Image processing-based integrated circuit visual calibration method and system for die bonder

Abstract

The present application relates to the technical field of image processing, solves the technical problems of low precision and low efficiency of integrated circuit visual calibration in the prior art, and provides an integrated circuit visual calibration method and system based on image processing, the method comprising: controlling a motion platform to move along a first preset direction and a second preset direction respectively, obtaining pulse displacement data and pixel position change data before and after each movement; performing calibration parameter calculation on the pulse displacement data and the pixel position change data to obtain initial calibration parameters; controlling the motion platform to perform movement verification of a preset distance according to the initial calibration parameters to obtain a verification result; adjusting the initial calibration parameters according to the verification result to obtain target calibration parameters that meet the preset calibration precision condition, so as to realize integrated circuit visual closed-loop calibration. The present application improves the precision and efficiency of calibration in integrated circuit packaging detection by constructing a visual closed-loop automatic calibration mechanism.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a visual calibration method and system for die bonder integrated circuits based on image processing. Background Technology

[0002] Visual calibration technology is a key component of integrated circuit die bonding machines. By establishing a precise conversion relationship between the pixel coordinate system and the machine's mechanical coordinate system, it forms the basis for achieving accurate chip positioning, measurement, and operation, and directly determines the accuracy and reliability of packaging and testing.

[0003] Currently, existing visual calibration methods typically rely on operators manually selecting feature points on the chip itself as calibration references. This method first requires controlling the movement of a motion platform, then manually aligning the feature points visually, collecting multiple sets of positional data, and finally calculating calibration parameters such as visual equivalent. However, this method is highly dependent on the operator's experience, and the entire process is cumbersome, requiring multiple manual interventions. The aforementioned existing technologies have the following main drawbacks: First, calibration efficiency is low; the time-consuming manual calibration process must be repeated every time the product is changed or the field of view is adjusted. Second, calibration accuracy and repeatability are poor; subjective errors are introduced during manual alignment, and calibration results from different operators or at different times are difficult to maintain consistency. Third, there is a lack of automatic compensation for installation angles; unavoidable small installation angles between the camera coordinate system and the motion platform coordinate system introduce systematic errors, and traditional methods struggle to automatically identify and compensate for these angles, limiting further improvements in calibration accuracy.

[0004] Chinese patent CN119295563A discloses a camera calibration method, system, and apparatus based on image processing, comprising: rotating a calibration plate around a rotating axis as the rotation center via a turntable; acquiring calibration plate images containing calibration points at multiple angles using an image acquisition device; performing image preprocessing on the calibration plate images to obtain calibration plate coordinates; wherein the image acquisition device includes an image acquisition unit, a rotating axis, and a turntable, with the rotating axis perpendicular to the turntable; the calibration plate having preset calibration points and being placed vertically on the turntable; acquiring distortion parameters and intrinsic parameters of the image acquisition device; performing distortion removal and coordinate standardization processing on the calibration plate coordinates using the distortion parameters and intrinsic parameters of the image acquisition device to obtain first calibration plate coordinates and performing coordinate transformation to obtain second calibration plate coordinates; and performing attitude analysis on the calibration plate based on the first and second calibration plate coordinates to obtain... The calibration board attitude transformation data is used to transform the coordinates of the first calibration board to obtain the first coordinates in the coordinate system of the image acquisition device. The translation position of the calibration board in the coordinate system of the image acquisition device is analyzed based on the calibration board attitude transformation data to obtain the translation amount and fit it to obtain the fitting plane and fitting transformation matrix. Based on the fitting transformation matrix, the first coordinates are projected onto the horizontal plane and fitted to obtain the center coordinates of the horizontal circle. The center coordinates of the horizontal circle are then inversely projected onto the fitting plane using the fitting transformation matrix to obtain the fitted center coordinates. The normal vector of the fitting plane is obtained. Based on the normal vector of the fitting plane and the fitted center coordinates, the equation of the rotation axis line is obtained. Based on the origin of the image acquisition device coordinate system, the fitting plane, and the equation of the rotation axis line, the motion radius and height of the image acquisition device are obtained. The aforementioned patented solutions can reduce the uncertainty caused by manual adjustments through image processing. The background section also points out that traditional camera calibration relies on manual angle and position adjustments, which are easily affected by subjective factors and environmental interference, and require repeated calibration. However, the aforementioned patents mainly focus on fitting the rotation trajectory of the calibration board and solving the camera's geometric parameters. They rely on dedicated mechanisms and calibration scenarios such as calibration boards, rotating axes, and turntables. Their output is biased towards the geometric relationship parameters of the camera relative to the rotating axis (such as the radius of motion and height). They do not provide a directional solution mechanism for the more critical needs in integrated circuit die bonding machines, such as online conversion between the pixel coordinate system and the mechanical coordinate system of the motion platform, as well as the closed-loop recognition and automatic compensation of the camera and motion platform installation angle in the actual work position. Therefore, it is difficult to directly meet the requirements of high-precision, high-efficiency, repeatable, and self-compensating visual closed-loop calibration in packaging and testing scenarios.

[0005] Therefore, how to provide a visual closed-loop calibration method for integrated circuits that combines high precision and high efficiency is an urgent technical problem to be solved. Summary of the Invention

[0006] In view of this, embodiments of the present invention provide a method and system for visual calibration of integrated circuits in die bonders based on image processing, in order to solve the problem of low accuracy and efficiency of visual calibration of integrated circuits in automated packaging and inspection in the prior art.

[0007] In a first aspect, embodiments of the present invention provide a visual calibration method for integrated circuits in a die bonder based on image processing, applied to a die bonder. The die bonder includes: an image processing module, a calibration board, a motion platform, a motion control module, and a calibration module. The calibration board is disposed on the motion platform, and the motion control module controls the motion platform to move. The method includes:

[0008] The calibration plate is fixed at a preset position on the motion platform. The motion control module controls the motion platform to move along a first preset direction and a second preset direction respectively. Pulse displacement data of each movement is obtained. The image processing module obtains pixel position change data before and after the movement. The first preset direction and the second preset direction are perpendicular to each other.

[0009] The calibration module calculates calibration parameters for the pulse displacement data and the pixel position change data to obtain initial calibration parameters. The initial calibration parameters include a first visual equivalent corresponding to the first preset direction, a second visual equivalent corresponding to the second preset direction, and a target rotation angle between the camera coordinate system and the motion platform coordinate system.

[0010] Based on the initial calibration parameters, the motion platform is controlled to perform a preset distance of movement verification to obtain the verification result;

[0011] Based on the verification results, the initial calibration parameters are adjusted to obtain target calibration parameters that meet the preset calibration accuracy conditions, so as to achieve integrated circuit vision closed-loop calibration.

[0012] Preferably, the calibration plate surface is provided with calibration features, which include alignment segments for adjusting the camera axis, a square array for visual equivalent calibration, a dot array for extracting rotation information, and positioning marks for positioning the calibration plate.

[0013] Preferably, the die bonder further includes an image acquisition module. The steps of fixing the calibration plate to a preset position on the motion platform, controlling the motion platform to move along a first preset direction and a second preset direction via the motion control module, acquiring pulse displacement data for each movement, and acquiring pixel position change data before and after the movement via the image processing module include:

[0014] The image acquisition module acquires the initial calibration plate image at the preset position and the target calibration plate image after the motion platform has moved.

[0015] The image processing module extracts pixel positions from the initial calibration board image and the target calibration board image to obtain the initial pixel position and the target pixel position.

[0016] The difference between the initial pixel position and the target pixel position is calculated to obtain the pixel position change data;

[0017] The motion control module acquires the pulse displacement data for each movement of the motion platform.

[0018] Preferably, the step of extracting pixel positions from the initial calibration board image and the target calibration board image using the image processing module to obtain the initial pixel positions and target pixel positions includes:

[0019] The initial calibration board image and the target calibration board image are respectively input into a pre-trained calibration feature recognition model to obtain the first calibration feature in the initial calibration board image and the second calibration feature in the target calibration board image;

[0020] Contour features are extracted from the first calibration feature and the second calibration feature respectively to obtain the first contour position information of the first calibration feature and the second contour position information of the second calibration feature;

[0021] Based on the first contour position information and the second contour position information, the geometric center positions of the first calibration feature and the second calibration feature are calculated respectively to obtain the first pixel coordinate position and the second pixel coordinate position;

[0022] The first pixel coordinate position is used as the initial pixel position, and the second pixel coordinate position is used as the target pixel position.

[0023] Preferably, the step of calculating calibration parameters for the pulse displacement data and the pixel position change data through the calibration module to obtain the initial calibration parameters includes:

[0024] Based on the pixel position change data, obtain the slope of the pixel displacement vector before and after the motion platform moves along the first preset direction multiple times;

[0025] Based on the coordinate axes of the preset pixel coordinate system, obtain the quadrant assignment of each pixel displacement vector in the pixel coordinate system;

[0026] Based on the motion direction of the first preset direction in the preset motion platform coordinate system and the correspondence between the first preset direction and the coordinate axes of the pixel coordinate system, the directional consistency constraint between the pixel displacement vector direction and the first preset direction is determined.

[0027] Based on the quadrant assignment and the direction consistency constraint, the range and sign of the angle values ​​corresponding to the slope of each pixel displacement vector are constrained.

[0028] Under the constraints of the angle value range and sign, the target rotation angle is calculated based on the slope of each pixel displacement vector;

[0029] Based on the pixel position change data and the pulse displacement data, the first visual equivalent and the second visual equivalent are calculated by least squares fitting.

[0030] The initial calibration parameters are determined based on the target rotation angle, the first visual equivalent, and the second visual equivalent.

[0031] Preferably, the step of calculating the target rotation angle based on the slope of each pixel displacement vector, under the constraints of the angle value range and sign, includes:

[0032] Based on the slope of each pixel displacement vector and the preset mapping relationship between the slope and the rotation angle, each initial rotation angle is calculated.

[0033] The average value of each initial rotation angle is calculated, and the average value of the calculated rotation angles is taken as the target rotation angle.

