JP7789310B2 - Surface defect inspection device and surface defect inspection method - Google Patents

Description

The present invention relates to a surface defect inspection device and method suitable for inspecting the condition of, for example, the paint on the surface of an automobile body.

Traditionally, the painted surface of automobile bodies has been visually inspected by inspectors during the painting process. However, visual inspection by inspectors is labor-intensive, and there is variation between individuals, which can lead to inspection errors or oversights. Furthermore, visual inspection by inspectors takes a lot of time, and labor costs are one factor that increase product production costs. Therefore, there is a demand for the automation of visual inspection, and in recent years, progress has been made in developing surface defect inspection equipment that can perform optical, automated inspections.

For example, Patent Document 1 describes a surface defect inspection device that photographs the surface to be inspected using the imaging means while moving the imaging means and the surface to be inspected relative to each other, extracts defect candidates from the multiple images captured, and determines that a defect exists if the movement distance and movement angle of the defect candidate between at least two or more images captured at different times are within a predetermined range.

Japanese Patent Application Laid-Open No. 2020-118572

However, the surface defect inspection device described in Patent Document 1 can make a judgment when the movement trajectory of a defect candidate is linear, but has difficulty making a judgment when the movement trajectory of the defect candidate is curved. For example, if the camera angle is changed in accordance with the movement of the surface to be inspected in order to maintain a high brightness that makes it easier to extract defect candidates, the movement trajectory of the defect candidate will become curved. Therefore, there has been a demand for the ability to make defect judgments with high accuracy even when the movement trajectory of the defect candidate is curved.

The present invention was developed in response to these problems, and aims to provide a surface defect inspection device and method that can determine defects with high accuracy even when the movement trajectory of the defect candidate is curved, and even when the camera angle changes in accordance with the movement of the surface being inspected.

The surface defect inspection device of the present invention comprises a light source that irradiates the surface to be inspected with light, an imaging means that captures the surface to be inspected illuminated by the light source to obtain an image, a moving means that moves the position of the surface to be inspected relative to the imaging means, a defect candidate extraction means that extracts defect candidates from multiple images captured by the imaging means at any given time while the moving means moves the imaging means and the surface to be inspected relatively, and a defect detection means that detects defects based on the defect candidates extracted by the defect candidate extraction means. The defect detection means divides the surface to be inspected into multiple sections, and determines that a defect exists if a defect candidate extracted by the defect candidate extraction means exists in each section image area corresponding to the same section of the surface to be inspected for a reference determination number or more of images captured by the imaging means at different capture times.

The surface defect inspection method of the present invention includes an imaging step of obtaining an image by imaging the surface to be inspected illuminated with light from a light source, a defect candidate extraction step of extracting defect candidates from the image captured by the imaging step, and a defect detection step of detecting defects based on the defect candidates extracted by the defect candidate extraction step. In the defect detection step, the surface to be inspected is divided into a plurality of sections, and for a number of images captured by the imaging means at different imaging times or greater than a reference determination number, a defect is determined to exist if a defect candidate extracted by the defect candidate extraction means exists in each section image area corresponding to the same section of the surface to be inspected.

According to this invention, the surface to be inspected is divided into multiple sections, and for images captured by the imaging means at different capture times equal to or greater than the reference determination number, if a defect candidate extracted by the defect candidate extraction means is present in each section image area corresponding to the same section of the surface to be inspected, it is determined that the defect candidate is a defect. Therefore, it is possible to determine whether a defect candidate is a defect based on the relationship between the movement trajectory of the defect candidate and the movement trajectory of the section of the surface to be inspected. Therefore, even if the movement trajectory of the defect candidate is curved, for example, if the camera angle is changed in accordance with the movement of the surface to be inspected relative to the camera in order to increase the brightness of the defect candidate, it is possible to determine whether the defect is a defect with high accuracy.

In addition, if a defect candidate exists within a specified range from the boundary line of a partition image area corresponding to a partition on the inspected surface, the defective partition image area and the adjacent partition image area are considered to be a single partition image area corresponding to a single partition on the inspected surface, thereby further improving the accuracy of defect determination.

