JP3296099B2 - Corresponding point position measurement method and apparatus for correlation matching - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a technique for determining a position (coordinate) of a corresponding point with a model in an input image by a correlation matching process in a so-called image processing apparatus, and more particularly, to a resolution of a pixel of the input image. The present invention relates to a method and an apparatus capable of measuring a position more precisely.

[0002]

2. Description of the Related Art When an object is inspected using an image processing apparatus, it is necessary to measure a position (coordinate) of the object in an image. For such a purpose, correlation matching processing is widely used in image processing apparatuses. In the correlation matching process, (1) First, as a preliminary stage before inspection, a part of an image of a sample of an object is cut out, and a pattern relating to the density of the part is stored as a model. (2) Next, at the time of inspection, a pattern relating to density having the same shape and the same size as the model is sequentially cut out for each point shifted by one pixel from the input image, and the correlation value between the model and the cut out pattern is obtained for each point Calculation is performed, and the position of the point where the pattern having the strongest correlation is cut out is set as the position of the corresponding point with the model. The magnitude of the correlation value and the strength of the correlation are different depending on the calculation formula for calculating the correlation value. The calculation formula that gives the strongest correlation when the correlation value is the minimum and the calculation formula that gives the strongest correlation when the correlation value is the maximum. There is.

The following equation (9) is an example of an equation for calculating a correlation value between two patterns. In the equation, P _i is the density of the i-th pixel of the model, and Q _i is the i-th pixel cut out. In the case of Expression (9), the position of a point at which the pattern with the smallest correlation value C is extracted from the extracted patterns is determined as the position of the corresponding point.

(Equation 9)

The following equation (10) is another example of the equation for calculating the correlation value. In this case, among the extracted patterns,
The position of the point at which the pattern with the maximum correlation value C is cut out is determined as the position of the corresponding point.

(Equation 10)

[0005] Here, a point where a pattern is cut out and a corresponding point with a model will be described with reference to FIG. 8 as an example. In the example illustrated in FIG. 8, a description is given on the assumption that a house is an object to be detected from an input image, and that the position of the house is represented by the position of the top of the roof. That is, in FIG. 8, the model 100 of this example is a rectangular density pattern including an image of the house 101, and the position of each point in the model 100 is determined from any one of the points. Can be expressed by position. Accordingly, since the vertex 102 of the roof represents the house 101, the vertex 102 is set as a reference point. In contrast, the input image 20
When cutting out a pattern 201 having the same shape and the same size as the model 100 from 0 at a certain point 202, a pattern 2
The positional relationship between 01 and the point 202 matches the positional relationship between the model 100 and its reference point (the roof vertex 102). Then, after the point 202 is sequentially shifted by one pixel to cut out the pattern, the model 10 is extracted with a certain pattern 201A.
If the correlation with 0 is found to be the strongest, the pattern 20
Since a point 202A obtained by cutting 1A corresponds to the vertex 102 of the roof of the model 100, the point 202A is a corresponding point with the model 100. Accordingly, the position of the point 202A obtained by cutting out the pattern 201A having the highest correlation is the position of the corresponding point with the model in the input image 200, and the house 20 in the input image 200
3 position.

[0006]

The above-described correlation matching processing is common sense and widely used, but strictly speaking, has the following problems. That is, since the pattern is cut out for each point shifted by one pixel, the position of the corresponding point obtained from the calculation of Expression (9) or Expression (10) is discrete, and in some cases, the true position is obtained. ± at maximum from corresponding point
There is a risk that the position will be shifted by 0.5 pixel. This 0.
The size of the displacement of the five pixels on the actual object depends on the resolution of the video camera used for photographing, and cannot be dealt with by the conventional correlation matching processing. However, in practice, a shift of 0.5 pixel has a large effect and cannot be used for precise positioning. For example, when the resolution of the video camera is 512 pixels in the horizontal direction and the viewing angle is 60 °, and the distance between the video camera and the object is 2 m, the deviation of ± 0.5 pixels is about 5 mm.

