JP4030812B2 - Appearance inspection apparatus and appearance inspection method - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to an appearance inspection apparatus and an appearance inspection method capable of inspecting an appearance even when an object to be inspected is uneven due to rotation and has unevenness on the surface.
[0002]
[Prior art]
In an apparatus that automatically inspects the appearance of a product using image recognition technology, a difference inspection is performed to compare an inspection image obtained by imaging a product to be inspected with a reference image registered in advance, A pass / fail decision is made based on the result. For inspection targets whose orientations are uneven due to rotation, pattern processing is performed after image processing is performed to rotate the inspection image so that the orientation is the same as the reference image, and so-called rotation is determined based on the difference. A matching difference check is performed.
[0003]
[Problems to be solved by the invention]
However, when the orientation becomes uneven due to rotation and the surface to be inspected is uneven, such as a lid with a tab, for example, the contrast difference caused by the surface unevenness is in the direction (rotation) of the inspection object. Even if the rotational position changes, it is extremely difficult to capture a uniform inspection image. Therefore, when the above-described rotational matching difference inspection is applied, there is a problem that the defect detection ability greatly changes depending on the rotational position. .
The present invention has been made in view of such circumstances, and it is possible to inspect an appearance even for an inspection object sensitive to illumination or the like whose orientations are uneven due to rotation and whose surface has irregularities. An object is to provide an appearance inspection apparatus and an appearance inspection method.
[0004]
[Means for Solving the Problems]
In order to achieve the above object, according to the first aspect of the present invention, the standard position and rotation angle of the inspection object are obtained from each captured image obtained by imaging a large number of normal inspection objects under the same conditions as the inspection. In accordance with the reference standard position for each rotation angle obtained, reference image data for each angle representing the luminance distribution characteristics of each pixel for each rotation angle is extracted, and optimal abnormality detection is performed under the same conditions as the inspection The coefficient for generating the upper limit reference image and the lower limit reference image that embody the level is calculated for each rotation angle, and the reference image data for each angle is stored, and the upper reference image is obtained from the reference image data for each angle in preparation for inspection. The standard position and rotation angle of the inspection object are detected from the generation process for generating the lower limit reference image and the inspection image obtained by imaging the inspection object at the time of performing the inspection, and corresponding to the rotation angle of the inspection object. The combined reference standard position of the reference image and the lower limit reference image to a standard position of the test object by comparing the brightness of the test image and the two reference images pixel by pixel, abnormality detection and if it exceeds the upper limit reference image or the lower reference image And a determination process for determining pass / fail.
[0005]
In the invention according to claim 2, the standard position and the rotation angle of the inspection object are obtained from each captured image obtained by imaging a large number of normal inspection objects under the same conditions as the inspection, and each rotation angle obtained Creation means for creating image management data in which each reference standard position and captured image are stored in association with each other, and each angle representing a reference standard position for each rotation angle and a luminance distribution characteristic in pixel units from the image management data Generating means for extracting reference image data, and generating an upper limit reference image and a lower limit reference image for realizing an optimum abnormality detection level under the same conditions as the inspection based on the reference image data for each angle for each rotation angle; Detecting means for detecting a standard position and a rotation angle of the inspection object from an inspection image obtained by imaging the inspection object at the time of performing the inspection, and an upper limit reference corresponding to the rotation angle of the inspection object detected by the detection means After comparing the reference standard position of the image and the lower limit reference image with the standard position of the inspection object, the brightness of both the reference image and the inspection image are compared in pixel units, and an abnormality is detected when the upper limit reference image or the lower limit reference image is exceeded. Determining means for determining pass / fail, and the generating means corrects the upper limit reference image and the lower limit reference image in accordance with an abnormal state based on an abnormality detection result of the determination means, and the corrected reference image is used as a new reference image. It is characterized by producing | generating as .
[0007]
In the present invention, an upper limit reference image and a lower limit reference image that realize an optimum abnormality detection level under the same conditions as the inspection are generated for each rotation angle based on each captured image obtained by imaging a large number of inspection objects in advance. When the standard position and rotation angle of the inspection object are detected from the inspection image obtained by imaging the inspection object, the reference standard positions of the upper limit reference image and the lower limit reference image corresponding to the detected rotation angle of the inspection object are determined. The brightness of both reference images and the inspection image are compared after matching the standard position of the inspection object, and an abnormality is detected when the upper reference image or the lower reference image is exceeded. It is possible to inspect the appearance even for an inspection object having unevenness.