1 is a diagram illustrating the overall configuration of a surface defect inspection device according to an embodiment of the present invention. FIG. 2 is a block diagram showing a configuration of a defect detection unit shown in FIG. 1 . FIG. 10 is a diagram illustrating an example of an image obtained by a preprocessing unit. FIG. 10 is a diagram illustrating an example of an image obtained by an image dividing unit. 10A and 10B are diagrams illustrating examples of binarized images binarized by changing the binarization threshold value. FIG. 10 is a diagram illustrating an example of a defect candidate extracted by a secondary extraction unit. FIG. 10 is a conceptual diagram in which the area around the reflected mirror image of the light source is divided according to the distance from the reflected mirror image of the light source. FIG. 1 is a diagram showing a state in which a surface to be inspected is divided into a plurality of sections. 10 is a diagram illustrating how, in the counting means, when a defect candidate exists within a predetermined range from the boundary line, the determination is made by expanding the range to include adjacent section image areas. FIG. 10 is a diagram for explaining how a determination unit determines a defect based on the number of times a defect candidate is detected for each block number. FIG. 1 is a diagram illustrating a procedure of a surface defect inspection method according to an embodiment of the present invention.

Embodiments of the present invention will be described in detail below with reference to the drawings.

Figure 1 shows the overall configuration of a surface defect inspection device 1 according to one embodiment of the present invention. This surface defect inspection device 1 detects defects present on the surface of the surface M to be inspected, which may be, for example, the painted surface of an automobile body.

The surface defect inspection device 1 includes, for example, a light source 10 that irradiates light onto the surface M to be inspected, an imaging means 20 that captures the surface M illuminated by the light source 10 to obtain an image, a moving means 30 that moves the position of the surface M to be inspected relative to the imaging means 20, a defect candidate extraction means 40 that extracts defect candidates from multiple images captured by the imaging means 20 at any given time while moving the imaging means 20 and the surface M to be inspected relative to each other using the moving means 30, a defect detection means 50 that detects defects based on the defect candidates extracted by the defect candidate extraction means, and a display means 60 that displays the detection results from the defect detection means 50.

The light source 10 is preferably a straight-tube lighting fixture, such as a straight-tube fluorescent lamp, LED light, or laser light, but is not limited to these types of lighting. Furthermore, since automobile bodies come in a wide variety of colors, it is preferable to use a white light source. It is preferable to arrange multiple light sources 10 relative to the surface M to be inspected, for example, so that the surface M can be observed from multiple directions.

The photographing means 20 has a camera 21, such as a CCD camera, that can capture digital images. The camera 21 is, for example, positioned facing the light source 10 and configured to photograph the reflected mirror image of the light source 10 and its surrounding area. It is also preferable that the angle of the camera 21 be configured so that it can be changed in accordance with the movement of the position of the surface M to be inspected relative to the camera 21. This is because if the angle of the camera 21 is fixed, the brightness of defect candidates may be low depending on the location of the surface M to be inspected, making it difficult to extract the defect candidates.

The moving means 30 moves the position of the surface M to be inspected relative to the imaging means 20 by moving at least one of the imaging means 20 and the surface M to be inspected. For example, the surface M to be inspected may be transported in one direction at a constant speed by a transport means such as a conveyor, or the camera 21 of the imaging means 20 may be moved in one direction at a constant speed. Note that Figure 1 shows a case in which the surface M to be inspected is moved by a conveyor or the like. The defect candidate extraction means 40 and defect detection means 50 are, for example, configured by a computer, and are configured to extract defect candidates or detect defects through image processing. The display means 60 is, for example, configured by a display, and is configured to display the center of gravity of the defect with a circle or the like.

(Defect candidate extraction means 40)
Fig. 2 shows the configuration of the defect candidate extraction means 40 and the defect detection means 50 shown in Fig. 1. The defect candidate extraction means 40 preferably includes an image storage means 41 such as a memory for storing a plurality of images captured by the photographing means 20 at different photographing times while moving the photographing means 20 and the surface M to be inspected relatively, a preprocessing means 42 for preprocessing the images captured by the photographing means 20, an image division means 43 for dividing an area including the reflected mirror image of the light source 10 into a plurality of divided images in the longitudinal direction of an extraction target image based on the image obtained by the photographing means 20, a primary extraction means 44 for binarizing the extraction target image based on the image obtained by the photographing means 20 and extracting primary defect candidates for extracting defect candidates, a primary defect candidate storage means 45 for storing the primary defect candidates, a secondary extraction means 46 for extracting defect candidates from the primary defect candidates, and a defect candidate storage means 47 for storing the defect candidates.

