Method for high-precision automatic calibration of three points and one line based on image processing

Abstract

The invention relates to the technical field of image processing, and discloses a high-precision automatic three-point-one-line calibration method based on image processing, and the method effectively improves the precision and robustness of image processing through the steps of refined image processing and data extraction. Especially under the influence of uncertain factors such as rotation angle change and image noise, accurate extraction and matching of feature points can be ensured. Compared with a traditional method, the geometric information in the image is captured more comprehensively and accurately through multi-dimensional data extraction and dimensionless processing, and then more reliable basic data is provided for subsequent angle correction, image alignment and three-point one-line calibration. The improvement not only improves the calibration precision of the system, but also significantly enhances the adaptability of the system in a complex environment, so that the system can process more types of images and cope with errors caused by different angles, thereby greatly improving the stability and flexibility of the calibration process.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a high-precision automatic calibration method of three points in a line based on image processing. Background Art

[0002] With the rapid development of computer vision and artificial intelligence technologies, image processing has been widely used in a variety of fields, including industrial automation, precision measurement, and robotic navigation. Image calibration, as a key technology for accurately matching image spatial information with actual physical space, has become an indispensable foundation for high-precision visual positioning systems. In image calibration technology, the "three-point-one-line" automatic calibration method, a typical geometric feature calibration mechanism, can effectively improve the system's posture recognition accuracy under rotational conditions. It is widely used in scenarios such as industrial measurement equipment, initial positioning of robotic arms, and visual navigation.

[0003] Traditional three-point-in-a-line image calibration methods typically rely on manual point selection or geometric registration algorithms based on simple corner detection. These methods suffer from numerous issues, including low feature extraction accuracy, inaccurate angle correction, and image registration sensitivity to noise. Calibration errors increase significantly when the image is rotated or the lighting changes, leading to feature point mismatches, angle estimation bias, and inaccurate final midpoint determination, reducing the stability and practicality of the entire system.

[0004] Therefore, we propose a high-precision automatic calibration method of three points in a line based on image processing to solve the above problems. Summary of the Invention

[0005] The present invention aims to provide a high-precision automatic three-point-one-line calibration method based on image processing, addressing the aforementioned background art issues. Conventional three-point-one-line image calibration methods typically rely on manual point selection or geometric registration algorithms based on simple corner detection, resulting in numerous problems such as low feature extraction accuracy, inaccurate angle correction, and image registration sensitivity to noise. In particular, calibration errors increase significantly when the image is rotated or the illumination changes, easily leading to feature point mismatches, angle estimation bias, and inaccurate final midpoint determination, thereby reducing the stability and practicality of the entire system.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for high-precision automatic calibration of three points and one line based on image processing, the specific steps of which are as follows: S1, image acquisition and processing, data acquisition and processing of the first positioning image and the rotated image; S2, corner detection, by analyzing the image data to generate corner feature coefficients; S3, matching corresponding points, analyzing corner feature coefficients, and identifying corner differences; S4, correcting the angle difference, using the matched feature points to perform geometric transformation, calculate and correct the rotation angle difference between the two images; S5. Determine the effective midpoint: After the two images are corrected, determine the effective midpoint between the two images based on geometric principles; S6. Feedback: Feedback the collected data and analysis results to the terminal.

[0007] Preferably, in step S1, the specific steps of image acquisition and processing are as follows: S1.1. Set the rotation angle, take a picture at the starting angle and the set rotation angle, perform denoising on the images, and mark them as the initial image and the rotated image respectively; S1.2. Extract data from the initial image and the rotated image, including corner response values, local texture contrast, local grayscale change, principal curvature ratio, and local gradient energy of the image; S1.3, preprocessing and dimensionlessizing the extracted data, and reorganizing them into an initial image data set and a rotated image data set; The initial image data set includes initial corner response value YA, initial local texture contrast YB, initial local grayscale change YC, initial principal curvature ratio YD and initial image local gradient energy YE; The rotation image data set includes the rotation corner response value RA, the rotation local texture contrast RB, the rotation local grayscale change RC, the rotation principal curvature ratio RD and the rotation image local gradient energy RE.

[0008] Preferably, in step S2, the specific steps of corner point detection are as follows: S2.1. Data coupling is performed on the initial image data set. Multiple data are input into a pre-trained deep learning framework. Feature fusion is performed through a multi-layer neural network to calculate the standard coefficient CSJ of the initial image corner points. The specific calculation formula is as follows;

[0009] Where: YA is the initial corner response value, YB is the initial local texture contrast, YC is the initial local grayscale change, YD is the initial principal curvature ratio, and YE is the initial image local gradient energy; S2.2. Data coupling is performed on the rotated image data set. Multiple data are input into a pre-trained deep learning framework. Feature fusion is performed through a multi-layer neural network to calculate the standard coefficient RSJ of the rotated image corner point. The specific calculation formula is as follows;

[0010] Where RA is the rotation corner response value, RB is the rotation local texture contrast, RC is the rotation local grayscale change, RD is the rotation principal curvature ratio, and RE is the rotation image local gradient energy.