[0008]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
(1) Overall Configuration FIG. 1 is a diagram showing an overall configuration of an appearance inspection apparatus according to an embodiment. In this figure, reference numeral 1 denotes a lid to be inspected, which is placed on the transport belt 2 and transported. The lid body 1 is not uniform in orientation and is uneven, and the surface has unevenness due to pressing. Reference numeral 3 denotes an imaging camera that images the lid 1, and reference numeral 4 denotes illumination means that emits strobe light during imaging.
[0009]
Reference numeral 5 denotes a housing that holds and fixes the imaging camera 3, and 6 denotes an imaging synchronization sensor that detects the transported lid body 1 and generates an imaging synchronization signal. Reference numeral 7 denotes a selection synchronization sensor that detects the transported lid body 1 and generates a selection synchronization signal. Denoted at 8 is a selecting means for discharging the lid body 1 determined to be rejected in the appearance inspection to the rejected product conveying belt 9 side. A sensor amplifier 11 shapes the imaging synchronization signal and the selection synchronization signal output from the imaging synchronization sensor 6 and the selection synchronization sensor 7, respectively, to a predetermined level and outputs them.
[0010]
Reference numeral 12 denotes synchronization control means. The synchronization control unit 12 gives an imaging instruction to the camera control unit 13 when an imaging synchronization signal is input via the sensor amplifier 11, and gives an emission instruction to the illumination control unit 14 to take an image of the lid 1. Let The synchronization control means 12 gives a sorting instruction to the rejected product sorting synchronization means 16 when a sorting synchronization signal is input via the sensor amplifier 11. The camera control unit 13 causes the imaging camera 3 to image the lid 1 in response to an imaging instruction from the synchronization control unit 12, and sends an inspection image obtained thereby to the image processing inspection unit 15.
[0011]
The illumination control unit 14 causes the illumination unit 4 to emit strobe light in response to an imaging instruction from the synchronization control unit 12. The image processing inspection unit 15 includes a CPU, an image memory, and the like, and performs a comparison inspection between an inspection image obtained by imaging the lid 1 and a reference image registered in advance, and makes a pass / fail determination based on the result. The processing operation of the image processing inspection means 15 according to the gist of the present invention will be described in detail later. Reference numeral 16 denotes a rejected product selection synchronization means. When a failure determination is input from the image processing inspection means 15, a rejected product discharge instruction is given to the selection control means 17 in accordance with a selection instruction given from the synchronization control means 12. give. The sorting control unit 17 instructs the sorting unit 8 to discharge the lid 1 to the rejected product transport belt 9 side in accordance with the rejected product discharge instruction.
[0012]
(2) Overall Operation Next, the overall operation of the appearance inspection apparatus having the above configuration will be described with reference to FIG. First, in step SA1, preparation processing is performed before the appearance inspection apparatus is actually operated. The preparation process is a process for generating a reference image for comparison with an inspection image, which will be described later.
After the preparation process is completed, the appearance inspection apparatus is operated, and for example, the lid body 1 transported at a speed of 23 pieces per second is detected by the imaging synchronization sensor 6, whereby the imaging camera is synchronized with the strobe emission of the illumination means 4. When the lid 1 to be inspected 3 is imaged, the image processing inspection means 15 captures the inspection image imaged in step SA2.
[0013]
Next, in step SA3, the center position of the lid 1 is detected based on the taken inspection image, and in the subsequent step SA4, the rotation angle of the lid 1 is detected. In step SA5, the reference image is compared with the inspection image after matching the reference center position of the reference image (corresponding to Max and Min reference images described later) corresponding to the detected rotation angle and the center position detected by the inspection image. To extract abnormal parts.
