JP2017187348A - Surface defect inspection system, method and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To prevent wrong detection of an engraved mark as a surface defect when detecting the surface defect of a measurement object part where the engraved mark may be provided, and enable detection of the surface defect without missing it.SOLUTION: Linear light is radiated along a height direction of a pillar part, and reflection light is photographed with a TDI camera to obtain a light cut-off image. An image processing device is configured to: prepare a shape image expressing a surface shape of the pillar part on the basis of the light cut-off image; implement shape correction processing of converting a curved-surface of the pillar part into a plane face with respect to the shape image; implement concavity binarization and convexity binarization with respect to an image subjected to the shape correction processing, and further implement expansion and contraction processing and then labelling processing, thereby detect an engraved mark to be provided in the pillar part, and prepare an engraved mark mask image; and implement mask processing of the image subjected to the shape correction processing, using the engraved mark mask image, and detect a surface defect of the pillar part.SELECTED DRAWING: Figure 9

Description

  The present invention relates to a surface defect inspection system, method, and program suitable for use in detecting a surface defect of a measurement target portion that may be marked.

In recent years, a light-cutting optical shape measuring method has been used in various fields. Even in a steel plate production line, the surface of a steel plate surface, such as dents and wrinkles, is emitted by irradiating the surface of the steel plate with linear laser light, imaging the surface of the steel plate with a delay integration camera, and outputting a light cut image. It has been proposed to detect defects (see Patent Document 1).
Here, for example, when a predetermined portion of a long material such as a rail, a shaped steel, a steel pipe, or the like is a measurement target portion, an end portion in a short direction (width direction or height direction) is not flat but a curved surface. Therefore, the surface defect detection accuracy may be lowered.
In view of this point, Patent Document 2 discloses that the edge position of the head is detected based on a luminance image with the head of the rail as a measurement target unit. Then, based on the light section image, a shape image representing the surface shape of the measurement target portion is created, a fitting curve is obtained from the shape image, and a difference calculation between the fitting curve and the shape image is performed, whereby the measurement target portion A configuration is disclosed in which shape correction processing is performed to re-bake a curved surface shape into a flat surface.

JP 2004-3930 A JP 2010-91514 A

The surface defect inspection method of Patent Document 2 can be applied not only to the detection of the surface defect of the rail pillar but also the rail head.
Here, in the rail 1, a rail number or the like may be stamped on the pillar portion 1a. As shown by a dotted line in FIG. 2, a stamped region (a stamped band) 11 having a certain width is set at a predetermined height position of the pillar portion 1 a, and the stamped band 11 is stamped. In addition, although the marking belt | band | zone 11 does not appear visually in the pillar part 1a, it shows with a dotted line for easy understanding in FIG. Thus, when detecting the surface defect of the pillar part 1a which may be marked, the marking is overdetected as a surface defect.

As a measure to suppress over-detection of engraving, it is considered that a range of size, height, or depth corresponding to the engraving is determined in advance, and the unevenness in that range is determined as engraving and is not handled as a surface defect. It is done. However, there is a possibility that a surface defect having a size, height, or depth corresponding to the stamp may be missed.
It is also conceivable that the marking band 11 is defined on the image and the unevenness detected by the marking band 11 is ignored. However, there is a risk of missing a surface defect appearing on the stamped band 11. In addition, when the marking band 11 itself cannot be appropriately determined due to bending or meandering of the rail, and the position of the determined marking band 11 deviates greatly from the actual marking band, the marking is detected as a surface defect. There is a fear.

Here, as a general marking detection method, a method is known in which convergent light is irradiated from the periphery of the marking so as to surround the marking so as to make a shadow by the marking, and the marking is detected from the brightness of the image. However, it is desirable that the surface of the measurement target portion is clean, and there is a problem that it cannot be applied when there is a change in luminance due to scale adhesion, a pattern, or the like like a rail.
In addition, as a method of measuring the step of the marking itself and detecting the marking, there are a focus method using the sharpness of the image by the focal length and an interference method of detecting the step from the interval of the interference fringes by the distance of the step. Has a problem that it is not suitable for real-time measurement.

  The present invention has been made in view of the above points, and when detecting a surface defect of a measurement target portion that may be engraved, the surface defect is not overdetected as a surface defect. The purpose is to enable detection without missing.

The gist of the present invention for solving the above problems is as follows.
[1] Among long materials, a predetermined part having a curved surface at an end in a short direction is used as a measurement target part.
Irradiation means for irradiating the measurement target portion with periodically modulated linear laser light along the short direction of the long material;
With a delay integration type imaging means for imaging reflected light from the measurement target part,
While continuously shifting the irradiation position of the linear laser light in the longitudinal direction of the long material, the reflected light from the measurement target portion is picked up by the imaging means to obtain a light cut image, and the measurement target portion A surface defect inspection system for detecting surface defects of
A shape image creating means for creating a shape image representing the surface shape of the measurement target part based on the light section image output from the imaging means;
Shape correction processing means for performing shape correction processing for converting the curved surface shape of the measurement target portion into a plane with respect to the shape image created by the shape image creation means;
By performing at least one of concave binarization and convex binarization on the image obtained by the shape correction processing means, and performing a labeling process, a mark having a known size is detected. A marking mask image creating means for creating a marking mask image;
Defect detection processing means for detecting a surface defect of the measurement target portion by masking the image obtained by the shape correction processing means using the marking mask image created by the marking mask image creating means. Surface defect inspection system characterized by that.
[2] The apparatus includes a boundary detection unit that evaluates a change in the degree of curvature of the surface shape of the measurement target portion in the shape image generated by the shape image generation unit and detects a boundary of the measurement target portion. The surface defect inspection system according to [1].
[3] The boundary detection unit performs second-order difference processing or Laplacian calculation processing on the shape image created by the shape image creation unit, and detects a peak position as a boundary of the measurement target unit. The surface defect inspection system according to [2], which is characterized.
[4] The marking mask image creating means detects the marking in a predetermined area defined on the basis of the boundary detected by the boundary detecting means. [2] or [3] Surface defect inspection system.
[5] The marking mask image creating means performs a concave binarization and a convex binarization on the image obtained by the shape correction processing means, and further performs an expansion and contraction process, followed by a labeling process The surface defect inspection system according to any one of [1] to [4], wherein the marking is detected by performing
[6] The long material is a rail,
In any one of [1] to [5], the measurement target part is a pillar part in the rail, and the pillar part is connected to the lower jaw at one end and to the leg part at the other end. The surface defect inspection system described.
[7] The shape image creating means includes:
Phase calculating means for calculating a phase shift at each position of the fringe image obtained by sequentially arranging the light sectioned images obtained by the imaging means;
Detecting the position where the phase shift is discontinuous based on the phase shift obtained by the phase calculation means, and connecting the phase shift at the detected position to make the phase shift continuous, The surface defect inspection system according to any one of [1] to [6], further comprising: a phase continuation processing unit that creates a shape image representing the surface shape of the measurement target portion.
[8] The shape correction processing unit obtains a fitting curve for the shape image created by the shape image creating unit or the shape image obtained by performing predetermined processing on the shape image created by the shape image creating unit. A fitting process is performed, a difference calculation between the fitting curve and the shape image created by the shape image creation means is performed, and the data obtained by the difference calculation has a predetermined length in the longitudinal direction of the long material. Any one of [1] to [7], wherein the difference calculation is performed again with respect to the value obtained by performing low-pass filter processing on the average value and the data obtained by the difference calculation. Surface defect inspection system described in 1.
[9] Of the long material, a predetermined part having a curved surface at the end in the short direction is used as a measurement target part.
Irradiation means for irradiating the measurement target portion with periodically modulated linear laser light along the short direction of the long material;
With a delay integration type imaging means for imaging reflected light from the measurement target part,
While continuously shifting the irradiation position of the linear laser light in the longitudinal direction of the long material, the reflected light from the measurement target portion is picked up by the imaging means to obtain a light cut image, and the measurement target portion A surface defect inspection method for detecting surface defects of
A shape image creating step for creating a shape image representing the surface shape of the measurement target part based on the light section image output from the imaging means;
A shape correction processing step for performing shape correction processing for converting the curved surface shape of the measurement target portion to a plane with respect to the shape image created in the shape image creation step;
By performing at least one of concave binarization and convex binarization on the image obtained in the shape correction processing step, and performing a labeling process, a mark with a known size is detected. A stamp mask image creating step for creating a stamp mask image;
Using a marking mask image created in the marking mask image creating step, masking the image obtained in the shape correction processing step, and detecting a surface defect in the measurement target portion. A method for inspecting a surface defect.
[10] Among long materials, a predetermined part having a curved surface at the end in the short direction is used as a measurement target part.
Irradiation means for irradiating the measurement target portion with periodically modulated linear laser light along the short direction of the long material;
With a delay integration type imaging means for imaging reflected light from the measurement target part,
While continuously shifting the irradiation position of the linear laser light in the longitudinal direction of the long material, the reflected light from the measurement target portion is picked up by the imaging means to obtain a light cut image, and the measurement target portion A program for detecting surface defects of
A shape image creating means for creating a shape image representing the surface shape of the measurement target part based on the light section image output from the imaging means;
Shape correction processing means for performing shape correction processing for converting the curved surface shape of the measurement target portion into a plane with respect to the shape image created by the shape image creation means;
By performing at least one of concave binarization and convex binarization on the image obtained by the shape correction processing means, and performing a labeling process, a mark having a known size is detected. A marking mask image creating means for creating a marking mask image;
Using a marking mask image created by the marking mask image creating means, masking the image obtained by the shape correction processing means, and a computer as a defect detection processing means for detecting a surface defect of the measurement target portion A program to make it work.