[0011] Preferably, in step S3, the specific steps of matching corresponding points are as follows: S3.1. Analyze the data of the standard coefficients CSJ ​​of the corner points of the initial image and RSJ of the corner points of the rotated image, and determine whether auxiliary processing of the image is required based on the analysis results. S3.2. Perform an integrated analysis of the corner point standard coefficients CSJ ​​of the initial image and RSJ of the rotated image after auxiliary processing. Based on the analysis results, determine whether there is a significant angular difference between the two images. The specific steps are as follows: when When , it means that there is no obvious angle difference in the current image and no correction is needed; when , it means that the current image has obvious angle difference and needs to be corrected.

[0012] Preferably, in step S3.1, the analysis method of the initial image corner point standard coefficient CSJ and the rotated image corner point standard coefficient RSJ is as follows: when or When , it means that the initial image or the rotated image needs to be enhanced; when or , it means that the original image or the rotated image does not need to be enhanced.

[0013] Preferably, in step S3.1, the image enhancement method is as follows: S3.11. Perform secondary data collection on the initial image and the rotated image, including image contrast, image clarity, noise level, edge density, and brightness level. Preprocess and dimensionlessly convert the extracted data into an enhanced initial image data set and an enhanced rotated image data set. The initial image enhancement data set includes initial image contrast YF, initial image clarity YG, initial noise level YH, initial edge density YI and initial brightness level YJ; The rotation image enhancement data set includes the rotation image contrast RF, the rotation image sharpness RG, the rotation noise level RH, the rotation edge density RI and the rotation brightness level RJ; S3.12. Data coupling is performed on the initial image enhancement data set and the rotated image enhancement data set. Multiple data are input into the pre-trained deep learning framework. Feature fusion is performed through a multi-layer neural network to calculate the initial image enhancement factor YZ and the rotated image enhancement factor RZ. The specific calculation formulas are as follows:

[0014]

[0015] Where: YF is the initial image contrast, YG is the initial image clarity, YH is the initial noise level, YI is the initial edge density, YJ is the initial brightness level, RF is the rotated image contrast, RG is the rotated image clarity, RH is the rotated noise level, RI is the rotated edge density, and RJ is the rotated brightness level; S3.13. Input the calculated initial image enhancement factor YZ and the rotation image enhancement factor RZ into the calculation formulas of the initial image corner point standard coefficient CSJ and the rotation image corner point standard coefficient RSJ for enhancement. The specific calculation formulas are as follows:

[0016]

[0017] Where: YA is the initial corner response value, YB is the initial local texture contrast, YC is the initial local grayscale change, YD is the initial principal curvature ratio, YE is the initial image local gradient energy, RA is the rotated corner response value, RB is the rotated local texture contrast, RC is the rotated local grayscale change, RD is the rotated principal curvature ratio, RE is the rotated image local gradient energy, YZ is the initial image enhancement factor, and RZ is the rotated image enhancement factor.

[0018] Preferably, in step S4, the specific steps of correcting the angle difference are as follows: S4.1. Determine the angle difference. Create a mirror image of the initial image at a standard rotation angle to generate a mirror image. Perform feature selection on the initial image, the mirror image, and the rotated image. Compare the feature coordinates of the mirror image with those of the rotated image. Obtain the angle deviation value JDC using the formula |mirror image feature coordinates - rotated image feature coordinates|. S4.2. Construct a rotation matrix. Use the obtained angle deviation value JDC to generate the rotation matrix using the following formula. Use OpenCV to perform scheduling adjustments to generate the actual rotated image.

[0019] S4.3. Crop the image and perform edge and scaling on the actual rotated image to ensure consistency with the initial image.

[0020] Preferably, in step S5, the specific steps of determining the effective midpoint are as follows: S5.1. Perform image registration on the initial image and the actual rotated image to establish a geometric mapping relationship between the two. After image registration is completed, extract the region of the initial image that is spatially aligned with the actual rotated image based on the geometric mapping relationship, thereby obtaining candidate center points corresponding to the overlapping regions of the images as a valid center point candidate set. S5.2. Verify the determined set of valid center point candidates. The verification process is based on a matching evaluation mechanism for the overlapping area of ​​image content. By calculating the content consistency index of the candidate center points in the overlapping area of ​​the image, the candidate center points that meet the preset similarity threshold conditions are screened out as the final valid center points.