[0014]
In the abnormal part extraction, the luminance of the inspection image is compared with the reference image in units of pixels, and when the luminance of the inspection image exceeds the Max and Min reference image ranges, it is determined as an abnormal pixel. Since the set of abnormal pixels indicates an abnormal portion, in step SA6, a pass / fail determination is performed to determine the presence or absence of a defect by comprehensively evaluating the size and shape of the set of abnormal pixels, the magnitude of the difference in brightness, and the like. In step SA7, when the pass determination is made, the corresponding lid body 1 is directly transported to the next process by the transport belt 2. On the other hand, when the pass determination is made, the sorting means 8 is switched by the sorting synchronization signal of the object. Driven to discharge the lid 1 to the rejected product transport belt 9 side.
[0015]
(3) Operation of Preparation Process Next, the operation of the preparation process performed in step SA1 will be described with reference to FIGS. The preparation process generates a reference image for comparison with the inspection image, and includes “image collection process”, “statistical information collection process”, “coefficient evaluation correction process”, and “inspection preparation process”. The contents of each of these processes will be described below.
[0016]
(1) Image collection process When the image collection process is executed, the process proceeds to step SB1 in FIG. 3, and a large number of lids 1 are imaged under the same conditions as in the inspection. In the captured image, if the inspection diameter is 300 pixels, the rotation direction is divided into 1000 (the divided number is referred to as the rotation angle) in order to set the perimeter length resolution of that portion to one pixel level. Therefore, in order to generate a reference image using 300 images per division, 300,000 images are captured. This can be realized, for example, by preparing 3000 normal lids 1, transporting them in a random orientation, capturing images 100 times, and saving them in an auxiliary storage device. You may image 300,000 sheets at a time. An example of the captured image 10 of the lid 1 captured in this way is shown in FIG.
[0017]
If processing time is allowed in the process of capturing 300,000 images, step SB2 may be performed every time one lid 1 is imaged. Otherwise, the image saved after imaging is read out from the auxiliary storage device. The position of the lid 1 is detected from this image, and the center position is determined. Specifically, in the case of the captured image 10 illustrated in FIG. 4, the outer periphery 101 serving as the inspection region is recognized to detect the position of the lid 1, and whether or not the small circle 102 is positioned at the center of the outer periphery 101. Check to make sure. Using the pixel recognized as the outer periphery 101, the center coordinate value and the radius of the outer periphery 101 are calculated by least square circle approximation to determine the center position.
[0018]
Next, in step SB3, the rotation angle of the lid 1 is determined. In the case of the captured image 10 illustrated in FIG. 4, the straight lines 103 and 104 formed by the press at the reference position for the rotation of the surface of the lid 1 are detected, and the boundary pixels of these lines are extracted. After removing abnormal points from the extracted boundary pixels, respective approximate lines are obtained by the method of least squares. The direction (rotation angle) of the lid 1 is calculated using the inclination of the central axis of the two approximate lines and the direction of the intersection of the two approximate lines. In step SB4, the captured image 10 is saved in the auxiliary storage device by sorting for each calculated rotation angle. A table in which the calculated rotation angle, center position, and captured image are associated with each other may be created and image management may be performed.
[0019]
(2) Statistical information collection processing When the captured images 10 are sorted for each rotation angle in the processing of steps SB1 to SB4, statistical information collection processing shown in FIG. 5 is executed. In steps SC1 to SC3 of this process, images of the same angle (referred to as group images) are read from the auxiliary storage device, and the position of the lid 1 is detected by recognizing the outer periphery 101 as in step SB2 described above. Using the coordinate value of the pixel corresponding to 101, the center coordinate value and radius of the outer periphery 101 are calculated by least square circle approximation (if the center position of the image has already been obtained, that data is used). Averaging is performed to determine the reference center position of the group image.
[0020]
Furthermore, in the group image, the difference between the reference center position and the center position of each image is corrected, and the luminance (including saturation and hue in the case of a color image) statistics for each pixel forming the inspection area, and differentiation processing The same statistics are taken for the group images, and characteristic data is generated. This characteristic data is the average value, standard deviation, maximum value, minimum value, cumulative 1 regarding the luminance (including saturation and hue in the case of a color image) of each pixel that forms the inspection region in both the raw image and the differential image. It consists of% point value and cumulative 99% point value.
If the cumulative 1% point value and the cumulative 99% point value are within the range of 2.5σ, this is an index indicating that the lid 1 and the imaging system used empirically are stable. In step SC4, the reference center position and characteristic data of the group image obtained as described above are obtained for the entire circumference, and this is saved in the auxiliary storage device as reference image information for each angle.
[0021]
(3) Coefficient evaluation correction processing When the reference image information for each angle is obtained in this way, the coefficient evaluation correction processing shown in FIG. 6 is executed. In the coefficient evaluation correction process, first, reference image information for each angle is read out in step SD1, and in the subsequent step SD2, each of the pixels forming the group image is subjected to calculations of the following expressions (1) and (2) to perform each rotation. A Max and Min reference image for each angle is generated. That is,
Max reference image = average value + standard deviation × Nmax (1)
Min reference image = average value−standard deviation × Nmin (2)
Here, as the average value and the standard deviation, the characteristic data for each angle obtained in the statistical information collection process described above is used. The coefficients Nmax and Nmin are set to “2.8” for the time being when there is no appropriate coefficient from the previous data. Thus, the Max reference image is an image that is 2.8σ higher than the average luminance, while the Min reference image is an image that is 2.8σ lower than the average luminance.
[0022]
Next, in step SD3, the captured image 10 sorted for each rotation angle in the image collection process described above is read, and in the subsequent step SD4, the corresponding Max reference image and Min of the same angle are read out from the captured image 10. Compared with the reference image, abnormal pixels that do not fall within this range are extracted. An image in which the abnormal pixel is detected is recorded. This is performed for all the target images.
After the target image processing, in step SD5, the captured image 10 extracted from the abnormal part is determined. If the image may be rejected, the process proceeds to step SD6, where the captured image 10 is sorted for each rotation angle. The captured image 10 is removed. Next, in step SD7, in response to the removal of the captured image 10, the above-described statistical information collection process is executed again to create new reference image information for each angle.
[0023]
Thereafter, the reference image information for each angle is updated, and the above steps SD1 to SD7 are repeated according to the updated reference image information for each angle. Then, as a result of automatic comparison between the captured image 10 and the Max and Min reference images over the entire circumference, if no image that may be rejected is extracted, the coefficient evaluation correction process is completed.
On the other hand, when abnormality is detected excessively, the process proceeds to step SD13, and the coefficient is reviewed according to the detection condition in step SD14. Specifically, the following measures (a) to (e) are followed.
[0024]
(A) When abnormal excessive detection is conspicuous on the whole, the Max and Min reference images are re-created by changing the above-described coefficients Nmax and Nmin and evaluated again.
[0025]
(B) When abnormal excessive detection is conspicuous at the information change point, Max and Min reference images are generated based on the following equations (3) and (4).
Max reference image = average value + standard deviation × Nmax + Bmax (3)
Min reference image = average value−standard deviation × Nmin−Bmin (4)
Bmax and Bmin are correction values obtained by multiplying the difference between the maximum luminance and the minimum luminance around the target pixel (including saturation and hue in the case of color) by constants fBmax and fBin, respectively.
[0026]
(C) When abnormal excessive detection is conspicuous in the deviation in the rotation direction, Max and Min reference images are generated based on the following equations (5) and (6).
Max reference image = average value + Kmax standard deviation × Nmax (5)
Min reference image = average value−Kmin standard deviation × Nmin (6)
For the Kmax and Kmin standard deviations, the maximum and minimum standard deviation values around the rotation direction of the target pixel are used.
[0027]
(D) When abnormal excessive detection is conspicuous in the positional deviation, Max and Min reference images are generated based on the following equations (7) and (8).
Max reference image = average value + Imax standard deviation × Nmax (7)
Min reference image = average value−Imin standard deviation × Nmin (8)
For the Imax and Imin standard deviations, the maximum value and the minimum value of the standard deviation values around the direction of displacement of the target pixel are used.
[0028]
(E) When any of the above (a) to (d) is combined, the Max and Min reference images are generated based on (9) and (10).
Max reference image = average value + KImax standard deviation × Nmax + Bmax (9)
Min reference image = average value−KImin standard deviation × Nmin−Bmin (10) Note that the KImax and KImin standard deviations are the maximum and minimum values of the peripheral standard deviation values obtained by combining the positional deviation of the target pixel with the rotational direction deviation. Value. In this case, correction values Bmax and Bmin introduced when abnormal excessive detection is conspicuous at a change point of information are corrected and used at a ratio according to the current state.