  According to the present invention, when detecting a surface defect of a measurement target portion that may be engraved, a marking mask image is created by detecting the imprint, so that the imprint is not overdetected as a surface defect. At the same time, it can be detected without missing a surface defect.

It is a schematic block diagram of the surface defect inspection system which concerns on embodiment. It is a figure for demonstrating the positional relationship of a delay integration type camera and a elongate material. It is a figure for demonstrating the structure and operation | movement of a delay integration type camera. It is a figure which shows an example of a fringe image. It is a figure for demonstrating the shift | offset | difference of the stripe in a fringe image. It is a figure which shows an example of a fringe image. It is a figure for demonstrating the relationship between the phase shift in slice fringe image data, and the depth of a measurement object part. It is a schematic block diagram of an image processing apparatus. 6 is a flowchart for explaining a processing operation of the image processing apparatus according to the embodiment. It is a figure which shows the photograph of an example of the image in each step. It is a characteristic view showing an example of the phase characteristic in each step. It is a figure for demonstrating the outline | summary of a centering process. It is a figure which shows the photograph of an example of the binarized image used for surface defect discrimination | determination. It is a figure which shows an example of the cross-sectional shape of a rail. It is a flowchart which shows a boundary detection process. It is a figure for demonstrating the outline | summary of the 2nd floor difference process in a boundary detection process. It is a characteristic view which shows an example of dimension drawing cross-sectional shape and 2nd floor difference data. It is a characteristic view which shows an example of the result of having performed the moving average process and the 2nd floor difference process with respect to the cross-sectional profile. It is a flowchart which shows a marking detection and mask process. It is a figure which shows the photograph of an example of the image in each step of engraving detection and a mask process. It is a figure which shows the photograph of an example of a depth image.

Preferred embodiments of the present invention will be described below with reference to the accompanying drawings.
FIG. 1 is a schematic configuration diagram of a surface defect inspection system according to an embodiment. The surface defect inspection system includes a laser device 10, a rod lens 20, a delay integration camera (TDI camera) 30, a timing signal generator 40, an image processing device 50, and a display device 60.

  The surface defect inspection system optically measures the surface shape of the measurement target part using a predetermined part having a curved surface at the end in the short direction of the long material as the measurement target part. In this embodiment, as shown in FIG. 2, the column part 1a of the rail 1 which is a long material is used as a measurement target part. The rail 1 is transported at a constant speed in the longitudinal direction (arrow X in FIG. 2), and the surface defect inspection system measures the surface shape of the pillar portion 1a of the rail 1 being transported, and the surface such as unevenness and wrinkles Detect defects.

  The laser device 10 generates a continuous wave laser beam. The rod lens 20 spreads the laser light emitted from the laser device 10 in a fan shape along the short side direction of the rail 1, specifically, the height direction of the pillar portion 1a. Thereby, the laser beam emitted from the laser device 10 is irradiated as a linear laser beam along the height direction of the column portion 1 a, that is, irradiated so as to cross the rail 1. At this time, the laser device 10 is installed so that the linear laser light is incident obliquely on the surface of the column portion 1a. Thus, on the surface of the column part 1a irradiated with the linear laser beam, a linear bright part is formed along the height direction. Further, since the rail 1 moves in the longitudinal direction, the linear bright part also moves along the longitudinal direction of the rail 1. The reflected light (linear reflected image) from this linear bright part is picked up by the TDI camera 30.

  The timing signal generator 40 generates a sine waveform signal having a predetermined frequency ω, and sends the sine waveform signal to the laser device 10. When the laser device 10 has its oscillation intensity continuously changed by an external signal and receives a sine waveform signal sent from the timing signal generator 40, the laser device 10 generates laser light whose output changes in a sine waveform. That is, the laser beam emitted from the laser device 10 is periodically modulated. The timing signal generator 40 generates a camera shift pulse signal having a frequency M times the frequency ω, and sends the camera shift pulse signal to the TDI camera 30.

  As shown in FIG. 2, the TDI camera 30 is arranged on the side of the rail 1 so as to face the pillar portion 1 a and captures a linear reflection image of a region including the pillar portion 1 a. FIG. 3 is a diagram for explaining the structure and operation of the TDI camera. In the TDI camera, as shown in FIG. 3A, a large number of photoelectric conversion elements 35 are arranged in a matrix. Here, for example, it is assumed that n photoelectric conversion elements 35 are arranged in the row direction and N in the column direction. Each row is numbered sequentially from the top with the top row as the first row, and each column is numbered sequentially from the left with the leftmost column as the first column. Here, the row direction of the photoelectric conversion elements 35 corresponds to the short direction of the long material (the height direction of the column portion 1a of the rail 1).