[0021] Compared with the prior art, the present invention has the following beneficial effects: 1. Through refined image processing and data extraction steps, the present invention effectively improves the accuracy and robustness of image processing, especially under the influence of uncertain factors such as rotation angle changes and image noise, and can ensure the accurate extraction and matching of feature points. Compared with traditional methods, the present invention captures the geometric information in the image more comprehensively and accurately through multi-dimensional data extraction and dimensionless processing, thereby providing more reliable basic data for subsequent angle correction, image alignment, and three-point-one-line calibration. This improvement not only improves the calibration accuracy of the system, but also significantly enhances its adaptability in complex environments, enabling the system to process a wider variety of images and deal with errors caused by different angles, thereby greatly improving the stability and flexibility of the calibration process.

[0022] 2. By comparing the initial image corner standard coefficients CSJ ​​with the rotated image corner standard coefficients RSJ, the system can accurately determine the angular difference between the images. This analysis process helps the system understand the alignment deviation between the images and determine whether rotation angle correction is necessary. Based on the analysis results of the corner standard coefficients, the system can adaptively determine whether auxiliary image processing is required to ensure more accurate corner matching and is not affected by noise or image blur. Based on preset judgment conditions, the system can efficiently determine whether the image needs angle correction. This not only avoids unnecessary calculations and improves the overall performance of the system, but also ensures accuracy during the calibration process. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A diagram showing the steps of the method of the present invention. DETAILED DESCRIPTION

[0024] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0025] Example 1: Please refer to Figure 1 , a high-precision automatic calibration method for three points and one line based on image processing, the specific steps are as follows: S1, image acquisition and processing, data acquisition and processing of the first positioning image and the rotated image; S2, corner detection, by analyzing the image data to generate corner feature coefficients; S3, matching corresponding points, analyzing corner feature coefficients, and identifying corner differences; S4, correcting the angle difference, using the matched feature points to perform geometric transformation, calculate and correct the rotation angle difference between the two images; S5. Determine the effective midpoint: After the two images are corrected, determine the effective midpoint between the two images based on geometric principles; S6. Feedback: Feedback the collected data and analysis results to the terminal.

[0026] In this embodiment, step S1 captures and processes the initial positioning image and the rotated image, laying the foundation for subsequent image analysis and feature extraction. Image processing includes routine operations such as denoising, grayscale conversion, and contrast enhancement, aiming to improve the accuracy of subsequent corner detection. This step successfully achieves precise conversion from physical space to digital imagery, providing the system with high-quality raw data input and ensuring the stability and effectiveness of the subsequent image analysis process.

[0027] In step S2, an algorithm analyzes the processed image, identifying and extracting corner features within the image and generating corner feature coefficients. Corner features typically possess high stability and recognition, becoming crucial for subsequent image matching and rotation correction. Completion of this step signifies that the system can accurately identify the geometric structure within the image, providing a mathematical basis for precise alignment and improving positioning consistency and repeatability.

[0028] In step S3, by comparing the corner point feature coefficients extracted from the two images, the system automatically identifies the positional differences between the same physical point in the different images and thus matches corresponding point pairs. This step significantly improves the efficiency and accuracy of feature point matching, providing high-quality reference coordinates for subsequent geometric transformations. Accurately matched corner point pairs lay the foundation for subsequent angle correction and spatial position calibration.

[0029] In step S4, the matched corner point pairs are used to calculate and correct the rotation angle difference between the two images using geometric transformation algorithms such as affine or perspective transformations. This step effectively addresses visual deviations caused by shooting angle, device errors, or object rotation, ensuring consistent image alignment. Accurately correcting the angle difference significantly improves calibration accuracy and provides more reliable data support for subsequent midpoint determination.

[0030] Through refined image processing and data extraction steps, the present invention effectively improves the accuracy and robustness of image processing, especially under the influence of uncertain factors such as rotation angle changes and image noise, and can ensure the accurate extraction and matching of feature points. Compared with traditional methods, the present invention captures the geometric information in the image more comprehensively and accurately through multi-dimensional data extraction and dimensionless processing, thereby providing more reliable basic data for subsequent angle correction, image alignment, and three-point-one-line calibration. This improvement not only improves the calibration accuracy of the system, but also significantly enhances its adaptability in complex environments, enabling the system to process a wider variety of images and deal with errors caused by different angles, thereby greatly improving the stability and flexibility of the calibration process.

[0031] In step S5, after image rotation correction is complete, the effective midpoint between the alignment points in the two images is calculated based on the geometric positional relationship of the matching corner points, following geometric principles. This midpoint is a key element in the construction of the three-point-one-line structure and directly affects the accuracy of the baseline. By accurately calculating the midpoint position, the system further achieves precise positioning of the three-point-one-line structure, improving the robustness and accuracy of the overall calibration system.

[0032] In step S6, data including image acquisition, corner analysis, matching results, angle correction, and midpoint determination are fed back to the terminal in real time, allowing users to monitor results, correct errors, and perform subsequent optimization. This step not only enhances the interactivity and visualization of the system but also improves the traceability of the calibration process, ensuring operational accuracy and quality control in industrial applications.

[0033] Compared to traditional three-point-one-line calibration methods, the image processing-based automatic calibration method of this invention achieves significant improvements in accuracy, efficiency, and automation. By introducing corner detection and geometric correction algorithms, the system effectively mitigates the impact of human and equipment errors on calibration results, ensuring a highly efficient and stable calibration process. The automated feature point matching and data feedback mechanism not only increases operation speed but also significantly enhances the system's fault tolerance and adaptability.

[0034] Example 2: Please refer to Figure 1 In step S1, the specific steps of image acquisition and processing are as follows: S1.1. Set the rotation angle, take a picture at the starting angle and the set rotation angle, perform denoising on the images, and mark them as the initial image and the rotated image respectively; S1.2. Extract data from the initial image and the rotated image, including corner response values, local texture contrast, local grayscale change, principal curvature ratio, and local gradient energy of the image; S1.3, preprocessing and dimensionlessizing the extracted data, and reorganizing them into an initial image data set and a rotated image data set; The initial image data set includes initial corner response value YA, initial local texture contrast YB, initial local grayscale change YC, initial principal curvature ratio YD and initial image local gradient energy YE; The rotation image data set includes the rotation corner response value RA, the rotation local texture contrast RB, the rotation local grayscale change RC, the rotation principal curvature ratio RD and the rotation image local gradient energy RE.

[0035] In this embodiment, in step S1.1, by setting the rotation angle and capturing images at both the starting angle and the rotation angle, data from different perspectives is provided for subsequent image processing. This approach effectively avoids image distortion or errors that may arise from a single angle, improving the accuracy of angular difference detection. Furthermore, multi-angle data acquisition increases the richness of the image, providing more dimensional support for feature point extraction and matching, and enhancing the reliability of subsequent geometric correction.

[0036] Step S1.2 enhances the ability to describe the geometric features of the image by extracting detailed multidimensional data from the initial and rotated images, including corner response values, local texture contrast, local grayscale variation, principal curvature ratio, and local image gradient energy. This data not only captures the spatial and structural characteristics of the image but also provides rich reference information for subsequent corner matching and geometric transformation. Compared to traditional methods that rely solely on simple feature points or image content, this method extracts data from multiple dimensions, improving the distinguishability of feature points and the stability of matching.

[0037] Step S1.3 introduces data preprocessing and dimensionless transformation techniques to eliminate noise from the image data, unify the scale, and improve data stability. This allows data extracted from different angles to be compared using the same standards, avoiding errors caused by varying data dimensions. This technique not only improves data consistency and contrast but also lays a more solid foundation for subsequent corner point analysis and matching.

[0038] The feature data for the initial and rotated images are organized into "initial image data sets" and "rotated image data sets," respectively, including corresponding data for each feature value, such as corner response values, local texture contrast, and local grayscale variations. This organized data division and management allows the system to more accurately locate each feature, thereby improving the accuracy of subsequent feature point matching. Unlike traditional methods that may involve data clutter or redundancy, the clear division of data sets ensures efficient and reliable feature extraction and processing.

[0039] Example 3: Please refer to Figure 1 In step S2, the specific steps of corner point detection are as follows: S2.1. Data coupling is performed on the initial image data set. Multiple data are input into a pre-trained deep learning framework. Feature fusion is performed through a multi-layer neural network to calculate the standard coefficient CSJ of the initial image corner points. The specific calculation formula is as follows;

[0040] Where: YA is the initial corner response value, YB is the initial local texture contrast, YC is the initial local grayscale change, YD is the initial principal curvature ratio, and YE is the initial image local gradient energy; S2.2. Data coupling is performed on the rotated image data set. Multiple data are input into a pre-trained deep learning framework. Feature fusion is performed through a multi-layer neural network to calculate the standard coefficient RSJ of the rotated image corner point. The specific calculation formula is as follows;

[0041] Where RA is the rotation corner response value, RB is the rotation local texture contrast, RC is the rotation local grayscale change, RD is the rotation principal curvature ratio, and RE is the rotation image local gradient energy.

[0042] In this embodiment, in S2.1 and S2.2, data coupling is performed on the initial image dataset and the rotated image dataset, respectively. This involves integrating multiple image feature data, such as corner response values, local texture contrast, local grayscale variation, principal curvature ratio, and local image gradient energy, into a unified data input. This data, as multidimensional features, is input into a pre-trained deep learning framework, which uses its multi-layer neural network to comprehensively analyze and process these features.

[0043] The deep learning framework uses a multi-layer neural network to fuse features. This process aims to extract deep connections between various data points and improve feature recognition. After obtaining the fused features, the system uses a specific formula to calculate the corner standard coefficients of the original image and the rotated image. These coefficients are called the original image corner standard coefficient CSJ and the rotated image corner standard coefficient RSJ.

