of spatial alternation function set up, that is,：Obtain Obtain the Mark point location information of the picture of Mark point to be positioned.

2. a kind of reference mark localization method based on Point matching according to claim 1 is it is characterised in that described step One includes：

Step is one by one：Set the Mark point template of SMT, determine the crucial summit of Mark point template；

Step one two：The Mark point template that comparison sets, obtains the information of corresponding Mark point offline and corresponds to key summit Information.

3. a kind of reference mark localization method based on Point matching according to claim 2 is it is characterised in that described step It is one by one：

The Mark point template setting SMT includes circle, triangle, rectangle, rhombus, butterfly, double square, cross and well shape Mark Point template；

The searching mode on crucial summit is：For Circular Mark point template, crucial summit is the center of Circular Mark point；For it His Mark point template, crucial summit is the summit of Mark point.

4. a kind of reference mark localization method based on Point matching according to claim 1 or 3 is it is characterised in that described Step 2 includes：

Step 2 one：When the picture of Mark point to be positioned is Circular Mark point, edge is carried out to the picture of Mark point to be positioned Extract, then obtain the center of the Mark point of picture of Mark point to be positioned using Hough circle detection method；

When the picture of Mark point to be positioned is non-circular Mark point, the information of the offline Mark point extracting is incorporated into angle point inspection In measuring and calculating, then Corner Detection is entered using corner detection operator, obtain the angle point of the Mark point of picture of Mark point to be positioned；

Step 2 two：The center of the Mark point of acquisition or angle point are changed to template coordinate system from image coordinate system；

Described image coordinate system be the picture of Mark point to be positioned left upper apex be initial point, level be to the right x-axis, vertically downward Coordinate system for y-axis；

Described template coordinate system be with picture centre as initial point, level to the right as x-axis, the coordinate system as y-axis vertically downward.

5. a kind of reference mark localization method based on Point matching according to claim 4 is it is characterised in that described step Three include：

Step 3 one：By the use of the offline Mark point obtaining parameter as threshold value, the angle point that step 2 obtains is polymerized：If The distance existing between angle point is less than threshold value, then be polymerized these angle points, asks for the corner location after being polymerized using the method for average；

Step 3 two：According to the parameter setting detection window of the offline Mark point obtaining, using detection window, step 3 one is gathered Corner location after conjunction is detected, obtains the image of each angle point region；

Step 3 three：The image zooming-out edge image of each angle point region that step 3 two is obtained；

Step 3 four：The edge image that step 3 three is extracted carries out fitting a straight line, obtains two straight lines, two straight lines through screening Intersection point be sub-pix angle point.

6. a kind of reference mark localization method based on Point matching according to claim 5 is it is characterised in that described step Four, set up the spatial alternation function that the sub-pix angle point extracting from step 3 is converted into the accordingly crucial summit that step one obtains For：

F (X)=XR^T+T

Wherein, X is the sub-pix angle point collection extracting in step 3, and R is the spin matrix with anglec of rotation θ as parameter for D × D dimension, Meet R^TR=I and | R |=1, T=1_Mt^T, t is translation vector.

7. a kind of reference mark localization method based on Point matching according to claim 6 is it is characterised in that described step Four, using method for registering, the parameter asking for the spatial alternation function of foundation includes：

Using registering energy function E, derivation is carried out to parameter θ and t：

$\begin{array}{c}\frac{\∂E}{\∂\mathrm{\θ}}=1_D^T\left(\right(G^{T}X)\⊗(\frac{\∂R}{\∂\mathrm{\θ}}\left)\right)1_D\\ \frac{\∂E}{\∂t}=G^T{1}_M\end{array};$

Wherein, X=[x₁,…,x_i,…,x_M]^T, x_iRepresent the element that sub-pix angle point is concentrated, i=1,2 ..., M, M represent sub- picture The quantity of plain angle point concentration element, 1_DIt is 1 D dimensional vector for all elements, 1_MIt is 1 M dimensional vector for all elements, Represent dot product；Registering energy function E is written as M × D dimension matrix form G=[G to the derivative of conversion point set U₁,…,G_i,…,G_M]； Conversion point set U=[u₁,…,u_i,…,u_M]^T, u_i=f (x_i)；

Try to achieve parameter θ and t.

8. a kind of reference mark localization method based on Point matching according to claim 7 is it is characterised in that described registration Energy function E：

$E(\mathrm{\Σ},f)=-\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}2{\mathrm{\Σ}}_{ij}N\left(u_i\right|y_j,2{\mathrm{\σ}}^{2}I)$

Wherein, E () is registering energy function, and Σ is the matrix that mates of sub-pix angle point collection and crucial vertex set, and N () expression is with u_i For average, with σ²I is the Gaussian function of covariance, and I is unit battle array, σ²For positive number, y_jRepresent the unit in corresponding crucial vertex set Element, j=1,2 ..., N, N represent the quantity of element in corresponding crucial vertex set.

9. a kind of reference mark localization method based on Point matching according to claim 8 is it is characterised in that described G_iFor：

$G_i=\frac{1}{{\mathrm{\σ}}^2}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}2{\mathrm{\Σ}}_{ij}N\left(u_i\right|y_j,2{\mathrm{\σ}}^{2}I)(u_i-y_j).$