  The photoelectric conversion element 35 accumulates charges corresponding to the intensity of the received light. It is assumed that the linear reflection image of the measurement target portion enters the photoelectric conversion element 35 with a width corresponding to one row through the lens 31. In the TDI camera 30, each photoelectric conversion element 35 transfers the accumulated charge to the photoelectric conversion element located in the same row as the photoelectric conversion element 35 and located in the next column. The timing of this transfer is the same for all the photoelectric conversion elements 35, and is controlled by a camera shift pulse signal sent from the timing signal generator 40. That is, each time a camera shift pulse signal is input, each photoelectric conversion element 35 transfers charges. The frequency of this camera shift pulse signal (camera shift frequency) is Mω. Then, when the camera shift pulse signal is input, the photoelectric conversion element 35 located in the Nth column (final column) sends the accumulated charge to the read register. Thereby, the light section image corresponding to the linear reflection image is output.

  In general, in the TDI camera 30, as shown in FIG. 3B, when light is incident on each photoelectric conversion element 35 while the charge is being transferred, a charge corresponding to the intensity of the incident light is added. Is done. However, as described above, a linear reflection image having a width corresponding to one column is incident on the photoelectric conversion element 35. For this reason, the charge is hardly added in each photoelectric conversion element 35 during the transfer of the charge. In particular, even when background light becomes a problem, this can be suppressed by installing an optical filter that transmits only the laser wavelength in front of the lens.

  Since the rail 1 is moving along the longitudinal direction, when a laser beam is irradiated from the laser device 10 onto the pillar portion 1a and a linear reflection image of a region including the pillar portion 1a is captured using the TDI camera 30 for a certain period of time. The light cut images at the respective positions in the longitudinal direction of the rail 1 can be obtained sequentially. Therefore, an image representing the entire rail 1 can be obtained by sequentially arranging the respective light cut images thus obtained.

  Further, the linear laser light is periodically modulated, and the intensity of the linear laser light changes with time, so that the amount of charge (received light intensity) accumulated in each photoelectric conversion element in the column direction in each row. The distribution also changes periodically. Therefore, an image obtained by sequentially arranging the light section images output from the TDI camera 30 becomes a fringe image in which the density (intensity) of each light section image periodically changes along the array direction. . FIG. 4 shows an example of a stripe image. Here, the light section image corresponding to one period of density change is referred to as “stripe”. In this striped image, the direction parallel to the strip corresponds to the short direction of the long material, and the direction orthogonal to the strip corresponds to the long direction of the long material. When the ratio between the camera shift frequency of the TDI camera 30 and the modulation frequency of the laser beam is M: 1, M light-cut images, that is, M pixels in the arrangement direction form one stripe.

  By the way, since the laser light is incident on the surface of the measurement target portion at an angle, for example, if there is a recessed portion in the measurement target portion, the reflection point of the laser light is shifted to the right in FIG. Therefore, the position of the light section image on the photoelectric conversion element 35 is also shifted in the column direction in the right side, that is, in FIG. For this reason, in the fringe image, the light cut image corresponding to the laser light reflected by the recessed portion is output earlier in time than the light cut image corresponding to the laser light reflected by the non-dented portion. Become. Therefore, in the image obtained by arranging the images output from the TDI camera 30 in order, the recessed portion can be clearly recognized as a stripe shift.

  This fringe shift will be described in more detail. FIG. 5A is a schematic enlarged view of a certain fringe image. In FIG. 5A, the position where the maximum density is given for each stripe is shown by a solid line. For example, in this fringe image, when the maximum density position is examined along the arrangement direction at the position A in the short direction, the maximum density positions are located at equal intervals, and no stripe deviation occurs. That is, the measurement target part has a flat shape along the arrangement direction at the position A in the short direction. In this case, the density distribution (slice fringe image data) of the stripe image along the arrangement direction at the position A in the short direction has a clean sine wave shape as shown in FIG.

  On the other hand, when the maximum density position is examined along the arrangement direction at the position B in the short direction shown in FIG. 5A, the interval between the maximum density positions gradually increases from the left to the right, and the deviation of the stripes is increased. Has occurred. That is, the measurement target portion has a dent along the arrangement direction at the short-side position B. In this case, the density distribution (slice fringe image data) of the fringe image along the arrangement direction at the position B in the short direction is compared with the sine wave shown in FIG. 5B as shown in FIG. Is out of phase. As described above, the fringe shift due to the dent of the measurement target portion appears as a phase shift in the slice fringe image data. In fact, this phase shift and the dent (depth) of the measurement target portion are in a proportional relationship. The greater the depth, the greater the phase shift in the slice fringe image data. In the surface defect inspection system, information related to the phase shift is calculated based on the fringe image, and the surface shape of the measurement target portion is measured based on the information related to the phase shift. If this method of calculating the phase shift is used, for example, when the ratio of the camera shift frequency to the modulation frequency of the laser beam is set to 4: 1, the fringes can be detected with unevenness even if the fringes are not clearly shifted by one pixel or more. There is sensitivity.

  Here, the fringe image shown in FIG. 4 is that of the measurement target portion having a flat surface. On the other hand, when a predetermined part of a long material having a curved surface at the end in the short direction is used as a measurement target part, a fringe image in which both ends of each stripe are curved as shown in FIG. The striped image shown in FIG. 6 is not that of the pillar portion 1a of the rail 1, but the striped image of the pillar portion 1a is similarly a striped image in which both ends of each stripe are curved.

Next, the relationship between the phase shift in the slice fringe image data and the depth of the measurement target portion will be described. FIG. 7 is a diagram for explaining the relationship between the phase shift in the slice fringe image data and the depth of the measurement target portion.
Now, as shown in FIG. 7, the incident angle at which the linear laser light is incident on the surface of the measurement target portion is denoted by θ. Further, it is assumed that the measurement target portion has a recess, and the linear laser light is reflected at a depth d from the surface of the measurement target portion and enters the TDI camera 30 when entering the recess. At this time, the linear laser beam reflected at the depth d has a reflection point by a distance h in the longitudinal direction (right direction) of the long material, compared to the linear laser beam reflected by the flat surface of the measurement target portion. Shift. Here, h = d · tan θ. As a result of the deviation of the reflection point of the linear laser beam by the distance h in the longitudinal direction, a phase shift occurs in the slice fringe image data. This phase shift is denoted by φ.

Assuming that the imaging resolution in the column direction of the photoelectric conversion element 35 in the TDI camera 30 is s (mm / pixel), the distance h that the reflection point of the linear laser beam is displaced in the longitudinal direction corresponds to the h / s pixel in the striped image. To do. When the ratio between the camera shift frequency of the TDI camera 30 and the modulation frequency of the laser beam is M: 1, M pixels in the arrangement direction form one stripe in the stripe image. That is, when the fringes are shifted by M pixels, the phase shift is 2π. Therefore, the phase shift φ in the slice fringe image data when the reflection point of the linear laser beam is shifted by the distance h in the longitudinal direction is
M / 2π = (h / s) / φ
Than,
d = {M · s / (2π · tan θ)} φ
It becomes. From this, it can be seen that the phase shift φ in the slice fringe image data is proportional to the depth d of the measurement target portion.

