JP7040325B2 - Concavo-convex part detection method and uneven part detection device - Google Patents

Description

The present invention relates to an uneven portion detecting method and an uneven portion detecting device.

For example, inclusions and cracks may occur on the surface of slabs, billets, and other slabs produced in ironworks due to various causes. Therefore, for example, scarfing is performed to remove defects existing on the surface of the slab by melting. However, depending on the scarfing method, unmelted portions may be generated on the surface of the slab. Therefore, it is also considered to acquire height information of the slab surface by a light cutting method of irradiating the slab surface with a laser beam, and to inspect uneven parts such as unmelted parts by using the height information (the height information). For example, see Patent Document 1). In such a conventional inspection method, for example, a surface shape image of a slab is created based on height information, and the surface shape image is binarized to detect uneven portions such as uncut portions.

Japanese Patent No. 5488953

However, the surface of the slab before laminating is not flat, and the shape before laminating remains even after laminating. Therefore, simply binarizing the surface shape image after laminating as in the conventional inspection method is sufficient. There is a problem that the unevenness caused by the unmelted portion is buried in the shape before the wrought, and the region of the uneven portion cannot be detected.

In view of the above circumstances, it is an object of the present invention to provide an uneven portion detecting method and an uneven portion detecting device capable of detecting a concave or convex region on the surface of an object to be inspected.

The uneven portion detection method of the present invention is a concave-convex portion detection method for detecting concave portions or convex portions existing on the surface of the inspected object from a surface shape image generated based on the measurement result of the surface of the inspected object by the optical cutting method. Therefore, the difference between the first difference value indicating the difference in brightness between adjacent pixels in the first direction in the surface shape image and the difference in brightness between adjacent pixels in the second direction orthogonal to the first direction. The difference strength calculation for calculating the difference strength based on the brightness difference value calculation step for calculating the second difference value indicating the above and the first difference value and the second difference value for each pixel of the surface shape image. The step, the difference direction calculation step of calculating the difference direction based on the first difference value and the second difference value for each pixel of the surface shape image, and the inside of the surface shape image based on the difference intensity. From the edge specific image generation step of generating an edge specific image showing the edge of the concave portion or the convex portion in the above and the edge specific image, a gradient image having a constant gradient magnitude is generated based on the difference direction. The gradient image generation step, the source term calculation step of calculating the source term of the two-dimensional Poisson equation from the difference of the gradient between adjacent pixels in the predetermined direction in the gradient image, and the above 2 using the source term. It has a solution step of finding a solution of a dimensional Poisson equation and a concavo-convex detection step of binarizing the solution of the two-dimensional Poisson equation and detecting a region of the concave portion or the convex portion.

The uneven portion detecting device of the present invention is an uneven portion detecting device that detects concave or convex portions existing on the surface of the inspected object from a surface shape image generated based on the measurement result of the surface of the inspected object by the optical cutting method. Therefore, the difference between the first difference value indicating the difference in brightness between adjacent pixels in the first direction in the surface shape image and the difference in brightness between adjacent pixels in the second direction orthogonal to the first direction. The difference strength calculation for calculating the difference strength based on the brightness difference value calculation unit for calculating the second difference value indicating the above and the first difference value and the second difference value for each pixel of the surface shape image. For each pixel of the surface shape image, the difference direction calculation unit that calculates the difference direction based on the first difference value and the second difference value, and the inside of the surface shape image based on the difference intensity. From the edge specific image generation unit that generates an edge specific image showing the edge of the concave portion or the convex portion in the above and the edge specific image, a gradient image having a constant gradient magnitude is generated based on the difference direction. The gradient image generation unit, the source term calculation unit that calculates the source term of the two-dimensional Poisson equation from the difference in the gradient between adjacent pixels in a predetermined direction in the gradient image, and the source term 2 above. It is provided with a solution unit for obtaining a solution of a dimensional Poisson equation and a concavo-convex detection unit for binarizing the solution of the two-dimensional Poisson equation and detecting a region of the concave portion or the convex portion.

The present invention calculates the source term of the two-dimensional Poisson's equation based on the gradient image in which the magnitude of the gradient of the edge pixel corresponding to the edge of the concave or convex portion is set to a constant value, and solves the two-dimensional Poisson's equation. Therefore, the edge shape of the concave portion or the convex portion can be restored. In the present invention, by binarizing the solution of the two-dimensional Poisson's equation, the concave or convex region can be clearly distinguished from other regions, so that the concave or convex region on the surface of the object to be inspected can be defined. Can be detected.

It is a schematic diagram which shows the whole structure of the unevenness part detection apparatus by this invention. It is a schematic diagram for demonstrating the outline of the uneven part detection method by this invention. It is a surface shape image used in the verification test. It is a block diagram which shows the circuit structure of the arithmetic unit by embodiment of this invention. It is an image in which each pixel of the unit gradient image created in the verification test is color-coded according to the size of the x component of the gradient G. It is an image showing an uncut portion created based on the solution data obtained in the verification test. It is an unevenness detection image created in the verification test.

