A visual guidance-based glass substrate rotary jig alignment method and system

Abstract

The application belongs to the technical field of image processing, and particularly relates to a glass substrate rotating jig alignment method and system based on visual guidance, which comprises the following steps: constructing a complex gradient vector field according to the gradient of each pixel point in a glass substrate gray image, performing fourth-order phase mapping on the complex gradient vector field, constructing a harmonic transformation field, calculating a square box saliency map according to the harmonic transformation field, calculating a circular saliency map based on the complex gradient vector field by using the gradient convergence characteristics of a solid circle point, performing dual fusion on the square box saliency map and the circular saliency map, constructing a target probability potential field, searching for an extreme point in the target probability potential field to determine a true fiducial point coordinate, and driving a rotating jig to perform alignment compensation. The application can resist the interference of metal grid textures, straight-line scratches and circular stains, has rotation invariance, and realizes high-precision visual alignment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing. More particularly, the present application relates to a visual guidance-based glass substrate rotary jig alignment method and system. BACKGROUND

[0002] In the automated production process of glass substrates, the accuracy of the visual alignment system directly determines the yield of subsequent hole insertion, screen printing or lamination processes. In order to ensure the uniqueness and robustness of alignment, the reference point is usually designed as a composite geometric structure of concentrically nested solid black dots and hollow black squares.

[0003] However, as the process requirements improve, the measured objects gradually develop in the direction of ultra-thin and high light transmission. In actual engineering scenarios, due to the high transparency of the glass substrate, the surface features of the metal jig table below are clearly imaged by the industrial camera when shooting the reference point. These features are usually regular metal grid textures, vacuum suction hole arrays or random straight-line scratches. In addition, the circular stains formed by ink droplets falling in the production environment and the shadows caused by uneven edge lighting are highly similar to the reference point in grayscale features.

[0004] Traditional visual algorithms face serious challenges when dealing with such complex scenes. A single circle detection algorithm cannot distinguish between true reference points with square constraints and independent circular stains. A pure square detection algorithm not only has a large amount of calculation, but is also easily disturbed by straight-line scratches, which are misjudged as square edges. Although existing feature extraction methods based on histogram of oriented gradients can distinguish between cluttered textures to some extent, they lack robustness when dealing with rotating targets and cannot effectively distinguish between one-way distributed straight-line interference and four-way orthogonal distributed square structures, resulting in a decrease in the recognition accuracy of composite reference points and positioning failure, affecting the accuracy and stability of glass substrate rotary jig alignment. SUMMARY

[0005] To solve the technical problems of large background interference of glass substrates, reference point recognition easily affected by scratches and stains, and lack of rotational invariance in traditional algorithms in the prior art, the present application provides solutions in the following aspects.

[0006] In a first aspect, the present application provides a visual guidance-based glass substrate rotary jig alignment method, comprising:

[0007] The gray scale image of the glass substrate is collected, a complex gradient vector field is constructed according to the gradient of each pixel point in the gray scale image of the glass substrate, the complex gradient vector field is subjected to fourth-order phase mapping to construct a harmonic transformation field, the box saliency value of the pixel point is calculated according to the fourth-order harmonic vector of the pixel point in the harmonic transformation field to form a box saliency map, the circular saliency map is calculated based on the complex gradient vector field by using the gradient convergence characteristics of the solid circular point, the box saliency map and the circular saliency map are dual fused based on a topological concentric constraint mechanism to construct a target probability potential field, the extreme point in the target probability potential field is searched to determine the true fiducial point coordinates, and the rotation tool is driven to perform alignment compensation according to the true fiducial point coordinates.