  Strictly speaking, when a normal lens is used, the imaging resolution s changes according to the depth d and needs to be corrected. However, as in the case of measuring a dent in a steel plate, the lens working distance is not affected. When the change in depth is small, this change in imaging resolution s can be ignored in practice. If a telecentric lens is used, the imaging resolution s can be made constant regardless of the depth d.

  The image processing device 50 creates an image representing the surface shape of the pillar portion 1a of the rail 1 based on each light cut image output from the TDI camera 30, and detects a defect based on the image. FIG. 8 shows a schematic block diagram of the image processing apparatus 50. The image processing apparatus 50 includes an A / D conversion unit 501, a pre-filter unit 502, an orthogonal sine wave generation unit 503, low-pass filter units 504a and 504b, a phase calculation unit 505, a phase continuation processing unit 506, It includes a boundary detection unit 507, a centering processing unit 508, a shape correction processing unit 509, a marking mask image creation unit 510, and a defect detection processing unit 511. The results processed by each part of the image processing device 50 are displayed on the screen of the display device 60.

Hereinafter, the processing operation of each unit of the image processing apparatus 50 will be described with reference to FIG. FIG. 9 is a flowchart for explaining the processing operation of the image processing apparatus 50.
First, the TDI camera 30 images a region including the pillar portion 1a of the rail 1 and outputs a light cut image (step S1). The A / D conversion unit 501 performs A / D conversion on each light-cut image output from the TDI camera 30 and outputs it as digital multilevel image data. This digital multilevel image data is stored in an image memory (not shown). By arranging these digital multivalued image data in order, a fringe image is formed.

  From the striped image (or digital multi-valued image data), data representing the density distribution of the striped image along the arrangement direction is generated at each position in the short direction. Data representing the density distribution of the stripe image along this arrangement direction is “slice stripe image data”. Slice stripe image data at each position in the short direction is sequentially output from the image memory.

  The pre-filter unit 502 removes noise and sharpens the stripe state by applying a predetermined filter process to each slice stripe image data. Note that the filtering process by the prefilter unit 502 is not necessarily performed. For example, it may be performed only when many fine noises are generated in the striped image.

The pre-filter unit 502 outputs two pieces of slice stripe image data I _j (k) at each position j (j = 0, 1, 2,...) In the short direction. k (k = 0, 1, 2,...) is a position in the arrangement direction. At this time, it is assumed that the slice fringe image data I _j (k) at the position j in the short direction changes sinusoidally. That is,
I _j (k) = A (j, k) {cos ((2πk / M) + φ (j, k)) + 1}
It is. Here, A (j, k) is the amplitude of the slice fringe image data at the pixel position (j, k), and φ (j, k) is the phase shift of the slice fringe image data at the pixel position (j, k). . The influence of the fringe shift generated in the fringe image due to the dent of the measurement target portion appears as a phase shift φ. Since the amplitude of the linear laser beam is constant, the amplitude A is usually constant. However, when the surface of the measurement target part is dirty, the amplitude A may decrease rapidly at the pixel position corresponding to the dirty position. Therefore, in the above equation, the amplitude A is written in a form depending on the pixel position (j, k).

The reason why “1” is added after the term of cos is to guarantee this because slice stripe image data (density value) I _j (k) does not become negative. Therefore, the slice fringe image data I _j (k) varies between 0 and 2A.

The orthogonal sine wave generation unit 503 generates two orthogonal reference sine wave data sin (2πk / M) and cos (2πk / M), which are created in advance on a memory such as a ROM. In particular, the former is also referred to as reference sin data, and the latter is also referred to as reference cos data. These two reference sine wave data are respectively multiplied by the slice fringe image data I _j (k) output from the prefilter unit 502. By this multiplication processing, the following two outputs Ia _j (k) and Ib _j (k) are obtained.

Each of the low-pass filter units 504a and 504b performs predetermined filter processing on the outputs Ia _j (k) and Ib _j (k) obtained by the above multiplication processing, thereby removing the fringe frequency component and its harmonic components. That is, a component including only the phase shift φ is extracted. If the output from the low-pass filter unit 504a is LPF (Ia _j (k)) and the output from the low-pass filter unit 504b is LPF (Ib _j (k)),
LPF (Ia _j (k)) = (A cos φ) / 2
LPF (Ib _j (k)) = − (A sin φ) / 2
It is.

  The phase calculation unit 505 calculates a phase shift φ (j, k) at each pixel position (j, k) based on the results output from the two low-pass filter units 504a and 504b (step S2). The phase shift φ (j, k) can be obtained from the following equation.

In the above equation, the arctan value range is set to −π / 2 to + π / 2, and information about the codes of LPF (Ia _j (k)) and LPF (Ib _j (k)) is used to shift the phase. φ is determined in the range of −π to + π. Here, the phase shift obtained in this range will be described again as φ ′. In this case, the phase shift φ ′ obtained by the above formula has a periodic relationship with the depth of the measurement target portion, and there are a plurality of depths that take a certain value of the phase shift φ ′. Therefore, accurate information on the surface shape of the measurement target portion cannot be obtained by using this phase shift φ ′. For this reason, it is necessary to obtain a phase shift φ that is proportional to the depth of the measurement target portion from the phase shift φ ′. The process of obtaining a phase shift φ proportional to the depth is performed by the phase continuation processing unit 506. FIG. 10A shows an example of a phase image created by the phase calculation unit 505. FIG. 11A shows an example of a characteristic diagram representing the phase characteristic in the phase image. 10 and 11 are not for the pillar portion 1a of the rail 1, but the same applies to the stripe image of the pillar portion 1a.

  The phase continuation processing unit 506 detects discontinuous points of the phase shift φ ′ based on the phase image created by the phase calculation unit 505, and the phase shift φ ′ so that the phase shift φ ′ is smoothly connected. 'Is corrected (step S3). As described above, since the value range of the phase shift φ ′ calculated by the phase calculation unit 505 is −π to + π, the phase shift φ ′ is discontinuous at −π and + π. For example, in the phase image shown in FIG. 10A, a portion where white (or black) changes to black (or white) corresponds to a discontinuous point of the phase shift φ ′. If this phase image is used as it is, it is difficult to recognize the surface shape of the measurement target portion. Therefore, it is necessary to correct the phase shift φ ′ so that the phase shift φ ′ is smoothly connected at the discontinuous points of the phase shift φ ′. This correction (phase jump correction) is a process for obtaining a unique phase shift φ proportional to the depth of the measurement target portion from the phase shift φ ′ defined in the range of 2π.