[0044] Through deep neural network processing, the framework is able to fuse different image features and use a multi-layer structure to extract nonlinear features of the image. Each layer of the neural network gradually strengthens the judgment of the corner area and optimizes the final corner standard coefficients CSJ ​​and RSJ, so that corner detection can not only effectively deal with complex image noise and rotation transformations, but also extract high-precision corner features in complex backgrounds. The process of calculating the corner standard coefficients involves the influence of the image rotation angle. Specifically, the formula captures the local texture contrast changes of the image at different angles through the operation of the cosine function, and then reflects the relationship between the corner response and the rotation angle. In the initial image and the rotated image, the relationship between each corner and the image rotation angle can be obtained through calculation. This provides effective data support for subsequent image matching and angle correction, ensuring the accurate alignment of corners under rotation transformation.

[0045] Example 4: Please refer to Figure 1 In step S3, the specific steps of matching corresponding points are as follows: S3.1. Analyze the data of the standard coefficients CSJ ​​of the corner points of the initial image and RSJ of the corner points of the rotated image, and determine whether auxiliary processing of the image is required based on the analysis results. S3.2. Perform an integrated analysis of the corner point standard coefficients CSJ ​​of the initial image and RSJ of the rotated image after auxiliary processing. Based on the analysis results, determine whether there is a significant angular difference between the two images. The specific steps are as follows: when When , it means that there is no obvious angle difference in the current image and no correction is needed; when , it means that the current image has obvious angle difference and needs to be corrected.

[0046] In this embodiment, the initial image corner standard coefficients CSJ ​​and the rotated image corner standard coefficients RSJ obtained in step S2 are analyzed separately. During the analysis, the system evaluates the characteristics and consistency of the two sets of data, such as whether there are significant differences in corner response, local texture contrast, and local grayscale variation.

[0047] By analyzing the difference between CSJ and RSJ, the system determines whether auxiliary processing is needed for the image. The key to this step is to analyze whether the difference in image features exceeds the preset threshold. If the difference is too large, it indicates that the image may be interfered with by factors such as noise, blur, and lighting changes. The system may initiate auxiliary processing such as denoising and image enhancement. By comparing the standard corner coefficients CSJ ​​of the initial image with the standard corner coefficients RSJ of the rotated image, the system can accurately determine the angular difference between the images. This analysis process helps the system understand the alignment deviation between the images and determine whether rotation angle correction is necessary. Based on the analysis results of the standard corner coefficients, the system can adaptively determine whether auxiliary image processing is required to ensure more accurate corner matching and is not affected by noise or image blur. Based on preset judgment conditions, the system can efficiently determine whether the image needs angle correction. This not only avoids unnecessary calculations and improves the overall performance of the system, but also ensures accuracy during the calibration process.

[0048] Example 5: Please refer to Figure 1 In step S3.1, the analysis method of the initial image corner point standard coefficient CSJ and the rotated image corner point standard coefficient RSJ is as follows: when or When , it means that the initial image or the rotated image needs to be enhanced; when or , it means that the original image or the rotated image does not need to be enhanced.

[0049] In this embodiment, in step S3.1, the system determines whether to perform image enhancement by comparing the values ​​of the initial image corner standard coefficients CSJ ​​and the rotated image corner standard coefficients RSJ. The core of this process is to assess image quality using corner feature strength, ensuring good image recognizability and stability before proceeding to the subsequent angular difference analysis and matching.

[0050] Only when the image quality reaches the preset standard, that is, the initial image corner standard coefficient CSJ and the rotated image corner standard coefficient RSJ are both greater than or equal to 1, the rotation features in the image, such as local texture contrast, image local gradient energy, etc., have higher stability and reliability, and the angle information calculated based on such features, such as the cosine direction component, is more accurate. If the image quality does not meet the standard and angle difference analysis is performed directly, the angle judgment result may be distorted due to the low CSJ and RSJ values, thereby affecting the overall accuracy of image matching and calibration. By pre-judging the image quality through this step and performing image enhancement processing as needed, the expression ability of corner features can be effectively improved, ensuring that the extracted rotation features can truly reflect the geometric structure of the image, especially in complex scenes such as large-angle rotation or local distortion, it can still maintain a high matching accuracy, thereby significantly reducing the risk of angle misjudgment and enhancing the system's adaptability and robustness to image rotation changes.

[0051] Example 6: Please refer to Figure 1In step S3.1, the image enhancement method is as follows: S3.11. Perform secondary data collection on the initial image and the rotated image, including image contrast, image clarity, noise level, edge density, and brightness level. Preprocess and dimensionlessly convert the extracted data into an enhanced initial image data set and an enhanced rotated image data set. The initial image enhancement data set includes initial image contrast YF, initial image clarity YG, initial noise level YH, initial edge density YI and initial brightness level YJ; The rotation image enhancement data set includes the rotation image contrast RF, the rotation image sharpness RG, the rotation noise level RH, the rotation edge density RI and the rotation brightness level RJ; S3.12. Data coupling is performed on the initial image enhancement data set and the rotated image enhancement data set. Multiple data are input into the pre-trained deep learning framework. Feature fusion is performed through a multi-layer neural network to calculate the initial image enhancement factor YZ and the rotated image enhancement factor RZ. The specific calculation formulas are as follows:

[0052]

[0053] Where: YF is the initial image contrast, YG is the initial image clarity, YH is the initial noise level, YI is the initial edge density, YJ is the initial brightness level, RF is the rotated image contrast, RG is the rotated image clarity, RH is the rotated noise level, RI is the rotated edge density, and RJ is the rotated brightness level; S3.13. Input the calculated initial image enhancement factor YZ and the rotation image enhancement factor RZ into the calculation formulas of the initial image corner point standard coefficient CSJ and the rotation image corner point standard coefficient RSJ for enhancement. The specific calculation formulas are as follows:

[0054]

[0055] Where: YA is the initial corner response value, YB is the initial local texture contrast, YC is the initial local grayscale change, YD is the initial principal curvature ratio, YE is the initial image local gradient energy, RA is the rotated corner response value, RB is the rotated local texture contrast, RC is the rotated local grayscale change, RD is the rotated principal curvature ratio, RE is the rotated image local gradient energy, YZ is the initial image enhancement factor, and RZ is the rotated image enhancement factor.

[0056] In this embodiment, unlike traditional image enhancement methods that rely solely on image grayscale or edge information, this step comprehensively collects multiple key image attributes—including image contrast, clarity, noise level, edge density, and brightness—to form a complete image enhancement dataset. This improvement enhances the system's ability to quantitatively analyze overall image quality and provides more objective and accurate basic data for subsequent calculation of enhancement factors.

[0057] The collected image quality data is analyzed through a multi-layer neural network coupling analysis within a deep learning framework, enabling the modeling and integration of nonlinear relationships between various indicators. The resulting image enhancement factors, YZ and RZ, fully reflect the image's adaptability to feature extraction during actual processing. This enhanced calculation method not only offers physical interpretation but also dynamically adjusts the enhancement strength based on the specific image conditions, significantly outperforming traditional static parameter methods.

[0058] Based on the existing corner point standard coefficients CSJ ​​and RSJ, the enhancement factors YZ and RZ are used to adjust them dually, adjusting both the image structure term and the local grayscale variation term, significantly amplifying the valuable geometric structure and texture changes in the image. This effectively improves the sensitivity and accuracy of corner point detection, especially in weak image quality or blurred details, and can still stably extract feature point information.

[0059] The introduction of enhancement factors ensures that the corner standard coefficients reflect not only the geometric features themselves but also implicitly reflect the adaptability to image conditions. Because the corner standard coefficients participate in the determination of rotation angle differences, their enhancement more accurately reflects the directional changes in the image under different rotational states, thereby improving the system's ability to cope with complex changes such as large-angle rotation, local distortion, and image blur. The entire enhancement process relies on the automatic acquisition of image quality parameters, network coupling, and dynamic factor generation. It does not rely on manual adjustment or empirical settings. It is highly intelligent and adaptable, suitable for a variety of imaging environments and image types, and enhances the system's versatility and engineering value.

[0060] Example 7: Please refer to Figure 1 In step S4, the specific steps for correcting the angle difference are as follows: S4.1. Determine the angle difference. Create a mirror image of the initial image at a standard rotation angle to generate a mirror image. Perform feature selection on the initial image, the mirror image, and the rotated image. Compare the feature coordinates of the mirror image with those of the rotated image. Obtain the angle deviation value JDC using the formula |mirror image feature coordinates - rotated image feature coordinates|. S4.2. Construct a rotation matrix. Use the obtained angle deviation value JDC to generate the rotation matrix using the following formula. Use OpenCV to perform scheduling adjustments to generate the actual rotated image.

[0061] S4.3. Crop the image and perform edge and scaling on the actual rotated image to ensure consistency with the initial image.

[0062] In this embodiment, unlike traditional methods that rely solely on direct comparison of the rotated image with the original image, the present invention constructs a mirror image with a standard rotation angle for angle difference calculation. This image is then used to calculate the feature coordinate difference between the mirror image and the rotated image, using the formula **|mirror image feature coordinates - rotated image feature coordinates|** to obtain the angle deviation value JDC. This method, with its stronger geometric symmetry as a reference, effectively mitigates the effects of errors caused by image distortion, occlusion, or noise, thereby improving the robustness and reliability of angle difference estimation.

[0063] Using image processing tools like OpenCV, the calculated JDC values ​​are used to generate a two-dimensional rotation matrix, XZJ, which is then applied to the image content for an affine transformation. This method of constructing a rotation transformation model based on measured deviation angles more accurately corrects image angles than traditional empirical settings or simple interpolation correction methods, ensuring consistency in the underlying geometric relationships of subsequent image registration and reconstruction.

[0064] Rotation often results in discontinuous edges or blank areas in the corrected image. This step involves cropping and scaling the image edges to ensure that the rotated image remains highly consistent with the original image in terms of size and boundary structure. This not only facilitates subsequent image registration and fusion processing but also avoids positioning errors caused by image deformation during the angle correction process.

[0065] Example 8: Please refer to Figure 1 In step S5, the specific steps for determining the effective midpoint are as follows: S5.1. Perform image registration on the initial image and the actual rotated image to establish a geometric mapping relationship between the two. After image registration is completed, extract the region of the initial image that is spatially aligned with the actual rotated image based on the geometric mapping relationship, thereby obtaining candidate center points corresponding to the overlapping regions of the images as a valid center point candidate set. S5.2. Verify the determined set of valid center point candidates. The verification process is based on a matching evaluation mechanism for the overlapping areas of image content. By calculating the content consistency index of the candidate center points in the overlapping areas of the images, the candidate center points that meet the preset similarity threshold conditions are screened out as the final valid center points.

[0066] In this embodiment, traditional image processing often uses the entire image or a pre-defined region to select the image midpoint. This is susceptible to image rotation, occlusion, or edge deformation, resulting in midpoint offset or distortion. The present invention performs precise image registration on the original image and the rotated image, establishing a geometric mapping relationship between the two. This allows direct extraction of spatially aligned overlapping regions, from which candidate center points are extracted. This approach effectively eliminates interference from non-aligned regions, improving the geometric validity and spatial consistency of midpoint extraction.

[0067] After obtaining the candidate center points, this step further evaluates their performance within the overlapping image regions through image content consistency analysis. This process incorporates a matching evaluation mechanism that calculates the similarity of the candidate points in the two images based on content features such as image grayscale, texture, or edges. Using a preset similarity threshold as a screening criterion, the candidate points with stable image representation are selected as the final valid center points.

[0068] By combining spatial position accuracy with content matching quality, this method achieves dual guarantees for midpoint selection. Compared to traditional methods that rely solely on coordinate centers or manual experience to select points, this method is more adaptable to applications with complex image content and significant local differences, and it can still ensure the stability of the selection results in the presence of rotation, scale transformation, and partial occlusion.

[0069] The contents not described in detail in this specification belong to the prior art known to those skilled in the art.

[0070] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A high-precision automatic calibration method for three points in a line based on image processing, characterized by: The specific steps are as follows: S1, image acquisition and processing, data acquisition and processing of the first positioning image and the rotated image; S2, corner detection, by analyzing the image data to generate corner feature coefficients; S3, matching corresponding points, analyzing corner feature coefficients, and identifying corner differences; S4, correcting the angle difference, using the matched feature points to perform geometric transformation, calculate and correct the rotation angle difference between the two images; S5. Determine the effective midpoint: After the two images are corrected, determine the effective midpoint between the two images based on geometric principles; S6. Feedback: Feedback the collected data and analysis results to the terminal.

2. The high-precision automatic three-point-one-line calibration method based on image processing according to claim 1, characterized in that: In step S1, the specific steps of image acquisition and processing are as follows: S1.

1. Set the rotation angle, take a picture at the starting angle and the set rotation angle, perform denoising on the images, and mark them as the initial image and the rotated image respectively; S1.

2. Extract data from the initial image and the rotated image, including corner response values, local texture contrast, local grayscale change, principal curvature ratio, and local gradient energy of the image; S1.3, preprocessing and dimensionlessizing the extracted data, and reorganizing them into an initial image data set and a rotated image data set; The initial image data set includes initial corner response value YA, initial local texture contrast YB, initial local grayscale change YC, initial principal curvature ratio YD and initial image local gradient energy YE; The rotation image data set includes the rotation corner response value RA, the rotation local texture contrast RB, the rotation local grayscale change RC, the rotation principal curvature ratio RD and the rotation image local gradient energy RE.

3. The high-precision automatic three-point-one-line calibration method based on image processing according to claim 2, characterized in that: In step S2, the specific steps of corner point detection are as follows: S2.

1. Data coupling is performed on the initial image data set. Multiple data are input into a pre-trained deep learning framework. Feature fusion is performed through a multi-layer neural network to calculate the standard coefficient CSJ of the initial image corner points. The specific calculation formula is as follows; Where: YA is the initial corner response value, YB is the initial local texture contrast, YC is the initial local grayscale change, YD is the initial principal curvature ratio, and YE is the initial image local gradient energy; S2.

2. Data coupling is performed on the rotated image data set. Multiple data are input into a pre-trained deep learning framework. Feature fusion is performed through a multi-layer neural network to calculate the standard coefficient RSJ of the rotated image corner point. The specific calculation formula is as follows; Where RA is the rotation corner response value, RB is the rotation local texture contrast, RC is the rotation local grayscale change, RD is the rotation principal curvature ratio, and RE is the rotation image local gradient energy.