In this situation, if the lid 1 that is desired to be rejected at the product inspection level is flowed many times in a random direction and all the lids are surely repelled, it will be evaluated as an appropriate inspection device.
When a defective product is detected abnormally, the process proceeds to step SD11, and in step SD12, (a) to (e) are reviewed to obtain an appropriate coefficient.
[0029]
(4) Inspection preparation process In the inspection preparation process, in step SE1, the reference image information for each angle corrected by the coefficient evaluation correction process described above is read, and in the subsequent step SE2, the reference image information for each angle is read based on the read reference image information for each angle. As shown in FIG. 1, an AVE image 55, a Max reference image 56, and a Min reference image 57 are generated for each angle. The AVE image is an image formed by average luminance in the characteristic data, and is used to verify at any time that the conditions of the imaging system and the illumination system have not changed.
[0030]
When the Max reference image 56 and the Min reference image 57 for each angle are generated to prepare for inspection and the appearance inspection apparatus is operated, the lid 1 conveyed by the conveyance belt 2 is detected and imaged by the imaging synchronization sensor 6. The Based on the captured inspection image, the center position and rotation angle of the lid body 1 are detected, and the reference center position of the Max and Min reference images corresponding to the detected rotation angle is matched with the detected center position of the lid body 1. The Max, Min reference image and the inspection image are compared, and an abnormal part is detected and inspected.
[0031]
That is, the brightness of the Max, Min reference image and the inspection image are compared on a pixel basis, and if it exceeds the Max, Min reference image, it is determined as an abnormal pixel. And, as a result of the pass / fail judgment that judges the presence or absence of defects by comprehensively evaluating the size and shape of the set of abnormal pixels, the size of the difference in brightness, etc., the orientation becomes uneven due to rotation, and the surface is uneven. Appearance inspection is possible even for existing inspection objects.
[0032]
As described above, according to the above-described embodiment, in the inspection preparation stage, a large number of normal lids 1 are imaged under the same conditions as during the inspection, and the center position of the lid 1 is determined from each captured image. After obtaining the rotation angle, and creating image management data in which the reference center position and the captured image obtained for each rotation angle are stored in association with each other, the image management data is subjected to statistical processing, and each rotation angle is obtained. Max reference image that extracts reference image information for each angle representing the reference position of each pixel and luminance distribution characteristics in units of pixels, and realizes an optimal abnormality detection under the same conditions as the inspection based on the extracted reference image information for each angle A Min reference image is generated for each rotation angle.
[0033]
Then, when entering the inspection stage, the center position and the rotation angle of the lid 1 are detected based on the imaged inspection image, and the Max, Min reference image corresponding to the detected rotation angle is compared with the inspection image, and the abnormal portion Extract and inspect. That is, the luminance of the Max, Min reference image and the inspection image are compared in pixel units, and when the range of the Max, Min reference image is exceeded, it is determined as an abnormal pixel, and the size, shape, and brightness of the abnormal pixel set are determined. Since the presence of defects is judged by comprehensively evaluating the size of the difference, etc., the appearance inspection that cannot be achieved by the conventional rotation matching difference inspection, that is, the direction becomes uneven due to rotation, and there are unevenness on the surface Even if it is an inspection object to be checked, an appearance inspection can be performed.
[0034]
In addition, according to the present embodiment, since the rotation matching difference inspection is not performed, an expensive image processor for rotation matching difference inspection is not essential, and an effect of contributing to a reduction in product cost is also invited.
Furthermore, in this embodiment, since the Max and Min reference images are generated under the same conditions as those at the time of inspection, there is an effect that nonuniformity of illumination and optical systems and distortion of the imaging system can be allowed.
In addition, since the user can arbitrarily adjust the coefficient for generating the Max and Min reference images and set it to a desired abnormality detection level, it can respond immediately to changes in the inspection object. An effect is also obtained.