  Specifically, the phase continuation processing unit 506 detects a discontinuous point of the phase shift φ ′ and corrects the phase shift φ ′ at the discontinuous point. Whether or not the phase shift φ ′ is discontinuous is not known by looking at only one pixel. It is necessary to judge by looking at adjacent pixels. First, the phase continuation processing unit 506 examines the phase image along the arrangement direction at each position in the short direction of the phase image, and compares the phase shift φ ′ between adjacent pixels. If the phase shift φ ′ differs greatly between the adjacent pixels, it is determined that the phase shift φ ′ is discontinuous between the pixels, and the phase shift φ ′ is corrected. Actually, the depth at the surface of the measurement target portion does not change abruptly. For this reason, it is considered that the phase shift φ ′ is largely different because the phase shift φ ′ is changed by ± 2π. Therefore, a pixel whose phase shift φ ′ is significantly different from the phase shift φ ′ of the adjacent pixels is examined, and the phase shift φ ′ may be smoothly connected (FIG. 11). (See arrow in (a)).

  For example, when a phase shift φ ′ is close to + π at a certain pixel position and the phase shift φ ′ is close to −π at a pixel position on the right side, the right adjacent pixel At the position, it is recognized that the phase shift φ ′ changes by + 2π. Then, the phase shift φ ′ is corrected by adding + 2π to the phase shift φ ′ at the right adjacent pixel position. Further, when the phase shift φ ′ is a value close to −π at a certain pixel position, and the phase shift φ ′ is a value close to + π at the pixel position on the right side, the pixel on the right side At the position, it is recognized that the phase shift φ ′ changes by −2π. The phase shift φ ′ is corrected by adding −2π to the phase shift φ ′ at the right pixel position.

  In this way, after examining pixels adjacent in the arrangement direction at each position in the short direction and correcting the phase shift φ ′, the phase continuation processing unit 506 next reduces the short position at each position in the arrangement direction. The adjacent pixels along the direction are examined, and similarly, the phase shift φ ′ is corrected. The phase shift at each pixel position after correction is a unique phase shift φ proportional to the depth of the measurement target portion.

Next, the phase continuation processing unit 506 creates a new phase image based on the corrected phase shift φ. This new phase image accurately represents the surface shape (cross-sectional profile) of the measurement target portion. This new phase image is referred to as a shape image. FIG. 10B shows an example of a shape image created by the phase continuation processing unit 506. FIG. 11B shows an example of a characteristic diagram showing the phase characteristic in the shape image.
The processing of steps S2 and S3 is an example of processing by the shape image creating means of the present invention.

On the other hand, the boundary detection unit 507 performs a moving average process and a second-order difference process on the shape image created by the phase continuation processing unit 506 to search for a peak, and determines the peak position in the height direction of the column part 1a. As a boundary, a column part region is extracted (step S4).
Details of this boundary detection processing will be described later with reference to FIGS. The process of step S4 is an example of the process by the boundary detection means of the present invention.

  The centering processing unit 508 performs centering processing on the shape image created by the phase continuation processing unit 506 based on the boundary of the column part 1a detected by the boundary detection unit 507 (step S5). When photographing a long material conveyed in the longitudinal direction, if the long material meanders left and right and up and down, or vibrates, it shifts in the short direction as shown in FIGS. 12 (a) and 12 (b). Taken as an image. FIG. 12A is a schematic diagram showing a state when the long material is meandering, and FIG. 12B is a schematic diagram showing a state when the long material vibrates, and the solid line indicates the boundary of the measurement target portion. The dotted line indicates the center position.

  For such an image, in the centering process, the boundary of the measurement target part is detected for each line, and the center thereof is calculated as the center position. This position is aligned with the image center in the vertical direction. FIG. 12C is a schematic diagram showing an image after the centering process. With this centering process, when a predetermined part of a long material having a curved surface at the end in the short direction is used as a measurement target part, even when the long material meanders or vibrates, surface defects Detection accuracy can be increased.

  The shape correction processing unit 509 performs shape correction processing on the shape image after the centering processing. In the shape images shown in FIG. 10B and FIG. 11B, the surface shape of the measurement target portion itself is not flat. Therefore, even if the binarization process for performing surface defect determination is performed, the surface of the measurement target portion Surface defects such as dents and wrinkles cannot be detected. Therefore, a shape correction process is performed in which the curved surface shape of the measurement target portion is rebaked into a flat surface, that is, converted into a flat surface.

Specifically, after pre-processing the shape image after the centering process (steps S6 to S10), a fitting process using a high-order function (for example, a fourth or higher order polynomial function) is performed for each line (step S11). .
In the pre-processing, prior to the fitting processing, primary shape correction (step S6), binarization (step S7), unevenness detection (step S8), expansion processing (step S9), and mask processing (step S10) are performed. Then, a shape image is created by removing the uneven portion from the shape image after the centering process based on the mask information. Since details of the pre-processing are detailed in Patent Document 2, detailed description thereof is omitted here.

The fitting process is a process for obtaining a curve (fitting curve) having the smallest residual square sum for a given data string, and a general least square method, a Chebyshev method, or the like can be used. The least square method is a process of calculating a coefficient by setting a partial differential coefficient based on a residual sum of squares coefficient to zero using a power of a variable as a basis function. At this time, if a Chebyshev polynomial, which is an orthogonal polynomial, is used as the basis function, a higher-order function can be fitted more accurately and stably. A fitting function based on a Chebyshev polynomial is expressed by the following equation, and coefficients a _i (i = 0, 1,..., N + 1) are obtained by fitting processing.

  Subsequently, a difference between the fitting curve generated in the fitting process (step S11) and the original data (the shape image after the centering process) is calculated for each line (step S12). In FIG. 11B, the dotted line is the fitting curve, and the solid line is the original data. By this fitting process, even when both ends of the measurement target portion have a steep shape, a sharp signal change at both ends is alleviated.

  Further, the data obtained by the difference calculation (step S12) is averaged (integrated) by a predetermined length in the running direction (longitudinal direction) of the long material (step S13), and subjected to low-pass filter processing (step S14). Then, a difference calculation (or division calculation) with the original data (data obtained by the difference calculation in step S12) is performed (step S15). By this shading correction, flattening is performed while retaining unevenness information that is a surface defect candidate.

As described above, as the shape correction process, a two-stage process such as a fitting process using a high-order function (for example, a polynomial function of a fourth order or higher) and further a shading correction is performed. The image after the shape correction process is referred to as a depth image. FIG. 10C shows an example of a depth image created by the shape correction processing unit 509. FIG. 11C shows an example of a characteristic diagram representing the phase characteristic in the depth image.
The process of steps S6 to S15 is an example of the process by the shape image creating means of the present invention.

The engraving mask image creation unit 510 performs concave portion binarization and convex portion binarization on the depth image created by the shape correction processing unit 509, and further performs labeling processing after performing expansion and contraction processing. By doing so, a stamp 11 having a known size is detected and a stamp mask image is created (step S16).
Details of this marking detection / mask processing will be described later with reference to FIGS. The process of step S16 is an example of the process performed by the marking mask image creating unit of the present invention.