[0008] The application maps the image gradient field to the complex domain, constructs a fourth-order harmonic transformation field, uses the four-way orthogonal gradient of the box structure to cause constructive interference in the four times frequency domain, and uses the continuous gradient of the circular structure to cause destructive interference in the four times frequency domain, realizes the extraction of the box structure, effectively suppresses the circular interference and the single-direction straight line scratch, uses the all-directional gradient convergence characteristics of the solid circular point to construct a circular saliency map which is only sensitive to the center of the circle, and through the topological concentric constraint mechanism, the box saliency map and the circular saliency map are dual fused, the center of the circle is forced to fall within the tolerance range of the box center, so that the circular stain and the straight line scratch are completely filtered out, the recognition accuracy is greatly improved, the application is based on complex vector operation and has rotation invariance, does not need to pre-correct the attitude angle of the glass substrate, and has simple parameters and high calculation efficiency, and meets the real-time alignment demand on the industrial high-speed production line.

[0009] Preferably, the complex gradient vector field is constructed, including: calculating the gradient of the glass substrate gray scale image in the horizontal direction and the vertical direction by using the Scharr operator respectively; calculating the complex gradient vector of each pixel point in the glass substrate gray scale image to form a complex gradient vector field, the complex gradient vector satisfies the expression: , wherein, represents the complex gradient vector at the pixel point . represents the gradient amplitude of the pixel point . represents the gradient direction angle of the pixel point . represents the horizontal direction gradient of the pixel point . represents the vertical direction gradient of the pixel point . is an imaginary unit.

[0010] This invention utilizes the Scharr operator to extract edge gradient information from glass substrate images and constructs it into a complex gradient vector field. This effectively reduces the angular error in rotating target edge detection. By using the complex form, it simultaneously preserves the intensity and direction information of the gradient, transforming traditional scalar image processing into vector field analysis. This provides a mathematical foundation for subsequent processing of geometric symmetry and frequency domain harmonic analysis using complex rotation operators.

[0011] Preferably, the fourth harmonic vector satisfies the expression: In the formula, Represents pixels The fourth harmonic vector at that location; Represents pixels Gradient magnitude at; Represents pixels The gradient direction angle at that location; 4 is the imaginary unit; 4 is the harmonic order.

[0012] This invention constructs a fourth-order harmonic vector to map the gradient field of an image to the fourth harmonic domain. By utilizing the unique fourfold rotational symmetry of the square geometric structure, the gradient directions corresponding to the four sides that are originally orthogonal in space are phase aligned in the frequency domain. This creates conditions for subsequent feature resonance enhancement through vector accumulation. At the same time, it causes interference features that do not have this symmetry to be phase discretized in the frequency domain, thus achieving spectral separation between target features and background noise from a mathematical level.

[0013] Preferably, the step of calculating the saliency value of a pixel includes: setting a first neighborhood centered on the current pixel, calculating the vector sum of the fourth harmonic vectors corresponding to all pixels in the first neighborhood, and using the magnitude of the vector sum as the saliency value of the current pixel.

[0014] This invention calculates the vector sum and magnitude of the fourth harmonic vectors in the first neighborhood as the salient value of the box. Utilizing the principle of vector superposition, the gradients of the four sides of the box structure undergo constructive interference after phase alignment, generating a high-intensity resonant response. Meanwhile, the gradients of circular structures or random textures, due to continuous changes in direction or random distribution, undergo destructive interference during the superposition process, with the response value approaching zero. Thus, without relying on complex geometric fitting, a high signal-to-noise ratio extraction of the reference point box components is achieved, and this process naturally possesses rotation invariance.

[0015] Preferably, calculating the circular saliency map includes: calculating the circular saliency value at each pixel to construct a circular saliency map, wherein the circular saliency value satisfies the expression: ,in, Represents pixels The significant value at the circular location; In pixels a second neighborhood centered at the center pixel point; denotes the gradient magnitude at the pixel point in the second neighborhood; denotes the gradient magnitude at the pixel point in the second neighborhood; denotes the gradient direction angle at the pixel point in the second neighborhood; denotes the gradient direction angle at the pixel point in the second neighborhood; denotes the displacement vector angle in the image coordinate system pointing to the center pixel point; denotes the displacement vector angle in the image coordinate system pointing to the center pixel point; denotes the displacement vector angle in the image coordinate system pointing to the center pixel point; is the ratio of a circle's circumference to its diameter.

