JP2879357B2 - Shape judgment method - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a shape judging method for judging the shape of an object to be inspected having a partial cylindrical surface, such as a press panel, using a light cutting method.

2. Description of the Related Art For example, in a press panel or the like in which a partial cylindrical surface is formed between two planes that are perpendicular to each other, it is often necessary to determine whether or not the bent shape of the partial cylindrical surface is good.

As a method for measuring the shape of such an object to be inspected, a light-section method has been studied, but a method for automatically determining the quality of the shape using the light-section method has not been proposed.

For this reason, conventionally, an operator visually judges the quality of the shape, but there is a problem that the determination is ambiguous, an error is large, and the operation takes time.

An object of the present invention is to solve the above-mentioned problem and to provide a method capable of automatically and accurately determining the quality of a shape of an inspection object by using a light section method.

Means for Solving the Problems A shape determination method according to the present invention is a method for determining the quality of a shape of an inspection object having a partial cylindrical surface by a light section method, based on image data captured by a two-dimensional imaging device. The end point of the partial cylindrical surface on the image data is obtained, the end point of the partial circumferential surface on the image data is estimated by the curve regression, and the end point of the partial cylindrical surface estimated by the curve regression on the image data is obtained. Is determined, and if the difference is equal to or larger than a predetermined value, it is determined that the shape is abnormal.

The end point of the partial cylindrical surface can be obtained from the image data captured by the two-dimensional imaging means in the light section method. Also, based on this image data, an exponential function, a fractional function,
By performing a curve regression using a regression function such as a suspension function, the end point of the partial cylindrical surface can be estimated. Further, when the shape of the inspection object is normal, the coordinates of the end point of the partial cylindrical surface on the image data and the end point of the partial cylindrical surface estimated by curve regression almost coincide, and the difference between these coordinates is small. When there is an abnormality in the shape of the inspection object, these two end points are separated, and the difference between these coordinates becomes large.

Therefore, when the difference between the coordinates of the end point of the partial cylindrical surface on the image data and the coordinates of the end point of the partial cylindrical surface estimated by curve regression is equal to or greater than a predetermined value, it is determined that the shape is abnormal, and the inspection object is determined. The quality of the shape can be determined.

Embodiments Hereinafter, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 shows an overall schematic configuration of a press panel (1), which is an inspection object, and a shape inspection apparatus.

The panel (1) has two planes perpendicular to each other, that is, a first plane (1a) and a second plane (1b) connected by a quarter cylindrical surface (1c). It is arranged such that the first plane (1a) is horizontal and the second plane (1b) is vertically downward.

The inspection device includes a light source (3), a CCD television camera (two-dimensional imaging means) (4), an image processing device (5), and an arithmetic processing device (6).

The light source (3) irradiates the surface of the panel (1) with a slit light beam orthogonal to the first plane (1a) and the second plane (1b) from directly above, and for example, a semiconductor laser or the like is used.

The television camera (4) is for capturing light that has hit the surface of the panel (1) facing the light source (3).

The image processing device (5) processes a video signal of the television camera (4) and outputs image data to be described later to the arithmetic device.

The arithmetic processing unit (6) obtains the shape of the panel (1) from the output of the image processing unit (5), and is configured by a computer.

FIG. 2 shows an example of a television image captured by the television camera (4). It should be noted that the upper, lower, left and right in FIG.