The defect detection processing unit 511 performs a binarization process by masking the depth image created by the shape correction processing unit 509 using the marking mask image created by the marking mask image creation unit 510 (step S17). ). In the binarization processing, a threshold value is set, and “0” is given to the portion within the threshold value, and “1” is given to the portion exceeding the threshold value, and imaged.
And the defect detection process part 511 performs surface defect discrimination | determination using a binarized image (step S18), and detects a surface defect (step S19). FIG. 13 shows a photograph of the binarized image. As shown in FIG. 13, the number of pixels of the “1” portion of the binarized image is calculated and multiplied by the imaging resolution to calculate the defect area, length, width, aspect ratio, etc., and surface defect discrimination Can be done. In addition to the binarized image, for example, a depth image may be used.
The processing of steps S17 to S19 is an example of processing by the defect detection processing means of the present invention.

(Details of boundary detection processing)
Here, the details of the boundary detection process in step S4 will be described.
In FIG. 14, an example of the cross-sectional shape of the rail 1 is shown. As shown in FIG. 14, the rail 1 has a head portion 1b, a head side portion 1c, a chin lower portion 1d, a column portion 1a, and a leg portion 1e from above, and the column portion 1a has a curved upper end in the height direction. The lower end of the chin is connected to the leg 1e while the lower end is curved.
In Patent Document 2, the edge position of the head 1b of the rail 1 (the boundary between the head 1b and the head side 1c) is detected based on the luminance image. The curve from the head 1b to the head side 1c is relatively steep, and the diffuse reflected light from the head side 1c does not return. Therefore, the edge position of the head 1b can be detected from the luminance image.
However, when the column part 1a is the measurement target part, the curve from the column part 1a to the lower jaw part 1d and the leg part 1e is gentle, and the diffuse reflected light of the linear laser light irradiated to the lower jaw part 1d and the leg part 1e. Therefore, it is difficult to accurately detect the boundary of the pillar portion 1a from the luminance image.

Therefore, as described above, the moving average process and the second-order difference process are performed on the shape image created by the phase continuation processing unit 506 to search for a peak, and the peak position is determined in the height direction of the column part 1a. It detects as a boundary and extracts a column part area | region.
FIG. 15 is a flowchart showing the boundary detection process.
In step S401, the boundary detection unit 507 performs a moving average process on the shape image representing the cross-sectional profile created by the phase continuation processing unit 506 to obtain a smoothed cross-sectional profile. The moving average process is performed in order to remove local noise so that the data after the next second-order difference process has a smooth peak.

In step S402, the boundary detection unit 507 performs second-order difference processing on the smoothed cross-sectional profile obtained in step S401. Specifically, as shown in FIG. 16, points A (x ₁ , y ₁ ), B (x ₂ , y ₂ ) where the distance between the x coordinates is separated by a predetermined distance D on the smoothed cross-sectional profile. against it, determine the midpoint of the line segment _{AB M (x m, y m} ) (xm = (x 1 + x 2) / 2, ym = (y 1 + y 2) / 2). The point on the smoothed cross-sectional profile corresponding to the x coordinate x _m is C (x _m , y _c ), and the length of the line segment MC, that is, MC = Δy = y _m −y _c is used as the second-order difference value. Ask. While moving the points A and B while keeping a predetermined distance D = | x ₂ −x ₁ | between the x coordinates constant like a moving average, the second-order difference processing is performed on the entire smoothed cross-sectional profile, Obtain second-order difference data.

In step S403, the boundary detection unit 507 searches for a convex peak on the second-order difference data obtained in step S402, and detects the x coordinate corresponding to the peak position as a boundary.
Fig. 17 shows the cross-sectional profile of the rail obtained by piecewise functioning with straight lines and arcs based on the start and end coordinates of the straight lines and the center coordinates and radius of the arcs included in the cross-sectional dimensions of the rails of the actual product. The graph showing the relationship between the obtained dimensional drawing cross-sectional shape (the shape of the rail of the actual product) and the second-order differential data obtained by performing the second-order differential processing after the moving average processing on the dimensional drawing cross-sectional shape is shown. . As indicated by the arrows in FIG. 17, convex peaks appear at the boundary between the column part 1a and the lower chin part 1d and at the boundary between the column part 1a and the leg part 1e. Thus, since the change in the curvature of the surface shape becomes large at the boundary between both ends of the pillar portion 1a, the change in the curvature of the surface shape is evaluated using the second-order difference value, and the boundary of the pillar portion 1a is determined. Can be detected.
In step S404, the boundary detection unit 507 extracts a region inside the boundary detected in step S403 as a columnar region.

  In addition, since the various dimensions of the rail 1 are determined in advance, only the boundary of one end of the pillar portion 1a may be detected, and the boundary of the other end may be detected using the known length of the pillar portion 1a. . For example, when the boundary between the column part 1a and the lower jaw 1d is not in the field of view or cannot be detected, after detecting the boundary between the column part 1a and the leg part 1e, the column part 1a is used using the known length of the column part 1a. And the boundary of the lower jaw 1d is detected.

  FIG. 18 shows an example of a result obtained by performing the moving average process and the second-order difference process on the cross-sectional profile obtained by the measurement on the actual rail. Length as a reference for performing moving average and second-order difference processing (for moving average, the interval between the maximum value and the minimum value regarding the x-coordinate of the target points. The distance D for second-order difference processing. 16), the depth of the curved portion shown as MC in FIG. 16 is sufficiently larger than the depth of the surface defect to be detected, and the curved portion that connects the column portion 1a to the lower jaw 1d, and the column portion 1a. It is preferable that the curved portion connected to the leg portion 1e is set to about 10 mm so that the reference length is not included at the same time, and the number of points corresponding to the length to be processed is an odd number. As a result of the second floor difference process, an upwardly convex peak corresponding to the boundary between the pillar portion 1a and the leg portion 1e appeared clearly. In the horizontal axis of FIG. 18, since the left side is the lower chin 1d side and the right side is the leg 1e side, the peak is searched from the right side, and the peak position is detected as the boundary between the column part 1a and the leg 1e. And after detecting the boundary of the column part 1a and the leg part 1e, the boundary of the column part 1a and the chin lower part 1d was detected using the length of the known column part 1a.

In this embodiment, the moving average processing with uniform weight is combined with the second-order difference processing. However, for example, the movement with non-uniform weight is such that the weight is reduced as the distance from the target point to be smoothed increases. You may combine an average process.
In the example of FIG. 18, the score of the moving average process is the same as the score of the second-order difference process. However, in steps S401 and S402 of FIG. If so, the moving average processing score may be less than the second-order difference processing score.
The second-order difference process is not limited as long as the change in the degree of curvature of the surface shape of the measurement target part is evaluated and the boundary of the measurement target part is detected.
In steps S401 and S402, a process in which the effects of both the smoothing calculation and the Laplacian calculation are expected such that a local peak such as a beard does not occur and a smooth peak appears may be used. Examples thereof include Laplacian / Gaussian filters, Savitzky / Golay filters, and the like. Further, the filter length may be determined so that a noise is suppressed as a processing result and a smooth peak is obtained.