The pre-processing means 42 converts the image obtained by the imaging means 20 into a density image such as a grayscale image to extract the area of the surface M to be inspected and reduce noise. Examples of noise reduction methods include a Gaussian filter and a median filter. Figure 3 shows an example of an image obtained by the pre-processing means 42. In Figure 3, the white portion is the reflected mirror image of the light source 10.

The image segmentation means 43 is preferably configured to, for example, cut out an area including one reflected mirror image of the light source 10 and its surrounding area from the image obtained by the preprocessing means 42, which is the image to be extracted, and segment the reflected mirror image into multiple segments along its length. Because the luminance differs between the center and ends of the reflected mirror image along its length, segmenting the image allows the primary extraction means 44 to perform binarization processing with high accuracy. Figure 4 shows an example of an image segmented by the image segmentation means 43. In Figure 4, the white areas represent the reflected mirror image of the light source 10. Note that Figure 4 shows gaps between each image to clearly show the segmentation. Also, Figure 4 shows the case where the reflected mirror image is segmented into six segments along its length by the image segmentation means 43, but the number of segments can be set as desired. The number of segments is preferably in the range of 2 to 12, for example.

The primary extraction means 44 is configured to, for example, sequentially perform binarization processing on images processed by the preprocessing means 42 (images to be extracted) while sequentially changing the binarization threshold in one direction from the binarization reference value, compare the multiple binarized images obtained in the order of processing, and determine as a threshold image the binarized image when the number of defect candidate points changes from less than the predetermined reference number to equal to or greater than the predetermined reference number, or the binarized image when the number of defect candidate points changes from greater than the predetermined reference number to equal to or less than the predetermined reference number, and extract primary defect candidates from this threshold image. In other words, the primary extraction means 44 automatically determines the binarization threshold based on the change in the number of defect candidate points that occurs as the binarization threshold changes, and performs binarization.

In the binarized image, defect candidate points appear, for example, in the peripheral region of the reflected mirror image of the light source 10. For example, as shown in Figure 4, if the reflected mirror image of the light source 10 appears white, defect candidate points appear as white points in the peripheral region of the reflected mirror image of the light source 10. The binarization reference value may be, for example, between 0 and 255, and may be the maximum value of 255, the minimum value of 0, or any value between the maximum and minimum values. When the preferred range of the binarization threshold is known in advance for the surface M to be inspected, the binarization threshold can be quickly determined by setting a value close to that range as the binarization reference value. When the preferred range of the binarization threshold is unknown, the binarization threshold can be determined by setting the maximum or minimum value as the binarization reference value.

The direction in which the binarization threshold is changed is toward a decrease when the binarization reference value is the maximum value of 255, toward an increase when the binarization reference value is the minimum value of 0, and can be either toward a decrease or an increase when the binarization reference value is a value between the maximum and minimum values.

For example, if the reflected mirror image of light source 10 and the defect candidate points appear in white as shown in Figure 4, changing the binarization threshold from a large value to a small value increases the number of defect candidate points in the binarized image, and at a certain binarization threshold value, the number of defect candidate points in the binarized image increases from less than a predetermined reference number to equal to or greater than the predetermined reference number. For reference, Figure 5(A) shows an example of a binarized image binarized with a binarization threshold of 255, the maximum value, and Figure 5(B) shows an example of a binarized image binarized with a binarization threshold smaller than 255, resulting in a number of defect candidate points greater than the predetermined reference number. For example, in the case where the reflected mirror image of the light source 10 and the defect candidate points appear in white as shown in Figure 4, conversely, changing the binarization threshold from a small value to a large value will decrease the number of defect candidate points that appear in the binarized image, and at a certain binarization threshold value, the number of defect candidate points that appear in the binarized image will go from being greater than a predetermined reference number to being equal to or less than the predetermined reference number. Also, in the case of the inverted image of Figure 4, where the reflected mirror image of the light source 10 and the defect candidate points appear in black, the direction of change will be opposite to the case where they appear in white as described above.