SUMMARY OF THE INVENTION In view of the above-mentioned prior art, an object of the present invention is to provide a method and an apparatus capable of measuring the position of a corresponding point more precisely than the resolution of an input image.

[0008]

According to a first aspect of the present invention, there is provided a method for measuring a corresponding point position of a correlation matching, wherein each pattern of an input image has a pattern having the same shape and the same size as a model. Is calculated, and the correlation value between the model and the cut pattern is calculated by a calculation formula in which the stronger the correlation, the smaller the correlation value is. From the calculation result of the correlation value at each point,
The position A ₀ of the first point having the strongest correlation, the correlation value C ₀ at the first point, and the second point located immediately before the first point on one dimension including the first point a correlation value C _-1 at a second point; the correlation values C _-1 in, obtaining a correlation value C ₊₁ at the third point located immediately after the first point on the same dimension Determining the magnitude relationship with the correlation value C _{+ 1} at the third point; from the determination result,
When C _-1 > C ₊₁ , the position _Ar of the corresponding point with the model in the input image is interpolated by the following equation (11), and when C _-1 <C ₊₁ , the following equation (12). by; obtaining
This is a corresponding point position measuring method for correlation matching, characterized in that the position of the corresponding point is measured more precisely than the image resolution .

[0009]

A _r = A ₀ − (1/2) (C ₋₁ −C ₊₁ ) / (C ₀ −C ₋₁ ) Equation (11)

[0010]

A _r = A ₀ + (１ ／) (C ₊₁ −C ₋₁ ) / (C ₀ −C ₊₁ ) Equation (12)

Further, in the method for measuring the corresponding point position of the correlation matching according to the second invention, a pattern having the same shape and the same size as the model is cut out for each point of the input image, and the pattern is compared with the model. Calculating the correlation value using a calculation formula in which the correlation value increases as the correlation increases; from the calculation result of the correlation value at each point, the position A _{0 of} the first point having the strongest correlation is calculated.
And a correlation value C ₀ at the first point and a correlation value C 2 at a second point located immediately before the first point on one dimension including the first point.
_-1 and a correlation value C _{+ 1} at a third point located immediately after the first point on the same dimension; a correlation value C at the second point
Determining the magnitude relationship between _-1 and the correlation value C ₊₁ at the third point; from the determination result, if C _-1 <C ₊₁ , the above equation (11) is used.
Accordingly, C _-1> C according to the equation in the case of ₊₁ (12), determining the position A _r of the corresponding point of the model in the input image by interpolation calculation; by, precisely corresponding points than the image resolution
Is a corresponding point position measuring method of correlation matching characterized by measuring the position of a corresponding point.

Further, the corresponding point position measuring apparatus for correlation matching according to the third invention cuts out a pattern having the same shape and the same size as the model for each point of the input image, and compares the pattern with the model. Means for calculating the correlation value by a calculation formula in which the stronger the correlation is, the smaller the correlation value is; and calculating the position A of the first point having the strongest correlation from the calculation result of the correlation value at each point.
₀ , a correlation value C ₀ at the first point, and a correlation value C ₋₁ at a second point immediately before the first point on one dimension including the first point, in the same dimension Means for obtaining a correlation value C ₊₁ at a third point located immediately after the first point above;
Means for storing 1) and equation (12); means for determining the magnitude relationship between the correlation value C _-1 at the second point and the correlation value C ₊₁ at the third point; determination results Therefore, when C _-1 > C _+1, the formula (11) is selected. When C _-1 <C ₊₁ , the formula (1) is selected.
Means for selecting 2); means for obtaining the position _Ar of the corresponding point with the model in the input image by interpolation by the selected formula; and determining the position of the corresponding point more precisely than the image resolution.
A corresponding point position measuring device for correlation matching characterized by measuring the position.