A television image is equally divided into a plurality of points by a horizontal scanning line (A) and a predetermined reference clock pulse, and each point is represented using a Y axis and a Z axis as follows. That is, the axis in the horizontal direction at the center of the television image is the Y axis, the axis in the vertical direction is the Z axis, and the intersection of these axes is the origin (0). Therefore,
The horizontal scanning line direction, that is, the horizontal scanning direction is the Z-axis direction, and the vertical scanning direction orthogonal thereto is the Y-axis direction. The right side of the television image is the positive direction of the Y axis, the left side is the negative direction, the lower side of the television image is the positive direction of the Z axis, and the upper side is the negative direction.
The coordinate value of each point is represented by an integer value obtained by multiplying the actual dimension (mm) of the panel (1) by 100. The relationship between the television image and the actual dimensions of the portion of the panel shown therein is determined by the relative positional relationship between the panel (1) and the television camera (4). Now, assuming that the horizontal width of the television screen corresponds to the actual dimension of the panel (1) of 30 mm, and the vertical width of the television screen corresponds to the actual dimension of the panel (1) of 20 mm, the Y of the origin (0) of the television image is obtained. The coordinate value and the Z coordinate value are both 0, the Y coordinate value at the right end is +1500 (+15 mm), the Y coordinate value at the left end is -1500 (-15 mm), and the Z coordinate value at the lower end is +10.
00 (+10 mm), and the Z coordinate value at the upper end is -1000 (-10 mm).

The Y coordinate value of each horizontal scanning line is stored in a variable Y [i]. Here, i is the number (scanning line number) of the horizontal scanning line (A). The number of horizontal scanning lines (A) is, for example, 484, and numbers are assigned in order with the rightmost one being 0.
That is, the number i of the rightmost horizontal scanning line (A) is 0, and the number i of the leftmost horizontal scanning line (A) is 483.

In the television image, a light portion (light portion) (H1) that hits the panel (1) appears. The light part (H1) is a linear part (L1) corresponding to the part of the light that hits the first plane (1a).
And a curved portion (C1) corresponding to the portion of light that hits the cylindrical surface (1c). Usually, the straight part (L1) appears on the left and the curved part (C1) appears on the right. The left end of the straight line part (L1) is the image start point (S1), the right end of the curved part (C1) is the image end point (E1), and the point (inflection point) from the straight line part (L1) to the curved part (C1) is the R start point. (RS). When the light hitting the panel (1) is imaged by the television camera (4), only the portion of the cylindrical surface (1c) on the first plane (1a) side is shown due to the relationship between the reflectance and the like, and the cylindrical surface (1c) And the second plane (1b) may not be captured. Accordingly, the R start point (RS) coincides with the point that transitions from the first plane (1a) to the cylindrical surface (1c), that is, the start point of the actual cylindrical surface (1c), but the image end point (E1), that is, the cylinder on the image data The end point of the plane (1c) is the point that transitions from the cylindrical plane (1c) to the second plane (1b), that is, the actual cylindrical plane (1c)
End point (R end point) (RE).

The image processing device (5) converts the video signals corresponding to the plurality of equally divided points as described above into luminance information, and outputs the light portion (H
Image data i, Y [i] and Z [i] as described below are generated only for a certain horizontal scanning line (A) having 1) and output to the arithmetic processing unit (6). i is the scanning line number of the horizontal scanning line (A) having the light portion (H1). Y [i] is the Y coordinate value of the horizontal scanning line (A) having the light portion (H1). Z
[I] is a light portion (H1) on the i-th horizontal scanning line i.
Are the Z coordinate values. A commercially available image processing device (5) having such a function can be used.

Next, an outline of an example of the operation of the arithmetic processing unit (6) at the time of inspection will be described with reference to the flowchart of FIG.

In FIG. 3, first, image data is read from the image processing device (5) (step 1), and the image data is rearranged (step 2). Next, an image start point (S1) is detected (step 3) and an image end point (E1) is detected (step 4), and an R start point (RS) is detected (step 5). Next, a regression function is determined (step 6).
An R end point (RE) is determined (step 7). Finally, the shape is determined (step 8), and the process ends.

As described above, the image data obtained from the image processing apparatus includes the Y data for the horizontal scanning line number i having the light portion (H1).
[I] and Z [i] only. Therefore, by the data rearrangement in step 2, the scanning line number i having no light portion exists.
, Set Y [i] and Z [i] as follows:

Z [i] = 1000 Y [i] = 0 The detection of the image start point (S1) in step 3 and the image end point (E1) in step 4 can be performed by changing the scanning line number i and examining the image data. it can.