The binarization threshold is preferably changed in increments of 1, but may also be changed in increments of any other value, such as 2 or 3. The reference number of occurrences for comparing the number of occurrences of defect candidate points can be set arbitrarily depending on the surface M to be inspected. As an example, it is preferable to set the reference number of occurrences in the range of 3 to 10.

The primary extraction means 44 also preferably performs binarization processing on each of the multiple divided images divided by the image division means 43 to extract primary defect candidates. This is because binarization processing can be performed in accordance with the difference in brightness between the center and ends in the longitudinal direction of the reflected mirror image of the light source 10, allowing defect candidates to be extracted with greater accuracy.

The primary defect candidate storage means 45 is configured, for example, by a memory, and is configured to store the centroid coordinates of the primary defect candidates extracted by the primary extraction means 44.

The secondary extraction means 46 is configured to extract, for example, from the primary defect candidates extracted by the primary extraction means 44, those for which the difference in brightness between the brightness value of the primary defect candidate and the average brightness value of the surrounding area of the primary defect candidate is equal to or greater than a predetermined brightness difference reference value, for an image processed by the preprocessing means 42, which is the extraction target image obtained by the imaging means 20. Specifically, for example, for the image processed by the division means 43, which is the extraction target image, the secondary extraction means 46 is configured to calculate the difference in brightness between the brightness value of the coordinates of the primary defect candidate stored in the primary defect candidate storage means 45 and the average brightness value of the coordinates of the surrounding area of the primary defect candidate, and to extract the defect candidate if the difference in brightness is equal to or greater than a predetermined brightness difference reference value.

In other words, by extracting defect candidates in two stages, using the primary extraction means 44 and the secondary extraction means 46, it is possible to extract them with higher accuracy. Furthermore, the secondary extraction means 46 makes a judgment based on the difference in brightness between the brightness value of the primary defect candidate and the average brightness value of its surroundings. Therefore, unlike when brightness values are used, the same brightness difference reference value can be used even if the paint color is different, making it possible to easily extract defect candidates. Figure 6 shows an example of a defect candidate extracted by the secondary extraction means 46. In Figure 6, the white dots within the circles are defect candidates.

The secondary extraction means 46 is preferably configured to, for example, divide the area around the reflected mirror image of the light source 10, which is the image to be extracted and processed by the division means 43, into multiple areas according to the distance from the reflected mirror image of the light source 10, and set a brightness difference reference value for each of the multiple distance areas. Since the brightness difference between the brightness value of a primary defect candidate and the average brightness value around that primary defect candidate changes depending on the distance from the reflected mirror image of the light source 10, setting a brightness difference reference value for each distance area enables defect candidates to be extracted with greater accuracy.

Figure 7 shows a conceptual diagram in which the area around the reflected mirror image of light source 10 is divided according to the distance from the reflected mirror image of light source 10. Figure 7 shows the area around the reflected mirror image of light source 10 divided into multiple regions R2, R3, R4, R5, and R6 at predetermined pixel intervals from region R1 of the reflected mirror image. Note that in Figure 7, each of regions R1, R2, R3, R4, R5, and R6 is shown with hatching to make it easier to understand.

The difference in distance between the reflected mirror image of adjacent regions R2, R3, R4, R5, and R6 and region R1, for example, the width of each region R2, R3, R4, and R5, can be set arbitrarily depending on the surface M to be inspected. It is preferable that the difference in distance between the reflected mirror image of adjacent regions R2, R3, R4, R5, and R6 and region R1 be in the range of 1 to 10 pixels, for example. Also, while Figure 7 shows a case where the region around the reflected mirror image of light source 10 is divided into five regions, the number of divisions can also be set arbitrarily depending on the surface M to be inspected. It is preferable that the number of divisions be, for example, 6 to 10.

The defect candidate storage means 47 is configured, for example, by a memory, and is configured to store the centroid coordinates of the defect candidates extracted by the secondary extraction means 46.