4. The high-precision automatic three-point-one-line calibration method based on image processing according to claim 3, characterized in that: In step S3, the specific steps of matching corresponding points are as follows: S3.

1. Analyze the data of the standard coefficients CSJ ​​of the corner points of the initial image and RSJ of the corner points of the rotated image, and determine whether auxiliary processing of the image is required based on the analysis results. S3.

2. Perform an integrated analysis of the corner point standard coefficients CSJ ​​of the initial image and RSJ of the rotated image after auxiliary processing. Based on the analysis results, determine whether there is a significant angular difference between the two images. The specific steps are as follows: when When , it means that there is no obvious angle difference in the current image and no correction is needed; when , it means that the current image has obvious angle difference and needs to be corrected.

5. The high-precision automatic three-point-one-line calibration method based on image processing according to claim 4, characterized in that: In step S3.1, the analysis method of the initial image corner point standard coefficient CSJ and the rotated image corner point standard coefficient RSJ is as follows: when or When , it means that the initial image or the rotated image needs to be enhanced; when or , it means that the original image or the rotated image does not need to be enhanced.

6. The high-precision automatic three-point-one-line calibration method based on image processing according to claim 5, characterized in that: In step S3.1, the image enhancement method is as follows: S3.

11. Perform secondary data collection on the initial image and the rotated image, including image contrast, image clarity, noise level, edge density, and brightness level. Preprocess and dimensionlessly convert the extracted data into an enhanced initial image data set and an enhanced rotated image data set. The initial image enhancement data set includes initial image contrast YF, initial image clarity YG, initial noise level YH, initial edge density YI and initial brightness level YJ; The rotation image enhancement data set includes the rotation image contrast RF, the rotation image sharpness RG, the rotation noise level RH, the rotation edge density RI and the rotation brightness level RJ; S3.

12. Data coupling is performed on the initial image enhancement data set and the rotated image enhancement data set. Multiple data are input into the pre-trained deep learning framework. Feature fusion is performed through a multi-layer neural network to calculate the initial image enhancement factor YZ and the rotated image enhancement factor RZ. The specific calculation formulas are as follows: Where: YF is the initial image contrast, YG is the initial image clarity, YH is the initial noise level, YI is the initial edge density, YJ is the initial brightness level, RF is the rotated image contrast, RG is the rotated image clarity, RH is the rotated noise level, RI is the rotated edge density, and RJ is the rotated brightness level; S3.

13. Input the calculated initial image enhancement factor YZ and the rotation image enhancement factor RZ into the calculation formulas of the initial image corner point standard coefficient CSJ and the rotation image corner point standard coefficient RSJ for enhancement. The specific calculation formulas are as follows: Where: YA is the initial corner response value, YB is the initial local texture contrast, YC is the initial local grayscale change, YD is the initial principal curvature ratio, YE is the initial image local gradient energy, RA is the rotated corner response value, RB is the rotated local texture contrast, RC is the rotated local grayscale change, RD is the rotated principal curvature ratio, RE is the rotated image local gradient energy, YZ is the initial image enhancement factor, and RZ is the rotated image enhancement factor.

7. The high-precision automatic three-point-one-line calibration method based on image processing according to claim 6, characterized in that: In step S4, the specific steps of correcting the angle difference are as follows: S4.

1. Determine the angle difference. Create a mirror image of the initial image at a standard rotation angle to generate a mirror image. Perform feature selection on the initial image, the mirror image, and the rotated image. Compare the feature coordinates of the mirror image with those of the rotated image. Obtain the angle deviation value JDC using the formula |mirror image feature coordinates - rotated image feature coordinates|. S4.

2. Construct a rotation matrix. Use the obtained angle deviation value JDC to generate the rotation matrix using the following formula. Use OpenCV to perform scheduling adjustments to generate the actual rotated image. S4.

3. Crop the image and perform edge and scaling on the actual rotated image to ensure consistency with the initial image.

8. The high-precision automatic three-point-one-line calibration method based on image processing according to claim 7, characterized in that: In step S5, the specific steps of determining the effective midpoint are as follows: S5.

1. Perform image registration on the initial image and the actual rotated image to establish a geometric mapping relationship between the two. After image registration is completed, extract the region of the initial image that is spatially aligned with the actual rotated image based on the geometric mapping relationship, thereby obtaining candidate center points corresponding to the overlapping regions of the images as a valid center point candidate set. S5.

2. Verify the determined set of valid center point candidates. The verification process is based on a matching evaluation mechanism for the overlapping area of ​​image content. By calculating the content consistency index of the candidate center points in the overlapping area of ​​the image, the candidate center points that meet the preset similarity threshold conditions are screened out as the final valid center points.

Images