JP4090895B2 - Component positioning method and apparatus - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a component positioning method and apparatus, and more specifically, the positioning of a component, that is, the position of the component (part center) and inclination, from the contour line of an image of an electronic component of an arbitrary shape captured by a component mounting apparatus or the like. The present invention relates to a component positioning method and apparatus for obtaining the same.
[0002]
[Prior art]
Various methods for positioning a pattern in an image, particularly a pattern of an electronic component sucked by a suction nozzle have been proposed and implemented. In particular, the template matching method for calculating the cross-correlation value by applying the template image of the component to be detected to the input image (photographed image of the sucked electronic component) and outputting the position and angle at which this value is the largest has been used for a long time. It is famous.
[0003]
In recent years, many matching operations using contour lines have been performed. For example, in Patent Document 1, in learning, a contour line is extracted from a reference image (learning image) that serves as a reference (template) for a component to be positioned, and coordinates on the contour line and the gradient vector of the image at that point are paired. In the positioning, the contour line is extracted from the input image, and matching is performed by calculating the degree of coincidence between the contour line and the learning data.
[0004]
Furthermore, for example, in Patent Document 2, in learning, a contour line is extracted from a learning image, and matching is performed using coordinates on the contour line and a gradient vector of the image at that point. In this patent document, there is a fixed reference coordinate, and for a straight line connecting the point on the contour line and the reference coordinate, the distance between the two points and the opening angle between the normal line of the contour line are used as learning data, and positioning is performed. The principle of extracting the contour line from the input image, recording the number of times it falls to the coordinates indicated by the angle and the distance by the so-called voting method at each point of the contour line, and the principle that the coordinates with the most votes are the reference coordinates Matching is performed.
[0005]
[Patent Document 1]
JP2001-251097
[Patent Document 2]
JP 2001-291105 A
[0006]
[Problems to be solved by the invention]
In all conventional methods using contour lines, the basic set of learning data is coordinates and angles. However, in the conventional method, since it is assumed that all sets have the same positioning ability (influence or positioning performance) at this time, there is a problem that positioning accuracy is lowered. That is, in general, the outer periphery of a part is rarely uniform, and there is a clear straight line part and, on the contrary, a blurred curve part. The positioning ability is high in the former and low in the latter. However, in the conventional method, pairs taken from anywhere are treated equally, so participation of a pair having a low positioning ability resulted in a decrease in positioning accuracy. However, on the other hand, in order to correctly position a general complex-shaped part, it is necessary to extract the set evenly from the outer periphery of the part.
[0007]
Therefore, the present invention solves such problems, and a component positioning method and apparatus capable of improving the component positioning accuracy performed based on the features extracted from the contour line and used for component positioning. The issue is to provide.
[0008]
[Means for Solving the Problems]
  In order to solve the above problems, in the present invention,
  In a component positioning method for extracting a feature from the contour line of a reference image of a component and positioning the component based on the extracted feature,
From the contour line, when a point on the contour line and a density gradient vector of the point are paired, a feature composed of two beams is extracted,
  A value indicating positioning performance is added to the extracted feature,
  Extract the contour line from the input image of the part to be positioned,
  A configuration is employed in which the component is positioned based on the feature having a large value indicating the positioning performance and the alignment of the contour line of the input image.
[0009]
  In the present invention,
  In a component positioning apparatus that extracts a feature from the contour line of a reference image of a component and positions the component based on the extracted feature,
  An outline extraction unit for extracting an outline from an image of a part;
  From the outline of the reference image of the partWhen the point on the contour line and the density gradient vector at that point are paired, it consists of two beamsMeans for extracting a feature and adding a value indicating positioning performance to the extracted feature (component learning unit);
  A component positioning unit for positioning the component from the alignment of the feature with a large value indicating the positioning performance and the contour line of the image of the component to be positioned;
  The structure which has is also employ | adopted.
[0010]
In such a configuration, a value indicating the positioning performance (positioning capability) is added to the feature used for positioning output from the contour line obtained from the reference image (learning image) of the component, and a large positioning performance is obtained during positioning. Since the positioning is performed using a feature having a high-precision positioning.