(Details of engraving detection and mask processing)
Next, details of the marking detection / mask processing in step S16 will be described.
As described above, the depth image created by the shape correction processing unit 509 is subjected to concave portion binarization and convex portion binarization, and further subjected to expansion and contraction processing and then labeling processing. Then, a stamp 11 having a known size range is detected to create a stamp mask image.
FIG. 19 is a flowchart showing stamp detection / mask processing. FIG. 20 shows an example of an image at each stage of engraving detection / mask processing.
In step S <b> 1601, the marking mask image creation unit 510 determines the marking band 11 on the depth image created by the shape correction processing unit 509 with reference to the boundary of the column part 1 a detected by the boundary detection unit 507. Since the height position and width of the marking strip 11 in the pillar portion 1a are known, if a boundary of the pillar portion 1a is known, a region having a constant width away from the border portion 1a can be defined as the marking strip 11.

FIG. 20A shows an enlarged view of the vicinity of the stamped band 11 in the depth image created by the shape correction processing unit 509. As shown in FIG. 20 (a), the letter “ALE” is stamped on the stamped band 11.
Note that the depth image shown in FIG. 20A is displayed in an unsigned 8-bit gray scale in which brightness is corrected by adding an offset of 128 to the original depth image for easy visual understanding. The concave portion is dark and the convex portion is brightly displayed.
Further, since the marking becomes a concave portion, the whole marking “ALE” should be displayed darkly, but a bright portion can also be seen. This is because the engraving has a very steeply inclined shape, so that there is a shadow portion where the linear laser beam does not reach or the diffuse reflected light is not imaged, and an error occurs in the phase detection processing or phase continuation processing. it is conceivable that. In addition, although the inscription itself is concave, it is considered that the bulge of the meat around it may become convex. In addition, since the width of the stamped band 11 is wide, a bounce caused by an approximation error in function fitting occurs and it is detected as a convex part.

In step S1602, the marking mask image creation unit 510 performs threshold processing at a preset depth on the marking band 11 determined in step S1601 out of the depth image created by the shape correction processing unit 509, and the concave portion Perform binarization. Similarly, the marking mask image creation unit 510 performs threshold processing at a preset height on the marking band 11 determined in step S1601 out of the depth image created by the shape correction processing unit 509, so that the convex portion 2 Perform valuation.
When the binarization binarization is performed on the depth image shown in FIG. 20A by whitening the concave pixels below a certain threshold and blackening the other pixels, as shown in FIG. A binarized image is obtained. In this concave portion binarized image, a portion detected as a convex portion is generated as described above, and hence a part of the marking is missing.
Moreover, when the convex part binarization which makes the pixel of the convex part more than a certain threshold value white with respect to the depth image shown to Fig.20 (a) and makes other pixels black is shown in FIG.20 (c). In this way, a convex binarized image is obtained. In the convex binarized image, a part of the mark missing from the concave binarized image is detected.
Since the marking becomes a concave portion in this way, in principle, only the concave portion binarization may be used. However, by performing the convex portion binarization together, the marking can be made to appear with high accuracy.

In step S1603, the marking mask image creation unit 510 performs a logical OR operation (OR operation) of the pixel bits of the concave portion binarized image and the convex portion binarized image obtained in step S1602 to binarize the uneven portion. A binarized image obtained by combining and synthesizing is obtained.
FIG. 20D shows an uneven portion binarized image in which white portions are synthesized by a logical sum operation.

In step S1604, the marking mask image creation unit 510 performs expansion and contraction processing on the uneven portion binarized image obtained in step S1603.
In the uneven portion binarized image shown in FIG. 20D, a defect is seen in a part of the inscription. Therefore, if a marking mask image is created in a narrow range of the white portion detected in the uneven portion binarized image, the uneven portion around the marking may be overdetected as a surface defect. Therefore, the large uneven portions among the binarized uneven portions are combined into one blob, and an expansion and contraction process for removing the small uneven portions is performed.
FIG. 20 (e) shows an uneven portion binarized image subjected to expansion and contraction processing. In order to connect the pixels constituting the inscription in the uneven portion binarized image, expansion processing was performed five times. Thereafter, in order to suppress fine noise, the shrinkage process was performed four times, and as shown in FIG. 20E, an uneven portion binarized image subjected to the expansion and contraction process was obtained. What is necessary is just to determine suitably the frequency | count of an expansion process and a contraction process according to a target image.
Note that the dilation processing is to perform processing for making the target pixel white if the adjacent pixel has at least one white pixel for the target pixel of the binarized image and output it. However, it is a well-known process in the image processing field that has the effect of grouping white parts that are separated. In addition, the contraction process is a process in which if the pixel of interest in the binarized image has at least one black pixel adjacent to the pixel of interest, the process is performed on the entire image and output as black. This processing is well known in the image processing field and has the effect of removing white noise pixels.

  In step S <b> 1605, the marking mask image creation unit 510 converts a pixel-connected portion into one blob for the concavo-convex portion binarized image subjected to the expansion and contraction processing obtained in step S <b> 1604. Perform the labeling process. For example, a circumscribed rectangle of a portion where pixels are connected is defined as one blob.

In step S1606, the marking mask image creation unit 510 determines, from among the blobs labeled in step S1605, a blob whose size (area of a circumscribed rectangle, etc.) corresponds to the marking, and selects it. . Since the size range of the marking is known, if the size of the blob is a size corresponding to the marking, the blob can be selected as the marking.
FIG. 20 (f) shows a labeling image in which the circumscribed rectangle of the blob selected as the marking is superimposed on the depth image of FIG. 20 (a).
Although the blob is a circumscribed rectangle here, other shapes (such as an ellipse) may be used. Moreover, it is good also as a large rectangle which encloses a circumscribed rectangle so that engraving can fully be covered.

In step S 1607, the marking mask image creation unit 510 creates a marking mask image with all pixel bits in the circumscribed rectangle being 0 and all pixel bits outside the circumscribed rectangle being 1 for the blob selected as the marking. create.
In this embodiment, the marking mask image is created by setting the pixel bits to 0 and 1 inside and outside the circumscribed rectangle of the blob selected as the marking. If the white portion of the digitized image can cover the marking, a marking mask image with pixel bits 0 and 1 inside and outside the white portion itself may be created.

Using the marking mask image created as described above, the defect detection processing unit 511 performs an AND operation (AND operation) of the depth image created by the shape correction processing unit 509 and the pixel bits of the marking mask image. To mask the markings appearing in the depth image. Thereby, since the unevenness corresponding to the marking is eliminated on the depth image, the marking can be detected without overdetecting the surface defect and without overlooking the surface defect.
FIG. 21 (a) shows a depth image in which deposits appear on the stamp band 11. In addition, although it does not appear in the position shown in FIG. 21 (a), the stamp band 11 is stamped. FIG. 21B shows an image drawn by superposing a circumscribed rectangle surrounding the deposit on the depth image. By masking the marking with the marking mask image in this way, the marking was not overdetected as a surface defect, and the surface defect appearing on the marking band 11 could be detected without missing it.