(1) Outline of the present invention First, an outline of the method for detecting uneven portions of the present invention will be described. In the present specification, the method for detecting the uneven portion of the present invention will be described by taking as an example the case of detecting the convex unmelted portion on the surface of the slab. The surface of the slab is melted by spraying a high-temperature combustion frame onto the surface using a device called a hot scarf. In the scarfing process to remove surface defects such as inclusions and cracks, a part of the surface of the slab is not melted, or the surface of the slab is not sufficiently melted, resulting in a convex shape on the surface of the slab. Unmelted part may be formed.

The uneven portion detecting method of the present invention is executed by the uneven portion detecting device 10 as shown in FIG. 1, and detects the uncut portion 7 formed on the surface of the slab 4. The slab 4 to be inspected has a rectangular shape having a predetermined thickness, and is conveyed in the conveying direction of arrow 9 in FIG. 1 by a conveying device (not shown). In the present specification, the transport direction of the slab 4 is defined as the Y direction as the longitudinal direction (also referred to as the vertical direction) of the slab 4, and the direction orthogonal to the transport direction on the surface 4a of the slab 4 is the width of the slab 4. It is defined as the X direction as the direction (also called the horizontal direction).

In the uneven portion detection method of the present invention, first, such a slab surface 4a is irradiated with linear light 2 by a laser light source 1, and a light cutting line 3 formed in the width direction of the slab surface 4a by the linear light 2. Is imaged by the area camera 5.

The area camera 5 captures an image of the optical cut line 3 formed on the surface of the slab 4, and generates a band-shaped optical cut line image including the optical cut line 3. The optical cut line image has a longitudinal direction along the X direction, which is the width direction of the slab 4, in accordance with the optical cut line 3 extending in the width direction of the slab 4, and is the longitudinal direction of the slab 4. It has a lateral direction along the Y direction. The area camera 5 sequentially images the optical cutting lines 3 formed on the surface of the conveyed slab 4, so that a plurality of optical cutting line images including the optical cutting lines 3 are imaged along the longitudinal direction of the slab 4. Can be acquired.

The area camera 5 sequentially sends out the obtained optical cut line image to the arithmetic unit 11. The arithmetic unit 11 measures the surface shape of the slab 4 by the optical cutting method using the optical cutting line image, and obtains a surface shape image showing the height of the slab surface 4a.

Here, since the uneven portion to be detected, such as the unmelted portion, has a sharp shape change in its contour, the surface shape image showing the height of the slab surface 4a is based on the difference in brightness of the surface shape image. Only the edge part of the uncut part can be extracted. In the present invention, the contour of the unmelted portion is detected from the extraction result of the edge portion and formulated as a problem of filling the inside of the contour. At this time, it is necessary to select a method that allows the defect of the contour or the case where the contour appears twice due to the influence of noise and the threshold value. Therefore, in the present invention, a method using Poisson's equation is adopted as a method having high robustness when restoring a shape from a gradient. Hereinafter, the method for detecting the uneven portion of the present invention, which detects the region of the unmelted portion by using the Poisson's equation, will be described with reference to FIG.

First, in step S0, as shown in FIG. 1, an optical cut line image is generated by the area camera 5. The optical cut line image generated by the area camera 5 is sent to the arithmetic unit 11, and the arithmetic unit 11 executes steps S1 to S13 shown in FIG. In step S1, the center of gravity is calculated to obtain the position of the center of gravity of the optical cutting line 3 extending in the longitudinal direction from the luminance distribution in the longitudinal direction at each position in the lateral direction of the optical cutting line image, and this is the optical cutting line. It is set to the reflection position of 3. Since this calculation of the center of gravity is disclosed in, for example, Japanese Patent No. 5488953, the description thereof will be omitted here.

In step S2, the light cut line image sequentially obtained from the area camera 5 and the reflection position of the light cut line 3 obtained by the calculation of the center of gravity in step S1 are used, as shown in FIG. 3 based on the light cut method. , Generates a surface shape image I showing the unevenness of the surface of the slab 4. Here, if a convex portion or a concave portion is present on the surface of the slab 4, the position of the optical cutting line 3 in the optical cutting line image is displaced according to the shape of the convex portion or the concave portion. In step S2, the height information of the surface of the slab 4 is obtained by obtaining the positional deviation of the optical cutting line 3 with reference to the reflection position of the optical cutting line 3 obtained by the calculation of the center of gravity in step S1.

Then, in step S2, a surface shape image I representing the height difference of the surface of the slab 4 as the difference in brightness is generated from the obtained height information. In this way, the surface shape image I in which the entire surface of the slab surface 4a is captured can be generated. FIG. 3 is an enlarged image of a part of the surface shape image I generated according to the above-mentioned steps S1 and S2 based on a plurality of optical cut line images acquired from the area camera 5 in time series. In FIG. 3, the width direction of the slab 4 is shown as the x direction, and the longitudinal direction of the slab is shown as the y direction. The surface shape image I of FIG. 3 shows the side portion 51 of the slab 4 extending in the longitudinal direction, and in FIG. 3, the right side region of the side portion 51 shows the surface of the slab 4. In FIG. 3, since the heights of the unmelted portion 7 and the melted portion of the slab 4 are different, the unmelted portion 7 on the surface of the slab 4 appears as a difference in brightness. In FIG. 3, the vertically elongated bright portion 52 is the unmelted portion 7.