[0012]
  In the present invention, as features used for positioning,The stability of the shape representative points at both ends of the beam is high, the beam is long, and the angle formed by the concentration gradient vectors at the shape representative points at both ends of the beam is close to 90 degrees.Coarse positioning features,The stability of the shape representative points at both ends of the beam is high, the beam is long, and the angle formed by the straight line is close to 90 degrees for any concentration gradient vector.Features for fine angle correction,Both the shape representative points at both ends of the beam are highly stable, the beam is long, and the angle formed by the straight line is close to 0 degrees for any of the concentration gradient vectors.Features for fine correction of magnification are extracted, and positioning is first performed using coarse positioning features in preference to positioning speed. If positioning accuracy cannot be obtained with only this rough positioning feature, fine correction is performed using the fine angle correction feature and / or the fine magnification correction feature. Each of the coarse positioning feature, angle fine correction feature, and magnification fine correction feature is given a value indicating positioning performance, and each feature having high positioning performance is selected to perform rough positioning or fine correction. Therefore, highly accurate positioning is possible.
[0013]
The positioning performance of the coarse positioning feature, angle fine correction feature, and magnification fine correction feature is determined based on the reliability of the beam, the length of the beam, and the angle formed by the density gradient vectors at both ends of the beam. The reliability of the beam is determined from the stability of the shape representative points constituting both ends of the beam. In addition, the stability of the shape representative point on the contour line is determined from the evaluation value of the clarity of the shape representative point, the evaluation value of the curvature of the shape representative point, and / or the evaluation value of the curvature holding length holding the curvature. Is done.
[0014]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, the present invention will be described in detail based on embodiments shown in the drawings.
[0015]
<Overall configuration>
FIG. 1 is a block diagram showing a schematic configuration of an apparatus for positioning a component according to the present invention. The control unit 10 is composed of a CPU or the like, and controls the entire flow. The component learning unit 11 acquires a learning image by the image input unit 13, outputs a component feature from the contour line of the learning image output by the contour line extraction unit 14, and an evaluation value indicating positioning performance with respect to the output feature Is added. The learning image is a reference image or basic image serving as a reference (template) for a component to be positioned, and a contour line is extracted from the learning image and a component feature is output. The component positioning unit 12 inputs an image of a component to be positioned by the image input unit 13, and performs matching between the contour line of the component image output from the contour line extraction unit 14 and the feature extracted from the contour line of the learning image. The position and inclination of the part are output, and the part is positioned.
[0016]
<Parts learning>
The component learning is performed by the component learning unit 11 according to the flow illustrated in FIG.
[0017]
In step S10, a part learning image, that is, a reference image is input. The input image is a grayscale image. For example, the pixel value takes a value from 0 (minimum brightness) to 255 (maximum brightness).
[0018]
In step S11, a contour line is extracted from the learning image by the contour line extraction unit 14. An appropriate threshold value is set in advance, and it is possible to trace the lightness boundary line based on the threshold value. A conventional contour line extraction method, for example, a method described in Japanese Patent Laid-Open No. 2001-251097 is used. Can do. The extracted contour line stores coordinates of vertices constituting the contour line and density gradient vectors at the coordinates. The density gradient vector is a vector in the vicinity of the coordinates, which is directed from a low lightness to a high lightness, and the magnitude of the vector represents a lightness difference.
[0019]
[Shape representative points and their stability]
Subsequently, in step S12, a set of shape representative points representing the shape is acquired from the extracted contour line. FIG. 3 illustrates a method for extracting a set of shape representative points. Reference numeral 20 denotes a contour line of the learning image extracted by the contour line extraction unit 14, and vertices R0 and R1 selected at appropriate intervals with respect to the position index attached to the vertex from the start point to the end point of the vertex line of the contour line. , R2,. . . . . Let Rn-2 and Rn-1 be shape representative points. For example, if the interval is 3, the start vertex [0] is taken as the starting point, and [3], [6],. . Will be chosen in the order. That is, the interval between the shape representative points is 3, and the point Rk is one of the vertices so selected. G is the concentration gradient vector. In the present invention, a value (0 to 1) called “stability” is generated at each point R, and the value is recorded at each shape representative point. Therefore, the shape representative point has three parameters: coordinates on the contour line, density gradient vector, and stability.