Further, in the apparatus for measuring the corresponding point position of the correlation matching according to the fourth invention, a pattern having the same shape and the same size as the model is cut out for each point of the input image. Means for calculating the correlation value using a calculation formula in which the stronger the correlation is, the larger the correlation value is; and calculating the position A of the first point having the strongest correlation from the calculation result of the correlation value at each point.
₀ , a correlation value C ₀ at the first point, and a correlation value C ₋₁ at a second point immediately before the first point on one dimension including the first point, in the same dimension Means for obtaining a correlation value C ₊₁ at a third point located immediately after the first point above;
Means for storing 1) and equation (12); means for determining the magnitude relationship between the correlation value C _-1 at the second point and the correlation value C ₊₁ at the third point; determination results Therefore, when C _-1 <C ₊₁ , the above equation (11) is selected. When C _-1 > C ₊₁ , the above equation (1) is selected.
Means for selecting 2); means for obtaining the position _Ar of the corresponding point with the model in the input image by interpolation by the selected formula; and determining the position of the corresponding point more precisely than the image resolution.
A corresponding point position measuring device for correlation matching characterized by measuring the position.

[0014]

In the vicinity of the true corresponding point with the model in the input image, as shown in FIG. 3, one-dimensional including the true corresponding point _Ar, that is, before and after on the straight line 11, the strength of the correlation is the true correspondence. It changes symmetrically with a good approximation around the point _Ar . That is, taking the correlation value C on the axis 12 perpendicular to the straight line 11, the change in linear 14 correlation values in the true and change linearly 13 of the correlation value in front of the corresponding points A _r, rear, together The signs of the slopes are opposite, and their absolute values are the same. In FIG. 3, it is assumed that the smaller the correlation value is, the stronger the correlation is.

By performing interpolation using the above-described symmetry, a precise corresponding point substantially equal to the true corresponding point _Ar can be obtained. If the above-described equations (11) and (12) are used for this interpolation, these are relatively simple equations.
There is an advantage that the time required for the interpolation calculation is short.

<Equations (11) and (12)> Next, the process of deriving the equations (11) and (12) and their proper use will be described with reference to FIGS.

First, when using a calculation formula such as the above formula (9) in which the correlation value C becomes smaller as the correlation becomes stronger, if an orthogonal coordinate system of A and C is introduced, as shown in FIG. 4 or FIG. Pairs (A ₀ , C ₀ ), (A ₋₁ , C ₋₁ ), (A ₊₁ ,
C ₊₁ ) can be plotted. Here, A ₀ is the position of the first point having the strongest correlation, C ₀ is the correlation value at the first point, and A ₋₁ is the one-dimensional A including the first point immediately before the first point. The position of the second point located, C _-1 is the correlation value at the second point, A ₊₁ is the position of the third point located immediately after the first point on the same dimension A, C ₊₁ Indicates the correlation value at the third point.

FIG. 4 shows C_-1> C₊₁This is an example where
In this case, (A₀, C₀) And (A _-1, C_-1) Passing through
Draw a straight line 15 of the line
Values have equal slopes and (A₊₁, C₊₁) Two-point chain passing
If the intersection of the line with the straight line 16 is found, the A coordinate is the corresponding point.
Position A_rBecomes In this case, the straight line 15 is expressed by the following equation (13).
Is represented by Therefore, the straight line 16 is expressed by the following equation (14).
It is. Equations (13) and (1) are used to determine the intersection of the two straight lines.
By eliminating C from 4), the following equation (15) is obtained. formula
A in (15) is the position A of the corresponding point_rIs it nothing but
By expanding the equation (15), the following equation (16) is obtained. So
Then, since the point where the pattern is cut out is shifted by one pixel,
A₀-A_-1= 1 pixel, A₊₁-A₀= 1 pixel, (A_-1+ A
₊₁) / 2 = A₀Therefore, formula (16) is rearranged.
And the above equation (11) is obtained. That is, calculate the correlation value
In the case of obtaining by equation (9) or the like, C_-1> C₊₁If
A is calculated by interpolation using (11).₀And C₀, C_-1Passing
And C₊₁And the exact position A of the corresponding point using_rSeeking
Can be.