Here, in the present specification, the positions of each pixel in such a surface shape image I are set by the x-axis (horizontal direction in FIG. 3) indicating the width direction (horizontal direction) of the slab 4 and the longitudinal direction of the slab 4. It is represented by Cartesian coordinates (x, y) using the y-axis (vertical direction in FIG. 3) indicating (vertical direction). The origin (0,0) of the Cartesian coordinates is, for example, a pixel at the leftmost and lowest positions in the surface shape image I, as shown in FIG. In this embodiment, the same applies to the other images described below.

In step S3, for each pixel of the surface shape image I, the luminance between adjacent pixels in the horizontal direction (first direction) using the equation Ix = I (x + 1, y) -I (x, y). The difference is calculated, and the first difference value Ix is obtained as the lateral difference. In step S4, for each pixel of the surface shape image I, the luminance between adjacent pixels in the vertical direction (second direction) using the equation Iy = I (x, y + 1) -I (x, y). The difference is calculated, and the second difference value Iy is obtained as the vertical difference. In steps S3 and S4, a luminance gradient is calculated in which the first difference value Ix is indicated by the x component and the second difference value Iy is indicated by the y component.

Next, in step S5, the square root of the sum of squares of the first difference value Ix and the second difference value Iy is calculated, and the difference intensity Gmag is calculated. Then, the luminance value of each pixel of the surface shape image I is converted into the differential intensity Gmag to generate the differential intensity image. The coordinate positions (x, y) of each pixel of the differential intensity image are the same as those of the surface shape image I.

In step S6, the inverse tangent of Iy / Ix (Gori = atan2 (Ix, Iy)) is calculated using the first difference value Ix and the second difference value Iy, and the difference direction (direction of brightness gradient and x) is calculated. It is the angle formed by the axis and is expressed as Gori). A difference direction image is created by converting the luminance value of each pixel of the surface shape image I into the difference direction Gori. Actually, the inverse tangent of Iy / Ix is calculated using the atan2 function, which is common in numerical calculation programs, and Gori is calculated. The coordinate positions (x, y) of each pixel of the difference direction image are the same as those of the surface shape image I.

Subsequently, in step S7, the binarization process and the thinning process are sequentially performed on the difference intensity image obtained in step S5. Specifically, first, the luminance value of the difference intensity image is binarized with a predetermined threshold value to generate a binarized image. In such a binarized image, a portion having a large difference intensity Gmag is shown. That is, in the binarized image, the edge (corresponding pixel) of the unmelted portion 7 in the surface shape image I can be shown based on the difference intensity Gmag. As the binarized image, for example, a first value such as 1 is assigned to a pixel having a luminance value equal to or greater than a predetermined threshold value, and a second value such as 0 is assigned to a pixel having a luminance value less than a predetermined threshold value. Assign the value of.

In the binarized image in which the brightness value of the differential intensity image is binarized in step S7, the edge of the uncut portion 7 can be represented, but the portion of the binarized image in which the line width of the edge is 2 pixels or more. Exists. Therefore, in step S7, the binarized image is subjected to the thinning process of narrowing the line width of the edge to one pixel while maintaining the connection relationship of each edge pixel, and the thinning image is generated. Since this thinning process is a general process for generating a thinned image by thinning the edge of the unmelted portion 7 using a general thinning algorithm as an image processing algorithm, the description thereof is described here. Is omitted. The coordinate positions (x, y) of each pixel of the thinned image are the same as those of the surface shape image I. Thus, in step S7, the binarization process and the thinning process are sequentially performed to obtain a thinned image in which the thinned edge pixels are indicated by the binarized luminance value Gbin (here, the first value).

Next, in step S8, the magnitude of the gradient G is determined from the thinned image in which the edge pixels are binarized with the luminance value Gbin (first value “1”) based on the difference direction Gori obtained in step S6. Generate a gradient image with a constant value. The gradient image is an image in which each pixel has two components of gradient G = (dx, dy) (dx and dy will be described later). Here, dx and dy indicate a component (x component) in the x direction and a component (y component) in the y direction of the gradient G. The magnitude of the gradient G is a value calculated by the square root of the sum of squares of dx and dy (that is, √ (dx ² + dy ² )) as shown in step S8 in FIG.

