A three-dimensional calibration device

Abstract

The application discloses a three-dimensional calibration device, comprising: a calibration ruler manufacturing unit, comprising a calibration ruler material design unit and a calibration ruler design unit; the calibration ruler material design unit comprises a design method of zero-expansion material; the calibration ruler design unit is designed based on a calibration ruler template set; a calibration layout unit: based on the first parameter, the calibration ruler is combined target layout to determine the first calibration combination; based on the high-precision motion mechanism, the first calibration combination is precision calibrated; a compensation unit: based on the combined target image, the positioning circle center of the combined target image is determined based on the gray information and the edge information; based on the circle center error prediction value, the positioning circle center is error compensated to determine the second parameter, and the second calibration combination is determined based on the second parameter and the first calibration combination; a calibration unit: the camera and the second calibration combination form a three-dimensional calibration device, and calibration is completed based on the three-dimensional calibration device combined with the second parameter. Through the application, a low-cost high-precision calibration device can be realized.

Description

Technical Field

[0001] This invention relates to the field of binocular camera calibration technology, and more specifically to a three-dimensional calibration device. Background Technology

[0002] Binocular cameras require calibration during use. However, existing technologies rely on calibration plates for calibration, which present dimensional challenges. For instance, large calibration plates are difficult to manufacture, easily affected by external forces, leading to deformation or damage. Furthermore, their storage and transportation are more difficult. Due to their large size, it is challenging to ensure consistent flatness across the entire surface, compromising calibration accuracy. Humidity also affects the plate's expansion and contraction, impacting its geometric properties and flatness, further affecting calibration accuracy. Therefore, this application proposes a three-dimensional calibration device. Summary of the Invention

[0003] To address the aforementioned issues, this application provides a three-dimensional calibration device that employs a zero-expansion material design method to ensure the stability of the calibration device. The calibration accuracy is guaranteed by the arrangement of combined targets. Furthermore, the calibration device of this application can adjust the size and complexity of the calibration scene according to actual needs to meet the requirements of different application scenarios.

[0004] To achieve the above objectives, this application provides the following technical solution:

[0005] This application provides a three-dimensional calibration device, specifically including:

[0006] Calibration ruler manufacturing unit: includes a calibration ruler material design unit and a calibration ruler design unit; the calibration ruler material design unit includes a design method for zero-expansion materials; the calibration ruler design unit includes designing the calibration ruler to be designed based on a calibration ruler template set;

[0007] Calibration layout unit: Based on the first parameter, the designed calibration ruler is combined with the target layout to determine the first calibration combination; the first calibration combination is calibrated for accuracy based on the high-precision motion mechanism;

[0008] Compensation unit: Based on the combined target image, determines the positioning center of the combined target image based on grayscale information and edge information; performs error compensation on the positioning center based on the predicted value of the center error determined by the positioning center of the combined target image; analyzes and determines a second parameter based on the compensated positioning center; and determines a second calibration combination based on the second parameter and the first calibration combination.

[0009] Calibration unit: The binocular camera and the second calibration combination are combined to form a three-dimensional calibration device, and the image information captured by the binocular camera is calibrated based on the three-dimensional calibration device and the second parameter.

[0010] As a preferred embodiment, the calibration ruler manufacturing unit is manufactured based on the zero-expansion material combined with the calibration ruler design unit. The design method for the zero-expansion material specifically includes:

[0011] Preset stacking method;

[0012] The preset laying angles are -40°, 40°, -40°, 40°, 0°, -40°, 40°, -40°, 40°, 90°, 40°, -40°, 40°, -40°, 0°, 40°, -40°, 40°, -40°;

[0013] Obtain carbon fiber synthetic fibers, and lay the carbon fiber synthetic fibers based on the preset laying angle and the preset stacking method to determine a zero-expansion material.

[0014] As a preferred embodiment, the calibration ruler design unit specifically includes:

[0015] Obtain application scenarios for numerous calibration ruler templates and assign values ​​to them to obtain the first attribute value of the calibration ruler template; obtain the feature attribute values ​​of numerous calibration ruler templates.

[0016] The representative attribute points of each calibration ruler template are determined by combining the first attribute value and the feature attribute value, denoted as X1Y1, X2Y2, X3Y3, ... XiYj, where X represents the first attribute value, Y represents the feature attribute value, i represents the number of existing first attribute values, and j represents the number of existing feature attribute values, and i... <j;

[0017] The calibration ruler template set is determined based on the representative attribute points of each calibration ruler template;

[0018] Obtain the first attribute value of the calibration ruler to be designed, and filter and match the calibration ruler template set based on the first attribute value of the calibration ruler to be designed to obtain several calibration ruler templates under the first attribute value of the calibration ruler to be designed, which are denoted as the calibration ruler to be designed.