[0016] The present application corrects the reverse gradient feature in the white background black point scene by calculating the circular significant value and introducing phase compensation, and can enhance the response intensity of the solid black circle by counting the gradient centripetal degree in the second neighborhood; the present application utilizes the geometric convergence feature of the omnidirectional gradient of the solid circle point, and forms a complement with the discrete gradient feature of the square box structure, so that the circular component of the reference point is effectively extracted, and the misjudgment of the hollow square box or non-circular interference is excluded.

[0017] Preferably, the target probability potential field is constructed, including: calculating the probability potential of each pixel point belonging to the center of the true reference point, and constructing the target probability potential field, wherein the probability potential of the pixel point belonging to the center of the true reference point satisfies the expression: , denotes the probability potential of the pixel point belonging to the center of the true reference point; denotes the probability potential of the pixel point belonging to the center of the true reference point; denotes the circular significant value at the pixel point; denotes the circular significant value at the pixel point; denotes the Gaussian kernel with a standard deviation of ; denotes the convolution operation; denotes the square significant value at the pixel point; denotes the square significant value at the pixel point; denotes the standard deviation parameter of the Gaussian kernel.

[0018] The present application establishes a tolerance area by smoothing and expanding the square significant value with the Gaussian kernel, and implements the topological concentric constraint by dual fusion of the circular significant value and the expanded square significant value, by constructing the target probability potential field; the present application requires that the high-response circle center must fall within the high-response square center tolerance range, so as to filter out the circular stains or straight-line scratches with only a single feature, and ensure that only the true reference point with the square-circle nested topological structure can produce a global extreme value, thereby greatly improving the accuracy and anti-interference ability of identification.

[0019] Preferably, the standard deviation parameter of the Gaussian kernel is , wherein is the allowed physical concentricity deviation, is the image resolution.

[0020] Preferably, the determining the true fiducial point coordinate comprises: searching for a global maximum value pixel coordinate in the target probability potential field; selecting a third neighborhood centered on the global maximum value pixel coordinate; and fitting a peak vertex of the response surface by using a binary quadratic polynomial to obtain a true fiducial point coordinate at a sub-pixel level.

[0021] The present application breaks through the limitation of the physical pixel resolution of the image sensor by searching for a global maximum value in the target probability potential field, selecting a third neighborhood for binary quadratic polynomial fitting, and using surface fitting method to analyze the sub-pixel position of the response peak, thereby further extracting a high-precision true fiducial point center coordinate on the basis of retaining the robustness of global search, and meeting the stringent requirements of micron-level alignment accuracy for precision manufacturing.

[0022] Preferably, the driving the rotary tool to perform alignment compensation according to the true fiducial point coordinate comprises: calculating a position deviation of the true fiducial point coordinate from a preset target position; converting the position deviation into a pulse signal in a mechanical coordinate system and sending the pulse signal to a motion control card to drive an X-axis, a Y-axis and a Z-axis motor of the rotary tool to perform linkage compensation until the position deviation is less than a preset process threshold.

[0023] In a second aspect, the present application provides a glass substrate rotary tool alignment system based on visual guidance, comprising a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned glass substrate rotary tool alignment method based on visual guidance is realized.

[0024] By using the above technical solution, the above-mentioned glass substrate rotary tool alignment method based on visual guidance is generated into a computer program and stored in the memory to be loaded and executed by the processor, so that a terminal device is manufactured according to the memory and the processor, and the use is convenient.

[0025] ​The beneficial effects of the present application are as follows: the present application maps the image edge features to the complex domain by constructing a complex gradient vector field, and uses a fourth-order harmonic transform field to make the four-way orthogonal gradients of the box component of the reference point phase-aligned and constructively interfere in the frequency domain, while the reference point circular component and the straight line interference destructively interfere due to phase dispersion, thereby realizing the extraction of the box structure without angle pre-correction; at the same time, the present application captures the center position of the solid black circle edge gradient using the omnidirectional convergence characteristics; on this basis, the present application uses a topological concentric constraint mechanism to couple the box saliency map and the circular saliency map, establishes an effective verification region with the box center as the reference, and mathematically and logically forces the high-response center to be located inside the high-response box, thereby filtering out circular stains, straight line scratches and background grid texture interference with only a single geometric feature; finally, in combination with sub-pixel level curved surface fitting and motion control feedback, high-precision and high-robustness automatic alignment of the glass substrate under complex background interference is realized. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 is a flow chart schematically showing a glass substrate rotation jig alignment method based on visual guidance in the present application;