In the present invention, for example, as a gradient image, a unit gradient image in which the gradient of the edge pixel is a unit gradient Gu having a magnitude of 1 and the magnitude of the gradient other than the edge pixel is 0 (dx = dy = 0) is generated. do. The dx, which is the x component of the unit gradient Gu, can be calculated by the equation dx = cos (Gori). Further, dy, which is the y component of the unit gradient Gu, can be calculated by the equation dy = sin (Gori). Here, the difference direction Gori of the formula dx = cos (Gori) and the formula dy = sin (Gori) corresponds to the coordinate position of the pixel for which the unit gradient Gu is obtained, and the difference direction of the pixel at the coordinate position in the difference direction image. Gori.

Next, in steps S9 and S10, the right-hand side (hereinafter referred to as a source term) of the two-dimensional Poisson equation (2) described later is calculated from the difference in gradient between adjacent pixels in a predetermined direction in the gradient image. For example, in step S9, the equation dxx = dx (x + 1, y) -dx (x, y) is used for each pixel of the unit gradient image, and the difference dxx of the unit gradient Gu between adjacent pixels in the horizontal direction is used. Is calculated. In step S10, the equation dyy = dy (x, y + 1) -dy (x, y) is used for each pixel of the unit gradient image to calculate the difference dyy of the unit gradient Gu between adjacent pixels in the vertical direction. do. Then, the source term of the two-dimensional Poisson's equation is calculated by adding dxx and dyy for each pixel.

In step S11, the source terms calculated in steps S9 and S10 are used to solve the two-dimensional Poisson's equation to calculate the solution data. Since the two-dimensional Poisson's equation is a partial differential equation as described later, the two-dimensional Poisson's equation can be solved by using a general method as a method for solving the partial differential equation. In the present invention, a two-dimensional Poisson's equation is solved by using a general difference method as a method for solving a partial differential equation. For the solution of the two-dimensional Poisson's equation, H (x, y) indicating the height H at the pixel at the coordinate position (x, y) is obtained as the solution data. By using the height H (x, y) of the pixel at the coordinate position (x, y) obtained as the solution data as the luminance value, an image representing the uncut portion 7 can be created, and the uncut portion can be created. Area 7 can be restored. In the present invention, since the source term is calculated from the unit gradient image whose width is one pixel by thinning, H (x, y) is approximately a value between 0 and 1.

Subsequently, in step S12, the solution data of the two-dimensional Poisson equation obtained in step S11 is binarized at a predetermined threshold value. Since the height H (x, y) at the pixel of the coordinate position (x, y) obtained in step S11 is obtained as a real number as a solution of the two-dimensional Poisson equation, the obtained height H (x, y). By binarizing with a threshold value between 0 and 1, the region of the uncut portion 7 can be obtained. Specifically, for example, when 0.5 is set as the threshold value, the value of the height H (x, y) above the threshold value is set as the first value such as 1, for example, and the height H (x, y) below the threshold value is set. The value of the height H (x, y) of the solution data is binarized with the value as a second value such as 0.

In step S13, using the binarized solution data, for example, an unevenness detection image showing H (x, y) divided into two values by luminance is generated. For example, a pixel having a height H (x, y) of 1 is set to white, and a pixel having a height H (x, y) of 0 is set to black to generate an unevenness detection image. In step S13, a region of pixels having a height H value of 1 (here, a white region) appearing in the unevenness detection image is detected as a region of the unmelted portion 7. In this way, in the surface shape image I, even if the contour (edge) of the unmelted portion is missing due to the influence of noise or the threshold value or the edge appears twice, the unevenness detection image in which the shape of the edge is restored from the gradient. Can be obtained, and the region of the uncut portion 7 can be detected from the unevenness detection image.

(2) Shape restoration from a gradient by a two-dimensional Poisson's equation In the present invention, the two-dimensional Poisson's equation is used based on the principle described below. Generally, if the gradient ∇H (x, y) at the coordinate position (x, y) is measurable instead of the height H (x, y) at the coordinate position (x, y), then the gradient If there is no noise in the measurement result of ∇H (x, y), the gradient ∇H (x, y) is lined by any route in order to satisfy the no-vortex condition (∇ × ∇H (x, y) = 0). By integrating, the height H (x, y) can be restored from the gradient ∇H (x, y) at each coordinate position (x, y). Here, ∇ is the so-called nabla operator.

However, when the gradient ∇H (x, y) is actually measured by a physical measuring means, noise is usually generated, so that the above-mentioned no-vortex condition is not satisfied, and the integration result differs depending on the path. The height H (x, y) cannot be restored from the gradient ∇H (x, y).

Therefore, it is necessary to find an approximate solution with some assumptions. In this regard, the present invention uses a method of deriving a two-dimensional Poisson's equation from a least squares solution. Specifically, the gradient ∇H (x, y) = (∂H / ∂x, ∂H / ∂y) of the desired height H (x, y) and the gradient image calculated from the surface shape image I We considered finding the least squares solution that minimizes the sum of squares of the difference from the gradient G = (dx, dy) of each pixel. The sum of squares of the differences between the gradient ∇H = (∂H / ∂x, ∂H / ∂y) and the actually calculated gradient G = (dx, dy) is expressed by the following equation (1).