(Defect detection means 50)
The defect detection means 50 is configured, for example, to divide the surface M to be inspected into a plurality of sections, and to judge a defect to exist when a defect candidate extracted by the defect candidate extraction means 40 exists in each section image area corresponding to the same section of the surface M to be inspected, for a number of images equal to or greater than the reference judgment number and captured at different times by the imaging means 20. The reference judgment number for judging a defect, i.e., the number of images in which a defect candidate exists in a section image area corresponding to the same section of the surface M to be inspected, can be set arbitrarily to 2 or more, and, for example, 3 or more is more preferable.

Specifically, the defect detection means 50 preferably includes a section discrimination means 51 that divides the surface M to be inspected into a plurality of sections, identifies section image areas corresponding to the sections of the surface M to be inspected in the images captured by the imaging means 20, and discriminates the sections of the surface M to be inspected in which defect candidates exist from the section image areas in which defect candidates exist; a counting means 52 that, based on the discrimination results of the section discrimination means 51, tallys up the number of times defect candidates have been detected for each section of the surface M to be inspected for each image captured by the imaging means 20 at different times; and a judgment means 53 that, if there is a section in which the number of times defect candidates have been detected tallied by the counting means 52 is equal to or exceeds a reference judgment number, judges that a defect exists in that section.

As shown in Figure 8, the section discrimination means 51 can be configured to divide the inspection surface M into multiple sections using a black-and-white grid or similar, perform binarization processing on a moving video of the inspection surface M, extract the movement trajectory of only the section lines, and synchronize the video captured by the imaging means 20 with the video of only the section lines to identify section image areas corresponding to the sections of the inspection surface M in the image captured by the imaging means 20. Furthermore, the size of the largest defect spot known in advance can be used as a reference value. If the size of the section is larger than the size of the candidate defect spot, the section area can be divided equally and smaller sections can be used as the section image area, thereby improving the detection accuracy of candidate defects. To prevent detection errors due to camera shake or other factors, it is preferable to set the size of the section approximately twice the size of the largest defect spot. On the other hand, if the size of the section is larger than the size of the candidate defect spot, there is a high possibility that multiple noises will occur simultaneously within the same section, resulting in reduced detection accuracy. For these reasons, the size of the compartments can be set arbitrarily, but when inspecting painted surfaces the size of a standard automobile, it is preferable that the length of one side be approximately 10 mm to 100 mm. Furthermore, it is preferable that the compartment discrimination means 51 assigns a compartment number to each compartment separated by compartment lines, and identifies compartments in which defect candidates exist by the compartment number.

The counting means 52 is preferably configured to, for example, count the block numbers in which defect candidates identified by the block discrimination means 51 exist for each image captured at different times, and determine the number of times the defect candidate is detected for each block number. Furthermore, for example, when a defect candidate extracted by the defect candidate extraction means 40 exists within a predetermined range from the boundary line of a block image area corresponding to a block on the inspected surface M in an image captured by the imaging means 20, the counting means 52 preferably determines that the defect block image area in which the defect candidate exists and the adjacent block image area located near the defect candidate are one block image area corresponding to one block on the inspected surface M. This is because, when a defect candidate exists near a block image area, i.e., near a block line, it is difficult to identify the block image area in which the defect candidate exists, i.e., the block, and the defect detection rate can be improved by expanding the determination to include adjacent block image areas.

For example, if there is one boundary line on which a defect candidate exists within a predetermined range, as shown in FIG. 9(A), it is preferable to consider the defective partition image area 52a where the defect candidate exists and the adjacent partition image area 52b adjacent to the defect candidate across the boundary line within the predetermined range as one partition image area corresponding to one partition on the inspected surface M. Furthermore, if there are two boundary lines on which a defect candidate exists within a predetermined range, as shown in FIG. 9(B), it is preferable to consider the defective partition image area 52a where the defect candidate exists, the two adjacent partition image areas 52b adjacent to the defect candidate across the boundary line within the predetermined range as one partition image area corresponding to one partition on the inspected surface. Note that in FIG. 9, defect candidates are indicated by black circles, the defective partition image area 52a is indicated by hatching with a lower right diagonal line, and the adjacent partition image area 52b is indicated by hatching with a lower left diagonal line.