[0020]
The stability of each shape representative point R is given as follows.
[0021]
First, the clarity g at the point R is defined as the magnitude of the density gradient vector G (g = | G |). The curvature of the R point is calculated, and the absolute value is k (k ≧ 0). In the present embodiment, for k, the curvature is classified into three, a zero band M0, an intermediate band M1, and an infinite band M2, as shown in FIG. Here, k0 and k1 are predetermined constants. However, k0 is a curvature corresponding to a gentle curve portion, and k1 is a curvature corresponding to a curve portion of a sharp curve. In order to determine the constants k0 and k1, each curve portion is represented by a circle. Each radius is described in the same coordinate system as the contour, and then the reciprocal is taken to obtain the desired curvature. The curvature k extracted from the contour line 20 falls into one of a zero band M0 where 0 ≦ k ≦ k0, an intermediate band M1 where k0 <k ≦ k1, and an infinite band M2 where k1 <k. Further, as shown in FIG. 5, an allowable band of zero band M0, intermediate band M1, and infinite band M2 widened by a constant ω0 or ω1 is added to each band.
[0022]
Further, the “curvature retention length” s can be obtained for the contour line 20 of FIG. s is the circumference from the vertices Zb to Zf. The vertex Zb starts from the vertex Rk and sequentially follows the vertex in the direction opposite to the vertex traveling direction, and stops when the curvature k of each point is outside the allowable band of the band to which k originally belonged. , And the vertex Zf is the last vertex obtained in the same way as in Zb, when the vertex Zf starts from the vertex Rk and sequentially follows the vertex in the same direction as the vertex moving direction. It is a vertex. Therefore, the curvature holding length s is the length when the curvature of a predetermined vertex is held in the permissible band before and after the curvature.
[0023]
The stability of the shape representative point R is considered to be higher as the intelligibility g is larger, higher as the curvature k is smaller, and higher as the curvature holding length s is larger. In this embodiment, α is an evaluation value of intelligibility ( 0 ≦ α ≦ 1), β is the evaluation value of curvature (0 ≦ β ≦ 1), γ is the evaluation value of curvature retention length (0 ≦ γ ≦ 1), and the stability r of point R is expressed by r = αβγ is doing. Therefore, the stability r is expressed by a real value 0 ≦ r ≦ 1 from the lowest 0 to the highest 1.
[0024]
A function for generating evaluation values of the above-described intelligibility g, curvature k, and curvature holding length s is shown in FIG. 4, and FIG. 4A shows a function E1 for generating an evaluation value α of intelligibility g. There are G_0 and G_1 determined in advance, and the function E1 generates an evaluation value in which g ≦ G_0 and α = 0, g ≧ G_1 and α = 1, and the middle is primary interpolation. FIG. 4B shows a function E2 for generating an evaluation value β of the curvature k, and there are predetermined K_0 and K_1. The function E2 is β = 1 when k ≦ K_1, β = 0 when k ≧ K_0, and intermediate. Generates an evaluation value for primary interpolation. FIG. 4C shows a function E3 for generating an evaluation value γ of the curvature retention length s. There are predetermined S_0 and S_1, and the function E3 is γ = 0 when s ≦ S_0, and γ = when s ≧ S_1. 1 and an intermediate value generate an evaluation value for primary interpolation.
[0025]
Thus, the shape representative points R0, R1, R2,. . . , Rn−1, by performing the above processing, a set of shape representative points {R0, R1, R2,. . . , Rn−1}, and this set is a set having data such as coordinates on the contour line, density gradient vector, and stability at each shape representative point. The above is the process shown in step S12 of FIG.
[0026]
[Rough positioning features]
Subsequently, coarse positioning features are output in step S13. As will be described later, at the time of component positioning, as shown in FIG. 8A, a positive number ε for positioning is prepared, and contour lines are provided on the left and right in the advancing direction of the contour line 30 of the component image to be positioned. Virtual boundary lines 30 ′ and 30 ″ indicated by dotted lines separated by a distance ε from 30 are set. In the region sandwiched between the boundary lines, the work of matching the features of the learned template (learned image) with the contour line 30 is rough. Positioning, speed is given priority over accuracy in coarse positioning.