In this case, the problem is to calculate the least squares solution that minimizes J in this equation (1). Here, this problem is a variational problem related to H (x, y), and the stationary point of J is calculated to calculate the solution, but the Euler-Lagrange equation for calculating the stationary point of J. Is the following equation (2).

Equation (2) is a two-dimensional Poisson's equation for height H (x, y). Therefore, by solving this two-dimensional Poisson's equation, the height H (x, y) can be obtained as the solution. In the present invention, the difference dxx of the x component of the gradient G obtained in step S9 is regarded as the partial derivative ∂dx / ∂x of the equation (2), and the difference dyy of the y component of the gradient G obtained in step S10 is expressed by the equation. The source term is calculated by regarding the partial differential of (2) as ∂dy / ∂y. In the present invention, as described above, the two-dimensional Poisson's equation is solved by using the difference method, but the method using FFT (Fast Fourier Transform), which is one of the solutions of the partial differential equation, also solves the two-dimensional Poisson's equation. Can be solved.

(3) Concavo-convex portion detection device according to the embodiment of the present invention Next, an arithmetic unit 11 that executes the above-mentioned uneven portion detection method will be described. As shown in FIG. 4, the arithmetic unit 11 includes a surface shape image generation unit 20, an image processing unit 30, an unevenness detection calculation unit 40, and a display unit 12, and each of these circuit units is via a bus 14. Is connected to the control unit 13. The control unit 13 comprehensively controls each circuit unit and executes various processes according to a surface shape image generation program, a gradient image generation program, an unevenness detection program, and the like.

The surface shape image generation unit 20 includes an acquisition unit 21, a center of gravity calculation unit 22, and an image generation unit 23, and generates a surface shape image I from the optical cut line images sequentially acquired from the area camera 5. The acquisition unit 21 acquires an optical cut line image from the area camera 5 and sends it to the center of gravity calculation unit 22 and the image generation unit 23. The center of gravity calculation unit 22 performs the center of gravity calculation described in step S1 above on the optical cut line image, identifies the reflection position of the optical cut line 3 from the optical cut line image, and sends this to the image generation unit 23. ..

The image generation unit 23 performs the process of step S2 above, and generates a surface shape image I showing the entire slab surface 4a based on the sequentially obtained optical cut line images and the reflection positions of the optical cut lines 3. The image generation unit 23 sends the surface shape image I to the luminance difference value calculation unit 31 of the image processing unit 30.

The image processing unit 30 includes a brightness difference value calculation unit 31, a difference intensity image generation unit 32, a difference direction image generation unit 33, a binarized image generation unit 34 as an edge specific image generation unit, and a thin line image generation unit 35. , A unit gradient image generation unit 36 as a gradient image generation unit is provided. Upon receiving the surface shape image I, the luminance difference value calculation unit 31 performs the processes of steps S3 and S4, and calculates the first difference value Ix and the second difference value Iy for each pixel of the surface shape image I. do. The brightness difference value calculation unit 31 uses the difference intensity image generation unit 32 and the difference intensity image generation unit 32 to obtain the coordinate positions (x, y) of each pixel in the surface shape image I and the first difference value Ix and the second difference value Iy of the pixel. It is sent to the difference direction image generation unit 33, respectively.

The difference intensity image generation unit 32 has a difference intensity calculation unit and an image generation unit (not shown), and performs the process of step S5 above. In the difference intensity image generation unit 32, the difference intensity calculation unit calculates the difference intensity Gmag, and the image generation unit generates a difference intensity image in which the luminance value of each pixel of the surface shape image I is the difference intensity Gmag. The difference intensity image generation unit 32 sends the difference intensity image to the binarized image generation unit 34. The difference direction image generation unit 33 has a difference direction calculation unit and an image generation unit (not shown), and performs the process of step S6 above. In the difference direction image generation unit 33, the difference direction calculation unit calculates the difference direction Gori, and the image generation unit generates a difference direction image in which the luminance value of each pixel of the surface shape image I is the difference direction Gori. The difference direction image generation unit 33 sends the difference direction image to the unit gradient image generation unit 36.

The binarized image generation unit 34 performs the edge specifying process described in step S7 above on the difference intensity image, and the brightness value is larger than a predetermined threshold value (for example, 0.5) for each pixel of the difference intensity image. A first value of 1 is assigned to the pixel, and a second value of 0 is assigned to a pixel having a brightness value less than a predetermined threshold value to generate a binarized image. For this threshold value, a differential intensity image is calculated from a sample shape image (surface shape image I) including the uneven portion (unmelted portion 7) to be detected, and the difference intensity image is binarized so that the contour of the uneven portion is extracted. It is desirable to decide in. The binarized image generation unit 34 sends a binarized image showing a portion having a large difference intensity to the thinned image generation unit 35.

The thinning image generation unit 35 performs the thinning process described in step S7 above on the binarized image, and uses the thinning algorithm to thin the edge pixels in the binarized image into a thinned image. Is generated, and this is sent to the unit gradient image generation unit 36.

