JP2002039949A - Method for regulating threshold used in coat defect inspecting step - Google Patents

Abstract

PROBLEM TO BE SOLVED: To realize a method for automatically obtaining a suitable value of a threshold though a threshold used for inspecting a coat defect is regulated by repeating a trial and an error at the present. SOLUTION: A method for regulating the threshold, used in a coat defect inspecting step, comprises a step S1 of photographing a coating surface 2 which does not have a defect to obtain a lightness of each pixel, a step S2 of space differentiating light information of the obtained each pixel to obtain a differential value 6 of each pixel, a step S3 of setting an initial threshold as the threshold, a step S4 of comparing the differential value of each pixel with the threshold, a step S5 of extracting a profile independent in a visual field of a profile of a pixel group in which the differential value is the threshold or larger, a step S6 of subtracting a first prescribed value from the threshold, a step S7 of repeating a cycle until an independent profile is extracted in the step S5 with the 'step S4, the step S5 and the step S6' as unit cycles, and a step S8 of adding a second prescribed value to the threshold, when the independent profile is extracted in the extracting step to decide the threshold.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

TECHNICAL FIELD The present invention relates to a painted surface,
For example, it relates to a preparation technique for a process of inspecting a painted surface of an automobile body after a painting process for a defect that may be present.

[0002]

2. Description of the Related Art In order to check for the presence of defects that may be present on a painted surface, the painted surface is usually photographed to obtain brightness for each pixel. Then, the obtained brightness of each pixel is spatially differentiated to obtain a differential value for each pixel. Then, the differential value of each pixel is compared with a threshold value, the contour of a pixel group whose differential value is equal to or greater than the threshold value is determined, and an independent contour is extracted in the visual field. If there is a defect on the painted surface, an independent contour is extracted, so that the presence or absence of a defect can be inspected by the above method.

[0003]

In order to obtain a correct result by the above method, the threshold value must be set to an appropriate value. If the threshold value is too large, the defect is overlooked, and if the threshold value is too small, even if there is no defect, it is erroneously determined that there is a defect. At present, there is no suitable method for adjusting the threshold to an appropriate value, and the threshold is adjusted by repeating trial and error. For this reason, it takes a lot of time to adjust the threshold value used in the coating surface defect inspection process to an appropriate value. Therefore, the present invention has created a method for automatically obtaining an appropriate threshold value.

[0004]

Means for Solving the Problems, Functions and Effects The method of the present invention is a method for adjusting a threshold used in a paint surface defect inspection step to an appropriate value. In this method, as schematically shown in FIG. 1, a step S1 of photographing a painted surface 2 having no defect to obtain a brightness 4 of each pixel, and a spatial differentiation of the obtained brightness of each pixel to differentiate each pixel. A step S2 for obtaining a value 6, a step S3 for setting an initial threshold value as a threshold value, a step S4 for comparing a differential value for each pixel with a threshold value, A step S5 of extracting a contour that has been performed, a step S6 of subtracting the threshold value by a first predetermined value, and an independent contour in the extraction step S5 as a unit cycle with the “comparison step S4, extraction step S5, and reduction step S6”. The method includes a step S7 of repeating the cycle until extraction is performed, and a step S8 of adding a second predetermined value to a threshold when the independent contour is extracted in the extraction step S5 to determine the threshold.

In this method, the extraction step of S5 is repeated while reducing the threshold value by a first predetermined value (Step S6). Initially, pixels having a differential value equal to or larger than the threshold are not extracted because the threshold is too large. While repeating the extraction step of S5 while reducing the threshold value by the first predetermined value, a pixel group having a differential value equal to or greater than the threshold value and having an independent contour starts to be extracted. FIG. 1 illustrates that a pixel group having a differential value equal to or larger than a threshold value and having an independent contour is extracted in the (n + 1) th extraction step. Here, it is known that no defect exists on the painted surface 2. Nevertheless, the group of independent pixels extracted corresponds to a degree of irregularity that does not hit a defect, called citron skin or orange peel. That is, while repeatedly performing the extraction step of S5, a pixel group having a contour independent of a differential value equal to or greater than the threshold value is extracted, and the threshold value at this time is barely a level at which citron skin can be detected. Has been adjusted. This threshold value is a value (level) at which the paint surface defect can be reliably extracted if it exists. However, it is too small in the sense that even the skin is detected.
In the step S8, a second predetermined value is added to the threshold value at this time. The threshold value adjusted in this manner is at a level at which the painted surface defect is extracted although the citron skin is not extracted, and is a level suitable for the inspection of the painted surface defect. The citron skin or orange peel referred to here is described in paragraph [0003] of JP-A-8-28557. The present invention is a method for automatically obtaining an appropriate threshold value using the presence of citron skin, and is extremely unique. According to the present invention, since the independent contour is extracted, the brightness pattern of the illumination does not adversely affect the process of determining the threshold.
Here, “independent” does not include a case where the extracted pixel groups are scattered on the same straight line and are substantially continuous. In the case where pixel groups equal to or larger than the threshold value are scattered on the same straight line, the boundary area of the light and dark pattern of the illumination is intermittently extracted instead of the citron skin. There is no.