[0027]
FIG. 6 is a diagram illustrating a state in which coarse positioning features are output from the contour line 20 of the learning image. The rough positioning feature means a feature of the beam_A (BEAM_A) type. That is, this type is a beam composed of two shape representative points A1 and A2 in the set of shape representative points. A beam suitable for rough positioning has high stability at the shape representative points at both ends of the beam, the beam is long, and the angle formed by the concentration gradient vectors U1 and U2 at the shape representative points at both ends of the beam is close to 90 degrees. Say things. Therefore, in this embodiment, λ is the evaluation value of the beam reliability (0 ≦ λ ≦ 1), μ is the evaluation value of the beam length (0 ≦ μ ≦ 1), and ν is the value at both ends of the beam. As an evaluation value (0 ≦ ν ≦ 1) of an angle formed by the concentration gradient vector, the positioning performance f of the beam _A is expressed as f = λμν. Accordingly, the positioning performance f of the beam_A is represented by a real value 0 ≦ f ≦ 1 from the lowest 0 to the highest 1.
[0028]
The reliability evaluation value λ is obtained by λ = MIN {r1, r2}. Here, r1 and r2 are the stability of the shape representative points A1 and A2, respectively. For example, as shown in FIG. 6, when the stability of the point A1 is r1 = 0.8 and the stability of the point A2 is r2 = 0.3, λ = 0.3. The length evaluation value μ is obtained by μ = h / H. Here, h is a distance between the shape representative points A1 and A2, and H is an appropriate positive number as large as the diameter of the part (= maximum distance from end to end). The angle evaluation value ν is obtained by ν = | sin θ |. Here, θ is an opening angle formed by the density gradient vectors U1 and U2 of the shape representative points.
[0029]
With respect to the characteristic beam, a beam having a large positioning performance value is selected. That is, if the f value indicating the positioning performance does not reach a predetermined level for a certain beam, the beam is discarded. As a result, beams of a certain level or more gather, so that stable positioning is guaranteed. There is a required number of coarse positioning features. In the present embodiment, for example, six beams (the number of shape representative points is twelve), but in principle, any arbitrary value of 1 or more may be used. The feature output continues until the required number is reached. In that case, it is important to perform control so that the output value is the one with the largest f value. To do so, follow the steps below.
[0030]
That is, first, all the shape representative points are unused. The “maximum value of f value” is set to 0 before outputting the next feature. Continue to select all unused pairs of points from the shape representative point set, calculate the f-value of the selected set, that is, the beam, and compare it with the maximum value so far. If it is larger, the set is updated as the maximum value. When all the combinations are completed, the group in which the maximum value is recorded is output, and the two shape representative points are used. By the above procedure, the required number of rough positioning features are output.
[0031]
Of course, the total number of combinations of two points selected from a finite set of points is limited, there are unused conditions, and there is an f-value level condition. May be depleted. That is, generally, the number of coarse positioning features finally output is less than the required number.
[0032]
[Features for fine angle correction]
Subsequently, the fine angle correction feature is output in step S14. The feature for fine angle correction refers to a feature of a beam_B (BEAM_B) type in FIG. That is, this die is a beam composed of two shape representative points B1 and B2. A beam that is good for fine angle correction is one in which the shape representative points at both ends of the beam are both highly stable, the beam is long, and the angle between the density gradient vectors V1 and V2 and the straight line B1B2 is close to 90 degrees. Say. Therefore, λ is a beam reliability evaluation value (0 ≦ λ ≦ 1), μ is a beam length evaluation value (0 ≦ μ ≦ 1), and ν is a concentration gradient vector at both ends of the beam. As an angle evaluation value (0 ≦ ν ≦ 1), the positioning performance f of the beam _B is expressed as f = λμν. The f value is represented by a real value 0 ≦ f ≦ 1 from the lowest 0 to the highest 1. λ and μ are obtained by the same evaluation formula as in the case of the coarse positioning feature, while ν is obtained by ν = MIN {| sinθ1 |, | sinθ2 |}. Here, θ1 and θ2 are the opening angles formed by the density gradient vectors V1 and V2 with the vector B1B2, respectively.