[0019]

(C−C ₋₁ ) = (C ₀ −C ₋₁ ) / (A ₀ −A ₋₁ ) (A−A ₋₁ ) Equation (13)

[0020]

(C−C ₊₁ ) = − (C ₀ −C ₋₁ ) / (A ₀ −A ₋₁ ) (A−A ₊₁ ) Equation (14)

[0021]

(C ₋₁ −C ₊₁ ) = − (C ₀ −C ₋₁ ) / (A ₀ −A ₋₁ ) (2A−A ₋₁ −A ₊₁ ) Equation (15)

[0022]

A _r = (1/2) (A ₋₁ + A ₊₁ ) − ((C ₋₁ −C ₊₁ ) / (C ₀ −C ₋₁ )) (A ₀ −A ₋₁ ) = A ₀ − (1/2) (C ₋₁ −C ₊₁ ) / (C ₀ −C ₋₁ ) Equation (16)

FIG._-1<C₊₁This is an example where
In this case, (A₀, C₀) And (A ₊₁, C₊₁) Passing through
Draw a straight line 17 of the line, and sign
Values have equal slopes and (A_-1, C_-1) Two-point chain passing
If the intersection of the line with the straight line 18 is found, the A coordinate is the corresponding point.
Position A_rBecomes In this case, the straight line 17 is expressed by the following equation (17).
Is represented by Therefore, the straight line 18 is expressed by the following equation (18).
It is. Equations (17) and (1) are used to determine the intersection of the two straight lines.
Eliminating C from 8) gives the following equation (19). formula
A in (19) is the position A of the corresponding point_rIs it nothing but
Then, when Expression (19) is expanded, the following Expression (20) is obtained. This
In the case of, the point where the pattern is cut out is shifted by one pixel.
And A₀-A_-1= 1 pixel, A₊₁-A₀= 1 pixel, (A_-1
+ A₊₁) / 2 = A₀Equation (20) is rearranged.
Then, the above equation (12) is obtained. That is, the correlation value
In the case of using the calculation formula (9) or the like, C_-1<C₊₁If,
A is calculated by interpolation using the above equation (12).₀And C₀, C
_-1And C₊₁And the exact position A of the corresponding point using_rAsk for
be able to.

[0024]

(C−C ₊₁ ) = (C _{+ 1−} C ₀ ) / (A _{+ 1−} A ₀ ) (A−A ₊₁ ) Equation (17)

[0025]

(C−C ₋₁ ) = − (C _{+ 1−} C ₀ ) / (A _{+ 1−} A ₀ ) (A−A ₋₁ ) Equation (18)

[0026]

Equation 19] _{(C +1 -C -1) = -} (C +1 -C 0) / (A +1 -A 0) (2A-A +1 -A -1) ... (19)

[0027]

Equation 20] _{A r = (1/2) (A} +1 + A -1) - ((C +1 -C -1) / (C +1 -C 0)) (A +1 -A 0)] = A ₀ + (1/2) (C ₊₁ −C ₋₁ ) / (C ₀ −C ₊₁ ) Equation (20)

Next, when the correlation is strong, as in the above equation (10),
When a calculation formula that increases the correlation value C is used, A and C
When the rectangular coordinate system is introduced, as shown in FIG. 6 or FIG.
The pair of the point at which the pattern was cut out and the correlation value at that point
(A₀, C₀), (A_-1, C _-1), (A₊₁, C₊₁)
Can be lotted. Where A₀Has the strongest correlation
Location of the first point, C₀Is the correlation value at the first point, A_-1Is the
The first position on the one-dimensional A including the point 1 located immediately before the first point
Position of point 2, C_-1Is the correlation value at the second point, A ₊₁Are the same
The position of a third point on the dimension A immediately after the first point;
C₊₁Indicates the correlation value at the third point.

FIG. 6 shows an example in which C _-1 <C ₊₁ .
In this case, (A ₀ , C ₀ ) and (A ₋₁ , C
_-1 ), a solid line 15 is drawn, the sign of which is opposite to that of this straight line 15, the slopes are equal, and (A ₊₁ , C ₊₁ )
Is obtained by finding the intersection of the two-dot chain line with the straight line 16 passing through
Coordinate is a position A _r of the corresponding points. In this case, the straight line 15 is represented by the equation (13). Therefore, the straight line 16 is also represented by the equation (14). Eliminating C from equations (13) and (14) to find the intersection of both straight lines yields equation (15)
Is obtained. Since the A in the formula (15) nothing but positions A _r of the corresponding points, Expanding equation (15), Equation (1
6) is obtained. Since the point at which the pattern is cut is shifted by one pixel, A ₀ −A ₋₁ = 1 pixel, A ₊₁ −A ₀ = 1 pixel, and (A ₋₁ + A ₊₁ ) / 2 = A ₀ , Equation (1)
By rearranging 6), the above equation (11) is obtained. That is, when the correlation value is obtained by the calculation formula (10) or the like, C _-1 <
If C ₊₁ , A is calculated by interpolation using equation (11).
_0, it is possible to determine the position A _r of precise corresponding points by using the C _0, C _-1 and C _+1.

FIG. 7 shows an example in which C _-1 > C ₊₁ .
In this case, (A ₀ , C ₀ ) and (A ₊₁ , C
₊₁ ), a straight line 17 is drawn, the sign of which is opposite to that of this straight line 17 and whose absolute value is equal to that of (A ₋₁ , C ₋₁ ).
Is obtained by finding the intersection of the two-dot chain line with the straight line 16 passing through
Coordinate is a position A _r of the corresponding points. In this case, the straight line 17 is represented by the equation (17). Therefore, the straight line 18 is also represented by the equation (18). Therefore, if C is eliminated from equations (17) and (18) in order to find the intersection of the two straight lines, equation (19) is obtained. Since the A in the formula (19) nothing but positions A _r of the corresponding points, obtained by expanding the formula (19), Equation (30). Also in this case, the point at which the pattern is cut is shifted by one pixel, so that A ₀ −A ₋₁ = 1 pixel and A ₊₁ −
Since A ₀ = 1 pixel and (A _-1 + A ₊₁ ) / 2 = A ₀ , the above equation (12) is obtained by rearranging equation (20). That is, when the correlation value is obtained by the calculation formula (10) or the like, if C ₋₁ > C ₊₁ , A ₀ , C ₀ , C _−1, and C ₀ are obtained by interpolation using the previous formula (12). it is possible to obtain the position a _r of precise corresponding points by using the _+1.

[0031]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 shows an example of a configuration of a corresponding point position measuring device for correlation matching. The device of this example includes a correlation value calculating unit 1, a model storage unit 2, a data extracting unit 3, a magnitude relation determining unit 4,
It is composed of an interpolation formula selection unit 5, an interpolation formula storage unit 6, and an interpolation calculation unit 7.

The correlation value calculating unit 1 uses the density pattern model information 2a stored in the model storage unit 2 in advance, and for each point of the input image 8 which is even one pixel at a time, has the same shape as the model. A pattern having the same size is cut out, and a correlation value between the model and the cut out pattern is calculated according to the calculation formulas (9) and (10) described above, and a calculation result 1a of a correlation value at each point is calculated as data. It is given to the extraction unit 3.

The data extracting unit 3 calculates the position A ₀ of the first point having the strongest correlation and the position A ₀ of the first point from the calculation result 1a of the correlation value at each point given from the correlation value calculating unit 1. Correlation value C ₀
And a correlation value C _-1 at a second point located immediately before the first point on one dimension including the first point, and a third value located immediately after the first point on the same dimension. A correlation value C ₊₁ at the point is obtained, and these data are given to the magnitude relationship determination unit 4 and the interpolation calculation unit 7.

In the present embodiment, the magnitude relation determining unit 4 determines the data
Correlation value C at the second point given from extraction unit 3_-1And the third
Correlation value C at point₊₁The magnitude relationship with the first point correlation
Value C ₀The absolute value of the difference with | C_-1-C₀|, | C₊₁-C₀|
And the result of determination 4a is calculated using an interpolation formula.
This is given to the selection unit 5.

The interpolation formula selection unit 5 selects one of the previous formulas (11) and (12) stored in the interpolation formula storage unit 6 in advance according to the judgment result given from the magnitude relation judgment unit 4. I do. Specifically, (1) | C _-1 -C ₀ |> | C ₊₁ -C
₀ |, equation (11) is selected, and (2) | C ₋₁ −C
_{0 | <| C +1 -C 0} | For selects equation (12).