It is preferable to determine whether a defect candidate exists within a predetermined range from the boundary line based on, for example, Equation 1. In Equation 1, d is the distance between the defect candidate and the boundary line, and A is the length of the boundary line in one section.
d<(1/8)A (Formula 1)
That is, if the distance d between the defect candidate and the boundary line is shorter than 1/8 of the length A of the boundary line, it is preferable to determine that the defect candidate is within the predetermined range.

Preferably, the determination means 53 determines that a defect exists in a section corresponding to a section number when the number of times a defect candidate has been detected for that section number, as tallied by the counting means 52, is equal to or exceeds the reference determination number. For example, as shown in FIG. 10, (A) is an image captured by the imaging means 20 at an arbitrary imaging time, (B) is an image captured by the imaging means 20 a predetermined time after (A), the direction of movement of the inspected surface M is X, and defect candidates are indicated by black circles. In (A), defect candidates exist in the section image areas corresponding to sections 32 and 33, and in (B), defect candidates exist in the section image areas corresponding to sections 32 and 34. Therefore, if the reference determination number for the number of times a defect candidate has been detected is set to 2 or more, it can be determined that the defect candidate in section 32 is a defect and the defect candidate in section 33 is a false defect.

This surface defect inspection device 1 is used, for example, as follows. Figure 11 shows the procedure for a surface defect inspection method using the surface defect inspection device 1. In this surface defect inspection method, for example, the inspection surface M illuminated with light from the light source 10 is first photographed by the imaging means 20, and the position of the inspection surface M relative to the imaging means 20 is moved by the moving means 30 to obtain multiple images photographed at different times (step S110; imaging procedure). In this case, it is preferable to change the angle of the camera 21 depending on the location of the inspection surface M, in accordance with the movement of the position of the inspection surface M relative to the camera 21, so that the brightness of defect candidates is increased. The images photographed by the imaging means 20 are stored in the image storage means 41.

Next, defect candidates are extracted from each of the multiple images captured at different times using the imaging procedure (step S110) (step S120; defect candidate extraction procedure). In the defect candidate extraction procedure (step S120), for example, the preprocessing means 42 first preprocesses the image obtained by the imaging means 20 as described above (step S121; preprocessing procedure). Next, for example, the image segmentation means 43 cuts out an area including one reflector image of the light source 10 and its surrounding area from the image preprocessed as described above, and segments the image into multiple parts along the length of the reflector image (step S122; image segmentation procedure).

Next, for example, the primary extraction means 44 performs binarization processing as described above for each of the multiple divided images obtained by the image division procedure (step S122) to extract primary defect candidates (step S123; primary extraction procedure). Specifically, for example, for a preprocessed image, which is the extraction target image based on the image obtained by the photographing procedure (step S110), binarization processing is performed for each of the divided images obtained by the image division procedure, while sequentially changing the binarization threshold value in one direction from the binarization reference value. The multiple binarized images obtained in the order of processing are compared, and the binarized image when the number of defect candidate points changes from less than the predetermined reference number to equal to or greater than the predetermined reference number, or when the number of defect candidate points changes from greater than the predetermined reference number to equal to or less than the predetermined reference number, is designated as a threshold image, and primary defect candidates are extracted from this threshold image. That is, for example, for one extraction target image, the binarization threshold is changed in one direction from the binarization reference value in order to repeatedly perform the binarization process, and the resulting binarized images are compared to find a threshold image based on the change in the number of defect candidate points that appear, thereby extracting primary defect candidates. The primary defect candidates extracted by the primary extraction procedure (step S123) are stored in the primary defect candidate storage means 45.

Then, for example, the secondary extraction means 46 extracts defect candidates from the primary defect candidates extracted in the primary extraction procedure (step S123) as described above for the segmented images that are the extraction target images based on the image obtained in the photography procedure (step S110) (step S124; secondary extraction procedure). Specifically, for example, for the segmented images that are the extraction target images, the luminance difference between the luminance value of the primary defect candidate and the average luminance value around that primary defect candidate is calculated, and if the luminance difference is equal to or greater than a predetermined luminance difference reference value, the defect candidate is extracted. The defect candidates extracted in the secondary extraction procedure (step S124) are stored in the defect candidate storage means 47.