[0033]
The output standard and output method of beam_B are the same as those of beam_A. However, the required number of features for fine angle correction was set to 16. Generally, the number of fine angle correction features that are finally output is less than or equal to this required number.
[0034]
[Features for fine correction]
In step S15, the feature for fine magnification correction is output. The feature for fine magnification correction refers to a feature of a beam_C (BEAM_C) type in FIG. That is, this mold is a beam composed of two shape representative points C1 and C2. A beam that is good for fine magnification correction is one in which the shape representative points at both ends of the beam are both highly stable, the beam is long, and the angle formed by the straight line C1C2 is close to 0 degrees in either of the gradient vectors W1 and W2. Say. Therefore, λ is a beam reliability evaluation value (0 ≦ λ ≦ 1), μ is a beam length evaluation value (0 ≦ μ ≦ 1), and ν is a concentration gradient vector at both ends of the beam. As an angle evaluation value (0 ≦ ν ≦ 1), the positioning performance f of the beam _C is expressed as f = λμν. The f value is represented by a real value 0 ≦ f ≦ 1 from the lowest 0 to the highest 1. λ and μ are obtained by the same evaluation formula as in the case of the rough positioning feature, and ν is obtained by ν = MIN {| cos θ1 |, | cos θ2 |}. Here, θ1 and θ2 are opening angles formed by the density gradient vectors W1 and W2 with the vector C1C2, respectively.
[0035]
The output standard and output method of beam_C are the same as those of beam_A. However, the required number of features for fine magnification correction was 10. In general, the number of magnification correction features to be finally output is less than or equal to this required number.
[0036]
As described above, the component learning unit 11 extracts the coarse positioning feature, the fine angle correction feature, and the fine magnification correction feature from the outline of the reference image, and indicates the positioning performance for each feature. Is added, and features with high positioning performance are selected.
[0037]
<Part positioning>
The component positioning is performed by the component positioning unit 12 according to the flow of FIG.
[0038]
In step S20, the image of the component to be positioned, which is sucked by the suction nozzle, is input via the image input unit 13, and the learning data (a set of shape representative points, high positioning performance) obtained by the component learning unit 11 is obtained. Coarse positioning features, angle fine correction features, magnification fine correction features, etc.). Subsequently, a contour line is extracted by the contour line extraction unit 14 from the image of the component to be positioned in step S21. The extracted outline fragment is shown at 30 in FIG.
[0039]
[Coarse positioning]
In the next step S22, rough positioning is performed. For rough positioning, for example, the method described in Japanese Patent Laid-Open No. 2001-251097 can be used, and the rough positioning feature is moved on the contour line 30 while matching the tangent vector (orthogonal vector of the density gradient vector), for rough positioning This is done by checking whether the shape representative points at both ends of the feature are on the contour line 30. For this purpose, the operator sets an allowable positioning width ε, and checks whether each coarse positioning feature has a distance from the contour line of ε or less and is within the range of the virtual boundary lines 30 ′ and 30 ″. In this way, in the rough positioning, the representative shape points 31a at both ends of the feature BEAM_A indicated by reference numeral 31 in FIG. Positioning is performed so that 31b exists on the contour line 30 with an allowable width ε.
[0040]
After the rough positioning, almost all of the shape representative points constituting each rough positioning feature (BEAM_A) are within the ε-band region. Therefore, when accuracy is not required, positioning can be completed only by this rough positioning. However, this situation (parallel movement / angle / magnification) does not indicate that the contour line 20 of the learning image and the contour line 30 of the component image are completely coincident with each other. The magnification still leaves room for fine correction.
[0041]
Therefore, the fine correction is repeated, and the repeat condition of the fine correction is inspected in step S23. In general, the more precise corrections are made, the higher the accuracy becomes. However, it takes much time, so it is necessary to stop at an appropriate place. In this embodiment, the accuracy condition is fixed three times. Of course, as will be described later, the conditions here can be changed according to the purpose of use of positioning.