The process of detecting the R start point (RS) in step 5 is performed, for example, as follows.

That is, first, image data is smoothed, first and second derivatives of the smoothed data are performed, and further, the second differential data is smoothed. The point at which the smoothed data becomes a predetermined value or less is determined. Let it be the R start point (RS).

FIG. 4 shows an outline of the process of determining the regression function in step 6 of FIG.

In this processing, the coefficients a and b of the regression function Z = ae ^bY for estimating the curved portion (C1) of the light portion (H1) are determined by the least square method, and the straight line portion of the light portion (H1) is determined. After performing coordinate transformation (rotation transformation) of the image data so that (L1) is parallel to the Y axis, coordinate transformation (parallel movement) of the image data is performed so that the R start point (RS) coincides with the Z axis. , And R are sequentially changed, and a and b are determined so that the weighted integrated value of the square of the error between the temporarily obtained regression function and the image data is minimized.

In FIG. 4, first, the inclination angle θ of the linear portion (L1) of the light portion (H1) with respect to the Y axis is determined (step 610). The details of this process are shown in FIG.

In FIG. 5, first, α and n are set to 0,
S1 (image start point) is set in data (step 611).
Then, α is obtained by the following equation, and n is increased by 1 (step 612).

In the above equation, dim3 [] is the tertiary smoothed data of the Z coordinate value. Next, it is checked whether data is larger than RS (R start point) +20 (step 613). If it is larger, data is decreased by 1 (step 614), and the process returns to step 612, where data>
Steps 612, 613 and 614 are repeated until RS + 20 is no longer reached. When data> RS + 20 is not satisfied in step 613, the inclination angle θ is obtained by the following equation (step 615).
The process ends.

θ = α / n In FIG. 4, when the process of step 610 is completed,
The coordinate transformation (rotation transformation) of the image data is performed so that the straight line portion (L1) is parallel to the Y axis, that is, the inclination angle θ is zero (step 620). The details of this process are shown in FIG.

In FIG. 6, first, data is set to 0 (step 621). Then, Y [data] and Z [data] are calculated by the following equation (step 622).

Y [data] = Y [data] × cos θ + Z [data] × sin θ Z [data] = Z [data] × cos θ−Y [data] × sin θ Next, it is checked whether data is greater than 483 (step 623). If not, increment data by 1 (step 624), return to step 622, and repeat steps 622, 623 and 624 until data> 483. If data> 483 in step 623, the process ends.

In FIG. 4, when the processing of step 620 is completed,
In order that the R starting point (RS) coincides with the Z axis,
The offset amount Yof in the Y-axis direction is obtained (step 630).

Yof = −Y [RS] Next, initialization for the least squares method is performed. That is, 0 is set to δ, 0.1 is set to d, RS (R start point) −E1 (image end point) +1 is set to m, and 10000 is set to MIN (step 640). δ is a variable for storing the Z coordinate value of the R starting point (RS), and is initially set to 0. d is a variable for storing the amount of change in the Z coordinate value of the R start point (RS), and is set to 0.1 at first. m is the number of data used for the least squares method, that is, from the R starting point (RS) to the image ending point (E1)
Is the number of data up to. MIN is a variable for storing the minimum value of the weighted integral value Σ of the square of the error.
A large value of 000 is set.

Next, the offset amount Zof in the Z-axis direction is set (step 650). The details of this process are shown in FIG.

In FIG. 7, first, δ + d is set to δ (step 651). Next, it is checked whether δ is 0 or less (step 652). If not, Zof is calculated by the following equation.
(Step 653), and the process ends.

Zof = −Z [RS] + δ When δ is 0 or less in step 652, MIN is set to 1000.
In addition to setting 00, d is set to -d / 10 (step 654), and the process returns to step 651.