[0034] Preferably, the step of calculating the first visual equivalent and the second visual equivalent by least squares fitting based on the pixel position change data and the pulse displacement data includes:

[0035] Based on the pixel position change data, obtain the displacement of each first pixel corresponding to the first preset direction and the displacement of each second pixel corresponding to the second preset direction;

[0036] Based on the pulse displacement data, obtain the first pulse displacement amount corresponding to the first preset direction and the second pulse displacement amount corresponding to the second preset direction;

[0037] The least squares method is used to linearly fit the displacement of each first pixel and the corresponding displacement of each first pulse, and the slope of the fitted first straight line is used as the first visual equivalent.

[0038] The least squares method is used to linearly fit each second pixel displacement and the corresponding second pulse displacement, and the slope of the fitted second straight line is used as the second visual equivalent.

[0039] Preferably, the step of controlling the motion platform to perform a preset distance movement verification based on the initial calibration parameters, and obtaining the verification result, includes:

[0040] Based on the first visual equivalent and the second visual equivalent, the preset distance is converted into the corresponding theoretical pixel displacement.

[0041] The motion platform is controlled to perform a preset distance of movement verification, and calibration plate images are acquired before and after the movement;

[0042] Pixel displacement is calculated on the calibration board images before and after the movement to obtain the actual pixel displacement.

[0043] The difference between the actual pixel displacement and the theoretical pixel displacement is calculated to obtain the verification result.

[0044] Preferably, adjusting the initial calibration parameters based on the verification results to obtain target calibration parameters that meet preset calibration accuracy conditions, thereby achieving integrated circuit vision closed-loop calibration, includes:

[0045] Based on the preset calibration accuracy conditions, the error threshold is obtained;

[0046] The verification result is compared with the error threshold. If the verification result is less than or equal to the error threshold, the initial calibration parameter is used as the target calibration parameter.

[0047] If the verification result is greater than the error threshold, the initial calibration parameters are adjusted according to the verification result to obtain the target calibration parameters.

[0048] In a second aspect, embodiments of the present invention provide a vision calibration system for a die bonder integrated circuit based on image processing, comprising: at least one processor, at least one memory, and computer program instructions stored in the memory, wherein when the computer program instructions are executed by the processor, the method of the first aspect described above is implemented.

[0049] In summary, the beneficial effects of the present invention are as follows:

[0050] This invention provides a method and system for visual calibration of integrated circuits in a die bonder based on image processing. The die bonder includes an image processing module, a calibration board, a motion platform, a motion control module, and a calibration module. The calibration board is disposed on the motion platform, and the motion control module controls the motion platform. The method includes: fixing the calibration board at a preset position on the motion platform; controlling the motion platform to move along a first preset direction and a second preset direction via the motion control module; acquiring pulse displacement data for each movement; and acquiring pixel position change data before and after the movement via the image processing module. The first preset direction... The first and second preset directions are perpendicular to each other; the calibration module calculates calibration parameters for the pulse displacement data and the pixel position change data to obtain initial calibration parameters, wherein the initial calibration parameters include a first visual equivalent corresponding to the first preset direction, a second visual equivalent corresponding to the second preset direction, and a target rotation angle between the camera coordinate system and the motion platform coordinate system; based on the initial calibration parameters, the motion platform is controlled to perform a preset distance movement verification to obtain a verification result; based on the verification result, the initial calibration parameters are adjusted to obtain target calibration parameters that meet the preset calibration accuracy conditions, so as to realize integrated circuit vision closed-loop calibration. This invention establishes a precise mapping relationship between physical motion quantities and visual measurements by rigidly fixing the calibration plate to the motion platform and, under the precise control of the motion control module, performing controlled displacements along two mutually perpendicular preset directions. Simultaneously, high-precision pulse displacement data and corresponding image pixel position change data are acquired. Based on this, the calibration module simultaneously solves for the visual equivalents in the first and second directions, as well as the rotation angle between the camera coordinate system and the motion platform coordinate system, avoiding the cumulative and coupling errors caused by manual annotation or unidirectional calibration in traditional methods. Furthermore, by introducing a preset distance movement verification mechanism based on initial calibration parameters, closed-loop verification of the calibration results is performed, and calibration parameters are automatically corrected according to the verification deviation. This transforms the calibration process from a one-time offline calibration to an iteratively optimized visual closed-loop calibration process, significantly improving the accuracy and stability of visual calibration in integrated circuit packaging inspection without complex manual intervention, while also shortening calibration time and improving overall automated inspection efficiency. Attached Figure Description

[0051] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments of the present invention will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, and these are all within the protection scope of the present invention.

[0052] Figure 1This is a schematic diagram of the overall process of the image processing-based integrated circuit visual calibration method for die bonders in Embodiment 1 of the present invention;

[0053] Figure 2 This is a flowchart illustrating the process of calculating calibration parameters for the pulse displacement data and pixel position change data using the calibration module in Embodiment 1 of the present invention to obtain initial calibration parameters.

[0054] Figure 3 This is a schematic diagram of the structure of the image processing-based integrated circuit vision calibration system for die bonders in Embodiment 2 of the present invention. Detailed Implementation

[0055] The features and exemplary embodiments of various aspects of the present invention will now be described in detail. To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only configured to explain the present invention and are not configured to limit the present invention. For those skilled in the art, the present invention can be practiced without some of these specific details. The following description of the embodiments is merely intended to provide a better understanding of the present invention by illustrating examples of the invention.

[0056] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0057] It should be noted that all actions involving the acquisition of signals, information, or data in this invention are carried out in compliance with the relevant data protection laws and regulations of the locality and with authorization from the owner of the relevant device.

[0058] Example 1

[0059] Please see Figure 1This invention provides a visual calibration method for integrated circuits in a die bonder based on image processing, which is applied to a die bonder. The die bonder includes: an image processing module, a calibration board, a motion platform, a motion control module, and a calibration module. The calibration board is disposed on the motion platform, and the motion control module controls the motion platform.

[0060] Specifically, this die bonder is structurally configured around the fundamental step of visual calibration. By integrating an image processing module, calibration board, motion platform, motion control module, and calibration module, a stable collaborative relationship is established between the vision system and the mechanical motion system. The calibration board is directly mounted on the motion platform, moving synchronously with it. Structurally, this ensures that the displacement changes of visual features originate entirely from the controlled motion of the platform, rather than from human placement or external interference, thus providing reliable reference conditions for subsequent visual equivalent and angle calculations. This configuration is particularly suitable for integrated circuit or LED packaging inspection scenarios with stringent requirements for positional accuracy and repeatability. The motion control module uniformly schedules and precisely controls the motion of the motion platform, enabling it to complete displacement actions according to preset directions, distances, and speeds during inspection or calibration, and synchronously outputting corresponding pulses or displacement feedback data. Because the calibration board and motion platform maintain a fixed installation relationship, each controlled movement of the platform manifests as a stable and measurable change in pixel position at the image acquisition end. This strong correlation between mechanical displacement and visual changes allows the calibration module to establish a mapping relationship between pixel coordinates and device coordinates without manual intervention. Through the above structural design, the equipment can obtain a unified and stable visual calibration basis before packaging and inspection, so that subsequent visual equivalent calculation, skew angle compensation, and accuracy verification are all based on controllable motion conditions. On the one hand, it reduces the efficiency loss caused by manual adjustment and repeated calibration; on the other hand, it also reduces the systematic errors introduced by inconsistent platform motion or unstable calibration references, thereby improving the overall accuracy consistency of visual measurements and the reliability of system operation during integrated circuit packaging and inspection.

[0061] The method includes:

[0062] The calibration plate is fixed at a preset position on the motion platform. The motion control module controls the motion platform to move along a first preset direction and a second preset direction respectively. Pulse displacement data of each movement is obtained. The image processing module obtains pixel position change data before and after the movement. The first preset direction and the second preset direction are perpendicular to each other.

[0063] Specifically, firstly, by fixing the calibration plate to a preset installation position on the motion platform, the spatial relationship between the calibration plate and the motion platform is maintained throughout the calibration process, thus ensuring good repeatability and consistency of the subsequently acquired data. Under the control of the motion control module, the motion platform performs controlled movements along two mutually perpendicular preset directions, which typically correspond to the main motion axis directions of the device. As the platform moves, the corresponding pulse displacement data of the platform and the pixel position change data of the features on the calibration plate before and after movement in the image are acquired synchronously. This establishes a basic data correlation between physical displacement and visual changes, providing the original basis for subsequent parameter calculations. This step, through a standardized motion and acquisition process, introduces clear and quantifiable input conditions to the visual calibration process.

[0064] The calibration module calculates calibration parameters for the pulse displacement data and the pixel position change data to obtain initial calibration parameters. The initial calibration parameters include a first visual equivalent corresponding to the first preset direction, a second visual equivalent corresponding to the second preset direction, and a target rotation angle between the camera coordinate system and the motion platform coordinate system.

[0065] Specifically, after obtaining the aforementioned multiple sets of motion and visual data, the calibration module performs unified processing and analysis on the pulse displacement data and pixel position change data, and calculates the initial calibration parameters through a mathematical model. The first visual equivalent and the second visual equivalent reflect the pixel change ratio corresponding to a unit pulse displacement of the motion platform in two orthogonal directions, respectively, and are used to describe the scale relationship between the device coordinate system and the pixel coordinate system. The target rotation angle is used to characterize the angular deviation between the camera coordinate system and the motion platform coordinate system caused by installation or assembly errors. By simultaneously introducing scale parameters and rotation angle parameters in the same calculation process, the calibration results can more realistically reflect the actual equipment state, avoiding systematic errors caused by calibration in a single direction or with a single parameter.

[0066] Based on the initial calibration parameters, the motion platform is controlled to perform a preset distance of movement verification to obtain the verification result;

[0067] Specifically, after calculating the initial calibration parameters, these parameters were not directly used as the final result. Instead, the motion platform was controlled to perform a preset distance movement verification based on these initial calibration parameters. This verification process uses the calculated visual equivalent and rotation angle to visually predict the actual motion result of the platform, and then compares and analyzes the predicted result with the actual pixel changes to obtain the corresponding verification result. In this way, the applicability and accuracy of the initial calibration parameters in real motion scenarios can be intuitively reflected, elevating the calibration process from simple parameter calculation to a verification process with self-checking capabilities.

[0068] Based on the verification results, the initial calibration parameters are adjusted to obtain target calibration parameters that meet the preset calibration accuracy conditions, so as to achieve integrated circuit vision closed-loop calibration.

[0069] Specifically, finally, based on the deviation results obtained during the verification process, the initial calibration parameters are adjusted in a targeted manner to gradually approach the target calibration parameters that meet the preset calibration accuracy conditions. This adjustment process can be completed through multiple iterations, allowing the visual equivalent and rotation angle to stabilize and become accurate through continuous correction, thus forming a complete visual closed-loop calibration mechanism. By introducing verification and correction steps, the problem of one-time calibration results failing to balance accuracy and stability is avoided, enabling visual calibration in integrated circuit packaging inspection to possess both high accuracy and adaptability to the actual needs of long-term equipment operation and status changes.

[0070] Preferably, the calibration plate surface is provided with calibration features, which include alignment segments for adjusting the camera axis, a square array for visual equivalent calibration, a dot array for extracting rotation information, and positioning marks for positioning the calibration plate.

[0071] Specifically, the calibration plate integrates multiple complementary calibration features on its surface, enabling the same carrier to simultaneously support multiple calibration needs, such as camera attitude adjustment, visual equivalent calculation, and coordinate relationship calibration. Alignment segments are arranged along the reference direction of the calibration plate; their simple geometry and clear directionality make them easily identifiable in images. These segments can be used to assist in adjusting the relative orientation between the camera's imaging axis and the motion platform's main axis, allowing the camera to quickly reach a near-orthogonal working state after initial installation or replacement, providing a good starting point for subsequent fine calibration. The square array, as the main feature for visual equivalent calibration, employs high-precision design in its side length and spacing, forming a stable and regular feature distribution in the image. By analyzing the pixel changes of the feature centers in the square array under different platform displacement states, the pixel scale change relationship corresponding to a unit mechanical displacement can be accurately reflected, thus enabling the calculation of visual equivalent parameters in different directions. The dot array, with its geometric center being insensitive to rotation, is used to extract rotation-related information. By analyzing the overall displacement direction and amplitude changes of the dots before and after platform movement, the rotation center and rotation angle between the camera coordinate system and the motion platform coordinate system can be effectively calculated, providing a reliable basis for subsequent angle compensation. Simultaneously, the positioning marks set in the calibration features define the installation position and orientation of the calibration board on the motion platform, ensuring that the calibration board can quickly return to a consistent spatial reference after repeated loading / unloading or equipment maintenance. By integrating these various patterns onto the same calibration board surface, not only is the complexity of changing calibration tools reduced during calibration, but also the calculation of various calibration parameters is ensured under the same coordinate reference, which helps improve the consistency and overall stability of calibration results, supporting the high-precision visual calibration requirements from both structural and algorithmic perspectives.

[0072] Preferably, the die bonder further includes an image acquisition module. The step of fixing the calibration plate to a preset position on the motion platform, and controlling the motion platform to move along a first preset direction and a second preset direction via the motion control module, to acquire pulse displacement data for each movement and pixel position change data before and after the movement, includes:

[0073] Specifically, this die bonder incorporates an image acquisition module, ensuring that the calibration process not only relies on the controllability of mechanical motion but also has a stable and quantifiable source of visual information. The image acquisition module typically consists of an industrial camera, an imaging lens, and a matching light source. It acquires clear, stable-contrast images of the calibration board when it is in different motion states, thus guaranteeing the consistency and repeatability of calibration features during the imaging process. The image processing module, based on a pre-defined visual algorithm, analyzes and processes the acquired images, identifying key features and extracting pixel coordinates. This transforms the raw image data into precise pixel position data that can be used in calculations. Together, these two modules form the core foundation for visual calibration data acquisition and analysis.

[0074] The image acquisition module acquires the initial calibration plate image at the preset position and the target calibration plate image after the motion platform has moved.

[0075] Specifically, in the calibration process, the image acquisition module first acquires an initial image of the calibration board when it is fixed at a preset position on the motion platform. This image serves as the visual reference when the platform is not displaced. Subsequently, after the motion control module drives the motion platform to complete controlled movement along a first or second preset direction, the corresponding target calibration board image is acquired again. By acquiring image data before and after motion under the same imaging conditions, the adverse effects of changes in illumination or fluctuations in imaging parameters on the calibration results can be effectively avoided, ensuring that subsequent pixel change analysis only reflects the visual changes caused by platform displacement.

[0076] The image processing module extracts pixel positions from the initial calibration board image and the target calibration board image to obtain the initial pixel position and the target pixel position.

[0077] Specifically, the image processing module processes the initial calibration board image and the target calibration board image respectively. By identifying and locating feature elements in the calibration features, it extracts the corresponding initial pixel positions and target pixel positions. This processing typically revolves around regular features such as square arrays and dot arrays, and obtains stable pixel coordinate results through edge detection, center fitting, or feature matching.

[0078] Preferably, the step of extracting pixel positions from the initial calibration board image and the target calibration board image using the image processing module to obtain the initial pixel positions and target pixel positions includes:

[0079] The initial calibration board image and the target calibration board image are preprocessed respectively to obtain corresponding preprocessed images;

[0080] Specifically, without altering the geometric structure of the calibrated features, noise and illumination fluctuations that affect the stability of subsequent feature detection are first suppressed. Preprocessing typically begins with the original image, smoothing and suppressing local salt-and-pepper noise, sensor readout noise, or fine textures caused by motion jitter. Simultaneously, grayscale normalization is performed on the entire image or local regions to ensure more consistent brightness levels across different frames. If there is slight unevenness in the ambient light, illumination compensation can be performed using background estimation to ensure similar contrast for the same type of calibrated features in both the initial and target images. This processing makes subsequent quality metrics such as edge gradients and grayscale contrast more reliable, and the confidence evaluation of candidate features is less likely to be misled by occasional brightness changes, thereby improving the repeatability of pixel location extraction under multiple sampling operations.

[0081] The calibration board region is located in each of the preprocessed images to obtain the region of interest of the calibration board;

[0082] Specifically, the calibration board region is located in each preprocessed image to obtain the region of interest (ROI). The key is to first limit the computational scope to the actual area where the calibration board is located, reducing interference from background structures, fixture reflections, or other device textures on candidate feature detection. In implementation, the stable appearance of the calibration board in the image can be used for region localization. For example, its approximate position and orientation can be determined by combining positioning marks or the geometric contours of the calibration board's edges, and then the ROI can be cropped accordingly. Alternatively, a consistency check can be added during the localization stage to ensure that the ROI of the initial image and the target image cover the same calibration board area, avoiding inconsistencies in the cropped area due to platform movement or field of view shift. By obtaining the ROI first, subsequent detection of calibration features becomes more focused and stable, which is particularly beneficial in maintaining the accuracy of candidate feature extraction even when there are localized bright reflections or stray textures in the packaging equipment environment.

[0083] The calibration features are detected in the region of interest of the calibration plate to obtain multiple candidate calibration features;

[0084] Specifically, detecting calibration features and obtaining multiple candidate calibration features within the region of interest on the calibration board transforms the actual geometric pattern on the board into a computable set of candidate targets, providing input for subsequent calculation of the center pixel coordinates. In implementation, feature enhancement is typically performed first based on preprocessed grayscale and edge information. Then, detection is performed on the expected geometric shapes on the calibration board. For example, line segment candidates can be obtained through line detection for line segments, square candidates can be obtained through corner point or rectangular contour detection for square arrays, and round / elliptical candidate candidates can be obtained through circular / elliptical contour detection for round dot arrays. During the detection process, more candidate features than the actual number are often generated, which may include both true calibration features and pseudo-features formed by stains, scratches, or reflections. Retaining these as candidates initially ensures recall. Subsequently, confidence-driven adaptive filtering removes low-quality candidates, thus balancing stable detection and anti-interference capabilities.

[0085] For each candidate calibration feature, an image quality index corresponding to the candidate calibration feature is extracted, and the feature confidence of the candidate calibration feature is determined based on the image quality index. The image quality index includes edge gradient magnitude, gray-level contrast, contour closure and shape matching.

[0086] Specifically, extracting image quality metrics and determining feature confidence for each candidate calibration feature transforms the candidate features from a set of questionable features into a prioritized and quantifiable set, making subsequent pixel location extraction more reliable. Among the image quality metrics, edge gradient magnitude reflects the clarity of feature boundaries, grayscale contrast characterizes the separation between the candidate region and the background, contour closure characterizes the completeness and continuity of the candidate contour, and shape matching measures the degree of fit between the candidate contour and the expected geometric model. In implementation, these metrics are calculated separately for each candidate region, normalized, and then fused to obtain a comprehensive quality score, which is used to determine the confidence level. The significance of this is that when some candidate features are affected by uneven local illumination, blurred edges, or occlusion, their gradient, closure, or matching will naturally deteriorate. The system will lower their confidence and weaken their impact in subsequent screening, thereby avoiding the introduction of systematic bias due to low-quality features participating in center coordinate calculations, and improving the stability and repeatability of the calibration process under conditions of multiple moving samples and closed-loop verification.

[0087] Preferably, the step of extracting an image quality index corresponding to each candidate calibration feature and determining the feature confidence level of the candidate calibration feature based on the image quality index includes:

[0088] Perform local grayscale analysis on the candidate feature region corresponding to each candidate calibration feature, and calculate the grayscale contrast index between the candidate feature region and its surrounding background region.

[0089] Specifically, local grayscale analysis is performed on the candidate feature region corresponding to each candidate calibration feature, and a grayscale contrast index is calculated. This is mainly to determine whether the candidate feature is sufficient to separate from the background under the current imaging conditions, avoiding the misclassification of background textures or reflections as effective features. In practice, grayscale distribution is typically first statistically analyzed within the candidate feature region, such as calculating the average grayscale, grayscale variance, or the main peak position of the grayscale histogram. Then, a neighboring background region is constructed outside the candidate feature region, and the background grayscale features are statistically analyzed in the same way. The grayscale contrast can be represented by the average grayscale difference, contrast ratio, or normalized difference, thus accommodating brightness variations under different exposure intensities. The resulting contrast index directly reflects whether the candidate region has a clear light-dark boundary. Subsequently, during feature confidence fusion, high-contrast candidate features are assigned higher confidence weights, making pixel location extraction less susceptible to grayscale drift caused by uneven lighting, local shadows, or surface contamination.

[0090] Edge detection is performed on the candidate feature regions to obtain edge gradient distribution information, and the mean or peak value of the edge gradient is calculated as an edge sharpness index.

[0091] Specifically, edge detection is performed on candidate feature regions to obtain edge gradient distribution information, which is used to characterize the sharpness and stability of candidate feature boundaries, thus forming an edge sharpness index. In implementation, gradient magnitude and direction are typically calculated within the candidate region. For example, the gradient magnitude of each pixel is obtained based on common gradient operators, and then the gradient magnitudes are statistically analyzed to obtain the mean or peak value as a sharpness index. The mean is more suitable for reflecting whether the overall boundary is generally sharp, while the peak value is more suitable for reflecting whether there is a sufficiently strong main boundary response. To reduce the impact of noise, edge detection is often combined with preliminary preprocessing results for moderate smoothing, and obviously outlier noise gradient points are filtered out when statistically analyzing the gradient distribution. Candidate features with high edge sharpness usually mean more stable contour positions and more accurate center calculations, and therefore are more worthwhile to retain in subsequent confidence-weighted selection, thereby improving the repeatability of pixel displacement measurements under multiple moving sampling conditions.

[0092] Contour extraction is performed on the candidate feature regions, and contour closure and contour continuity indices are calculated.

[0093] Specifically, contour extraction and contour closure and continuity indices are performed on candidate feature regions to determine whether candidate features have complete and coherent geometric boundaries, preventing incomplete contours caused by occlusion, reflection breaks, or dirt from entering subsequent center calculations. In implementation, contour point sets are first extracted from the candidate region based on edge results or binarized segmentation results. Then, the topological integrity of the contours is analyzed. For example, closure can be characterized by the distance between the beginning and end of the contour, whether a closed loop is formed, or whether the contour area is stable. Continuity can be evaluated by the discontinuity length between contour points, the number of discontinuities, or the number of connected segments of the contour curve. For calibration features such as square arrays and dot arrays, the more closed and continuous the contour, the smaller the error in solving the geometric center or fitted center. Therefore, by introducing closure and continuity indices, false features with broken boundaries can be preferentially screened out during the confidence evaluation stage, reducing center offset caused by incomplete contours, thereby improving the stability and accuracy of calibration parameter calculations.

[0094] Based on the preset geometric model of the candidate calibration features, the shape matching degree of the candidate feature region is calculated to obtain the shape consistency index.

[0095] Specifically, a shape matching degree is calculated for candidate feature regions based on a preset geometric model of candidate calibration features to obtain a shape consistency index. This mainly uses the known geometric rules of the calibration board features to filter out false features and improve the reliability of center localization. The preset geometric model can correspond to the rectangular boundaries and corner relationships of a square array, or to the circular or elliptical contour features of a dot array. During shape matching, contour point sets or edge point sets are usually obtained from the candidate feature regions first, and then fitted or aligned with the target model. For example, for dot features, circle fitting or ellipse fitting can be performed and the fitting residual can be calculated. For square features, minimum bounding rectangle fitting can be performed and the side length ratio, right angle deviation, and contour overlap can be calculated. The shape consistency index can be derived from the fitting residual, the overlap area ratio, or the shape similarity. The higher the value, the more the candidate features match the real pattern of the calibration board, thus enabling the extraction of more calibration features that rely on the real structure and geometric stability for subsequent pixel positions, reducing the false detection impact caused by stains, reflections, or background textures.

[0096] The grayscale contrast index, edge sharpness index, contour closure index, and shape consistency index are normalized to obtain the corresponding normalized quality parameters.

[0097] Specifically, the four indicators—grayscale contrast, edge sharpness, contour closure, and shape consistency—are normalized to obtain normalized quality parameters. This is to ensure that quality indicators with different dimensions and numerical ranges can be compared within the same fusion framework. Since contrast may be expressed as grayscale difference or ratio, and gradient magnitude, closure, and matching degree have different scales, direct addition could lead to one indicator dominating the score due to its larger dimension. Therefore, in implementation, each indicator needs to be mapped to a unified scale range, for example, by linearly scaling it to the 0-1 interval, or by truncating and normalizing it based on the upper and lower limits obtained from historical statistics. This ensures that each indicator stably reflects the quality level without amplifying noise due to extreme values. The normalized quality parameters not only facilitate subsequent weighted fusion but also make the confidence calculation more robust to different imaging conditions. Even with slight changes in exposure or lighting, the indicators can still reflect relative superiority or inferiority at a unified scale, thus maintaining the consistency of the candidate feature selection strategy.

[0098] The normalized quality parameters are weighted and fused according to preset weights to obtain the comprehensive quality score of the candidate calibration features;

[0099] Specifically, a comprehensive quality score is obtained by weighted fusion of normalized quality parameters according to preset weights. This aggregates quality information from different dimensions into a unified evaluation metric that can be directly used for screening and ranking, thereby achieving confidence-driven adaptive feature selection. In implementation, weights can be set for grayscale contrast, edge sharpness, contour closure, and shape consistency, and the normalized parameters are fused by weighted summation or weighted product. The weights can be differentiated according to the type of calibrated feature. For example, for a dot array, the weights of shape consistency and closure can be increased to emphasize the stability of circle fitting; for a square array, the weights of edge sharpness and right-angle shape consistency can be increased to emphasize the reliability of corner point positioning. The comprehensive score can also be smoothed or truncated after fusion to avoid distortion caused by a single abnormal indicator. The comprehensive quality score uniformly characterizes the usability of candidate features. Subsequently, high-scoring features can be retained using a fixed threshold, or the top features can be selected according to the score ranking to participate in pixel position calculation and calibration parameter fitting. This allows the image processing module to automatically select a more reliable feature set when feature quality fluctuates, thereby improving the stability and accuracy of calibration parameter calculation.

[0100] The overall quality score is compared with a preset reliability threshold, or the feature confidence of the candidate calibration feature is determined based on the ranking of the overall quality score among all candidate calibration features.

[0101] Specifically, comparing the comprehensive quality score with a pre-set confidence threshold, or determining the feature confidence of candidate calibration features based on the ranking of the comprehensive quality score among all candidate calibration features, is a key step in transforming the aforementioned multidimensional quality analysis results into an executable screening strategy. In implementation, a corresponding comprehensive quality score is first assigned to each candidate calibration feature. This score comprehensively reflects factors such as grayscale separation, boundary clarity, contour integrity, and geometric consistency. If a threshold-based approach is used, a confidence threshold is pre-set based on historical calibration data or equipment operating experience. Candidate calibration features with scores above the threshold are marked as high-confidence features, while those below the threshold are downweighted or eliminated. If a ranking approach is used, all candidate calibration features are arranged from highest to lowest score, and the top few are selected as the effective feature set, with the remaining features serving as alternatives or not participating in the center coordinate calculation. During execution, the strategy can be dynamically adjusted based on the number of candidate features within the current field of view. For example, when the number of candidate features is small, ranking can be prioritized to retain the top few highest-scoring features to ensure sample size; when the number of candidate features is sufficient, a combination of thresholding and ranking can be used to simultaneously ensure both quality and quantity. In this way, the image processing module no longer simply relies on a fixed number of features to participate in the calculation. Instead, it adaptively selects more reliable features to participate in subsequent geometric center calculation and visual equivalent fitting based on the current image quality. This reduces the disturbance of low-quality features to the calibration results and enables the calibration parameters to maintain better stability and consistency during multiple rounds of moving verification and closed-loop adjustment.

[0102] Based on the feature confidence level, the candidate calibration features are adaptively filtered to obtain the target calibration feature set;

[0103] Specifically, adaptively filtering candidate calibration features based on feature confidence to obtain the target calibration feature set is the process of truly applying confidence from a scoring result to usable samples for calibration calculation. In implementation, each candidate calibration feature is first assigned its corresponding feature confidence. Then, the filtering method is selected based on the number and distribution of candidate features in the current image. For example, when the number of candidate features is sufficient, a confidence threshold is used to eliminate low-confidence features, ensuring that features entering subsequent calculations have clear boundaries, complete contours, and consistent shapes. When the number of candidate features is insufficient to support stable fitting, the top few features can be selected by sorting by confidence, while stricter retention conditions are set for features at edges or in areas easily affected by reflection. For calibration features arranged in a regular pattern, such as square arrays or dot arrays, spatial consistency constraints can be added during the filtering stage. For example, the retained features should meet the expected array spacing relationship or the overall distribution should cover multiple areas, thus avoiding calibration bias caused by only locally clustered high-confidence points remaining. Through adaptive filtering, all subsequent pixel position calculations are based on a more stable feature set, which can significantly reduce the disturbance of false detections, missed detections, and local poor-quality features on the calibration results.

[0104] Contour extraction is performed on each target calibration feature in the target calibration feature set to obtain a contour point set;

[0105] Specifically, extracting contours from each target calibration feature in the target calibration feature set to obtain a contour point set further refines the target features from region-level candidate boxes to a set of boundary points, facilitating subsequent center localization or geometric fitting. In practice, edge detection or binarization segmentation is typically performed within the local region containing the target features, and then connected component boundaries are extracted from the segmentation results to obtain a sequentially arranged contour point set. For a dot array, the contour point set generally forms a closed curve, suitable for circle or ellipse fitting; for a square array, the contour point set can be used to extract the four sides or four corner points, and the geometric center can be calculated accordingly. To avoid jagged boundaries introducing center offset, simple smoothing or resampling can be performed on the contour point set to make the contour points more evenly distributed and improve fitting stability. By transforming target features into a contour point set, the subsequent calculation of center pixel coordinates can utilize more accurate boundary information, rather than relying on grayscale fluctuations within the region, thereby improving the repeatability of center localization.

[0106] The center pixel coordinates of each target calibration feature are calculated based on the contour point set to obtain the feature center pixel coordinates.

[0107] Specifically, calculating the center pixel coordinates of each target calibration feature based on the contour point set and obtaining the feature center pixel coordinates is a process of transforming geometric boundary information into a single center point parameter. The center point will directly participate in pixel displacement calculation and visual equivalent fitting. In implementation, the center calculation method can be selected according to the target calibration feature type. For example, for a circle array, the contour point set can be fitted with a circle or an ellipse, and the center of the fitted circle or ellipse can be used as the center pixel coordinates. For a square array, the intersection of the four sides can be obtained through minimum bounding rectangle fitting or edge fitting, and then the intersection of the diagonals can be taken as the center pixel coordinates. If the contour point set has local gaps or slight deformations, the previous confidence screening has already eliminated low-quality features as much as possible, but robust fitting or removal of outlier contour points can still be used to further reduce the impact of noise on the center when calculating the center. The feature center pixel coordinates obtained in this way are not only closer to the true geometric center but also more stable, providing a consistent measurement benchmark for subsequent pixel displacement calculations through multiple platform movements.

[0108] Based on the feature center pixel coordinates, the initial pixel position in the initial calibration board image and the target pixel position in the target calibration board image are determined.

[0109] Specifically, the initial pixel positions in the initial calibration board image and the target pixel positions in the target calibration board image are determined based on the feature center pixel coordinates. The key is to correctly match the same calibration feature in the two frames and form a set of pixel positions that can be used for difference calculation. In implementation, the feature center pixel coordinate sets of the initial and target calibration board images are first obtained separately. Then, matching rules are established based on the type and array structure of the calibration features. For example, for square arrays, matching can be performed according to row and column indices or spatial adjacency; for dot arrays, matching can be performed according to their relative position in the array or their relative coordinates with the positioning markers, ensuring a one-to-one correspondence between a center point in the initial image and its corresponding center point in the target image. After matching, the corresponding center points in the initial image are used to form the initial pixel positions, and the corresponding center points in the target image are used to form the target pixel positions. This allows the difference calculation of subsequent pixel position change data to be performed under strict feature alignment, avoiding displacement anomalies caused by mismatches. Through this process, the calculation basis of the pixel displacement vector becomes more reliable, further supporting the accuracy and stability of subsequent rotation angle calculation, visual equivalent fitting, and closed-loop verification adjustment.

[0110] The difference between the initial pixel position and the target pixel position is calculated to obtain the pixel position change data;

[0111] Specifically, after extracting the corresponding pixel positions before and after the calibration board's movement, the pixel displacement of the calibration board in the image coordinate system can be directly obtained by calculating the difference between the initial pixel position and the target pixel position. This pixel position change data is not isolated but corresponds one-to-one with the actual motion state of the motion platform. Its direction and magnitude of change reflect the visual response of the calibration board during the controlled movement of the platform. Through this difference method, the pixel coordinate information originally scattered in two frames of images is transformed into a unified displacement data form, which is beneficial for subsequent centralized analysis and modeling of multiple motion results in different directions.

[0112] The motion control module acquires the pulse displacement data for each movement of the motion platform.

[0113] Specifically, during the execution of each displacement command by the motion platform, the motion control module synchronously acquires and outputs corresponding pulse displacement data. This data characterizes the actual movement of the motion platform in the device coordinate system. Since the pulse displacement data originates directly from the motion control system, its accuracy and stability are closely related to the platform's motion control capabilities, thus serving as a reliable physical displacement reference. By pairing this pulse displacement data with the aforementioned pixel position change data, a direct mapping relationship between mechanical displacement and visual displacement can be established.

[0114] Preferably, the step of extracting pixel positions from the initial calibration board image and the target calibration board image using the image processing module to obtain the initial pixel positions and target pixel positions includes:

[0115] The initial calibration board image and the target calibration board image are respectively input into a pre-trained calibration feature recognition model to obtain the first calibration feature in the initial calibration board image and the second calibration feature in the target calibration board image;

[0116] Specifically, during pixel location extraction, the initial calibration board image and the target calibration board image are respectively input into a pre-trained calibration feature recognition model for processing. This model is trained on a predetermined pattern structure on the calibration board and can stably distinguish and locate calibration feature regions under complex backgrounds or minor imaging differences. Through the recognition output of this model, the corresponding first calibration feature can be obtained in the initial image, and the matching second calibration feature can be obtained in the moved image. This ensures that the visual features involved in subsequent calculations are always of the same type and with the same positional relationship, avoiding deviations introduced by misidentification or feature drift.

[0117] Contour features are extracted from the first calibration feature and the second calibration feature respectively to obtain the first contour position information of the first calibration feature and the second contour position information of the second calibration feature;

[0118] Specifically, after obtaining the calibration feature region, contour feature extraction is further performed on the first and second calibration features respectively. By analyzing the geometric shape of the pattern edge, its contour position information in the image is extracted. This contour information can accurately reflect the spatial distribution of the calibration features in the imaging plane. Compared with directly using pixel grayscale or single-point features, the contour-based description method is more robust to noise and local defects, which is beneficial to improving the stability of pixel positioning.

[0119] Based on the first contour position information and the second contour position information, the geometric center positions of the first calibration feature and the second calibration feature are calculated respectively to obtain the first pixel coordinate position and the second pixel coordinate position;

[0120] Specifically, based on the extracted first and second contour position information, the geometric center position of the corresponding calibration feature is calculated to obtain the first pixel coordinate position and the second pixel coordinate position. The geometric center, as a comprehensive expression of the overall contour features, can effectively reflect the overall displacement of the calibration feature, avoiding measurement errors caused by local edge distortion or anomalies of individual feature points. By uniformly adopting the geometric center as the position expression form, good consistency is achieved in the results of motion in different directions and with different numbers of repetitions.

[0121] The first pixel coordinate position is used as the initial pixel position, and the second pixel coordinate position is used as the target pixel position.

[0122] Specifically, the first pixel coordinates are used as the initial pixel position, and the second pixel coordinates are used as the target pixel position, providing direct input for subsequent pixel position change calculations. Through this hierarchical processing method, the pixel position acquisition process unfolds step by step from pattern recognition and contour extraction to center calculation, ensuring the accuracy of feature matching and improving the precision and reliability of pixel position extraction, thus laying a stable data foundation for visual equivalent calculation and closed-loop calibration.

[0123] Preferably, please refer to Figure 2 The initial calibration parameters obtained by calculating calibration parameters for the pulse displacement data and pixel position change data through the calibration module include:

[0124] Based on the pixel position change data, obtain the slope of the pixel displacement vector before and after the motion platform moves along the first preset direction multiple times;

[0125] Specifically, in the calibration parameter calculation stage, the pixel displacement results formed before and after multiple movements of the motion platform along a first preset direction are first analyzed based on pixel position change data. The displacement of each movement in the image coordinate system is represented as a pixel displacement vector. Since there may be installation angle deviations between the camera coordinate system and the motion platform coordinate system, these pixel displacement vectors are usually not strictly distributed along a single coordinate axis in the image; their directional characteristics can be quantified by slope. By statistically analyzing the slopes of the pixel displacement vectors corresponding to multiple movements, basic data that stably reflects the relative attitude relationship of the coordinate systems can be obtained.

[0126] Based on the coordinate axes of the preset pixel coordinate system, obtain the quadrant assignment of each pixel displacement vector in the pixel coordinate system;

[0127] Specifically, the pixel coordinate system is a defined coordinate system with the top-left corner of the image as the origin. Once the coordinate axes are fixed, the positive and negative combinations of the horizontal and vertical components of the pixel displacement vector will fall within a specific quadrant. In implementation, the image processing module typically first locates stable features on the calibration board, such as extracting the center pixel coordinates of a square array or a dot array, obtaining the feature point pixel coordinates before and after platform movement, and then subtracting them to obtain the pixel displacement vector. To reduce single-point errors, similar to the angle compensation approach described in the document, pixel coordinates of multiple feature points can be calculated, or the displacements of multiple feature points can be statistically fused to obtain a more stable displacement vector before determining the quadrant. Determining the quadrant assignment first is equivalent to locking the candidate range for subsequent slope-to-angle conversion within a certain interval, avoiding multiple angle solutions corresponding to the same slope in different quadrants, eliminating systematic errors caused by unstable angle signs at the root, and supporting the reliability of subsequent coordinate transformations and automatic angle compensation.

[0128] Based on the motion direction of the first preset direction in the preset motion platform coordinate system and the correspondence between the first preset direction and the coordinate axes of the pixel coordinate system, the directional consistency constraint between the pixel displacement vector direction and the first preset direction is determined.

[0129] Specifically, the first preset direction in engineering typically corresponds to a certain axis of the machine coordinate system. The machine coordinate system takes a fixed point preset by the platform as its origin and provides an automatic angular compensation path that triggers the controllable pixel response by controlling the platform to move a fixed number of pulses in the first preset direction. The motion control module first issues a forward or reverse displacement command in the predetermined direction and outputs the corresponding number of pulses. At the same time, the image acquisition module acquires images before and after the motion under the same imaging conditions. After the image processing module calculates the pixel displacement vector, it compares the sign of its components on the pixel coordinate axis with the forward and reverse directions of the platform's motion to establish a consistency rule. For example, when the platform moves along the first preset direction, the pixel displacement vector in the pixel coordinate system should show the principal component sign and change trend that matches the direction of motion. By introducing this directional consistency constraint derived from the platform's motion direction, the pixel displacement vector is no longer an isolated image measurement result, but rather establishes a verifiable correspondence with the controlled motion of the machine tool. In this way, even if there is an installation deviation angle, the angle branch consistent with the machine tool's motion can be selected by using this constraint when converting the slope to the angle later, thereby improving the determinism and repeatability of the deviation angle solution and meeting the goals of automation and automatic deviation angle compensation.

[0130] Based on the quadrant assignment and the direction consistency constraint, the range and sign of the angle values ​​corresponding to the slope of each pixel displacement vector are constrained.

[0131] Specifically, the slope itself only reflects the tangent of the pixel displacement direction, which can easily lead to problems such as ambiguous angle signs or non-unique angle ranges. Therefore, quadrant assignment and direction consistency are used together as constraints to limit the legal range of the angle corresponding to the slope. In implementation, the slope is first calculated from the pixel displacement vector, and the basic range of the angle is given according to the quadrant assignment, for example, restricting the angle to the range corresponding to the quadrant; then, the direction consistency constraint is used to further eliminate angle branches that do not match the platform motion direction, so that the sign and value range of the angle are locked at the same time. After this processing, the slope is still the input for the deflection angle calculation, but it no longer directly determines the deflection angle. Instead, it obtains a unique angle interpretation under the combined effect of the two types of constraints, avoiding the deflection angle from jumping or flipping its sign in multiple calibrations, thereby reducing the accumulation of rotation matrix error in the coordinate transformation model and improving the stability of subsequent visual equivalent fitting and whole-machine coordinate conversion.

[0132] Under the constraints of the angle value range and sign, the target rotation angle is calculated based on the slope of each pixel displacement vector;

[0133] Specifically, under the constraints of the angle's range and sign, the rotational relationship between the camera coordinate system and the motion platform coordinate system is calculated using the slopes of the multiple sets of pixel displacement vectors, thereby obtaining the target rotation angle. This method of multiple sampling and comprehensive calculation helps to reduce the impact of single-motion errors or image noise on the results, ensuring that the obtained rotation angle accurately reflects the current installation state of the equipment. This rotation angle serves as an important parameter for subsequent coordinate transformations, describing the actual projection relationship of the mechanical motion direction in the image coordinate system.

[0134] Based on the pixel position change data and the pulse displacement data, the first visual equivalent and the second visual equivalent are calculated by least squares fitting.

[0135] Specifically, after completing the rotation relationship analysis, pixel position change data and corresponding pulse displacement data are introduced as paired samples. The linear relationship between the two is fitted and calculated using the least squares method to obtain the visual equivalent parameters in the first and second preset directions, respectively. The least squares method solves multiple sets of data uniformly by minimizing the overall error, which can effectively suppress the influence of random errors on the calculation of visual equivalent in a single direction, making the obtained scale parameters smoother and more stable, and suitable for subsequent accuracy verification and compensation calculations.

[0136] The initial calibration parameters are determined based on the target rotation angle, the first visual equivalent, and the second visual equivalent.

[0137] Specifically, the calculated target rotation angle, first visual equivalent, and second visual equivalent are then comprehensively determined to form a complete set of initial calibration parameters. This parameter set contains both scale and angle information, comprehensively describing the spatial mapping relationship between the camera coordinate system and the motion platform coordinate system, providing a unified and readily available basic calibration result for subsequent motion verification, parameter correction, and visual closed-loop calibration.

[0138] Preferably, the step of calculating the target rotation angle based on the slope of each pixel displacement vector, under the constraints of the angle value range and sign, includes:

[0139] Based on the slope of each pixel displacement vector and the preset mapping relationship between the slope and the rotation angle, each initial rotation angle is calculated.

[0140] Specifically, in the calculation of the rotation angle, the directional characteristics embodied by the pixel displacement vector are first utilized to establish a relationship between its slope and the relative rotation relationship between the camera coordinate system and the motion platform coordinate system. Based on the pre-defined mapping relationship between the slope and the rotation angle, the slope of the pixel displacement vector corresponding to each movement of the platform along a preset direction can be converted into a corresponding initial rotation angle. This mapping relationship originates from the geometric description of the rotation matrix in the coordinate transformation model, enabling the directional changes at the pixel level to be quantified as angle parameters. This transforms the motion direction information in the image into a rotation amount that can be directly used for coordinate compensation, by controlling the platform to move a fixed number of pulses in the X direction. Collect the pixel coordinates of feature points before and after movement. and Calculate the initial rotation angle The calculation formula is as follows:

[0141]

[0142] The average value of each initial rotation angle is calculated, and the average value of the calculated rotation angles is taken as the target rotation angle.

[0143] Specifically, due to the potential influence of image noise, feature extraction errors, or minor platform fluctuations during a single movement, there is usually a certain degree of dispersion among different initial rotation angles. Therefore, comprehensively processing the initial rotation angles obtained from multiple calculations is more beneficial to improving the stability of the results. By calculating the average of each initial rotation angle, the influence of outliers or random errors can be effectively reduced, making the final rotation angle closer to the actual installation state of the system. Using the rotation angle obtained from the above average calculation as the target rotation angle not only gives this parameter good repeatability and reliability but also provides a stable angular benchmark for subsequent visual equivalent calculations and coordinate transformations. By introducing multiple sampling and statistical fusion, the process of obtaining the rotation angle is transformed from a single estimation to parameter determination based on the overall trend, which is beneficial to improving the consistency and long-term stability of the visual calibration results in actual packaging inspection operations.

[0144] Preferably, the step of calculating the first visual equivalent and the second visual equivalent by least squares fitting based on the pixel position change data and the pulse displacement data includes:

[0145] Based on the pixel position change data, obtain the displacement of each first pixel corresponding to the first preset direction and the displacement of each second pixel corresponding to the second preset direction;

[0146] Specifically, in the visual equivalent calculation process, the pixel position change data is first distinguished according to the direction of motion. The pixel displacement results generated when the motion platform moves along the first preset direction and the second preset direction are extracted respectively to form the corresponding first pixel displacement and second pixel displacement. Since the two preset directions are orthogonal to each other, this distinction in direction helps to decouple the visual responses of different axes, so that subsequent calculations can reflect the scale characteristics of each motion direction in the image coordinate system, avoiding calculation errors caused by the mixing of displacements in different directions.

[0147] Based on the pulse displacement data, obtain the first pulse displacement amount corresponding to the first preset direction and the second pulse displacement amount corresponding to the second preset direction;

[0148] Specifically, corresponding to the process of acquiring pixel displacement, the pulse displacement data is also organized according to the direction of motion to obtain the first pulse displacement corresponding to the first preset direction and the second pulse displacement corresponding to the second preset direction. This step ensures that each set of pixel displacement data can form a clear pairing relationship with the mechanical displacement data corresponding to its generation, so that the visual data and motion data are consistent in direction and quantity, providing a clear data structure for subsequent fitting calculations.

[0149] The least squares method is used to linearly fit the displacement of each first pixel and the corresponding displacement of each first pulse, and the slope of the fitted first straight line is used as the first visual equivalent.

[0150] Specifically, after data pairing is completed, the displacement of each first pixel and its corresponding first pulse displacement are used as a sample set and introduced into the least squares method for linear fitting. The optimal linear relationship between the two is solved by minimizing the overall error. The slope of the fitted line is used to characterize the pixel change caused by a unit pulse displacement in the image, i.e., the visual equivalent in the first preset direction. This calculation method can integrate the results of multiple motions, reduce the impact of single measurement errors on the visual equivalent, and make the scale parameters smoother and more stable.

[0151] The least squares method is used to linearly fit each second pixel displacement and the corresponding second pulse displacement, and the slope of the fitted second straight line is used as the second visual equivalent.

[0152] Specifically, similarly, the displacement of each second pixel is linearly fitted with the corresponding displacement of the second pulse to obtain the slope of the fitted straight line in the second preset direction, which is then used as the second visual equivalent. By calculating the visual equivalent parameters in the two orthogonal directions respectively, the mapping relationship between the motion platform's axial motion in the plane and the changes in image pixels can be fully described, thus providing an accurate scale basis for subsequent coordinate transformation, accuracy verification, and visual closed-loop calibration.

[0153] Specifically, collect multiple sets of data. (Pixel displacement) and (Pulse displacement), the first visual equivalent was calculated using the least squares fitting method. Second visual equivalent ,in,

[0154] .

[0155] This represents the pixel displacement in the x-direction. This represents the pixel displacement in the y-direction. This represents the pulse displacement in the x-direction. This represents the pulse displacement in the y-direction.

[0156] Preferably, the step of controlling the motion platform to perform a preset distance movement verification based on the initial calibration parameters, and obtaining the verification result, includes:

[0157] Based on the first visual equivalent and the second visual equivalent, the preset distance is converted into the corresponding theoretical pixel displacement.

[0158] Specifically, in the motion verification phase, the first and second visual equivalents are used to transform the pre-set verification distance from the device coordinate space to the image coordinate space, and the corresponding theoretical pixel displacement is calculated. This theoretical pixel displacement reflects the pixel change that the calibration board should produce in the image when the motion platform moves a preset distance in a specified direction under ideal calibration conditions, thus providing a clear reference for subsequent verification.

[0159] The motion platform is controlled to perform a preset distance of movement verification, and calibration plate images are acquired before and after the movement;

[0160] Specifically, under the control of the motion control module, the motion platform is driven to move at the preset distance, and corresponding calibration board images are acquired before and after the movement. By acquiring verification images under the same imaging parameters and lighting conditions, it can be ensured that the differences between the images before and after the movement mainly come from the platform displacement itself, rather than changes in the external environment, thus making the verification process highly comparable.

[0161] Pixel displacement is calculated on the calibration board images before and after the movement to obtain the actual pixel displacement.

[0162] Specifically, after acquiring the verification image, pixel displacement is calculated for the calibration board images before and after movement. Image processing methods are used to extract the positional changes of calibration board features within the images, yielding the actual pixel displacement. This actual pixel displacement reflects the platform's true visual response under the current calibration parameters and is a crucial basis for evaluating the effectiveness of the initial calibration parameters.

[0163] The difference between the actual pixel displacement and the theoretical pixel displacement is calculated to obtain the verification result.

[0164] Specifically, the difference between the actual pixel displacement and the theoretical pixel displacement is calculated to obtain the corresponding verification result. This result is used to quantify the degree of deviation of the initial calibration parameters in actual motion scenarios, providing a basis for judging whether subsequent corrections to the visual equivalent or rotation angle are needed, thus making the calibration process verifiable and evaluable.

[0165] Preferably, adjusting the initial calibration parameters based on the verification results to obtain target calibration parameters that meet preset calibration accuracy conditions, thereby achieving integrated circuit vision closed-loop calibration, includes:

[0166] Based on the preset calibration accuracy conditions, the error threshold is obtained;

[0167] Specifically, after completing the motion verification and obtaining the corresponding verification results, a preset calibration accuracy condition is introduced as an evaluation criterion. This condition determines the allowable error threshold, which is used to limit the acceptable range of visual calibration results in practical applications. This error threshold is usually set comprehensively based on the positioning accuracy requirements of the die bonder, the repeatability of the motion platform, and the resolution capability of visual measurement. This ensures that the calibration evaluation standard can meet the accuracy requirements of integrated circuit packaging inspection while also being engineering-feasible, thus providing a unified standard for subsequent parameter judgment.

[0168] The verification result is compared with the error threshold. If the verification result is less than or equal to the error threshold, the initial calibration parameter is used as the target calibration parameter.

[0169] Specifically, after obtaining the error threshold, the verification result is compared with the error threshold. When the verification result is less than or equal to the error threshold, it indicates that the pixel-device coordinate mapping relationship established based on the current initial calibration parameters can meet the expected accuracy requirements. In this case, the initial calibration parameters can stably reflect the actual relationship between the camera coordinate system and the motion platform coordinate system, and can be directly identified as the target calibration parameters and used in subsequent packaging and inspection processes, thereby avoiding unnecessary repetitive calculations or corrections and improving overall calibration efficiency.

[0170] If the verification result is greater than the error threshold, the initial calibration parameters are adjusted according to the verification result to obtain the target calibration parameters.

[0171] Specifically, when the verification result exceeds the error threshold, it indicates that the initial calibration parameters are insufficient to meet the accuracy requirements under the current conditions. Targeted adjustments to the relevant parameters are needed, taking into account the deviations reflected in the verification results. This adjustment process can focus on correcting the visual equivalent or rotation angle, enabling the corrected parameters to more accurately describe the relationship between actual motion and visual response. Through this judgment and adjustment mechanism, the calibration process forms a closed-loop structure, gradually approaching the target calibration parameters that meet the accuracy requirements through continuous verification and correction, thereby achieving stable and reliable visual closed-loop calibration in integrated circuit packaging inspection.

[0172] Preferably, if the verification result is greater than the error threshold, adjusting the initial calibration parameters based on the verification result to obtain the target calibration parameters includes:

[0173] If the verification result is greater than the error threshold, then obtain the initial parameter adjustment step size corresponding to the initial calibration parameters, the first verification result corresponding to the previous round of moving verification, and the second verification result obtained in the current round of moving verification;

[0174] Specifically, if the verification result exceeds the error threshold, an initial parameter adjustment step size and two consecutive rounds of verification results are introduced. This is mainly to establish a controllable update benchmark and comparison benchmark for subsequent iterative corrections. The initial parameter adjustment step size can be understood as the upper limit of the allowable change or a proportional coefficient when incrementally correcting the target rotation angle, first visual equivalent, and second visual equivalent, ensuring that parameter updates do not overshoot due to a single excessive error. In implementation, after completing the current round of preset distance movement verification, the system obtains a quantified verification result, such as the error formed by the difference between the theoretical pixel displacement and the actual pixel displacement, and compares it with the error threshold to trigger iteration. Once iteration is triggered, the first verification result of the previous round of movement verification and the second verification result of the current round of movement verification are read together, along with the initial parameter adjustment step size bound to the current initial calibration parameters, as input for subsequent calculation of error trends and dynamic adjustment step sizes. The effect of this approach is to transform closed-loop calibration from a one-time correction into a traceable iterative process. Combined with the calibration model for visual equivalent and skew angle compensation in the scheme, it is more conducive to maintaining stable convergence of the update process under noise, illumination fluctuations, or feature extraction jitter.

[0175] The error reduction rate is calculated based on the first verification result and the second verification result;

[0176] Specifically, calculating the error reduction rate essentially involves using the results of two consecutive rounds of verification to characterize whether the error is decreasing and at what rate, thus determining whether the current parameter correction direction and intensity are appropriate. In implementation, the first and second verification results can be unified into an error index of the same dimension, such as the magnitude of the error vector or the combined component error value along the pixel coordinate axis. Then, the difference between the two is compared with the first verification result to obtain the reduction rate, making it comparable across different initial error scales. If there are concerns about occasional fluctuations in a single verification, the reduction rate can be calculated by averaging or medianing multiple sampling results within each round of verification. This aligns with the approach in the scheme of improving the accuracy of the skew angle calculation by averaging multiple samples. By explicitly quantifying the error change trend, the system not only knows whether the error is large or small, but also whether it is decreasing as expected, oscillating, or even rebounding. This provides a basis for the next adaptive step size update, avoiding blindly increasing the correction and causing the skew angle or visual equivalent to oscillate back and forth during iteration.

[0177] The error decrease rate is compared with a preset decrease rate threshold, and the initial parameter adjustment step size is updated based on the comparison result to obtain the parameter update step size;

[0178] Specifically, comparing the error reduction rate with a preset reduction rate threshold and updating the initial parameters to adjust the step size addresses the contradiction between oscillation and convergence speed, a common problem in iterative updates. In implementation, a reduction rate threshold is first set as a benchmark for the desired convergence speed. When the reduction rate is too small, indicating slow or stagnant error reduction, the system can reduce the step size to obtain finer search granularity, avoiding crossing the optimal point near the error trough due to an excessively large step size. When the reduction rate is too large, indicating an effective update direction and rapid error reduction, the system can appropriately increase the step size to reduce the number of iterations and improve calibration efficiency. Step size updates can employ proportional scaling, segmented mapping, or upper / lower limit pruning to ensure the step size always remains within a controllable range, preventing parameter mutations that could distort the theoretical pixel displacement prediction in the next round. This adaptive step size mechanism, combined with the fully automatic calibration and closed-loop verification path emphasized in the scheme, allows parameter adjustments to quickly approach the region meeting accuracy requirements while maintaining stable refinement as they approach the target, improving the repeatability and reliability of the final target calibration parameters.

[0179] The initial calibration parameters are corrected based on the updated step size of the parameters to obtain the updated calibration parameters;

[0180] Specifically, by using the step size as a constraint, the correction action is transformed into a controllable incremental update, rather than a one-time replacement. The initial calibration parameters typically include the target rotation angle, the first visual equivalent, and the second visual equivalent. The aforementioned scheme has already determined the parameter update step size using the error reduction rate. This step size is applied to the corresponding parameter correction amount. For example, while maintaining the existing calibration model structure, a small-angle incremental correction is made to the rotation angle, and a small-proportional incremental correction is made to the visual equivalent, ensuring that the corrected parameters still meet the continuity requirements of the coordinate transformation model. First, the correction direction is determined based on the error direction of the second verification result. Then, the parameter update step size is used to limit the correction magnitude, and a new set of calibration parameters is formed after correction and written to memory or the calibration file cache. This incremental correction avoids parameter jumps caused by error fluctuations and allows subsequent verification to accurately reflect the improvement effect brought about by this small correction, thus supporting the gradual convergence of closed-loop calibration.

[0181] Based on the updated calibration parameters, the motion platform is controlled to perform the next round of movement verification over a preset distance to obtain a third verification result;

[0182] Specifically, based on the updated calibration parameters, the motion platform is controlled to perform the next round of movement verification at a preset distance and obtain a third verification result. Essentially, the new parameters are immediately incorporated into the same verification mechanism for self-consistency testing, ensuring the adjustment direction is correct and genuinely reduces error. In implementation, the system converts the preset distance into theoretical pixel displacement based on the updated first and second visual equivalents, then controls the motion platform to move at the same preset distance. The image acquisition module acquires images of the calibration board before and after the movement, and the image processing module calculates the actual pixel displacement and subtracts it from the theoretical pixel displacement to obtain the third verification result. To enhance repeatability, the preset distance is usually kept consistent with the previous round, or multiple preset distances are repeatedly verified within the same round, and a statistical value is taken to reduce the impact of feature recognition noise on a single verification. This approach tightly binds parameter updates and effect verification within the same closed loop. The third verification result directly reflects whether the incremental correction in this round is effective, avoiding updates only at the algorithm level without closed-loop evidence from physical motion and image measurement.

[0183] If the third verification result is less than or equal to the error threshold, the iteration ends and the updated calibration parameter is determined as the target calibration parameter;

[0184] Specifically, the iteration ends when the third verification result is less than or equal to the error threshold, and the updated calibration parameters are determined as the target calibration parameters. This step is the implementation of the convergence criterion, used to switch the closed-loop calibration from the iterative state to a stable state with usable output parameters. The error threshold corresponds to the preset calibration accuracy condition. The third verification result reaching this threshold means that, under the motion verification at the current preset distance, the deviation between the theoretical pixel displacement and the actual pixel displacement has been compressed to an acceptable range, so there is no need to continue changing the rotation angle or visual equivalent. In implementation, the system will permanently save the calibration parameters updated in the current round, for example, by writing them to a calibration file and marking the version number, while recording the last verification result and key intermediate quantities for traceability. This is consistent with the idea in the scheme that the calibration process should be recordable and traceable. Through a clear stopping condition, meaningless over-iteration can be avoided, calibration efficiency can be improved, and the output parameters can be ensured to meet the accuracy requirements and have repeatability.

[0185] If the third verification result is greater than the error threshold, then the third verification result is taken as the new second verification result, and the second verification result is taken as the new first verification result. The process returns to the step of calculating the error reduction rate based on the first verification result and the second verification result, until the third verification result is less than or equal to the error threshold. Then the iteration ends and the updated calibration parameter is determined as the target calibration parameter.