[0006]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, an embodiment in which a threshold value adjusting method of the present invention is applied to a preparation stage of a coating surface defect inspection process of an automobile body will be described. For this purpose, a painted surface for threshold adjustment (not shown) is prepared under the same conditions as the painted automobile body to be inspected. The painted surface for threshold adjustment is a painted surface that is known in advance to be free from defects. The painted surface of the automobile body, which is previously confirmed to be free from defects, can be used as the painted surface for threshold adjustment.

[0007] A camera (not shown) for photographing a threshold adjustment painted surface (hereinafter simply referred to as a test painted surface) outputs the brightness of each pixel as information that can be processed by an electronic computer. For example, it refers to a CCD camera or the like. The camera according to the present embodiment outputs a signal of the intensity level of light at 8 bits per pixel. As a result, the brightness of each pixel becomes 0
Can be taken at 256 gradations from 255 to 255. As shown in FIG. 12, the computer operates according to a program to store pixel-by-pixel brightness information sent from the camera for each view, and stores per-view pixel-unit brightness information storage device 122; And a device 123 for storing the differential value of each pixel
A comparison operation device 126 for comparing the differential value with a threshold value, an independent contour extracting device 128 for extracting a contour of a pixel group whose differential value is equal to or more than the threshold value and independent in the visual field, and a predetermined value from the threshold value. Calculation device 130 for reducing or adding
Is composed. Naturally, CPU 130 and interface 1
34. An illumination device (not shown) for illuminating a painted surface is disclosed in Japanese Patent No. 263812 and Japanese Patent Application Laid-Open No. 9-280.
As described in Japanese Patent No. 845, the illumination is performed so that a light and dark stripe pattern is formed on the painted surface. The positional relationship between the test painted surface and the lighting device and the lighting method are the same as those used in the actual painted surface defect inspection process.

[0008] Using the flowchart shown in FIG.
The threshold adjustment procedure will be described. First, the lighting device described above is turned on. Next, adjust the aperture of the camera. The amount of light captured by the camera during shooting differs depending on the body color of the car body. Therefore, the following processing is performed with the aperture of the camera adjusted according to the body color of the vehicle body so that a substantially uniform light amount can be obtained regardless of the vehicle type.

In step S12, a test painted surface is photographed.
As a result, the brightness of each pixel is obtained for the test painted surface. The brightness information for each pixel is stored in the brightness information storage device 122.
Is stored. As clearly shown in FIG. 3, the test painted surface reflects the illumination light clearly, so that a clear striped light-dark pattern is photographed. The graph F4 in FIG.
FIG. 4 is a diagram illustrating a lightness distribution on a straight line perpendicular to a boundary between light and dark patterns. The horizontal axis corresponds to the position on the straight line. The vertical axis corresponds to lightness. Curve 52 is a position-brightness curve. In the diagram of F2, the brightness of each pixel is binarized by a predetermined value. Pixels having brightness equal to or higher than a predetermined value are set to pure white, and pixels having brightness smaller than the predetermined value are set to black.

In step S14 of FIG. 2, the differentiation operation device 124 of the electronic computer operates to perform a differentiation process. As a result, a differential value, that is, an amount of change in brightness is obtained for each pixel.
Hereinafter, “differential value” and “brightness change amount” are used synonymously. FIG.
The middle graph F8 is a diagram showing a differential value distribution on a straight line perpendicular to the boundary between the light and dark patterns. The horizontal axis corresponds to the position on the straight line. The vertical axis corresponds to the differential value or the brightness change amount. The same applies to the vertical and horizontal axes of graphs F12, 16, and 20 described later. Curve 54 is a position-differential value curve.
The curves 56, 60, 64 described below are equal to the curve 54. The differential value for each pixel obtained by performing the differentiation processing is stored in the differential value storage device 123 for each pixel of the electronic computer (step S14). The differential value storage device 123 for each pixel stores differential value distribution map information in the visual field.

In step S16, an initial threshold is set as the threshold. This initial threshold is set in advance based on past data. That is, a large value is set so as not to detect a change in the brightness of the illumination.

The comparison operation unit 126 compares the differential value for each pixel obtained in step S14 with a threshold value. Then, a pixel group whose differential value is equal to or larger than the threshold value and whose contour is independent is extracted (step S18).

Since the initial threshold value is large, when step S18 is first executed, the independent contour extraction device 12
8 does not extract independent contours. As is well shown in the graph F12 of FIG.
, The image obtained by binarizing the differential value distribution map with the initial threshold value 58 becomes black (see F10), as shown in the image F10, and the brightness change amount or the differential value exceeds the threshold value. Are not extracted. FIG. 4 is an actual image obtained by binarizing the differential value with the initial threshold value 58 and is black. Since there is no pixel having a differential value equal to or larger than the threshold value, step S20 is initially NO, and the process proceeds to step S22.

In step S22, since the threshold value is too large, it is reduced by a first predetermined value. As the first predetermined value, a value set in advance based on past data is used. The threshold calculator 130 derives a new threshold.

Using the new threshold, steps S18 and S20 described above are repeated. When the independent contour extraction device 128 does not extract an independent contour even when the second threshold is used, a third threshold is derived by subtracting the first threshold from the second threshold in step S22. As described above, steps S18, S20, and S22 are set as a unit cycle, and the cycle is repeated until an independent contour is extracted in step S20. When the threshold value decreases to around the peak value of the position-differential value curve 54 in FIG. 2, a white area starts to appear on the binarized differential value distribution map screen. FIG. 5 shows a binarized screen in this state. In this case, a pixel group whose differential value is equal to or larger than the threshold value starts to exist (the pixel group is shown in white). The pixel group detected at this time corresponds to the boundary region of the light and dark pattern of the illumination, and rides substantially in the same straight line. In this technique, even though the pixels are divided, the pixel groups on the same straight line are substantially continuous, so that the pixel groups are not independent pixel groups. The threshold for providing the binarized screen in FIG. 5 does not mean that the pixel group providing an independent contour has been extracted in step S20 in FIG. Therefore, in step S22, the threshold value is further reduced. When the threshold value that is gradually reduced falls below the peak level of the position-differential value curve 60 as shown in a graph F16, the differential value distribution map binarized by the threshold value 62 is schematically shown in an image F14. As shown in FIG. In this case, a pixel group having a differential value equal to or larger than the threshold value is linearly continuous, and independent contours are not extracted in the visual field.

Incidentally, the position-differential value curve is obtained by overlapping a gentle wave corresponding to an illumination pattern with a fine wave. These fine waves are caused by the citron skin on the painted surface. No defects are present on the test painted surface. The threshold value gradually decreases while repeating the extraction processing in which steps S18, S20, and S22 are performed as a unit cycle.
The position at 20 approaches the bottom level of the differential value curve 64.
In this case, when the differential value is compared with the threshold value, a pixel group having a differential value equal to or larger than the threshold value and whose contour enclosing the pixel group is independent in the visual field starts to be extracted. here,
It is assumed that an independent contour is extracted in the (n + 1) -th step S22. As schematically schematically shown in the graph F20, the threshold value 66 at the (n + 1) th time is at the bottom level of the position-differential value curve 64, and is the level at which the lightness change caused by yuzu skin starts to be detected. is there. At this time, the image F
As schematically shown in FIG. 18, the independent contour 70 is extracted in a screen obtained by binarizing the differential value distribution diagram. FIG.
Is an actual binarized image corresponding to the image F18. Independent of the boundary area of the light and dark pattern, an independent contour is extracted. It has been previously determined that the test painted surface is free of defects, and the independent contours that begin to be extracted in this way correspond to citron skin.

The threshold value of the (n + 1) th time at which the independent contour is started to be extracted is a value (level) at which a paint surface defect can be reliably extracted if it exists. However, as described above, even the yuzu skin is extracted. Therefore, step S in FIG.
At 26, the second predetermined value is added to the (n + 1) th threshold value. This second predetermined value is set in advance based on past data. The threshold value calculation device 130 sets the (n + 1) th threshold value to the second
The threshold is determined by adding a predetermined value. The threshold value adjusted in this manner (hereinafter simply referred to as a determination threshold value) is a level at which a painted surface defect is extracted although citron skin is not extracted, and is a level suitable for inspecting a painted surface defect. If the pixel-differential value distribution map is binarized using the decision threshold, a paint surface defect having a lightness change larger than that of citron skin is reliably detected as an independent contour in the visual field. According to the decision threshold,
Coating surface defects can be reliably extracted.

An embodiment in which the determined threshold value adjusted by the threshold value adjusting method according to the present embodiment is applied to a coating surface defect inspection process of an automobile body will be described. As schematically shown in FIG. 8, an automobile body 20 to be inspected is a pallet 3
It is sent in the direction of the arrow with being placed on 0. A plurality (for example, 10) of cameras 1-1 in a direction perpendicular to the transport direction
0 is arranged. The cameras 1 to 10 are the same as the cameras that photographed the test painted surface described above. The cameras 1 to 10 are connected to the above-described computer. The imaging device shown in FIG. 8 is also provided with a device (not shown) for illuminating the inspection object. This lighting device is similar to the above-described lighting device.

An automobile body 2 placed on a pallet 30
0 is conveyed in the direction of the arrow in FIG. During transportation, the vehicle body 20 reaches the shooting range of the cameras 1 to 10. In the present embodiment, ten cameras are provided in the width direction of the automobile. The car body 20 is photographed while being divided into ten ranges in the width direction. In this embodiment, the photographing range of each camera is set to 20 cm × 20 cm at each time. Therefore, each time the vehicle body 20 is conveyed slightly less than 20 cm, an image is taken with each camera. In practice, images are taken so that the fields of view slightly overlap. That is, cameras 1 to 10 are arranged at intervals of 19 cm, and each camera is
Is photographed every time 19 cm is transported. Actually, when the pallet 30 advances to a predetermined position, the first photographing is performed,
Thereafter, every time the camera advances 19 cm, a picture is taken. The number of times of shooting is called an address. The shooting is repeated until the rear end of the vehicle body 20 comes out of the shooting range of the camera. As described above, the entire area of the vehicle body 20 is covered by the number of cameras multiplied by the number of addresses, as shown in FIG. As schematically shown in FIG. 11, the field of view is photographed so as to overlap by d (d = 1 cm in this embodiment) in FIG. 11, so that the entire painted surface is reliably covered. .

FIG. 7 shows a binarized image obtained by binarizing the obtained differential value with respect to a certain visual field of the automobile body 20 using a decision threshold. The decision threshold is a threshold obtained by adding a second predetermined value to the (n + 1) th threshold. In this case, citron skin is not extracted. In FIG. 7, the independent contour is a paint surface defect. Using this determined threshold value, when inspecting the paint surface defect, it was confirmed that the defect-free one was not determined to be defective, and the defective one was not determined to be no defect, and that the determination could be made accurately. .

In the present embodiment, when a defect happens to be in the boundary area between light and dark of the illumination, the defect may not be extracted. Figure 10 shows a technique for avoiding this.
This will be described with reference to FIG.

FIG. 10 shows a pixel-differential value distribution diagram binarized using a decision threshold. In FIG. 10, reference numeral 72 denotes a pixel group having a differential value equal to or larger than the determination threshold, which is a brightness sudden change area. The brightness change area 72 is shown in pure white. In FIG. 10, reference numeral 74 denotes a pixel group having a differential value smaller than the determination threshold, which is a lightness slowly changing region. The lightness change region 74 is shown in black. When the defect is in the suddenly changing brightness region 72, an independent contour corresponding to the defect cannot be extracted. Therefore, in the present embodiment, 1
Shoot the field of view several times. At this time, the image is shot twice with a distance c larger than the width a in the traveling direction of the sharply changing area 72 in FIG. 10 and smaller than the pitch b in the traveling direction. Here, a set of visual fields shifted by c is referred to as “one visual field”. FIG. 11 schematically shows the overlapping of the photographing ranges.
Shoot twice. The overlapping area is the field of view 1. Similarly, field of view 2
Are photographed twice in the range 2-1 and 2-2. The overlapping area is the field of view 2. The visual field 1 and the visual field 2 overlap by d (about 1 cm as described above). As described above, when the image is shot twice with a slight distance (c) shifted for each field of view, the image is shot with a shift of the distance c even if the defect is located in the suddenly changing brightness area 72 on any of the shooting screens. On other photographing screens, the defect appears in the slowly changing brightness region 74. FIG. 10 exemplifies that the paint surface defect 76 which was in the non-brightness changing area 72 in one photographing screen is in the slowly changing area 74 in another photographing screen. As a result, an independent contour having a differential value equal to or larger than the threshold value is extracted from the other lightness screen in the lightness variation 74, and a paint surface defect is extracted. By photographing a plurality of times while slightly shifting one field of view, it is possible to extract, without leaking, if a paint surface defect exists in the field of view.

The illumination pattern may be in the form of parallel stripes as shown in FIG. 10, or may be in the form of lattice stripes. In the case of a grid pattern, it is preferable to take two or more images in one visual field. By setting all the positions of the painted surface to be within the slowly changing brightness region 74 on any of the photographing screens, it is possible to extract defects without leaking.

In step S16, an initial threshold value set in advance based on past data is used, but the present invention is not limited to this. The setting may be made using the peak value of the differential value obtained in step S14.

In step S22, a first predetermined value set in advance based on past data is used, but the present invention is not limited to this. The mode may be set using the difference between the peak value and the bottom value of the differential value.

In step S26, the second predetermined value set in advance based on past data is used, but the present invention is not limited to this. The mode may be set by using the height of a wave having a fine cycle appearing in the distribution map information of the differential value.

In the present embodiment, when adjusting the threshold value, an automobile body that has been confirmed in advance to be free from defects is used. Alternatively, the determination threshold may be adjusted using a sample that has been specifically prepared for threshold adjustment.

[Brief description of the drawings]

FIG. 1 is a diagram schematically illustrating the method of the present invention.

FIG. 2 is a flowchart illustrating a threshold adjustment procedure and a schematic image of the embodiment.

FIG. 3 is a camera photographing screen of a painted surface having no defect.

FIG. 4 is a binarized screen of differential values of a paint surface having no defect when a threshold value is large.

FIG. 5 is a binarized screen of differential values of a paint surface having no defect when a threshold value is large.

FIG. 6 is a binarized screen of a differential value of a paint surface having no defect when a threshold value is small.

FIG. 7 is a screen obtained by binarizing a differential value of a paint surface having a defect with a determined threshold.

FIG. 8 is a view schematically showing a coating surface defect inspection process of an automobile body.

FIG. 9 is a diagram showing a shooting range of each camera for each time.

FIG. 10 is a diagram schematically showing a binarized screen of a painted surface having a defect.

FIG. 11 is a diagram showing overlap of visual fields.

FIG. 12 is a diagram schematically showing a configuration of an electronic computer.

[Explanation of symbols]

 54, 56, 60, 64: position-differential value curve 66: n + 1th threshold 70: independent contour

 ──────────────────────────────────────────────────続 き Continued on the front page F term (reference) 2G051 AA89 AB07 AB12 BB20 CA03 CA07 CC07 DA06 EA08 EA11 EA14 EB01 EB02 ED04 ED07 4D075 AA81 DB02 DC12 4F042 AA09 DH09 5B057 AA04 BA02 CA08 CA12 CA16 CB16 CE03 DA03 DB02 DB09 DC16

Claims (1)

[Claims] A step of obtaining a lightness of each pixel by photographing a paint surface having no defect; a step of spatially differentiating the obtained lightness of each pixel to obtain a differential value for each pixel; Setting, and comparing the differential value of each pixel with a threshold,
A step of extracting a contour that is a contour of a pixel group whose differential value is equal to or larger than a threshold and is independent in the field of view; a step of subtracting the threshold from the threshold by a first predetermined value; And repeating the cycle until an independent contour is extracted in the extraction step as a unit cycle, and determining a threshold value by adding a second predetermined value to a threshold value when the independent contour is extracted in the extraction step. A threshold adjustment method used in the paint surface defect inspection process.