[0027] Figure 2 is a glass substrate grayscale image;

[0028] Figure 3 is a gradient amplitude image of the glass substrate grayscale image;

[0029] Figure 4 is a box saliency map;

[0030] Figure 5 is a circular saliency map;

[0031] Figure 6 is a visualization image of the target probability potential field;

[0032] Figure 7 is a true reference point positioning result schematic diagram. DETAILED DESCRIPTION

[0033] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the present application.

[0034] The specific embodiments of the present application will be described in detail below with reference to the drawings.

[0035] The embodiment of the application discloses a visual guidance-based glass substrate rotating jig alignment method Figure 1 , comprising steps S1-S5:

[0036] S1, collect a glass substrate gray image, and construct a complex gradient vector field according to gradients of each pixel point in the glass substrate gray image.

[0037] It should be noted that the target of the application is a black reference point under high-light backlight, in order to extract a weak geometric edge from a background containing a metal grid texture and random noise, and provide a complete vector basis for subsequent frequency domain analysis, the application no longer processes the gradient in the Euclidean space, but maps the gradient information of the image to the complex domain, processes the symmetry of the geometric structure through a complex rotation operator, and meanwhile, the amplitude intensity and direction information of the gradient are retained.

[0038] Specifically, a backlight illumination unit located below the glass substrate is turned on, an industrial camera installed above the rotating jig is used to collect a glass substrate gray image containing a reference point, in order to obtain better rotation invariance, a Scharr operator is used to calculate a gradient of the glass substrate gray image in the horizontal direction and a gradient of the glass substrate gray image in the vertical direction, respectively. Exemplarily, Figure 2 is the glass substrate gray image, Figure 3 is a gradient amplitude image of the glass substrate gray image.

[0039] For each pixel point in the glass substrate gray image, a complex gradient vector is calculated to form a complex gradient vector field, and the complex gradient vector satisfies the expression:

[0040]

[0041] In the formula, represents the complex gradient vector at the pixel point ; represents the gradient amplitude of the pixel point , and the calculation method is , which reflects the intensity of the edge, when the pixel point is located on the edge of the reference point or the edge of the background texture, is larger, and vice versa; represents the gradient direction angle of the pixel point , and the calculation method is , and the value range is ; represents the horizontal direction gradient of the pixel point ; represents the vertical direction gradient of the pixel point ; is an imaginary unit. It is a two-parameter arctangent function.

[0042] The complex gradient vectors of all pixels are used to construct a complex gradient vector field.

[0043] Thus, this invention transforms a scalar grayscale image of a glass substrate into a dense complex gradient vector field, providing a mathematical basis for subsequent harmonic analysis.

[0044] S2. Perform fourth-order phase mapping on the complex gradient vector field to construct a harmonic transform field. Calculate the saliency value of the pixel based on the fourth-order harmonic vector of the pixel in the harmonic transform field to form a saliency map.

[0045] It should be noted that the reference point box component exhibits significant 4-fold rotational symmetry, meaning that the normal directions of its four sides differ by 90 degrees in space. In the conventional gradient domain, these four directions are mutually exclusive and difficult to directly accumulate for enhancement. However, in the fourth harmonic domain, the phases of all orthogonal directions will resonate. For example, suppose the overall rotation angle of the box is... Then the gradient directions of its four sides are respectively After multiplying by 4, these four phases become By utilizing the periodicity of complex phases, they point in the same direction in the complex plane. However, for circular or linear structures, this resonance phenomenon does not occur or is extremely weak. This invention utilizes this physical property to perform fourth-order phase mapping on the complex gradient field to construct a harmonic transformation field, thereby enhancing the rectangular structure and suppressing circular and linear structures.

[0046] Specifically, a fourth-order phase mapping is performed on the complex gradient field to construct a harmonic transform field:

[0047]

[0048] In the formula, Represents pixels The fourth harmonic vector at that location; Represents pixels Gradient magnitude at; Represents pixels The gradient direction angle at that location; 4 is the imaginary unit; 4 is the harmonic order, corresponding to the fourfold symmetry of the square.

[0049] Furthermore, define a first neighborhood. The saliency value of the bounding box of the center pixel is calculated based on the fourth harmonic vectors of the neighboring pixels within the first neighborhood of the center pixel:

[0050]

[0051] In the formula, represents the block saliency value at pixel point ; represents the first neighborhood centered at pixel point with a side length of Since the block saliency value is generated depending on the phase resonance of the four edges of the block, the value of needs to be able to cover the block component of the reference point, in the embodiment, the value of is set to 1.0 to 1.2 times the block length of the reference point, for example, 100 pixels, in other embodiments, the implementer can set the value of according to the actual pixel size of the reference point in the image; represents the fourth-order harmonic vector at pixel point in the first neighborhood centered at pixel point ; represents the accumulation of complex vectors; represents the length of the complex module.

[0052] When the first neighborhood covers the window at the center of the block, the gradient vectors of the four edges of the block are phase-aligned in the fourth-order harmonic domain, the length of the accumulation reaches a maximum value, resulting in constructive interference, so that is large; when the first neighborhood covers the window at the center of the circular component or the circular stain of the reference point, the gradient direction of the first neighborhood changes continuously from 0 to , which is mapped to , and then rotates rapidly from 0 to , so that the length of the vector after integration tends to 0, resulting in destructive interference, so that is extremely small; when the first neighborhood covers the window on the straight line scratch, there is only one-way energy, lacking multi-directional superposition, so that is significantly lower than the center of the block.

[0053] Figure 4 Thus, a block saliency map that is only sensitive to block structure is obtained, which is the block saliency map.

[0054] S3, using the gradient convergence characteristics of the solid circle, a circular saliency map is calculated based on the complex gradient vector field.

[0055] It should be noted that in order to accurately locate the reference point and exclude the interference of the empty block, a complementary algorithm to harmonic resonance is needed to detect the solid circle, since the present application is aimed at a white background black point scene, the gradient direction of the edge of the solid black circle points to the outside of the circle, i.e. the high gray area, while the displacement vector pointing to the center of the circle points to the inside of the circle, both of which are in opposite relationship. Therefore, the present application constructs a circular structure saliency model with a phase compensation mechanism to correctly capture the black circle feature.

[0056] Specifically, the circular saliency value of each pixel point is calculated as follows:

[0057]

[0058] wherein, represents the circular saliency value at the pixel point ; is a second neighborhood with the pixel point as the center and a side length of Since the reference point circular component is nested in the square frame, its size is small, and in order to avoid introducing unnecessary background gradient noise, should be less than the side length of the first neighborhood In the present embodiment, is set to 1.2 to 1.5 times the diameter of the solid black circle inside the reference point, for example, 40 pixels, and in other embodiments, the implementer can set according to the actual pixel size of the reference point in the image; represents the gradient amplitude at the pixel point in the second neighborhood; represents the gradient direction angle at the pixel point in the second neighborhood; represents the displacement vector angle of the pixel point in the second neighborhood in the image coordinate system pointing to the center pixel point , and the calculation formula is ; is the ratio of the circumference of a circle to its diameter, used to realize 180° phase compensation to adapt to the gradient reverse characteristic in the white background black circle scenario; is used to measure the consistency of the gradient direction and the centripetal direction; is used to eliminate the gradient component away from the center and only keep the centripetal convergent component; is the coordinate of the center pixel point in the image coordinate system, is the coordinate of the pixel point in the image coordinate system.

[0059] When the pixel point is located at the center of the solid black circle, the edge gradient is omnidirectional to the outside of the circle, and after phase compensation , the equivalent direction is directed to the center, the accumulated centripetal component is the most, and is extremely high; when the pixel point is located at the center of the square frame, only a small amount of gradient at the four corners is directed to the center, and is weak; when the pixel point is located on the straight line scratch, there is no convergent center, and is extremely low.

[0060] Thus, a circle saliency map sensitive to the circle structure only is obtained, Figure 5 a circle saliency map.

[0061] S4, based on the topological concentric constraint mechanism, dual fusion of the box saliency map and the circle saliency map is carried out to construct a target probability potential field.

[0062] It should be noted that after steps S2 and S3, two characteristic maps with completely orthogonal properties are obtained, that is, the box saliency map responding to the box and suppressing the circle, and the circle saliency map responding to the circle and suppressing the box. The topological definition of the true reference point is that a strong response circle is surrounded and concentric with a strong response box. Simple pixel-level multiplication may cause signal attenuation due to printing concentricity error, therefore, the present application introduces a Gaussian relaxation constraint to establish an effective verification area with the box center as the reference, as long as the circle center falls within this area, it can be determined as a true reference point.

[0063] Specifically, the probability potential of each pixel point belonging to the center of the true reference point is calculated to construct the final target probability potential field, and the probability potential of the pixel point belonging to the center of the true reference point satisfies the expression:

[0064]

[0065] In the formula, represents the pixel point The probability potential of the pixel point belonging to the center of the true reference point; represents the circle saliency value at the pixel point ; represents a Gaussian kernel with a standard deviation of , which is used for smoothing and expanding operation; represents convolution operation; represents the box saliency value at the pixel point ; represents the standard deviation parameter of the Gaussian kernel, in the embodiment, the value of is set to 5 pixels, in other embodiments, the implementer can set according to the concentricity error range allowed by the printing process, for example, if the allowed physical concentricity deviation is mm, and the image resolution is pixels / mm, then should be set to .

[0066] When the pixel point is located at the circle stain, although is high, but tends to 0 due to the lack of box structure around it, so that is suppressed; when the pixel point When located at the intersection of the empty square or grid, Although higher, the center without a circular structure leads to Approaching 0, so that Is suppressed; only when the pixel point Located at the true fiducial point, both are high, the product produces a global extreme value.

[0067] Exemplarily, Figure 6 The visualization image of the target probability potential field.

[0068] S5, searching for extreme points in the target probability potential field to determine the true fiducial point coordinates, and driving the rotary tool to perform alignment compensation according to the true fiducial point coordinates.

[0069] Specifically, the global maximum pixel coordinates are searched in the target probability potential field, and a third neighborhood is selected with the coordinates as the center, the probability potential that all pixel points in the third neighborhood belong to the true fiducial point center is fitted by using a binary quadratic polynomial, the peak vertex of the fitted surface is obtained, and the accurate coordinates at the sub-pixel level are obtained. It should be noted that the binary quadratic polynomial is used for fitting in the present application, which is aimed at sub-pixel refinement of the global extreme value point that has been locked, and only the local surface features near the peak value need to be concerned, therefore, the scale of the third neighborhood should be much smaller than the first neighborhood and the second neighborhood, and is usually set to or pixels.

[0070] Exemplarily, Figure 7 The true fiducial point positioning result schematic diagram.

[0071] The accurate coordinates are compared with the preset theoretical reference coordinates, the positional deviation between the two is calculated, the positional deviation is converted into a pulse signal in the mechanical coordinate system, and is sent to the motion control card to drive the X-axis, Y-axis and Z-axis motors of the rotary tool to perform linkage compensation until the deviation is less than the preset process threshold.

[0072] The embodiment of the present application also discloses a glass substrate rotary tool alignment system based on visual guidance, comprising a processor and a memory, the memory stores computer program instructions, and when the computer program instructions are executed by the processor, a glass substrate rotary tool alignment method based on visual guidance according to the present application is realized.

[0073] The above system also includes a communication bus and a communication interface and other components familiar to those skilled in the art, the settings and functions of which are known in the art, and therefore will not be repeated here.