The interpolation operation unit 7 uses the equations (11) or (12) selected by the interpolation equation selection unit 6 to provide A ₀ , C ₀ , C _-1 and C ₊₁ given from the data extraction unit 3. from obtains the position a _r of the precision corresponding points.

In the case of | C ₋₁ −C ₀ | = | C ₊₁ −C ₀ |, since the first point is the true corresponding point, A ₀ corresponds without the need for interpolation calculation. The position of the point. However, in order to avoid the complexity of the processing, an equal sign may be included in one of the determination conditions, for example, as described below. (1)
When | C ₋₁ −C ₀ | ≧ | C _{+ 1} −C ₀ |, the equation (11) is used.
And (2) If | C ₋₁ −C ₀ | <| C ₊₁ −C ₀ |, the equation (12) is selected.

By the way, | C_-1-C
₀| And | C₊₁-C₀To determine the magnitude relationship between |
The reason is that the correlation is calculated by the correlation value calculation unit 1 as shown in Expression (9).
Equation (10), where the correlation value decreases as the value increases.
Whichever formula is used, the stronger the correlation, the larger the correlation value.
Even if it is, the appropriate interpolation formula (11) or
Is because (12) can be selected. That is, (1) When calculating the correlation value by the calculation formula (9), | C_-1-C
₀｜ ＞ ｜ C₊₁-C₀| Is C_-1> C₊₁Equation (1)
1) Select | C_-1-C₀| <| C₊₁-C₀| Is C_-1
<C₊₁Equation (12) is selected. (2) On the contrary, when the correlation value is calculated by the formula (10),
｜ C_-1-C₀｜ ＞ ｜ C ₊₁-C₀| Is C_-1<C₊₁Is equivalent to
Expression (11) is selected and | C_-1-C₀| <| C₊₁-C
₀| Is C_-1> C₊₁Equation (12) is selected.

Of course, the correlation value calculator 1 calculates the equation (9).
For example, when using only a calculation formula in which the correlation value becomes smaller as the correlation becomes stronger, the magnitude relationship between C ₋₁ and C ₊₁ is simply determined by the magnitude relationship determination unit 4, and the interpolation formula selection unit 5 determines C ₋₁ ≧ C select the expression (11) in the case of ₊₁ may be selected equation (12) in the case of C _-1 <C _+1.

Similarly, the correlation value calculation unit 1 calculates the equation (10)
If, for example, only a calculation formula that increases the correlation value as the correlation becomes stronger is used, the magnitude relationship between C ₋₁ and C ₊₁ is simply determined by the magnitude relationship determination unit 4, and the interpolation formula selection unit 5 determines C ₋₁ ≦ C select the expression (11) in the case of ₊₁ may be selected equation (12) in the case of C _-1> C _+1.

FIG. 2 shows an operation flow of the embodiment apparatus shown in FIG.