[0186] Specifically, when the third verification result is still greater than the error threshold, a sliding window-like trend judgment mechanism is constructed by using the third verification result as the new second verification result, the original second verification result as the new first verification result, and returning to the error reduction rate calculation step. This ensures that the step size update is always dynamically adjusted based on the results of the two most recent verification rounds. In implementation, the system updates internal cache variables or registers, moves the latest error index forward and saves it, then recalculates the new error reduction rate and updates the parameter update step size for the next round, thus performing the next incremental correction and verification. Because each round uses the two most recent verification results to judge the error trend, when the error reduction slows down or rebounds, the step size is automatically compressed to suppress oscillations; when the error continues to decrease rapidly, the step size is amplified to accelerate convergence, thus forming an adaptive iterative adjustment loop. This loop continues until the third verification result meets the error threshold, giving the closed-loop calibration process stable convergence and noise resistance. It is particularly suitable for reliably obtaining target calibration parameters even when there are light fluctuations, platform micro-vibrations, or feature recognition jitter in the packaging equipment environment.

[0187] Example 2

[0188] In addition, combined Figure 1 The image processing-based visual calibration method for die bonder integrated circuits described in this embodiment of the invention can be implemented by an image processing-based visual calibration system for die bonder integrated circuits. Figure 3 A schematic diagram of the hardware structure of the image processing-based die bonder integrated circuit vision calibration system provided in an embodiment of the present invention is shown.

[0189] Image processing-based die bonder integrated circuit vision calibration system may include a processor and a memory storing computer program instructions.

[0190] Specifically, the processor may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement embodiments of the present invention.

[0191] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, like read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0192] Computer-readable media include both permanent and non-permanent, removable and non-removable media, which can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient media, such as modulated communication signals and carrier waves.

[0193] The processor reads and executes computer program instructions stored in the memory to implement any of the image processing-based integrated circuit visual calibration methods in the above embodiments.

[0194] In one example, the image processing-based die bonder integrated circuit vision calibration system may also include a communication interface and a bus. For example, Figure 3 As shown, the processor 401, memory 402, and communication interface 403 are connected through bus 410 and complete communication with each other.

[0195] The communication interface is mainly used to enable communication between various modules, devices, units and / or equipment in the embodiments of the present invention.

[0196] A bus, including hardware, software, or both, couples components of an image processing-based die-bonded integrated circuit vision calibration system together. For example, and not limitingly, a bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, a bus may include one or more buses. While specific buses are described and illustrated in embodiments of the invention, the invention contemplates any suitable bus or interconnect.

[0197] In summary, the embodiments of the present invention provide a visual calibration method and system for die bonder integrated circuits based on image processing.

[0198] It should be clarified that the present invention is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of the present invention.

[0199] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0200] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0201] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0202] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0203] It should also be noted that the exemplary embodiments mentioned in this invention describe methods or systems based on a series of steps or apparatus. However, this invention is not limited to the order of the steps described above; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0204] The above description is merely a specific embodiment of the present invention. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the protection scope of the present invention.

Claims

1. A visual calibration method for die bonder integrated circuits based on image processing, characterized in that, The method is applied to a die bonder, which includes: an image processing module, a calibration board, a motion platform, a motion control module, and a calibration module. The calibration board is disposed on the motion platform, and the motion control module controls the motion platform. The calibration plate is fixed at a preset position on the motion platform. The motion control module controls the motion platform to move along a first preset direction and a second preset direction respectively. Pulse displacement data of each movement is obtained. The image processing module obtains pixel position change data before and after the movement. The first preset direction and the second preset direction are perpendicular to each other. The calibration module calculates calibration parameters for the pulse displacement data and the pixel position change data to obtain initial calibration parameters. The initial calibration parameters include a first visual equivalent corresponding to the first preset direction, a second visual equivalent corresponding to the second preset direction, and a target rotation angle between the camera coordinate system and the motion platform coordinate system. Based on the initial calibration parameters, the motion platform is controlled to perform a preset distance of movement verification to obtain the verification result; Based on the verification results, the initial calibration parameters are adjusted to obtain target calibration parameters that meet the preset calibration accuracy conditions, so as to achieve integrated circuit vision closed-loop calibration.

2. The image processing-based visual calibration method for integrated circuits in a die bonder according to claim 1, characterized in that, The calibration plate surface is provided with calibration features, including alignment segments for adjusting the camera axis, a square array for visual equivalent calibration, a dot array for extracting rotation information, and positioning marks for positioning the calibration plate.

3. The image processing-based visual calibration method for integrated circuits in a die bonder according to claim 1, characterized in that, The die bonder also includes an image acquisition module. The process of fixing the calibration plate to a preset position on the motion platform, controlling the motion platform to move along a first preset direction and a second preset direction via the motion control module, acquiring pulse displacement data for each movement, and acquiring pixel position change data before and after the movement via the image processing module includes: The image acquisition module acquires the initial calibration plate image at the preset position and the target calibration plate image after the motion platform has moved. The image processing module extracts pixel positions from the initial calibration board image and the target calibration board image to obtain the initial pixel position and the target pixel position. The difference between the initial pixel position and the target pixel position is calculated to obtain the pixel position change data; The motion control module acquires the pulse displacement data for each movement of the motion platform.

4. The image processing-based visual calibration method for integrated circuits in a die bonder according to claim 3, characterized in that, The step of extracting pixel positions from the initial calibration board image and the target calibration board image using the image processing module to obtain the initial pixel positions and target pixel positions includes: The initial calibration board image and the target calibration board image are respectively input into a pre-trained calibration feature recognition model to obtain the first calibration feature in the initial calibration board image and the second calibration feature in the target calibration board image; Contour features are extracted from the first calibration feature and the second calibration feature respectively to obtain the first contour position information of the first calibration feature and the second contour position information of the second calibration feature; Based on the first contour position information and the second contour position information, the geometric center positions of the first calibration feature and the second calibration feature are calculated respectively to obtain the first pixel coordinate position and the second pixel coordinate position; The first pixel coordinate position is used as the initial pixel position, and the second pixel coordinate position is used as the target pixel position.

5. The image processing-based visual calibration method for integrated circuits in a die bonder according to claim 1, characterized in that, The initial calibration parameters obtained by calculating calibration parameters for the pulse displacement data and the pixel position change data through the calibration module include: Based on the pixel position change data, obtain the slope of the pixel displacement vector before and after the motion platform moves along the first preset direction multiple times; Based on the coordinate axes of the preset pixel coordinate system, obtain the quadrant assignment of each pixel displacement vector in the pixel coordinate system; Based on the motion direction of the first preset direction in the preset motion platform coordinate system and the correspondence between the first preset direction and the coordinate axes of the pixel coordinate system, the directional consistency constraint between the pixel displacement vector direction and the first preset direction is determined. Based on the quadrant assignment and the direction consistency constraint, the range and sign of the angle values ​​corresponding to the slope of each pixel displacement vector are constrained. Under the constraints of the angle value range and sign, the target rotation angle is calculated based on the slope of each pixel displacement vector; Based on the pixel position change data and the pulse displacement data, the first visual equivalent and the second visual equivalent are calculated by least squares fitting. The initial calibration parameters are determined based on the target rotation angle, the first visual equivalent, and the second visual equivalent.

6. The image processing-based visual calibration method for integrated circuits in a die bonder according to claim 5, characterized in that, Under the constraints of the angle value range and sign, the target rotation angle is calculated based on the slope of each pixel displacement vector, including: Based on the slope of each pixel displacement vector and the preset mapping relationship between the slope and the rotation angle, each initial rotation angle is calculated. The average value of each initial rotation angle is calculated, and the average value of the calculated rotation angles is taken as the target rotation angle.

7. The image processing-based visual calibration method for integrated circuits in a die bonder according to claim 5, characterized in that, The step of calculating the first visual equivalent and the second visual equivalent by least squares fitting based on the pixel position change data and the pulse displacement data includes: Based on the pixel position change data, obtain the displacement of each first pixel corresponding to the first preset direction and the displacement of each second pixel corresponding to the second preset direction; Based on the pulse displacement data, obtain the first pulse displacement amount corresponding to the first preset direction and the second pulse displacement amount corresponding to the second preset direction; The least squares method is used to linearly fit the displacement of each first pixel and the corresponding displacement of each first pulse, and the slope of the fitted first straight line is used as the first visual equivalent. The least squares method is used to linearly fit each second pixel displacement and the corresponding second pulse displacement, and the slope of the fitted second straight line is used as the second visual equivalent.

8. The image processing-based visual calibration method for integrated circuits in a die bonder according to claim 1, characterized in that, The step of controlling the motion platform to perform a preset distance movement verification based on the initial calibration parameters, and obtaining the verification result includes: Based on the first visual equivalent and the second visual equivalent, the preset distance is converted into the corresponding theoretical pixel displacement. The motion platform is controlled to perform a preset distance of movement verification, and calibration plate images are acquired before and after the movement; Pixel displacement is calculated on the calibration board images before and after the movement to obtain the actual pixel displacement. The difference between the actual pixel displacement and the theoretical pixel displacement is calculated to obtain the verification result.

9. The image processing-based visual calibration method for die bonder integrated circuits according to any one of claims 1-8, characterized in that, The step of adjusting the initial calibration parameters based on the verification results to obtain target calibration parameters that meet the preset calibration accuracy conditions, in order to achieve integrated circuit vision closed-loop calibration, includes: Based on the preset calibration accuracy conditions, the error threshold is obtained; The verification result is compared with the error threshold. If the verification result is less than or equal to the error threshold, the initial calibration parameter is used as the target calibration parameter. If the verification result is greater than the error threshold, the initial calibration parameters are adjusted according to the verification result to obtain the target calibration parameters.

10. A vision calibration system for integrated circuits on a die bonder based on image processing, characterized in that, include: At least one processor, at least one memory, and computer program instructions stored in the memory, which, when executed by the processor, implement the method as described in any one of claims 1-9.

Images