In FIG. 4, when the processing of step 650 is completed, Z
The constants a and b of the exponential function corresponding to the value of, ie, δ, are calculated (step 660). The details of this processing are shown in FIG.

In FIG. 8, first, 0 is set to y1, y2, z1 and z2 (step 661), and RS is set to data (step 662). Then, the calculation of yy and zz is performed by the following equation (step 663).

yy = Y [data] + Yof zz = log (Z [data] + Zof) Next, y1, y2, z1, z2 and y2 are calculated by the following equations (step 664).

y1 = y1 + yy y2 = y2 + yy ² z1 = z1 + zz z2 = z2 + zz ² yz = yz + yy × zz Next, it is checked whether data is equal to or greater than E1 (the end point of the image) (step 665). Then, the process returns to step 663, and steps 663, 664, 665, and 666 are repeated until data ≧ E1.
When data ≧ E1 is not satisfied in step 665, Vr, Vy and Vz are calculated by the following equation (step 667).

Vr = m * yz-y1 * z1 Vy = m * y2-y1 * y1 Vz = m * z2-z1 * z1 Then, b and a are calculated by the following equations (step 668), and the process is terminated.

b = Vr / Vy a = (Z1−b × y1) / m a = exp (a) In FIG. 4, when the processing of step 660 is completed,
A weighted integral value Σ of the square of the error is obtained (step 67)
0). The details of this process are shown in FIG.

In FIG. 9, first, Σ is set to 0 (step
671), and set RS to data (step 672). Then, Σ is calculated by the following equation (step 673).

Py = log ｛(Z [data] + Yof) / a｝ / bΣ = Σ + (RS−data + 1) × | Y [data] −Yof + Py | Next, check whether data is greater than or equal to E1 (end point of image). (Step 674) If so, decrement data by 1 (Step 675) and return to Step 673 to repeat Steps 673, 674 and 675 until data ≥ E1 is no longer satisfied. When data ≧ E1 is not satisfied in step 674, the process ends.

In FIG. 4, when the processing of step 670 is completed,
It is checked whether Σ is less than MIN (step 680). If so, Σ at that time is set to MIN (step 690), and the process returns to step 650 until steps 650 and 660 are not satisfied until Σ ≦ MIN. , 670, 680 and 690 are repeated. If Σ ≦ MIN is not satisfied in step 680, the absolute value of d | d |
Is greater than 0.0001 (step 700)
If so, set d to -d / 10 and MIN
Is set to 100000 (step 710), the process returns to step 650, and the processes from step 650 onward are repeated until | d |> 0.0001 is not satisfied. Then, in step 700, | d |> 0.0001
If not, the process ends.

In the above-described processing, the process proceeds from step 680 to step 690 when the currently obtained Σ is equal to or less than the minimum value MIN. By setting Σ to MIN in step 690, MIN is updated. Each time the process returns from step 690 to step 650, δ is set at step 651 in FIG.
= Δ + d is performed, so that δ changes by d. Therefore, as long as Σ is equal to or less than MIN, Σ is changed by d, Σ is calculated, and MIN is updated. Step 6
The process proceeds from step 80 to step 700 when the obtained い ま becomes larger than the previous minimum value MIN, that is, δ becomes Σ
Immediately after the point where is minimized, the Z coordinate value of the R starting point (RS) at that time is set to δ. Step 70
The process proceeds from 0 to step 710 when | d | is greater than 0.0001. Initially, since d is set to 0.1 in step 640, the process proceeds from step 700 to step 710, but since d is multiplied by −1/10 in step 710,
When the process proceeds to step 700 for the fourth time, d is -0.0001.
And the process ends. Every time the process proceeds to step 710, the sign of d changes and the absolute value is 1/10.
Σ is obtained by changing the Z coordinate value of the R starting point (RS).
By changing the coordinate value in the opposite direction to that of the previous one by one-tenth by an amount, a point past the point where Σ becomes minimum is searched, and finally a when と き becomes minimum And b are required.

The process of determining the R end point (RE) in step 7 of FIG. 3 is performed as follows.

When the exponential function Z = ae ^bY which is a regression function is determined as described above, the Y coordinate value on the exponential function is calculated using this, while gradually changing the Z coordinate value from the R starting point (RS). A point at which the first derivative of the Y coordinate value becomes smaller than a predetermined threshold value T is defined as an R end point (RE). Conversely, the Z coordinate value on the exponential function is calculated using the exponential function Z = ae ^bY while gradually changing the Y coordinate value from the R starting point (RS), and the first derivative of the Z coordinate value is The point at which the value exceeds the predetermined threshold value T may be set as the R end point (RE).

As a regression function for estimating the shape of the cylindrical surface (1c) of the panel (1), various functions such as a fractional function and a suspension function can be used in addition to the above-described exponential function.

The determination of the shape in step 8 is for determining the quality of the shape of the panel (1), and is performed, for example, as follows.

That is, first, the R end point (R
The difference between the coordinates of E) and the coordinates of the image end point (E1) is obtained. If the difference is smaller than a predetermined threshold value, it is determined that the shape of the panel (1) is normal, and if the difference is equal to or larger than the threshold value, it is determined that the shape of the panel (1) is abnormal. When the shape of the panel (1) is normal, the R end point (RE) almost coincides with the image end point (E1), and the difference between these coordinates is small. The R end point (RE) and the image end point (E1) are separated, and the difference between these coordinates increases. Therefore, the quality of the panel (1) can be determined as described above.

Depending on the relative positional relationship between the panel (1) and the television camera (4), the television image may be different from that in FIG. However, even in such a case, the inspection can be performed by processing the image data in substantially the same manner as described above.
Also, by performing appropriate coordinate transformation on the image data,
It can be converted into image data as shown in FIG. 2 and processed.

According to the shape determination method of the present invention, as described above, the quality of the shape of the inspection object having the partial cylindrical surface can be automatically and accurately determined using the light cutting method.

[Brief description of the drawings]

FIG. 1 is a schematic perspective view of an inspection object and a shape inspection apparatus showing an embodiment of the present invention, FIG. 2 is a view showing an example of a television image,
FIG. 3 is a flowchart showing an example of a shape inspection process;
FIG. 4 is a flowchart showing an example of the regression curve determination processing of FIG. 3, FIG. 5 is a flowchart showing an example of the inclination angle calculation processing of the linear portion of FIG. 4, and FIG. 6 is an image of FIG. 7 is a flowchart showing an example of a data coordinate conversion process, FIG. 7 is a flowchart showing an example of an offset amount determination process in the Z-axis direction in FIG. 4, and FIG. 8 is an exponential function constant determination process in FIG. FIG. 9 is a flowchart showing one example, and FIG. 9 is a flowchart showing one example of the error integral value calculation processing of FIG. (1) Press panel (inspection object), (1a) (1b)
... plane, (1c) ... partial cylindrical surface, (3) ... light source,
(4) CCD TV camera (two-dimensional imaging means),
(5) ... image processing device, (6) ... arithmetic processing device.

Claims (1)

(57) [Claims] 1. A method for judging the quality of a shape of an object to be inspected having a partial cylindrical surface by a light section method, the method comprising: determining a shape of a partial cylindrical surface on image data based on image data captured by two-dimensional imaging means. While calculating the end point, the end point of the partial cylindrical surface is estimated by curve regression, and the difference between the coordinates of the end point of the partial cylindrical surface on the image data and the coordinates of the end point of the partial cylindrical surface estimated by curve regression is calculated. A shape determination method characterized by determining that the shape is abnormal when the value is equal to or more than a predetermined value.