<Extension to 2D, 3D, etc.>
Akira is about measuring the position of corresponding points in one dimension
Measures the position of corresponding points in 2D or 3D
To do so, do the following. (1) Two-dimensional case: In this case, the apparatus shown in FIG.
Then, two portions 9 surrounded by a two-dot chain line are prepared, and XY rectangular coordinates are prepared.
Considering the system, on the other hand, A is X, A₀To X₀, A _rTo X_r
From the calculation result of the correlation value at each point.
Strong first point position X₀And the correlation value C at the first point₀
And on the X-axis including the first point, immediately before the first point
Correlation value C at the second point_-1And immediately after the first point on the same X axis
The correlation value C at the third point located at₊₁And the second point
Correlation value C at_-1And the correlation value C at the third point₊₁Relationship with
Is determined, and equation (11) or equation (1) is determined according to the determination result.
2) Select the corresponding point of the input image
Position X on X axis_rAsk for. On the other hand, A is Y, A
₀Is Y₀, A_rIs Y_rAnd calculate the correlation value at each point.
From the calculation result, the position Y of the first point having the strongest correlation₀And the same
Correlation value C at the first point₀And on the Y axis including the first point
Correlation value C at a second point located immediately before point 1_-1And the same
Correlation value at a third point located immediately after the first point on the Y axis
C₊₁, And the correlation value C at the second point_-1And the phase in the third point
Function value C₊₁Judge the magnitude relationship with
Select (11) or (12) to select
Y on the Y-axis of the corresponding point with the model_rAsk for. this
Gives (X_r, Y_r) Is the position of the corresponding point in two dimensions
Is determined. (2) Three-dimensional case: In this case, the part surrounded by the two-dot chain line
9 is added, A is Z, A₀To Z₀, A_rTo Z_r
From the calculation result of the correlation value at each point.
Strong first point position Z₀And the correlation value C at the first point
₀And on the Z axis including the first point, immediately before the first point.
The correlation value C at the second point_-1And the first point on the Z axis
Correlation value C at the third point located later₊₁Find the second point
Correlation value C at_-1And the correlation value C at the third point₊₁Relationship with
Is determined, and equation (11) or equation (1) is determined according to the determination result.
2) is selected and Z of the corresponding point with the model in the input image is selected.
Position Z on axis_rAsk for. Thus, (X_r, Y_r,
Z_r) Is obtained as the position of the corresponding point in three dimensions. (3) In the above two examples, parallel processing is used.
Enclosed part 9 is only one, and X_r, Y_r, Z_rTo
You may ask for it.

[0043]

As can be understood from the above description, the position of the corresponding point with the model can be measured only with an accuracy equivalent to the resolution of the image in the past. The position of the corresponding point can be measured more precisely than the resolution of the image. Further, since the equation used for the interpolation calculation is relatively simple, there is an advantage that the time required for the calculation is short.

[Brief description of the drawings]

FIG. 1 is a diagram showing an apparatus configuration according to an embodiment of the present invention.

FIG. 2 is a diagram showing a flow of operation of the apparatus.

FIG. 3 is a diagram showing a symmetry of a correlation value change which is a principle of the present invention.

FIG. 4 is an explanatory diagram of derivation of an interpolation expression.

FIG. 5 is an explanatory diagram of derivation of an interpolation formula.

FIG. 6 is an explanatory diagram of derivation of an interpolation formula.

FIG. 7 is an explanatory diagram of derivation of an interpolation equation.

FIG. 8 is an explanatory diagram of a correlation matching process and corresponding points.

[Explanation of symbols]

 Reference Signs List 1 correlation value calculation unit 2 model storage unit 3 data extraction unit 4 magnitude relation determination unit 5 interpolation formula selection unit 6 interpolation formula storage unit 7 interpolation calculation unit 8 input image

──────────────────────────────────────────────────続 き Continuation of front page (58) Fields investigated (Int. Cl. ⁷ , DB name) G01B 11/00-11/30 102 G06T 1/00 G06T 7/00 G06T 7/60

Claims (4)