After defect candidates are extracted using the defect candidate extraction procedure (step S120), defects are detected based on the defect candidates extracted using the defect candidate extraction procedure (step S120) (step S130; defect detection procedure). In the defect detection procedure (step S130), for example, it is preferable to divide the inspected surface M into multiple sections, and for multiple images captured by the imaging means (step S110) at different capture times (a reference determination number or more), determine that a defect exists if the defect candidates extracted using the defect candidate extraction procedure (step S120) exist in each section image area corresponding to the same section of the inspected surface M.

Specifically, for example, the partition discrimination means 51 first divides the surface M to be inspected into multiple partitions as described above, identifies partition image areas corresponding to the partitions of the surface M in the images captured by the photographing procedure (step S110), and discriminates the partitions of the surface M in which defect candidates exist from the partition image areas in which defect candidates exist (step S131; partition discrimination procedure). Next, for example, the count tallying means 52 tally the number of times defect candidates have been detected for each partition of the surface M to be inspected for each image captured at different times by the photographing procedure (step S110) based on the discrimination results by the partition discrimination procedure means 51 as described above (step S132; count tallying procedure). Next, for example, the judgment means 53 determines that a defect exists in a partition if the number of times defect candidates have been detected as counted by the count tallying procedure is equal to or exceeds the reference judgment number as described above (step S133; judgment procedure).

Then, for example, the display means 60 displays the detection results obtained in the defect detection procedure (step S130) (step S140; display procedure). The display means 60, for example, displays the defect's center of gravity on a display or the like by adding a circle mark or the like. This allows surface defect inspection to be performed.

As described above, according to this embodiment, the surface M to be inspected is divided into multiple sections, and for images captured by the imaging means 20 at different capture times equal to or greater than the reference determination number, if a defect candidate extracted by the defect candidate extraction means 40 is present in each section image area corresponding to the same section of the surface M to be inspected, it is determined that the defect candidate is a defect. Therefore, it is possible to determine whether a defect candidate is a defect based on the relationship between the movement trajectory of the defect candidate and the movement trajectory of the section of the surface M to be inspected. Therefore, even if the movement trajectory of the defect candidate is curved, for example, if the angle of the camera 21 is changed in accordance with the movement of the surface M to be inspected relative to the camera 21 in order to increase the brightness of the defect candidate, it is possible to determine the defect with high accuracy.

Furthermore, when a defect candidate exists within a predetermined range from the boundary line of a partition image area corresponding to a partition on the inspected surface M, the defective partition image area 52a and the adjacent partition image area 52b are deemed to be one partition image area corresponding to one partition on the inspected surface M, thereby further improving the accuracy of defect determination.

The present invention has been described above using embodiments, but the present invention is not limited to the above embodiments and can be modified in various ways. For example, while each component has been specifically described in the above embodiments, it is not necessary to include all components, and other components may also be included.

In the above embodiment, the defect candidate extraction means 40 and the defect candidate extraction procedure (step S120) were specifically described, but defect candidates may be extracted using other configurations and procedures. For example, the defect candidate extraction means 40 and the defect candidate extraction procedure (step S120) may extract defect candidates using conventional methods such as global binarization and local binarization. Furthermore, defect candidates may be extracted based on gradient images generated by, for example, performing difference calculations in two orthogonal directions on images obtained by the imaging means 20. In this case, difference calculations may be performed in both the positive and negative directions in the two orthogonal directions, and gradient images may be generated from the calculation results. Alternatively, difference calculations may be performed in the two orthogonal directions, and gradient images may be generated from the absolute values of gradient vectors calculated using only positive values from the calculation results, and the absolute values of gradient vectors calculated using only negative values from the calculation results.

Furthermore, while the above embodiment specifically describes the case of inspecting the painted surface of an automobile body, the present invention is not limited to automobile bodies and can also be applied to the surface inspection of other painted products. Furthermore, it can also be applied to the inspection of reflective surfaces, not limited to painted surfaces.