[0035]
【The invention's effect】
According to the first and second aspects of the invention, the standard position and the rotation angle of the inspection object are obtained from each captured image obtained by imaging a large number of normal inspection objects under the same conditions as the inspection, and further obtained. The reference image data for each angle representing the reference standard position for each rotation angle and the luminance distribution characteristic in pixel units is extracted according to the reference standard position and the captured image for each rotation angle, and the reference image for each angle is extracted. Based on the data, an upper limit reference image and a lower limit reference image that realize an optimum abnormality detection level under the same conditions as in the inspection are generated for each rotation angle. When the standard position and rotation angle of the inspection object are detected from the inspection image obtained by imaging the inspection object, the reference standard positions of the upper limit reference image and the lower limit reference image corresponding to the detected rotation angle of the inspection object are inspected. The brightness of both reference image and inspection image is compared in pixel units after matching the target standard position, and if the upper limit reference image or the lower limit reference image is exceeded, an abnormality is detected and pass / fail judgment is made, so the orientation is uneven due to rotation In addition, the appearance can be inspected even if there is an unevenness on the surface, or even if there is some illumination, optical system nonuniformity, or imaging system distortion.
[Brief description of the drawings]
FIG. 1 is a diagram showing a configuration of an appearance inspection apparatus according to an embodiment of the present invention.
FIG. 2 is a flowchart showing the overall operation of the appearance inspection apparatus.
FIG. 3 is a flowchart showing an operation of image collection processing.
FIG. 4 is a diagram illustrating an example of a captured image.
FIG. 5 is a flowchart showing the operation of statistical information collection processing;
FIG. 6 is a flowchart showing the operation of coefficient evaluation correction processing.
FIG. 7 is a flowchart showing the operation of an inspection preparation process.
FIG. 8 is a diagram illustrating an example of an AVE image, a Max reference image, and a Min reference image.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 Cover body 2 Conveyance belt 2a Conveyance guide 3 Imaging camera 4 Illumination 5 Housing 6 Imaging synchronization sensor 7 Sorting synchronization sensor 8 Sorting means 9 Defective article conveyance belt 11 Sensor amplifier 12 Synchronization control means 13 Camera control means 14 Illumination control means 15 Image Processing inspection means 16 Defective product selection synchronization means 17 Selection control means

Claims (2)

Obtain the standard position and rotation angle of the inspection object from each captured image obtained by imaging a large number of normal inspection objects under the same conditions as the inspection, and adjust the obtained standard reference position for each rotation angle. By extracting reference image data for each angle representing the luminance distribution characteristics of each pixel for each rotation angle, an upper limit reference image and a lower limit reference image that realize the optimum abnormality detection level under the same conditions as the inspection are displayed for each rotation angle. A process for determining the coefficient to be generated and storing the reference image data for each angle, and a generation process for generating an upper limit reference image and a lower limit reference image from the reference image data for each angle in preparation for inspection,
The standard position and rotation angle of the inspection object are detected from the inspection image obtained by imaging the inspection object at the time of performing the inspection, and the reference standard position of the upper and lower reference images corresponding to the rotation angle of the inspection object is inspected. And comparing the brightness of both the reference image and the inspection image in pixel units in accordance with the standard position of the image, and determining whether or not an abnormality is detected when the upper limit reference image or the lower limit reference image is exceeded. Appearance inspection method. Determine the standard position and rotation angle of the inspection object from each captured image obtained by imaging many normal inspection objects under the same conditions as the inspection, and obtain the standard reference position and the captured image for each rotation angle obtained. Creating means for creating image management data stored in association with each other;
  The reference image data for each angle representing the reference standard position for each rotation angle and the luminance distribution characteristics in pixel units is extracted from the image management data, and the optimum under the same conditions as at the time of inspection based on the reference image data for each angle. Generating means for generating an upper limit reference image and a lower limit reference image for realizing an abnormal detection level for each rotation angle;
  Detecting means for detecting a standard position and a rotation angle of the inspection object from an inspection image obtained by imaging the inspection object at the time of performing the inspection;
  The reference standard positions of the upper limit reference image and the lower limit reference image corresponding to the rotation angle of the inspection object detected by this detection means are matched with the standard position of the inspection object, and then the brightness of both the reference image and the inspection image is compared in pixel units. And determining means for detecting pass / fail when the upper limit reference image or the lower limit reference image is exceeded,
The visual inspection apparatus, wherein the generation unit corrects the upper limit reference image and the lower limit reference image according to an abnormal state based on an abnormality detection result of the determination unit, and generates the corrected reference image as a new reference image. .

Images