[0042]
If the learned component-specific coordinate system (reference coordinate system) is the UV coordinate system and the coordinate system of the image of the component to be positioned is the XY coordinate system, the coordinates (u, v) and (x, y) are expressed by the following equations: It is known to be tied together.
[0043]
(X, y) = S · {(a11, a12) (a21, a22)} (u, v) + (Tx, Ty)
However, S is a magnification, {(a11, a12) (a21, a22)} is a 2 × 2 rotation matrix, and (Tx, Ty) is a translation.
[0044]
[Fine correction of translation]
The first fine correction to be repeated is a parallel movement fine correction performed in step S24. In the present embodiment, all the beams, that is, all the beams for coarse positioning, fine angle correction, and fine magnification correction are allowed to participate in the fine correction of translation. This is aimed at the statistical effect obtained when the number is large. In FIG. 8A, a beam (BEAM_A) 31 shows a situation after the end of rough positioning or after several fine corrections, and the coordinates of both ends 31a and 31b of the beam 31 are the current coordinate conversion formulas. The generated coordinates. Each coordinate has outline coordinates, that is, coordinates closest to the outline (indicated by black dots). The vectors from the beam coordinates to the contour line coordinates are represented by e0 and e1.
[0045]
The beam-specific correction vector v is v = e0 when | e0 | ≧ | e1 |, and v = e1 when | e0 | <| e1 |. For example, in the case of FIG. 8A, since e1 is larger, v = e1. When there are a total of N beams participating in the translational fine correction, correction vector sets {v0, v1,. . . , VN−1} is obtained.
[0046]
Let the component parallel movement fine correction vector ΔT be ΔT = Σ (| vi | · vi) / Σ | vi |. Here, the numerator is the N totals (vector) of | vi | · vi, and the denominator is the N totals (scalar) of | vi |. This equation is a weighted average equation based on the magnitude of Vi, and the larger the vi, the larger the ratio of vi to the result. The validity is shown in FIG. 8 (d). FIG. 8D shows N = 3 beams BEAM0, BEAM1, and BEAM2, and beam-specific correction vectors are v0, v1, and v2, respectively. There is a shift in the X direction and a shift in the Y direction, but since the shift in the Y direction is conspicuous, the Y direction must be greatly improved. If a simple arithmetic average (weight = 1) is taken, the Y component of ΔT is approximately v0 / 3, and thus the purpose cannot be achieved. However, in the method of this embodiment, the Y component of ΔT is approximately v0. Therefore, effective fine correction can be performed and the object can be achieved.
[0047]
In this way, in the fine correction of the translation, by setting the translation fine correction vector ΔT of the part to ΔT = Σ (| vi | · vi) / Σ | vi | Accurate corrections can be made.
[0048]
Regarding the translation term of the current coordinate conversion formula, the term obtained by adding ΔT is newly referred to as a translation term.
[0049]
[Fine angle correction]
In step S25, the angle is finely corrected. The angle fine correction beam is used for fine angle correction. In FIG. 8B, a beam (BEAM_B) 32 shows a situation after the end of rough positioning or after the end of several fine corrections. The coordinates of both ends 32a and 32b of the beam 32 are coordinates generated by the current coordinate conversion formula. Each coordinate has a coordinate closest to the contour line (indicated by a black dot). A dotted line indicates the direction of a vector connecting two points on the contour line. If the signed opening angle from the beam to this vector is δ radians, δ is a correction angle unique to this beam. When there are a total of N beams participating in angle fine correction, a set of correction angles {δ0, δ1,. . . , ΔN−1} is obtained.
[0050]
The component angle fine correction ΔΘ is ΔΘ = Σ (| fi | · δi) / Σ | fi |, where fi is the beam positioning performance (0 to 1). Here, the numerator is N sums (radians) of | fi | · δi, and the denominator is N sums (scalar) of | fi |. This equation is a weighted average equation based on the magnitude of fi, and the larger the fi, the larger the proportion of δi in the result. As the beam performance is higher (the f value is larger), the degree of influence becomes stronger, thereby enabling accurate fine correction.
[0051]
Thus, in the fine angle correction, the angle fine correction ΔΘ of the component is ΔΘ = Σ (| fi | · δi) / Σ | fi |, where fi is the beam performance (0 to 1). Accurate corrections can be made.