1...Surface defect inspection device, 10...Light source, 20...Photographing means, 21...Camera, 30...Moving means, 40...Defect candidate extraction means, 41...Image storage means, 42...Preprocessing means, 43...Image segmentation means, 44...Primary extraction means, 45...Primary defect candidate storage means, 46...Secondary extraction means, 47...Defect candidate storage means, 50...Defect detection means, 51...Division discrimination means, 52...Counting means, 52a...Defective division image area, 52b...Adjacent division image area, 53...Determination means, 60...Display means, M...Inspected surface

Claims (3)

a light source that irradiates light onto the surface to be inspected;
an imaging means for capturing an image of the surface to be inspected illuminated by the light source;
a moving means for moving the position of the inspection surface relative to the imaging means;
a defect candidate extracting means for extracting defect candidates from a plurality of images taken by the photographing means at arbitrary time intervals while the photographing means and the inspection surface are moved relatively by the moving means;
a defect detection means for detecting defects based on the defect candidates extracted by the defect candidate extraction means,
The defect detection means divides the surface to be inspected into a plurality of sections, and when a defect candidate extracted by the defect candidate extraction means exists in each section image area corresponding to the same section of the surface to be inspected for a reference determination number or more of images photographed by the photographing means at different photographing times, the defect detection means determines that a defect exists in the same section, and
The defect detection means
a section discrimination means for dividing the surface to be inspected into a plurality of sections in advance, extracting a moving trajectory of only the section lines from a moving image of the divided surface to be inspected, and synchronizing an image photographed by the photographing means with the moving image of only the section lines to identify a section image area corresponding to the section of the surface to be inspected, and discriminating a section of the surface to be inspected in which a defect candidate exists from the section image area in which a defect candidate exists;
a counting means for counting the number of times defect candidates are detected for each section of the inspection surface for each image taken by the photographing means at different photographing times based on the discrimination result by the section discriminating means;
a determination means for determining that a defect exists in a section when the number of times the defect candidate has been detected, as counted by the counting means, is equal to or exceeds a reference determination number.
A surface defect inspection device characterized by: When a defect candidate extracted by the defect candidate extraction means exists within a predetermined range from a boundary line of a partition image area corresponding to a partition of the inspected surface in the image photographed by the photographing means, and there is only one boundary line within the predetermined range, the defect detection means determines that the defective partition image area in which the defect candidate exists and one adjacent partition image area adjacent to the defect candidate located within the predetermined range via the boundary line are one partition image area corresponding to the partition of the inspected surface; and when there are two boundary lines within the predetermined range, the defect detection means determines that the defective partition image area in which the defect candidate exists, two adjacent partition image areas adjacent to the defect candidate located within the predetermined range via the boundary lines, and one adjacent partition image area adjacent to the defect candidate at a corner of the two boundary lines located within the predetermined range are one partition image area corresponding to the partition of the inspected surface .
The partition image area is a polygon with straight or curved sides, and each side is a boundary line.
2. The surface defect inspection device according to claim 1. an imaging procedure in which an image of the inspection surface illuminated with light from a light source is captured by an imaging means, and a position of the inspection surface relative to the imaging means is moved by a moving means to obtain a plurality of images captured at different times ;
a defect candidate extraction step of extracting defect candidates from the images captured by the photographing step;
a defect detection step for detecting defects based on the defect candidates extracted by the defect candidate extraction step,
In the defect detection procedure, the surface to be inspected is divided into a plurality of sections, and for images photographed in the photographing procedure at different photographing times equal to or greater than a reference determination number, if a defect candidate extracted in the defect candidate extraction procedure exists in each section image area corresponding to the same section of the surface to be inspected, it is determined that a defect exists in the same section;
The defect detection procedure includes:
a section discrimination procedure for dividing the surface to be inspected into a plurality of sections in advance, extracting a movement trajectory of only the section lines from a captured video of the sectioned surface to be inspected being moved, and synchronizing the video of only the section lines with the image captured by the photographing procedure to identify section image areas corresponding to the sections of the surface to be inspected, and discriminating the sections of the surface to be inspected in which defect candidates exist from the section image areas in which defect candidates exist;
a counting step of counting the number of times defect candidates are detected for each section of the inspection surface for each image taken at a different photographing time in the photographing step, based on a discrimination result in the section discrimination step;
and a determination step of determining that a defect exists in a section when the number of times the defect candidate is detected, which is counted by the counting step, is equal to or exceeds a reference determination number.
A surface defect inspection method characterized by: