Multi-camera position calibration method and system, and wafer bonding detection device

Abstract

A multi-camera position calibration method and system, and a wafer bonding detection device. The multi-camera position calibration method comprises the following steps: using a plurality of cameras to collect a plurality of edge images of a same wafer (S110); respectively determining multiple groups of edge information of the wafer on the basis of the edge images, wherein the edge information comprises a first coordinate of at least one edge point in each edge image (S120); on the basis of a second coordinate of the center of a theoretical circle of the wafer and the first coordinate of each edge point, respectively determining a feature distance between the center of the theoretical circle and each edge point (S130); respectively solving, by means of a gradient descent model, an optimization result that minimizes a difference between a feature distance of each edge image and the radius of the wafer, so as to respectively determine the relative position of each camera and the center of the theoretical circle (S140); and determining a relative position relationship between the plurality of cameras on the basis of the relative position of each camera and the center of the theoretical circle (S150).

Description

Multi-camera position calibration method, system and wafer bonding detection equipment Technical Field

[0001] The present invention relates to the field of chip manufacturing technology, and in particular to a multi-camera position calibration method, a multi-camera position calibration system, a computer-readable storage medium, and a wafer bonding detection device. Background Art

[0002] In the wafer bonding process, it is inevitable that the wafers being bonded are large-format wafers. In processes such as melt bonding, if a single camera is used to capture the overall outline of the wafer, the image captured by the camera often fails to meet the required accuracy. Therefore, when the melt bonding process requires capturing the outline of a large-format wafer, multiple cameras with non-overlapping fields of view are often required to jointly capture the edge of the wafer to meet the required accuracy of the captured wafer image. The accuracy of the captured wafer edge image is related to the accuracy of the subsequent bonding process, and the edge position represented by the captured image is closely related to the relative positions of the multiple cameras. Therefore, calibrating the relative position relationship of multiple cameras is crucial for the wafer alignment process in the subsequent bonding process.

[0003] In the prior art, a calibration plate is typically used to determine the relative positions of multiple cameras. However, the accuracy and specifications of existing calibration plates cannot meet the accuracy requirements required for relative position calibration between cameras.

[0004] In order to overcome the above-mentioned defects of the existing technology, the field urgently needs a multi-camera position calibration technology, which is used to realize the calibration of the relative positions of multiple cameras in the absence of a high-precision, large-size calibration plate, thereby improving the accuracy of the wafer bonding detection equipment and the accuracy of the wafer bonding process to achieve good wafer bonding quality.

[0005] Summary of the Invention

[0006] The following is a brief summary of one or more aspects to provide a basic understanding of these aspects. This summary is not an exhaustive overview of all conceivable aspects and is neither intended to identify key or critical elements of all aspects nor to define the scope of any or all aspects. Its sole purpose is to present some concepts of one or more aspects in a simplified form as a prelude to the more detailed description that will be provided later.

[0007] In order to overcome the above-mentioned defects of the prior art, the present invention provides a multi-camera position calibration method, a multi-camera position calibration system, a computer-readable storage medium, and a wafer bonding detection device, which can realize the calibration of the relative positions of multiple cameras without a high-precision, large-size calibration plate, thereby improving the accuracy of the wafer bonding detection equipment and the accuracy of the wafer bonding process to achieve good wafer bonding quality.

[0008] Specifically, the above-mentioned multi-camera position calibration method provided in accordance with the first aspect of the present invention includes the following steps: using multiple cameras to capture multiple edge images of the same wafer; determining multiple sets of edge information of the wafer based on each edge image, wherein the edge information includes the first coordinates of at least one edge point in each edge image; determining the characteristic distance between the center of the theoretical circle and each edge point based on the second coordinate of the center of the theoretical circle of the wafer and the first coordinates of each edge point; solving the optimization result that minimizes the difference between the characteristic distance of each edge image and the radius of the wafer through a gradient descent model, so as to determine the relative position of each camera and the center of the theoretical circle; and determining the relative position relationship between the multiple cameras based on the relative position of each camera and the center of the theoretical circle.

[0009] Preferably, in one embodiment of the present invention, the step of capturing multiple edge images of the same wafer using multiple cameras includes: correcting the zoom factor and / or placement angle of at least one of the cameras; and photographing the same wafer through each of the corrected cameras to capture multiple edge images with the same zoom factor and the same offset angle.

[0010] Preferably, in one embodiment of the present invention, the step of determining multiple groups of edge information of the wafer based on each of the edge images includes: performing Gaussian smoothing filtering on the acquired edge image to eliminate noise in the edge image; calculating the gradient size and gradient direction of the value of each pixel point in the noise-eliminated edge image; and using a non-maximum suppression method to screen the pixel points with the maximum gradient value along the gradient direction according to the gradient size and gradient direction of the value of each pixel point to determine the edge information of the wafer.

[0011] Preferably, in one embodiment of the present invention, the first coordinate is in an image coordinate system, and the second coordinate is in a world coordinate system. The step of determining the characteristic distance between the theoretical circle center and each edge point based on the second coordinate of the theoretical circle center of the wafer and the first coordinate of each edge point comprises: determining the third coordinate of at least one edge point in each edge image in the world coordinate system based on the relative positional relationship between the world coordinate system and each image coordinate system; and determining the characteristic distance between the theoretical circle center and each edge point based on the second coordinate and each third coordinate.

[0012] Preferably, in one embodiment of the present invention, the step of respectively solving the optimization result that minimizes the difference between the characteristic distance of each edge image and the radius of the wafer through the gradient descent model to respectively determine the relative positional relationship between each camera and the center of the theoretical circle includes: constructing a cost function of the difference between the characteristic distance and the radius of the wafer:

[0013] Wherein, (x, y) is the coordinate of the camera in the world coordinate system, (E X , E y ) is the first coordinate of the edge point in the image coordinate system, (c x , c y ) is the second coordinate of the center of the theoretical circle in the world coordinate system; the first-order partial derivative of the cost function is calculated to determine the gradient d of the difference along the X direction x And the gradient d of the difference along the Y direction y ,in,

[0014] The gradient d x and the gradient d y The gradient descent model is introduced to obtain an optimization result that minimizes the difference between the characteristic distance of each edge image and the radius of the wafer.

[0015] Preferably, in one embodiment of the present invention, the gradient descent model involves a learning rate α. x and the gradient d y The step of importing the gradient descent model to solve the optimization result of minimizing the difference between the characteristic distance of each edge image and the radius of the wafer includes: determining the learning rate α of the gradient descent model; x and the learning rate α, iteratively optimize the X coordinate of the camera in the world coordinate system: X′=X-α*d x ; and according to the gradient d yand the learning rate α, iteratively optimize the Y coordinate of the camera in the world coordinate system: Y′=Y-α*d y .

[0016] Preferably, in one embodiment of the present invention, the step of determining the learning rate α of the gradient descent model includes: initializing a variable s=0, and defining a stability parameter ∈=10 -10 ; According to the variable s, the stability parameter ∈ and the preset learning rate α0, the learning rate α of the current round is determined:

[0017] as well as

[0018] The gradient d in the gradient descent model x and gradient d y The squares are summed and added to the variable s to determine the learning rate α' for the next round.

[0019] Preferably, in one embodiment of the present invention, the gradient descent model involves a learning rate α and a momentum value β. x and the gradient d y The step of importing the gradient descent model to solve the optimization result of minimizing the difference between the characteristic distance of each edge image and the radius of the wafer further includes: x , the gradient d y , the momentum value β and the initial momentum velocity v dx0 and v dyo , determine the momentum speed v of the current round dx and v dy :v dx =β*v dx0 +(1-β)*d x , v dy =β*v dyo +(1-β)*d y ; According to the learning rate α and the momentum speed v dx , iteratively optimize the X coordinate of the camera in the world coordinate system: X′=X-α*v dx ; According to the learning rate α and the momentum speed v dy , iteratively optimize the Y coordinate of the camera in the world coordinate system: Y′=Y-α*v dy ; and according to the gradient d x , the gradient d y , the momentum value β and the momentum speed v of the current round dx and v dy , determine the momentum velocity v of the next round dx ′、v dy ′:vdx ′=β*v dx +(1-β)*d x , v dy ′=β*v dy +(1-β)*d y .

[0020] Furthermore, the multi-camera position calibration system provided in accordance with a second aspect of the present invention includes a memory and a processor. The memory stores computer instructions. The processor is connected to the memory and configured to execute the computer instructions stored in the memory to implement the multi-camera position calibration method provided in any of the above embodiments.

[0021] Furthermore, the computer-readable storage medium provided according to the third aspect of the present invention stores computer instructions, which, when executed by a processor, implement the multi-camera position calibration method provided in any one of the above embodiments.

[0022] In addition, the wafer bonding detection equipment provided according to the fourth aspect of the present invention includes multiple cameras with non-overlapping fields of view, which are used to jointly capture images of the wafer edge, wherein the relative position of each of the cameras is determined by the multi-camera position calibration method provided by any one of the above embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The above features and advantages of the present invention will be better understood after reading the detailed description of the embodiments of the present disclosure in conjunction with the following drawings. In the drawings, the components are not necessarily drawn to scale, and components with similar related properties or characteristics may have the same or similar reference numerals.

[0024] FIG1 shows a flowchart of a method for calibrating multi-camera positions according to some embodiments of the present invention;

[0025] 2A and 2B are schematic diagrams showing correction of positions of multiple cameras according to some embodiments of the present invention;

[0026] 3A to 3D illustrate edge images of the same wafer captured by various cameras according to some embodiments of the present invention;

[0027] FIG4 shows a schematic diagram of the screening principle of the non-maximum suppression method provided according to some embodiments of the present invention;

[0028] FIG5 is a schematic diagram showing a method for calibrating multi-camera positions according to some embodiments of the present invention;

[0029] FIG6 is a schematic diagram showing a method for calibrating multiple camera positions according to some embodiments of the present invention; and

[0030] FIG7 shows a schematic diagram of a gradient descent model provided according to some embodiments of the present invention.

[0031] Reference numerals: 100: multi-camera position calibration method; S110-S150: steps; 21, 22, 23, 24: cameras; 31: edge; 41, 42, 43: pixels; O0, O1, O2, O3, O4: positions; and A: edge point. DETAILED DESCRIPTION

[0032] The following specific embodiments illustrate the embodiments of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. Although the description of the present invention will be introduced in conjunction with the preferred embodiment, this does not mean that the features of this invention are limited to this embodiment. On the contrary, the purpose of introducing the invention in conjunction with the embodiment is to cover other options or modifications that may be extended based on the claims of the present invention. In order to provide a deep understanding of the present invention, the following description will include many specific details. The present invention can also be implemented without using these details. In addition, in order to avoid confusion or blurring the focus of the present invention, some specific details will be omitted in the description.

[0033] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood in a broad sense. For example, they may refer to fixed, detachable, or integral connections; mechanical or electrical connections; direct or indirect connections through an intermediate medium; and internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on the specific circumstances.

[0034] Furthermore, the terms "upper," "lower," "left," "right," "top," "bottom," "horizontal," and "vertical" used in the following description should be understood to refer to the orientations depicted in that section and the accompanying drawings. These relative terms are used solely for convenience of description and do not necessarily imply that the devices described herein must be manufactured or operated in a specific orientation. Therefore, they should not be construed as limiting the present invention.

[0035] It will be understood that although the terms "first," "second," "third," etc. may be used herein to describe various components, regions, layers, and / or portions, these components, regions, layers, and / or portions should not be limited by these terms, and these terms are merely used to distinguish different components, regions, layers, and / or portions. Thus, a first component, region, layer, and / or portion discussed below may be referred to as a second component, region, layer, and / or portion without departing from some embodiments of the present invention.

[0036] As mentioned above, the accuracy of the captured wafer edge image is crucial to the accuracy of the subsequent bonding process. The edge position represented by the captured image is closely related to the relative positions of multiple cameras. Therefore, calibrating the relative positions of multiple cameras is crucial for wafer alignment during the subsequent bonding process. Existing technologies use calibration plates to determine the relative positions of multiple cameras. However, the accuracy and specifications of existing calibration plates cannot meet the precision requirements for relative camera position calibration.

[0037] In order to overcome the above-mentioned defects of the prior art, the present invention provides a multi-camera position calibration method, a multi-camera position calibration system, a computer-readable storage medium, and a wafer bonding detection device, which can realize the calibration of the relative positions of multiple cameras without a high-precision, large-size calibration plate, thereby improving the accuracy of the wafer bonding detection equipment and the accuracy of the wafer bonding process to achieve good wafer bonding quality.

[0038] In some non-limiting embodiments, the multi-camera position calibration method provided in the first aspect of the present invention can be implemented via the multi-camera position calibration system provided in the second aspect of the present invention. Specifically, the multi-camera position calibration system can be configured with a memory and a processor. The memory includes, but is not limited to, the computer-readable storage medium provided in the third aspect of the present invention, having computer instructions stored thereon. The processor is connected to the memory and configured to execute the computer instructions stored in the memory to implement the multi-camera position calibration method provided in the first aspect of the present invention.

[0039] The following will describe the working principle of the multi-camera position calibration system in conjunction with some embodiments of the multi-camera position calibration method. Those skilled in the art will understand that these embodiments of the multi-camera position calibration method are only some non-limiting implementation methods provided by the present invention, which are intended to clearly demonstrate the main concept of the present invention and provide some specific solutions that are convenient for the public to implement, rather than to limit all functions or all working methods of the calibration system. Similarly, the calibration system is also only a non-limiting implementation method provided by the present invention, and does not constitute a limitation on the execution subject and execution order of each step in these calibration methods.

[0040] First, please refer to Figure 1 and Figures 2A-2B. Figure 1 shows a flow chart of a method for calibrating the positions of multiple cameras according to some embodiments of the present invention. Figures 2A-2B show schematic diagrams of correcting the positions of multiple cameras according to some embodiments of the present invention.

[0041] As shown in FIG1 , the multi-camera position calibration method 100 provided by the present invention includes:

[0042] Step S110: using multiple cameras to capture multiple edge images of the same wafer.

[0043] In some embodiments, when calibrating the relative positions of multiple cameras, the present invention can use the captured wafer as a calibration feature for the multi-camera position calibration method. Here, the wafer dimensions are known; it can be a 12-inch wafer with a radius of 150 mm. If four cameras are used to image the wafer edge, each camera can produce an edge image approximately 5-6 mm in size, with a pixel size of approximately 3 μm, thus achieving the requirements of non-overlapping fields of view and high accuracy.

[0044] Furthermore, as shown in FIG2A , the original camera positions may have inconsistent placement angles and zoom factors. For example, the placement angle of camera 21 in the upper left corner of FIG2A is offset (offset angle θ) compared to the placement angles of cameras 22, 23, and 24. For another example, the actual image zoom factor captured by camera 22 in the upper right corner of FIG2A (shown by the dotted line in the figure) is greater than the image zoom factors captured by cameras 21, 23, and 24 (the ideal image zoom factor that camera 22 should capture is shown by the solid line in the figure).

[0045] Therefore, the multi-camera position calibration system provided by the present invention can correct the placement angle of camera 21 in Figure 2A and the zoom factor of camera 22 in Figure 2A. Furthermore, if the placement angle and zoom factor of a camera are inconsistent with those of other cameras, both the placement angle and zoom factor of that camera can be corrected.

[0046] As shown in FIG2B , after completing the correction of the camera zoom factor and / or placement angle, the calibration system can photograph the same wafer through the corrected cameras 21, 22, 23, and 24 to collect multiple edge images with the same zoom factor and the same offset angle, thereby reducing the deviation in the positioning results caused by the zoom factor and placement angle of each camera.

[0047] Please refer to FIG. 3A to FIG. 3D , which illustrate edge images of the same wafer captured by various cameras according to some embodiments of the present invention.

[0048] As shown in Figures 3A to 3D, the present invention can capture images of the edge of the same wafer using four calibrated cameras, and extract multiple edge images of the same wafer as shown in Figures 3A to 3D. Here, each edge image contains at least one edge 31 of the wafer.

[0049] Continuing to refer to FIG. 1 , the multi-camera position calibration method 100 shown in FIG. 1 further includes:

[0050] Step S120 : determining multiple sets of edge information of the wafer according to each edge image, wherein the edge information includes a first coordinate of at least one edge point in each edge image.

[0051] In a preferred embodiment, the calibration system can first perform Gaussian smoothing filtering on the acquired edge image to eliminate noise in the edge image, and then calculate the gradient size and gradient direction of the value of each pixel point in the noise-eliminated edge image to separately determine multiple groups of edge information of the wafer.

[0052] Specifically, the calibration system can use the following edge detection operator to calculate the gradient:

[0053] Among them, d x Represents the gradient in the X direction, d y Represents the gradient in the Y direction.

[0054] Continuing to refer to the multiple edge images shown in Figures 3A to 3D, each edge image contains two inner and outer edge lines. To this end, the calibration system can use non-maximum suppression to obtain accurate edge information, thereby retaining the maximum value in the gradient direction and excluding interference areas.

[0055] Please refer to FIG4 for details, which shows a screening principle diagram of the non-maximum suppression method provided according to some embodiments of the present invention.

[0056] As shown in FIG4 , in the process of implementing the non-maximum suppression method, the calibration system can first screen the pixel points with the maximum gradient value along the gradient direction according to the gradient size and gradient direction of each pixel point to determine the edge information of the wafer.

[0057] Specifically, the direction of the arrow shown in Figure 4 is the gradient direction of pixel 41. Through the non-maximum suppression method provided by this preferred embodiment, the present invention can compare the gradient value of pixel 41 with the gradient values ​​of the two pixels 42 and 43 before and after the pixel 41 in the gradient direction. If the gradient value of pixel 41 is greater than the gradient values ​​of pixel 42 and pixel 43, the value of pixel 41 is retained, otherwise the value of pixel 41 is set to 0. In this way, through continuous comparison, the calibration system can screen out the pixel point with the maximum gradient value along the gradient direction to locate the actual edge (as shown in edge 31 in Figure 3), thereby determining the edge information and edge data of the wafer.

[0058] Furthermore, after acquiring the edge information of the wafer, the calibration system may perform position matching of the edge information with a theoretical circle at a preset position.

[0059] Please refer to FIG5 for details, which shows a schematic diagram of a multi-camera position calibration method provided according to some embodiments of the present invention.

[0060] The white circle shown in Figure 5 is a theoretical circle, which is located in the world coordinate system and has a radius r. The wafer edge information obtained by positioning includes at least one edge point (white dot) in each edge image, and has a first coordinate in the corresponding image coordinate system. To facilitate the mutual operation between the two, the calibration system can determine the third coordinate of at least one edge point in each edge image in the world coordinate system based on the relative position relationship between the world coordinate system and each image coordinate system, thereby converting each edge point in Figure 5 to the world coordinate system.

[0061] Please refer to FIG6 for details, which shows a schematic diagram of a multi-camera position calibration method provided according to some embodiments of the present invention.

[0062] As shown in Figure 6, the images captured by each camera can each have a corresponding image coordinate system, where the origin of the world coordinate system is located at position O0, and the origins of the image coordinate systems are located at positions O1, O2, O3, and O4, respectively. Thus, during the coordinate conversion process, the calibration system can determine the third coordinate of at least one edge point in each edge image in the world coordinate system based on the relative positional relationship between the world coordinate system and each image coordinate system.

[0063] Specifically, taking the edge point A in the first image coordinate system as an example, assuming that the origin of the first image coordinate system is O1, the coordinates of the edge point A in the first image coordinate system are (x′, y′), and the coordinates of the origin O1 of the first image coordinate system relative to the world coordinate system are (x1, y1), then the third coordinate of the edge point A in the world coordinate system is (x1+x′, y1+y′).

[0064] Similarly, the calibration system can determine the third coordinate of at least one edge point in each edge image in the world coordinate system one by one, which will not be described in detail here.

[0065] Continuing to refer to FIG. 1 , the multi-camera position calibration method 100 shown in FIG. 1 further includes:

[0066] Step S130 : determining a characteristic distance d between the theoretical circle center and each edge point according to the second coordinate of the theoretical circle center of the wafer and the first coordinate of each edge point.

[0067] As described above, the wafer serves as the calibration feature for the multi-camera position calibration method. The second coordinate of its theoretical circle center, O, lies in the world coordinate system. Therefore, by converting the first coordinate of each edge point in each edge image into its third coordinate in the world coordinate system, the calibration system can efficiently determine the characteristic distance d between the theoretical circle center, O, and each edge point based on the positional relationship between the second coordinate of the theoretical circle center and the third coordinate of each edge point.

[0068] Continuing to refer to FIG. 1 , the multi-camera position calibration method 100 shown in FIG. 1 further includes:

[0069] Step S140 : Using a gradient descent model, respectively solving the optimization results that minimize the difference between the feature distance of each edge image and the radius of the wafer, so as to respectively determine the relative position of each camera and the center of the theoretical circle.

[0070] In some reference examples of the present invention, technicians can use a global search model to traverse the pixels one by one from the origin of the world coordinate system, from left to right and from top to bottom, until the entire image is traversed. The obtained characteristic distance d values ​​are then sorted to determine the position at which the characteristic distance d reaches its minimum value. However, wafer bonding inspection requires high image acquisition accuracy. Therefore, when the image resolution is too high, the traversal time of the global search model may be too long, making it unsuitable for practical use.

[0071] In other reference examples of the present invention, technicians can use a jump search model to jump through some pixels of the edge image according to a preset step size (for example, 2 pixels), effectively reducing the traversal range and increasing the search speed. However, because it ignores most pixels, the jump search model generally has low search accuracy, thus failing to meet the precision requirements of the wafer bonding process and not meeting practical needs.

[0072] To this end, the present invention adopts a gradient descent model that can quickly locate the target position and reduce calculation time while meeting accuracy requirements. Please refer to Figure 7, which shows a schematic diagram of a gradient descent model provided according to some embodiments of the present invention.

[0073] As shown in Figure 7, in the process of finding the optimization result that minimizes the difference between the characteristic distance of each edge image and the radius of the wafer, the calibration system can input the edge information obtained in the above steps into the gradient descent model, which takes the center of the theoretical circle as the initial point, descends along the gradient direction of the difference, and continuously iterates and optimizes until the optimization result that minimizes the difference between the characteristic distance of each edge image and the radius of the wafer is found.

[0074] Specifically, the calibration system can first construct a cost function for the difference between the feature distance d and the radius r of the wafer:

[0075] Among them, (x, y) is the coordinate of the camera in the world coordinate system, (E X , E y ) is the first coordinate of the edge point in the image coordinate system, (c x , c y ) is the second coordinate of the center of the theoretical circle in the world coordinate system.

[0076] Afterwards, the calibration system can find the first-order partial derivative of the cost function to determine the gradient d of the difference along the X direction x And the gradient d of the difference along the Y direction y ,in,

[0077] Afterwards, the calibration system can calculate the gradient d x and gradient d y A gradient descent model is introduced to find the optimization result that minimizes the difference between the feature distance of each edge image and the radius of the wafer.

[0078] Furthermore, in some embodiments, the gradient descent model may involve a pre-set or adaptively adjusted learning rate α. The calibration system may preferably be based on the learning rate α and the gradient d x d y And through continuous iterative optimization, the X and Y coordinates of the camera in the world coordinate system are obtained: X′=X-α*d x Y′=Y-α*d y .

[0079] Furthermore, given that a too large learning rate α will cause the gradient descent model to skip the optimal point, and a too small learning rate α will cause the gradient descent model to converge too slowly, the present invention also preferably introduces an AdaGrad adaptive learning rate algorithm for adaptively adjusting the learning rate α at each iteration, thereby obtaining a rapidly decreasing learning rate when the parameter has the largest partial derivative of loss, and obtaining a slowly decreasing learning rate when the parameter has a small partial derivative.

[0080] Specifically, based on the introduced AdaGrad adaptive learning rate algorithm, the calibration system can set the initialization variable s = 0 and define the stability parameter ∈ = 10 -10 , and determine the learning rate α of the current round based on the variable s, stability parameter ∈ and preset learning rate α0:

[0081] Afterwards, the calibration system can iteratively optimize the X and Y coordinates of the camera in the world coordinate system according to the learning rate α of the current round as described above, and convert the gradient d in the gradient descent model into x and gradient d y The squares are summed and added to the variable s to determine the learning rate α' for the next round. In this way, the present invention can adjust the variable s and the obtained learning rate α accordingly based on the different parameters and gradients, thereby achieving adaptive adjustment of the learning rate α.

[0082] In addition, considering that the traditional gradient descent model has the problem of easily falling into local optimality, in some embodiments, the present invention may also preferably introduce a momentum value β (generally taking a value of 0.9) to facilitate the training of the gradient descent model.

[0083] Specifically, during the iterative optimization process of the camera's X and Y coordinates in the world coordinate system, the calibration system can first calculate the value of the camera's X and Y coordinates according to the gradient d. x , gradient d y , momentum value β and initial momentum velocity v dx0 and v dy0 , determine the momentum speed v of the current round dx and v dy : v dx =β*v dxo +(1-β)*d x v dy =β*v dyo +(1-β)*d y

[0084] Afterwards, the calibration system can be adjusted based on the learning rate α and momentum velocity v as described above. dx , iteratively optimize the X coordinate of the camera in the world coordinate system: X′=X-α*v dx

[0085] And as described above, according to the learning rate α and momentum speed v dy , iteratively optimize the Y coordinate of the camera in the world coordinate system: Y′=Y-α*v dy

[0086] After that, the calibration system can be adjusted based on the gradient d x , gradient d y , momentum value β and momentum speed v of the current rounddx and v dy , determine the momentum velocity v of the next round dx ′、v dy ′: v dx ′=β*v dx +(1-β)*d x v dy ′=β*v dy +(1-β)*d y .

[0087] In this way, by introducing the momentum value for weighting in the direction of the next gradient optimization based on the historical gradient optimization direction, the present invention can effectively prevent the gradient descent model from falling into the local optimum, thereby more quickly and accurately finding the optimization result that minimizes the difference between the feature distance of each edge image and the radius of the wafer.

[0088] Continuing to refer to FIG. 1 , the multi-camera position calibration method 100 shown in FIG. 1 further includes:

[0089] Step S150: Determine the relative position relationship between the multiple cameras based on the relative position of each camera and the center of the theoretical circle.

[0090] Specifically, based on the optimization results determined by the above steps, the present invention can respectively determine the relative position of each camera and the center of the theoretical circle, and determine the relative position between multiple cameras accordingly, thereby calibrating the relative positions between multiple cameras in the absence of a high-precision, large-size calibration plate, so as to improve the accuracy of the wafer bonding detection equipment.

[0091] In addition, in some embodiments, those skilled in the art can also construct the wafer bonding detection equipment provided by the fourth aspect of the present invention based on the relative positions between the multiple cameras obtained by the calibration method and calibration system provided by the present invention. Specifically, the wafer bonding detection equipment includes multiple cameras with non-overlapping fields of view, which are used to jointly capture images of the wafer edge. Here, the relative positions of the multiple cameras are determined by the multi-camera position calibration method provided by the first aspect of the present invention, and thus have higher accuracy and can improve the accuracy of the wafer bonding process.

[0092] In summary, the multi-camera position calibration method, multi-camera position calibration system, computer-readable storage medium, and wafer bonding detection equipment provided by the present invention can all use a gradient descent model to determine the relative positions of multiple cameras and the theoretical circle and between multiple cameras without a high-precision, large-size calibration plate, and can achieve fast and effective positioning while ensuring the accuracy and precision of the position, thereby reducing calculation time and improving the utilization of hardware resources.

[0093] Although the above methods are illustrated and described as a series of acts for simplicity of explanation, it is to be understood and appreciated that these methods are not limited by the order of the acts, as some acts may occur in a different order and / or concurrently with other acts from those illustrated and described herein or not illustrated and described herein but understandable to those skilled in the art according to one or more embodiments.

[0094] The previous description of the disclosure is provided to enable any person skilled in the art to make or use the disclosure. Various modifications to the disclosure will be apparent to those skilled in the art, and the general principles defined herein may be applied to other variations without departing from the spirit or scope of the disclosure. Thus, the disclosure is not intended to be limited to the examples and designs described herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for calibrating the positions of multiple cameras, characterized in that: The following steps are involved: Use multiple cameras to capture multiple edge images of the same wafer; Determine, according to each of the edge images, a plurality of sets of edge information of the wafer, wherein the edge information includes a first coordinate of at least one edge point in each of the edge images; Determine the characteristic distance between the center of the theoretical circle and each of the edge points according to the second coordinate of the center of the theoretical circle of the wafer and the first coordinate of each of the edge points; By using a gradient descent model, respectively solving the optimization results that minimize the difference between the characteristic distance of each edge image and the radius of the wafer, so as to respectively determine the relative position of each camera and the center of the theoretical circle; and The relative position relationship between the multiple cameras is determined according to the relative position of each camera and the center of the theoretical circle.

2. The calibration method according to claim 1, characterized in that: The step of using multiple cameras to collect multiple edge images of the same wafer includes: Correcting the zoom factor and / or the positioning angle of at least one of the cameras; and The same wafer is photographed by each of the calibrated cameras to acquire multiple edge images with the same zoom factor and the same offset angle.

3. The calibration method according to claim 1, characterized in that: The step of respectively determining multiple sets of edge information of the wafer according to each of the edge images comprises: Performing Gaussian smoothing filtering on the acquired edge image to eliminate noise in the edge image; Calculate the gradient magnitude and gradient direction of each pixel value in the edge image that has been de-noised; and The non-maximum suppression method is adopted to select the pixel points with the maximum gradient value along the gradient direction according to the gradient size and gradient direction of each pixel point, so as to determine the edge information of the wafer.

4. The calibration method according to claim 1, characterized in that: The first coordinate is located in the image coordinate system, the second coordinate is located in the world coordinate system, and the step of determining the characteristic distance between the center of the theoretical circle and each of the edge points according to the second coordinate of the center of the theoretical circle of the wafer and the first coordinate of each of the edge points comprises: Determining respectively the third coordinate of at least one edge point in each of the edge images in the world coordinate system according to the relative positional relationship between the world coordinate system and each of the image coordinate systems; and The characteristic distance between the center of the theoretical circle and each of the edge points is determined respectively according to the second coordinate and each of the third coordinates.

5. The calibration method according to claim 4, characterized in that: The step of respectively solving the optimization result that minimizes the difference between the characteristic distance of each edge image and the radius of the wafer through the gradient descent model to respectively determine the relative position relationship between each camera and the center of the theoretical circle includes: Construct a cost function of the difference between the characteristic distance and the radius of the wafer: Wherein, (x, y) is the coordinate of the camera in the world coordinate system, (E X , E y ) is the edge point The first coordinate in the image coordinate system, (c x , c y ) is the second coordinate of the center of the theoretical circle in the world coordinate system; The first-order partial derivative of the cost function is calculated to determine the gradient d of the difference along the X direction. x And the gradient d of the difference along the Y direction y ,in, as well as The gradient d x and the gradient d y The gradient descent model is introduced to solve the optimization result that minimizes the difference between the feature distance of each edge image and the radius of the wafer.

6. The calibration method according to claim 5, characterized in that: The gradient descent model involves a learning rate α, and the gradient d x and the gradient d y The step of importing the gradient descent model to solve the optimization result that minimizes the difference between the feature distance of each edge image and the radius of the wafer includes: Determining a learning rate α of the gradient descent model; According to the gradient d x and the learning rate α, iteratively optimize the X coordinate of the camera in the world coordinate system: X′=X-α*d x ;as well as According to the gradient d y and the learning rate α, iteratively optimize the Y coordinate of the camera in the world coordinate system: Y′=Y-α*d y .

7. The calibration method according to claim 6, characterized in that: The step of determining the learning rate α of the gradient descent model comprises: Initialize the variable s = 0 and define the stability parameter ε = 10 -10 ; According to the variable s, the stability parameter ε and the preset learning rate α0, the learning rate α of the current round is determined: as well as The gradient d in the gradient descent model x and gradient d y The squares are summed and added to the variable s to determine the learning rate α' for the next round.

8. The calibration method according to claim 5, characterized in that: The gradient descent model involves a learning rate α and a momentum value β. x and the gradient d y The step of importing the gradient descent model to solve the optimization result that minimizes the difference between the feature distance of each edge image and the radius of the wafer further includes: According to the gradient d x , the gradient d y , the momentum value β and the initial momentum velocity v dx0 and v dy0 , determine the momentum speed v of the current round dx and v dy :v dx =β*v dx0 +(1-β)*d x , v dy =β*v dy0 +(1-β)*d y ; According to the learning rate α and the momentum speed v dx , iteratively optimize the X coordinate of the camera in the world coordinate system: X′=X-α*v dx ; According to the learning rate α and the momentum speed v dy , iteratively optimize the Y coordinate of the camera in the world coordinate system: Y′=Y-α*v dy ;as well as According to the gradient d x , the gradient d y , the momentum value β and the momentum speed v of the current round dx and v dy , determine the momentum velocity v of the next round dx ′、v dy ′:v dx ′=β*v dx +(1-β)*d x , v dy ′=β*v dy +(1-β)*d y .

9. A multi-camera position calibration system, characterized in that: include: a memory having computer instructions stored thereon; as well as A processor is connected to the memory and is configured to execute computer instructions stored in the memory to implement the multi-camera position calibration method according to any one of claims 1 to 8.

10. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the computer instructions are executed by a processor, the multi-camera position calibration method according to any one of claims 1 to 8 is implemented.

11. A wafer bonding inspection device, comprising a plurality of cameras with non-overlapping fields of view, for jointly capturing images of wafer edges, wherein: The relative position of each of the cameras is determined by the multi-camera position calibration method according to any one of claims 1 to 8.

Images