[0052]
A rotation matrix obtained by multiplying the rotation matrix of the current coordinate transformation formula by a matrix corresponding to ΔΘ from the front is defined as a rotation matrix.
[0053]
[Fine correction of magnification]
Subsequently, in step S26, the magnification is finely corrected. A beam for fine correction of magnification is used for fine correction of magnification. In FIG. 8C, a beam (BEAM_C) 33 shows a situation after the rough positioning is completed or after several fine corrections are completed. The coordinates of both ends 33a and 33b of the beam 33 are coordinates generated by the current coordinate conversion formula. Each coordinate has a coordinate closest to the contour line (indicated by a black dot). Assuming that the length of the beam 33 is d0 and the distance between the outline black points indicated by the dotted line is d1, s = d1 / d0 is a correction magnification specific to this beam. When there are a total of N beams participating in the magnification fine correction, the beam-specific correction magnification set {s0, s1,. . . , SN-1}.
[0054]
The component magnification fine correction ΔS is ΔS = Σ (| fi | · si) / Σ | fi |, where fi is the beam positioning performance (0 to 1). Here, the numerator is N totals (scalar) of | fi | · si, and the denominator is N totals (scalar) of | fi |. This equation is a weighted average equation based on the size of fi, and the larger the fi, the larger the proportion of si in the result. As the beam performance is higher (the f value is larger), the degree of influence becomes stronger, thereby enabling accurate fine correction.
[0055]
In this way, in the fine correction of the magnification, the component fine correction ΔS is ΔS = Σ (| fi | · si) / Σ | fi |, where fi is the beam performance (0 to 1). Accurate corrections can be made.
[0056]
The magnification obtained by multiplying the current coordinate conversion formula by ΔS is set as a new magnification, and then the process returns to step S23.
[0057]
In this way, by repeating the fine correction of the alignment (matching) between the features 31, 32, and 33 and the contour line 30, the alignment between the features and the contour line 30 is completed.
[0058]
<Other embodiments>
In step S12, the coordinates of the shape representative point and the density gradient vector are the coordinates and density gradient vector obtained directly from the outline, respectively. However, the effect is not changed even if the interpolated coordinates and the interpolated density gradient vector are used instead.
[0059]
In addition, although the curvature classification is set to 3 in step S12, the same effect can be obtained even if it is 2 or more, and the function of FIG. 4 is used as the evaluation functions E1, E2, and E3 in step S12. The effect does not change even if the intermediate portion is changed to something other than the primary interpolation taking a value of 0-1.
[0060]
Further, in each of the steps S13 to S15, the beam reliability λ is set to the function λ = MIN {r1, r2} of the stability r1, r2, but it is effective even if other functions, for example, λ = r1 × r2. does not change.
[0061]
In addition, in step S23, the repetition condition for fine correction is fixed three times, but the effect is not changed even if other conditions are used instead. For example, paying attention to the distance between the contour line and each end point of the beam, it may be possible to exit the loop when the maximum distance is within a predetermined distance, or the degree of coincidence between each point of the template and each point of the contour line It is also possible to prepare a function for measuring and to exit the loop when the function value exceeds a predetermined value.
[0062]
In addition, regarding the execution order of the fine corrections in steps S24 to S26, in the embodiment, parallel movement → rotation → magnification is used, but the effect does not change even in an arbitrary order.
[0063]
As for the fine correction, both the angle fine correction and the magnification fine correction are not performed, but only one of them can be performed.
[0064]
【The invention's effect】
As described above, in the present invention, a value indicating the positioning performance is added to the feature used for positioning output from the contour line obtained from the reference image of the component, and the feature having high positioning performance at the time of positioning is added. Since positioning is performed using this, positioning with high accuracy becomes possible.
[0065]
In the present invention, the coarse positioning feature, the fine angle correction feature, and the fine magnification correction feature are extracted as the features used for positioning, and positioning is performed using the coarse positioning feature giving priority to the positioning speed. Therefore, high-speed positioning is possible. If positioning accuracy cannot be obtained with this rough positioning feature alone, fine correction is performed using the fine angle correction feature and / or the fine magnification correction feature, thus improving positioning accuracy. In this case, a value indicating the positioning performance is assigned to each of the rough positioning feature, angle fine correction feature, and magnification fine correction feature, and each feature having high positioning performance is selected. As a result, rough positioning or fine correction is performed, so that the positioning accuracy can be dramatically improved.