[0019] Several undetermined calibration scales are sorted according to priority, and the design is carried out based on the sorted undetermined calibration scales.

[0020] As a preferred embodiment, the calibration layout unit further includes a combined target comprising a precision target and a coded target, wherein the precision target and the coded target are bonded together based on adhesive-backed alumina and the calibration ruler; the alumina is marked using a photolithographic combined target.

[0021] As a preferred embodiment, the compensation unit specifically includes:

[0022] 5.1 Obtain the initial combined target image. Based on the similarity processing of the image structure, determine the optimized initial combined target image, which is denoted as the combined target image.

[0023] 5.2 Determine a first candidate circle center based on the grayscale information of the combined target image, determine a second candidate circle center based on the edge information of the combined target image, and determine the positioning circle center of the combined target image based on the first candidate circle center and the second candidate circle center;

[0024] 5.3 Determine the center error prediction value based on the center of the positioning circle of the combined target image, compensate the center of the positioning circle of the combined target image based on the center error prediction value, and determine the compensated center of the positioning circle.

[0025] 5.4 Determine the positional parameter relationship between each combined target image based on the compensated positioning center, denoted as the first positional parameter. Perform coordinate transformation based on the first positional parameter to determine the positional parameter relationship between the actual combined targets, denoted as the second positional parameter. The second positional parameter is also known as the second parameter.

[0026] 5.5 Determine the second calibration combination based on the second parameter and the first calibration combination.

[0027] As a preferred embodiment, step 5.2, determining the center of the circle, specifically includes:

[0028] Obtain the pixel and grayscale information of the combined target image;

[0029] The combined target image is divided into N sub-image regions on average, and the center pixel information of the sub-image region is determined based on the grayscale information of the pixels in the sub-image region.

[0030] The center pixel of the combined target image, i.e., the first candidate circle center, is determined based on the center pixel information of the sub-image region;

[0031] Edge extraction is performed on the combined target image to determine the edge information of the combined target image. After classifying the edge information, edge information is filtered based on pixel continuity.

[0032] The center pixel of the combined target image, i.e., the second candidate circle center, is determined by fitting the filtered edge information;

[0033] The pixel association relationship between the first candidate circle center and the second candidate circle center is determined based on the first candidate circle center and the second candidate circle center.

[0034] If the pixel association relationship meets the preset conditions, the center of the combined target image is determined based on the first candidate center and the second candidate center.

[0035] As a preferred embodiment, 5.3 specifically includes:

[0036] The combined target image is rotated to determine the combined target image set, and the center error set is determined based on the combined target image set and the positioning center of the combined target image;

[0037] The prediction model is trained based on the circle center error set to determine the trained prediction model, which is denoted as the circle center error prediction model.

[0038] The predicted center error value of the positioning circle of the combined target image is determined based on the center error prediction model.

[0039] When the predicted center error value is greater than zero, the first compensation method is performed based on the predicted center error value to determine the center of the compensated combined target image.

[0040] When the predicted center error value is less than zero, a second compensation method is performed based on the predicted center error value to determine the center of the compensated combined target image.

[0041] As a preferred embodiment, the construction and training of the prediction model specifically includes:

[0042] The obtained circle center error set is randomly divided into a training set and a test set, with the training set accounting for 80% of the circle center error set and the test set accounting for 20% of the circle center error set, and the training set and the test set are normalized.

[0043] The input of the prediction model is determined based on the pixel information of the positioning center and the rotation information, and the output layer of the prediction model is determined based on the error prediction value of the positioning center.

[0044] An optimization function is constructed to optimize the initial parameters of the prediction model and determine the optimal parameters.

[0045] The prediction model is trained based on the optimal parameters and the training set, and the training results of the prediction model are compared with the preset error based on the test set.

[0046] Iterative calculations are performed until the training results meet the preset error to obtain the trained prediction model, i.e., the center error prediction model.

[0047] As a preferred embodiment, section 5.4 specifically includes:

[0048] The positional relationship of the combined target images in the first coordinate system is determined based on the compensated positioning center, and is denoted as the first positional relationship. The positional relationship between each combined target image in the first coordinate system is determined based on the first positional relationship, and is denoted as the second positional relationship.

[0049] The second positional relationship is used to determine the actual positional relationship of the combined target based on a preset transformation relationship, and is denoted as the third positional relationship.

[0050] The second parameter is determined by updating and compensating the first parameter based on the third positional relationship.

[0051] As a preferred embodiment, the calibration unit further includes:

[0052] The three-dimensional calibration device performs posture transformation based on a preset number of transformations M and combines it with a binocular camera to complete the calibration, where M≥15.

[0053] The three-dimensional calibration device provided in this application has the following beneficial effects:

[0054] 1. This application obtains zero-expansion material by stacking 19 layers of carbon fiber material at different angles, overcoming the influence of external factors such as temperature and humidity, maintaining constant volume and size, and using rectangular carbon fiber tubes to change the rigidity problem from the cross-sectional shape, thus possessing strong stability and improving calibration accuracy.