The unit gradient image generation unit 36 performs the process of step S8 above to set the unit gradient Gu in which the magnitude of the gradient in the edge pixels is 1, and the magnitude of the gradient in the pixels other than the edge pixels is 0 (dx = dy = 0). ) Unit gradient image is generated. The unit gradient image generation unit 36 sends the unit gradient image to the source term calculation unit 41 of the unevenness detection calculation unit 40.

The unevenness detection calculation unit 40 includes a source term calculation unit 41, a solution unit 42, and an unevenness detection unit 43. The source term calculation unit 41 performs the processes of steps S9 and S10, and for each pixel of the unit gradient image, the difference dxx of the gradient between the adjacent pixels in the horizontal direction and the difference dxx between the adjacent pixels in the vertical direction. Calculate the difference dyy of the gradient. Then, the source term calculation unit 41 calculates the source term of the two-dimensional Poisson equation by adding dxx and dyy for each pixel, and sends the source term to the solution unit 42.

The solution unit 42 performs the process of step S11, solves the two-dimensional Poisson's equation using the source term received from the source term calculation unit 41, and calculates the solution data. The solution unit 42 sends the calculated solution data to the unevenness detection unit 43. The unevenness detection unit 43 first performs the process of step S12, and sets the value of the height H at each coordinate position (x, y) of the solution data to 1 and 0 with a predetermined value (for example, 0.5) as a threshold value. To become.

The unevenness detection unit 43 subsequently performs the process of step S13, and uses the binarized solution data to generate an unevenness detection image in which the value of the height H (x, y) is represented as the brightness. In this case, the unevenness detection unit 43 generates, for example, an unevenness detection image in which a pixel having a brightness of 1 is set to white and a pixel having a brightness of 0 is set to black. The unevenness detection unit 43 detects the white region in the unevenness detection image as the region of the unmelted portion 7.

The display unit 12 is a general monitor such as a liquid crystal monitor. The display unit 12 displays the unevenness detection image created by the unevenness detection unit 43, and causes the operator to confirm the region of the unmelted portion 7. Further, the display unit 12 displays the surface shape image I, the unit gradient image, and the like each time, causes the operator to confirm the image generated in the detection process of the unmelted portion 7, and verifies the detection process of the unmelted portion 7. You can also do it.

(4) Action and Effect In the above configuration, the uneven portion detecting device 10 of the present invention acquires the surface shape image I generated based on the measurement result of the slab surface 4a by the optical cutting method. The uneven portion detecting device 10 determines the difference in luminance between adjacent pixels in the horizontal direction (first direction) in the surface shape image I as the first difference value Ix = I (x + 1, y) -I (x,, Calculated as y), and the difference in brightness between adjacent pixels in the vertical direction (second direction) orthogonal to the horizontal direction is the second difference value Iy = I (x, y + 1) -I (x, y). (Brightness difference value calculation process). The uneven portion detecting device 10 calculates the difference strength Gmag for each pixel of the surface shape image I based on the first difference value Ix and the second difference value Iy (difference strength calculation step), and each of the surface shape image I. For the pixel, the difference direction Gori is calculated based on the first difference value Ix and the second difference value Iy (difference direction calculation step).

The uneven portion detecting device 10 generates a thinned image in which the edge of the unmelted portion 7 in the surface shape image I is specified based on the difference intensity Gmag (edge specifying image generation step). The uneven portion detecting device 10 is configured to generate a gradient image (gradient image generation step) in which the magnitude of the gradient G is a constant value (1 in the present embodiment) based on the difference direction Gori from the thinned image.

The uneven portion detection device 10 of the present invention has a difference in the x component of the gradient between adjacent pixels in the horizontal direction in the gradient image, dxx = dx (x + 1, y) -dx (x, y), and the vertical direction. From the difference dyy = dy (x, y + 1) -dy (x, y) of the y component of the gradient between adjacent pixels, the source term of the two-dimensional Poisson equation is calculated (source term calculation step). The solution of the two-dimensional Poisson's equation was obtained using the source term (solving step), the solution of the two-dimensional Poisson's equation was binarized, and the region of the uncut portion 7 was detected (unevenness detection step).

Therefore, the uneven portion detecting device 10 of the present invention calculates the source term of the two-dimensional Poisson's equation based on the gradient image in which the magnitude of the gradient of the edge pixel corresponding to the edge of the uncut portion 7 is set to a constant value. By solving the two-dimensional Poisson's equation, the edge shape of the uncut portion 7 can be restored. In the present invention, by binarizing the solution of the two-dimensional Poisson's equation, the region of the unmelted portion 7 can be clearly distinguished from other regions, so that the unmelted portion 7 on the surface of the object to be inspected can be clearly distinguished. Area can be detected.