[Brief description of the drawings]
FIG. 1 is an overall block diagram showing an embodiment of an apparatus of the present invention.
FIG. 2 is a flowchart schematically showing a flow of parts learning.
FIG. 3 is an explanatory diagram illustrating a principle of extracting shape representative points in a part learning unit.
FIG. 4 is a diagram showing characteristics of an evaluation formula for obtaining the stability of a shape representative point in a part learning unit.
FIG. 5 is an explanatory diagram showing curvature classification in a part learning unit.
FIG. 6 is an explanatory diagram showing types of feature data output from a part learning unit.
FIG. 7 is a flowchart schematically showing a flow of positioning in a component positioning unit.
FIG. 8 is an explanatory diagram showing the principle of various fine corrections in the component positioning unit.
[Explanation of symbols]
10 Control unit
11 Parts learning department
12 Parts positioning part
13 Image input section
14 Outline extraction unit
20 Outline of learning image
30 Contour line of part image to be positioned

Claims (6)

In a component positioning method for extracting a feature from the contour line of a reference image of a component and positioning the component based on the extracted feature,
From the contour line, when a point on the contour line and a density gradient vector of the point are paired, a feature composed of two beams is extracted,
A value indicating positioning performance is added to the extracted feature,
Extract the contour line from the input image of the part to be positioned,
A component positioning method comprising positioning a component based on a feature having a large value indicating positioning performance and alignment of a contour line of the input image. The above features are features for positioning that the shape representative points at both ends of the beam are both highly stable, the beam is long, and the angle formed by the density gradient vectors at the shape representative points at both ends of the beam is close to 90 degrees . Besides, the stability of the shape representative points at both ends of the beam is both high, the beam is long, and for any of the concentration gradient vectors, an angle fine correction feature whose angle to a straight line is close to 90 degrees , and / or Both the shape representative points at both ends of the beam are highly stable, the beam is long, and for any density gradient vector, a feature for fine magnification correction whose angle to the straight line is close to 0 degrees is extracted to make it a rough positioning feature 2. The component positioning method according to claim 1, wherein the positioning performed based on the angle fine correction feature and / or the magnification fine correction feature is finely corrected. 3. The component positioning method according to claim 2 , wherein the positioning performance of each feature is determined based on the reliability of the beam, the length of the beam, and the angle formed by the concentration gradient vectors at both ends of the beam. . In a component positioning apparatus that extracts a feature from the contour line of a reference image of a component and positions the component based on the extracted feature,
An outline extraction unit for extracting an outline from an image of a part;
When a point on the contour line and a density gradient vector of the point are paired from the contour line of the reference image of the part , a feature composed of two beams is extracted, and positioning performance is obtained with respect to the extracted feature. Means for adding a value indicating
A component positioning unit for positioning the component from the alignment of the feature with a large value indicating the positioning performance and the contour line of the image of the component to be positioned;
An apparatus for positioning a part, comprising: The above features are features for positioning that the shape representative points at both ends of the beam are both highly stable, the beam is long, and the angle formed by the density gradient vectors at the shape representative points at both ends of the beam is close to 90 degrees . Besides, the stability of the shape representative points at both ends of the beam is both high, the beam is long, and for any of the concentration gradient vectors, an angle fine correction feature whose angle to a straight line is close to 90 degrees , and / or Both the shape representative points at both ends of the beam are highly stable, the beam is long, and for any density gradient vector, a feature for fine magnification correction whose angle to the straight line is close to 0 degrees is extracted to make it a rough positioning feature 5. The component positioning apparatus according to claim 4 , wherein the positioning performed based on the angle fine correction and / or the magnification fine correction feature is finely corrected. 6. The component positioning apparatus according to claim 5 , wherein the positioning performance of each feature is determined based on the reliability of the beam, the length of the beam, and the angle formed by the concentration gradient vectors at both ends of the beam. .

Images