Claims

1. A method for aligning a glass substrate using a vision-guided rotating fixture, characterized in that, include: Acquire grayscale images of the glass substrate and construct a complex gradient vector field based on the gradient of each pixel in the grayscale image of the glass substrate; A fourth-order phase mapping is performed on the complex gradient vector field to construct a harmonic transformation field. The saliency value of the pixel is calculated based on the fourth-order harmonic vector of the pixel in the harmonic transformation field. This includes: setting a first neighborhood with the current pixel as the center, calculating the vector sum of the fourth-order harmonic vectors corresponding to all pixels in the first neighborhood, and using the magnitude of the vector sum as the saliency value of the current pixel to form a saliency map. Utilizing the gradient convergence property of solid circles, a circular saliency map is calculated based on a complex gradient vector field, including: calculating the circular saliency value at each pixel to construct the circular saliency map, wherein the circular saliency value satisfies the expression: ,in, Represents pixels The circular significant value at the location; In pixels The second neighborhood centered on; Represents the pixels in the second neighborhood. Gradient magnitude at; Represents the pixels in the second neighborhood. The gradient direction angle at that location; Represents the second neighboring pixel. Pointing to the center pixel The displacement vector angle in the image coordinate system; Pi; Based on the topological concentricity constraint mechanism, the box saliency map and the circular saliency map are dually fused to construct the target probability potential field, including: calculating the probability potential of each pixel belonging to the center of the true reference point, and constructing the target probability potential field. The probability potential of a pixel belonging to the center of the true reference point satisfies the expression: , Represents pixels The probabilistic potential energy belonging to the true reference point center; Represents pixels The circular significant value at the location; The standard deviation is expressed as Gaussian kernel; This represents the convolution operation; Represents pixels The significance value of the box at the location; The standard deviation parameter represents the Gaussian kernel. The extreme point is searched in the target probabilistic potential energy field to determine the true reference point coordinates, and the rotating fixture is driven to perform alignment compensation based on the true reference point coordinates.

2. The method for aligning a glass substrate using a vision-guided rotating fixture according to claim 1, characterized in that, The construction of the complex gradient vector field includes: The Scharr operator is used to calculate the gradients in the horizontal and vertical directions of the grayscale image of the glass substrate. The complex gradient vector of each pixel in the grayscale image of the glass substrate is then calculated to form a complex gradient vector field, which satisfies the expression: ,in, Represents pixels The complex gradient vector at point; Represents pixels Gradient magnitude at; Represents pixels The gradient direction angle at that location; Represents pixels The horizontal gradient at that location; Represents pixels The vertical gradient at that location; It is the imaginary unit.

3. The method for aligning a glass substrate using a vision-guided rotating fixture according to claim 1, characterized in that, The fourth harmonic vector satisfies the expression: ； In the formula, Represents pixels The fourth harmonic vector at that location; Represents pixels Gradient magnitude at; Represents pixels The gradient direction angle at that location; 4 is the imaginary unit; 4 is the harmonic order.

4. The method for aligning a glass substrate using a vision-guided rotating fixture according to claim 1, characterized in that, The standard deviation parameter of the Gaussian kernel is: ,in To allow for physical concentricity deviation, This refers to the image resolution.

5. The method for aligning a glass substrate using a vision-guided rotating fixture according to claim 1, characterized in that, Determining the true reference point coordinates includes: Search for the global maximum pixel coordinates in the target probabilistic potential energy field; select a third neighborhood centered on the global maximum pixel coordinates, and use a bivariate quadratic polynomial to fit the peak vertex of the response surface to obtain the sub-pixel level true reference point coordinates.

6. The method for aligning a glass substrate using a vision-guided rotating fixture according to claim 1, characterized in that, The step of driving the rotating fixture to perform alignment compensation based on the true reference point coordinates includes: Calculate the positional deviation between the true reference point coordinates and the preset target position; The positional deviation is converted into pulse signals in the mechanical coordinate system and sent to the motion control card to drive the X-axis, Y-axis, and... The shaft motor performs linkage compensation until the position deviation is less than the preset process threshold.

7. A vision-guided glass substrate rotation fixture alignment system, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement a vision-guided glass substrate rotation fixture alignment method according to any one of claims 1-6.

Images