Although the present invention has been described together with the embodiments, the above-described embodiments are merely examples of implementation in carrying out the present invention, and the technical scope of the present invention is interpreted in a limited manner by these. It must not be. That is, the present invention can be implemented in various forms without departing from the technical idea or the main features thereof.
In this embodiment, the hot stamping which becomes a concave portion is taken as an example. For example, the column portion 1a may be provided with information having a convex portion called a roll mark. The present invention can be similarly applied. In this case, in principle, only the convex portion binarization may be used, but the roll mark can be made to appear with high accuracy by performing the concave portion binarization together. The term “engraved” as used in the claims of the present application means information such as characters and marks that are broadly represented by unevenness on the premise that the size range is known.

The image processing apparatus 50 to which the present invention is applied is realized by a computer system including a CPU, a ROM, a RAM, and the like, and the functions or processing of the image processing apparatus 50 are realized by the CPU executing a computer program. Note that the image processing apparatus 50 does not have to be a single apparatus, and may include a plurality of devices.
The present invention also provides software (program) that implements the functions of the present invention to a system or apparatus via a network or various storage media, and the system or apparatus computer reads out and executes the program. It is feasible.

  DESCRIPTION OF SYMBOLS 10: Laser apparatus, 20: Rod lens, 30: Delay integration type camera, 40: Timing signal generation part, 50: Image processing apparatus, 60: Display apparatus, 501: A / D conversion part, 502: Pre filter part, 503 : Orthogonal sine wave generation unit, 504a, 504b: low-pass filter unit, 505: phase calculation unit, 506: phase continuation processing unit, 507: boundary detection unit, 508: centering processing unit, 509: shape correction processing unit, 510: Imprint mask image creation unit, 511: Defect detection processing unit

Claims (10)

Among long materials, a predetermined part having a curved surface at the end in the short direction is used as a measurement target part.
Irradiation means for irradiating the measurement target portion with periodically modulated linear laser light along the short direction of the long material;
With a delay integration type imaging means for imaging reflected light from the measurement target part,
While continuously shifting the irradiation position of the linear laser light in the longitudinal direction of the long material, the reflected light from the measurement target portion is picked up by the imaging means to obtain a light cut image, and the measurement target portion A surface defect inspection system for detecting surface defects of
A shape image creating means for creating a shape image representing the surface shape of the measurement target part based on the light section image output from the imaging means;
Shape correction processing means for performing shape correction processing for converting the curved surface shape of the measurement target portion into a plane with respect to the shape image created by the shape image creation means;
By performing at least one of concave binarization and convex binarization on the image obtained by the shape correction processing means, and performing a labeling process, a mark having a known size is detected. A marking mask image creating means for creating a marking mask image;
Defect detection processing means for detecting a surface defect of the measurement target portion by masking the image obtained by the shape correction processing means using the marking mask image created by the marking mask image creating means. Surface defect inspection system characterized by that.   2. A boundary detection unit that detects a boundary of the measurement target part by evaluating a change in curvature of the surface shape of the measurement target part in the shape image generated by the shape image generation unit. Item 10. The surface defect inspection system according to Item 1.   The boundary detection unit performs second-order difference processing or Laplacian calculation processing on the shape image created by the shape image creation unit, and detects a peak position as a boundary of the measurement target unit. The surface defect inspection system according to claim 2.   4. The surface defect inspection system according to claim 2, wherein the marking mask image creating unit detects the marking in a predetermined region determined based on a boundary detected by the boundary detection unit. 5.   The engraving mask image creating means performs a concave portion binarization and a convex portion binarization on the image obtained by the shape correction processing means, and further performs a labeling process after performing expansion and contraction processing. 5. The surface defect inspection system according to claim 1, wherein the marking is detected by: The long material is a rail,
The said measurement object part is a pillar part in the said rail, Comprising: The said pillar part is connected to a chin lower part at one end, and is connected to a leg part at the other end. Surface defect inspection system. The shape image creating means includes:
Phase calculating means for calculating a phase shift at each position of the fringe image obtained by sequentially arranging the light sectioned images obtained by the imaging means;
Detecting the position where the phase shift is discontinuous based on the phase shift obtained by the phase calculation means, and connecting the phase shift at the detected position to make the phase shift continuous, The surface defect inspection system according to claim 1, further comprising a phase continuation processing unit that creates a shape image representing a surface shape of a measurement target portion.   The shape correction processing unit performs a fitting process for obtaining a fitting curve for a shape image created by the shape image creating unit or a shape image obtained by performing a predetermined process on the shape image created by the shape image creating unit. And performing a difference calculation between the fitting curve and the shape image created by the shape image creating means, and averaging the data obtained by the difference calculation by a predetermined length in the longitudinal direction of the long material. The surface according to any one of claims 1 to 7, wherein a difference calculation between a value obtained by performing a low-pass filter process on the averaged value and the data obtained by the difference calculation is performed again. Defect inspection system. Among long materials, a predetermined part having a curved surface at the end in the short direction is used as a measurement target part.
Irradiation means for irradiating the measurement target portion with periodically modulated linear laser light along the short direction of the long material;
With a delay integration type imaging means for imaging reflected light from the measurement target part,
While continuously shifting the irradiation position of the linear laser light in the longitudinal direction of the long material, the reflected light from the measurement target portion is picked up by the imaging means to obtain a light cut image, and the measurement target portion A surface defect inspection method for detecting surface defects of
A shape image creating step for creating a shape image representing the surface shape of the measurement target part based on the light section image output from the imaging means;
A shape correction processing step for performing shape correction processing for converting the curved surface shape of the measurement target portion to a plane with respect to the shape image created in the shape image creation step;
By performing at least one of concave binarization and convex binarization on the image obtained in the shape correction processing step, and performing a labeling process, a mark with a known size is detected. A stamp mask image creating step for creating a stamp mask image;
Using a marking mask image created in the marking mask image creating step, masking the image obtained in the shape correction processing step, and detecting a surface defect in the measurement target portion. A method for inspecting a surface defect. Among long materials, a predetermined part having a curved surface at the end in the short direction is used as a measurement target part.
Irradiation means for irradiating the measurement target portion with periodically modulated linear laser light along the short direction of the long material;
With a delay integration type imaging means for imaging reflected light from the measurement target part,
While continuously shifting the irradiation position of the linear laser light in the longitudinal direction of the long material, the reflected light from the measurement target portion is picked up by the imaging means to obtain a light cut image, and the measurement target portion A program for detecting surface defects of
A shape image creating means for creating a shape image representing the surface shape of the measurement target part based on the light section image output from the imaging means;
Shape correction processing means for performing shape correction processing for converting the curved surface shape of the measurement target portion into a plane with respect to the shape image created by the shape image creation means;
By performing at least one of concave binarization and convex binarization on the image obtained by the shape correction processing means, and performing a labeling process, a mark having a known size is detected. A marking mask image creating means for creating a marking mask image;
Using a marking mask image created by the marking mask image creating means, masking the image obtained by the shape correction processing means, and a computer as a defect detection processing means for detecting a surface defect of the measurement target portion A program to make it work.

Images