(5) Modification Example In the above embodiment, the case where the region of the unmelted portion 7 of the slab 4 is detected as the convex portion has been described, but the present invention is not limited to this, and the unmelted portion 7 is like the above. It is also possible to detect the region of the concave portion formed on the surface instead of the convex portion. For example, the present invention can also detect areas of defects such as wrinkles and flaws on the surface of the slab 4. Further, the present invention can detect regions of recesses or protrusions formed on the surface of other steel products, products of other materials, etc., in addition to the slab 4.

In the above-described embodiment, the thinned image is applied as the edge specific image, but the present invention is not limited to this, and the binarized image before the thinned image is generated may be applied as the edge specific image.

In the above embodiment, the width direction of the slab 4 and the lateral direction of the surface shape image I are parallel to each other, and the surface shape image I showing the entire surface of the slab 4 is used to use the uncut portion 7 on the surface of the slab 4. However, the present invention is not limited to this, and the width direction of the slab 4 and the lateral direction of the surface shape image I do not have to be parallel, and a part of the surface of the slab 4 is enlarged. The captured surface shape image I may be used.

In the above embodiment, the case where the transport direction of the slab 4 is the vertical direction of the surface shape image I and the direction orthogonal to the transport direction of the slab 4 is the horizontal direction of the surface shape image I has been described. The transport direction of the slab 4 may be the horizontal direction of the surface shape image I, and the direction perpendicular to the transport direction of the slab 4 may be the vertical direction of the surface shape image I.

In the above embodiment, the source term calculation unit 41 determines the difference dxx of the x component dx of the gradient G between the pixels adjacent in the horizontal direction and the pixels adjacent to each other in the y direction for each pixel of the unit gradient image. The case where the difference dyy of the y component dy of the gradient G is calculated and the source term is calculated by adding these is described, but the present invention is not limited to this. For example, either one of the difference dxx or the difference dyy may be calculated, and the calculated difference may be used as the source term.

In this case, the unit gradient image generation unit 36 substitutes the angle Gori at the same coordinate position in the angle image into the equation dx = sgn (cos (Gori)) for the edge pixels in the thinned image, and dy = 0. (When calculating only dxx) or dx = 0 and substituting the angle Gori into the formula dy = sgn (sin (Gori)) (when calculating only dyy), the unit gradient Gu = with a magnitude of 1 Calculate (dx, dy) and create a unit gradient image.

Here, the unmelted portion 7 of the slab 4 often has an elongated shape. Therefore, when the source term is calculated using only the difference dxx or only the difference dyy, the direction orthogonal to the longitudinal direction of the uncut portion 7 (in the case of the surface shape image I in FIG. 3, the width direction of the slab 4) x It is better to calculate the difference dxx of the x component of the gradient G with one adjacent pixel in the direction). By doing so, the two-dimensional Poisson's equation can be solved accurately.

In the above embodiment, the case where the optical cutting line 3 is imaged by using the area camera 5 and the surface shape image I created by the optical cutting method using the obtained optical cutting line image is used has been described. The invention is not limited to this, and a surface created by an optical cutting method using an optical cut line image obtained by imaging an optical cut line using an integration delay type camera by the method described in Japanese Patent No. 3845354. A shape image may be used.

In the above-described embodiment, the optical cut line image, the surface shape image I, the difference intensity image, the difference direction image, the binarized image, the thinning image, the unit gradient image, the unevenness detection image and the like will be described in the present specification. As for the image, not only the form as a concrete image that can be displayed on the display unit but also the data before being generated as an image is naturally included.

Further, in the above embodiment, the unevenness detection unit 43 uses the binarized solution data to generate an image with the value of the height H as the brightness, and among the pixels of the image, the pixel with the brightness of 1 is made white. The case where the pixel with the brightness of 0 is set to black and the unevenness detection image is generated has been described, but the present invention is not limited to this, and the unevenness detection image in which a predetermined color is set for each pixel can be generated. Instead, a pixel having a height H value of 1 may be detected as a region of the uncut portion 7 from the binarized solution data.

(6) Verification test A slab whose surface was melted was prepared, and the surface shape image I of the slab was created by the surface shape image generation unit 20 of the uneven portion detection device 10. A part of the created surface shape image I was trimmed to prepare the surface shape image I shown in FIG. The x direction in FIG. 3 is the width direction of the slab.

In the surface shape image I shown in FIG. 3, an elongated uncut portion 7 along the longitudinal direction of the slab 4 was visually confirmed as a bright portion 52. Using the surface shape image I subjected to such trimming treatment, the bright portion 52 in the image was detected as the region of the unmelted portion 7 by using the uneven portion detecting device 10. In the verification test, only the lateral difference dxx was used as the source term, and the two-dimensional Poisson equation was solved to obtain the solution.

First, the prepared surface shape image I was sent to the image processing unit 30 to create a unit gradient image. FIG. 5 is an image in which the color of each pixel of the unit gradient image is set to three colors according to the x component of the gradient. Pixels with an x component of +1 are set to white, pixels with an x component of 0 are set to gray, and pixels with an x component of -1 are set to black. In FIG. 5, it was confirmed that the edge 53 was color-coded in white and the edge 54 was color-coded in black, and that only the pixel of the edge had a unit gradient.