(57) [Claims] 1. A pattern having the same shape and the same size as a model is cut out for each point of an input image, and a correlation value between the model and the cut out pattern is calculated using a calculation formula in which the stronger the correlation, the smaller the correlation value. From the calculation result of the correlation value at each point, the position A ₀ of the first point having the strongest correlation, the correlation value C ₀ at the first point, and the one-dimensional position including the first point To calculate a correlation value C ₋₁ at a second point located immediately before the first point and a correlation value C ₊₁ at a third point located immediately after the first point on the same dimension. Determining the magnitude relationship between the correlation value C ₋₁ at the second point and the correlation value C ₊₁ at the third point; from the determination result, if C ₋₁ > C ₊₁ , By (1),
In the case of C _-1 <C _{+ 1} , the position _Ar of the corresponding point with the model in the input image is obtained by interpolation by the following equation (2); thereby, the position of the corresponding point is measured more precisely than the image resolution. You
The corresponding point position measuring method of the correlation matching characterized by Rukoto. A _r = A _0- (1/2) (C _-1 -C ₊₁ ) / (C ₀ -C _-1 ) Equation (1) _Ar = A ₀ + (1) / 2) (C ₊₁ −C ₋₁ ) / (C ₀ −C ₊₁ ) Equation (2) 2. A pattern having the same shape and the same size as the model is cut out for each point of the input image, and the correlation value between the model and the cut out pattern is calculated by a calculation formula in which the stronger the correlation, the larger the correlation value. From the calculation result of the correlation value at each point, the position A ₀ of the first point having the strongest correlation, the correlation value C ₀ at the first point, and the one-dimensional position including the first point To calculate a correlation value C ₋₁ at a second point located immediately before the first point and a correlation value C ₊₁ at a third point located immediately after the first point on the same dimension. Determining the magnitude relationship between the correlation value C _-1 at the second point and the correlation value C ₊₁ at the third point; from the determination result, if C _-1 <C ₊₁ , (3)
In the case of C _-1 > C _{+ 1} , the position _Ar of the corresponding point with the model in the input image is obtained by interpolation by the following equation (4); thereby, the position of the corresponding point is measured more precisely than the image resolution. You
The corresponding point position measuring method of the correlation matching characterized by Rukoto. Equation 3] _{A r = A 0 - (1/2} ) (C -1 -C +1) / (C 0 -C -1) ... Equation (3) [Expression 4] A _r = A ₀ + (1 / 2) (C ₊₁ −C ₋₁ ) / (C ₀ −C ₊₁ ) Equation (4) 3. A pattern having the same shape and the same size as the model is cut out for each point of the input image, and the correlation value between the model and the cut out pattern is calculated by a calculation formula in which the stronger the correlation, the smaller the correlation value becomes. A position A ₀ of the first point having the strongest correlation, a correlation value C ₀ at the first point, and a one-dimensional pattern including the first point based on the calculation result of the correlation value at each point. A correlation value C _-1 at a second point located immediately before the first point above and a correlation value C ₊₁ at a third point located immediately after the first point on the same dimension are calculated as follows. Means for obtaining; means for storing the following two equations (5) and (6); magnitude of correlation value C _-1 at the second point and correlation value C ₊₁ at the third point Means for judging the relationship; means for selecting the following equation (5) if C _-1 > C ₊₁ and selecting the following equation (6) if C _-1 <C ₊₁ from the judgment results: The selected expression More, means for determining the interpolation calculation the position A _r of the corresponding point of the model in the input image; equipped with, than the image resolution to measure the position of the precisely corresponding points
The corresponding point position measuring device of the correlation matching characterized by Rukoto. A _r = A _0- (1/2) (C _-1 -C ₊₁ ) / (C ₀ -C _-1 ) Equation (5) _Ar = A ₀ + (1) / 2) (C ₊₁ -C _-1 ) / (C ₀ -C ₊₁ ) Equation (6) 4. A pattern having the same shape and the same size as the model is cut out for each point of the input image, and the correlation value between the model and the cut out pattern is calculated by a calculation formula in which the stronger the correlation, the larger the correlation value. A position A ₀ of the first point having the strongest correlation, a correlation value C ₀ at the first point, and a one-dimensional pattern including the first point based on the calculation result of the correlation value at each point. A correlation value C _-1 at a second point located immediately before the first point above and a correlation value C ₊₁ at a third point located immediately after the first point on the same dimension are calculated as follows. Means for obtaining; means for storing the following two equations (7) and (8); magnitude of the correlation value C _-1 at the second point and the correlation value C ₊₁ at the third point Means for judging the relationship; means for selecting the following equation (7) if C _-1 <C ₊₁ and selecting the following equation (8) if C _-1 > C ₊₁ from the judgment results: The selected expression More, means for determining the interpolation calculation the position A _r of the corresponding point of the model in the input image; equipped with, than the image resolution to measure the position of the precisely corresponding points
The corresponding point position measuring device of the correlation matching characterized by Rukoto. A _r = A ₀ − (１ ／) (C ₋₁ −C ₊₁ ) / (C ₀ −C ₋₁ ) Equation (7) _Ar = A ₀ + (1) / 2) (C ₊₁ −C ₋₁ ) / (C ₀ −C ₊₁ ) Equation (8)