Next, the unit gradient image was sent to the unevenness detection calculation unit 40, and the region of the bright part 52 was detected. FIG. 6 is an image created by converting the value of the height H of each pixel into a luminance value based on the solution data obtained by the solution unit 42. In FIG. 6, the region of the unmelted portion 7 appears as the white region 56, but it can be confirmed that there is a blur 55 in a part of the edge of the region 56.

In FIG. 7, the unevenness detection unit 43 binarizes the height value of the solution data into 1 and 0 at a threshold value of 0.5, and makes the pixel with a height of 1 white and the pixel with a height 0 black. It is a created unevenness detection image. As shown in FIG. 7, the blur 55 of the edge of the white region 56 seen in FIG. 6 is also eliminated, and the region of the unmelted portion 7 can be clearly distinguished from other regions as the white region 56. It was confirmed that the region of the unmelted portion 7 could be detected.

1 Laser light source 3 Optical cut line 4 Slab 5 Area camera 7 Unmelted part 10 Uneven part detection device 31 Luminance difference value calculation part 34 2 Quantified image generation part 35 Thinning image generation part 36 Unit gradient image generation part 41 Source item Calculation unit 42 Solving unit 43 Concavo-convexity detection unit

Claims (6)

A method for detecting unevenness, which detects concave or convex portions existing on the surface of the inspected object from a surface shape image generated based on the measurement result of the surface of the inspected object by the optical cutting method.
A first difference value showing the difference in brightness between adjacent pixels in the first direction in the surface shape image and a second difference in brightness between adjacent pixels in the second direction orthogonal to the first direction. The luminance difference value calculation process for calculating the two difference values and the
A difference strength calculation step of calculating the difference strength based on the first difference value and the second difference value for each pixel of the surface shape image.
A difference direction calculation step of calculating a difference direction based on the first difference value and the second difference value for each pixel of the surface shape image.
An edge specific image generation step of generating an edge specific image showing the edge of the concave portion or the convex portion in the surface shape image based on the difference intensity.
A gradient image generation step of generating a gradient image having a constant gradient magnitude from the edge specific image based on the difference direction.
A source term calculation step for calculating the source term of the two-dimensional Poisson's equation from the difference in the gradient between adjacent pixels in a predetermined direction in the gradient image.
A solution step for finding a solution to the two-dimensional Poisson's equation using the source term, and
A method for detecting unevenness, which comprises a step of binarizing the solution of the two-dimensional Poisson's equation and detecting a region of the concave or convex. In the gradient image generation step, a unit gradient image in which the magnitude of the gradient of the edge pixel indicating the edge in the edge specific image is set to 1 and the magnitude of the gradient of the pixels other than the edge pixel is set to 0 is obtained. The uneven portion detection method according to claim 1, which is generated. The edge specific image generation step is
A binarized image generation step of generating a binarized image showing the magnitude of the difference intensity, and a binarized image generation step.
The uneven portion detection method according to claim 1 or 2, further comprising a thinning image generation step of generating a thinning image in which the edge is thinned from the binarized image as the edge specific image. The surface of the object to be inspected is the surface of the slab.
The source term calculation step is any of claims 1 to 3, wherein the source term is calculated from the difference in the gradient between adjacent pixels in the width direction orthogonal to the longitudinal direction of the slab in the gradient image. The method for detecting uneven portions according to item 1. The surface of the object to be inspected is the surface of the slab.
The source term calculation step is described in any one of claims 1 to 3, wherein the source term is calculated from the difference in the gradient between adjacent pixels in the longitudinal direction of the slab in the gradient image. Unevenness detection method. An uneven portion detecting device that detects concave portions or convex portions existing on the surface of the inspected object from a surface shape image generated based on the measurement result of the surface of the inspected object by the optical cutting method.
A first difference value showing the difference in brightness between adjacent pixels in the first direction in the surface shape image and a second difference in brightness between adjacent pixels in the second direction orthogonal to the first direction. 2 Luminance difference value calculation unit to calculate the difference value,
A difference strength calculation unit that calculates the difference strength based on the first difference value and the second difference value for each pixel of the surface shape image.
A difference direction calculation unit that calculates a difference direction based on the first difference value and the second difference value for each pixel of the surface shape image.
An edge specific image generation unit that generates an edge specific image showing the edge of the concave portion or the convex portion in the surface shape image based on the difference intensity.
A gradient image generation unit that generates a gradient image having a constant gradient magnitude from the edge specific image based on the difference direction.
A source term calculation unit that calculates the source term of the two-dimensional Poisson's equation from the difference in the gradient between adjacent pixels in the gradient image in a predetermined direction.
A solution unit for finding a solution to the two-dimensional Poisson's equation using the source term,
An uneven portion detecting device including a concave-convex detecting portion that binarizes a solution of the two-dimensional Poisson's equation and detects a concave-convex portion or a convex portion region.

Images