[0055] 2. In this application, by designing a combination of targets, the precision target is used for position calibration and the coded target is used for spatial position determination, which can ensure that the calibration ruler can complete the accuracy measurement and graphic matching even when the calibration ruler moves and rotates during the calibration process.

[0056] 3. In this application, a high-precision motion mechanism is used to locate and compensate for the center of the combined target. This improves the positioning accuracy of the center and enables high-precision compensation of the positional parameters between the determined combined targets, thereby enhancing the calibration accuracy of the calibration ruler and realizing a high-precision calibration device. Attached Figure Description

[0057] Figure 1 This is a schematic diagram of a three-dimensional calibration device according to this application;

[0058] Figure 2 This is a flowchart illustrating the compensation unit in a three-dimensional calibration device according to this application;

[0059] Figure 3 This is an example diagram of a combined target in a three-dimensional calibration device according to this application;

[0060] Figure 4 This is an example diagram of a calibration ruler in an embodiment of this application; Detailed Implementation

[0061] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0062] See Figure 1 A three-dimensional calibration device specifically includes:

[0063] Calibration ruler manufacturing unit: includes a calibration ruler material design unit and a calibration ruler design unit; the calibration ruler material design unit includes a design method for zero-expansion materials; the calibration ruler design unit includes designing the calibration ruler to be designed based on a calibration ruler template set;

[0064] Calibration layout unit: Based on the first parameter, the designed calibration ruler is combined with the target layout to determine the first calibration combination; the first calibration combination is calibrated for accuracy based on the high-precision motion mechanism;

[0065] Compensation unit: Based on the combined target image, determines the positioning center of the combined target image based on grayscale information and edge information; performs error compensation on the positioning center based on the predicted value of the center error determined by the positioning center of the combined target image; analyzes and determines a second parameter based on the compensated positioning center; and determines a second calibration combination based on the second parameter and the first calibration combination.

[0066] Calibration unit: The binocular camera and the second calibration combination are combined to form a three-dimensional calibration device, and the image information captured by the binocular camera is calibrated based on the three-dimensional calibration device and the second parameter.

[0067] In an optional implementation, the calibration ruler manufacturing unit completes the manufacturing based on the zero-expansion material combined with the calibration ruler design unit, and the design method of the zero-expansion material specifically includes: a preset stacking method;

[0068] A preset laying angle is used, which is -40°, 40°, -40°, 40°, 0°, -40°, 40°, -40°, 40°, 90°, 40°, -40°, 40°, -40°, 0°, 40°, -40°, 40°, -40°; carbon fiber synthetic fibers are obtained, and the carbon fiber synthetic fibers are laid based on the preset laying angle and the preset stacking method to determine a zero-expansion material;

[0069] In an optional embodiment, if the number of stacked layers used is too small, the resulting rectangular tube will have insufficient rigidity and an increased expansion coefficient; if the number of stacked layers used is too large, the resulting rectangular tube will be excessively heavy, the expansion coefficient will also increase, leading to poor stability of the rectangular tube. The stacking method set in the present application adopts 19 layers with the aforementioned preset laying angle, which can ensure the optimal expansion coefficient in this case, reduce the repeatability of the expansion coefficient, and obtain the rectangular tube with the best rigidity. Through the design method of the zero-expansion material in the present application, the influence of related factors such as temperature can be overcome, the volume and size can be kept unchanged under temperature changes, and strong stability is achieved; moreover, the use of rectangular carbon fiber tubes solves the rigidity problem by changing the cross-sectional shape, so as to improve the calibration accuracy.

[0070] In an optional embodiment, the calibration ruler design unit specifically includes: acquiring the application scenarios of a plurality of calibration ruler templates and performing assignment processing to obtain the first attribute value of the calibration ruler template; if the application scenarios are the same, the assignments of the calibration ruler templates for the application scenarios are the same, that is, the first attribute values are the same. It can be understood that among the plurality of calibration ruler templates, if the application scenarios are the same, the first attribute values are the same, and the same first attribute value indicates that the application scenarios are consistent;

[0071] acquiring the characteristic attribute values of the plurality of calibration ruler templates; the characteristic attribute values include structural attributes, material attributes, dimensional attributes, combined target attributes, and the like;

[0072] determining the representative attribute point of each calibration ruler template by combining the first attribute value and the characteristic attribute value, which are recorded as X1Y1, X2Y2, X3Y3, ... XiYj, where X represents the first attribute value, Y represents the characteristic attribute value, i represents the number of existing first attribute values, that is, the number of application scenarios, j represents the number of existing characteristic attribute values, that is, the number of calibration rulers, and i < j. At least one calibration ruler exists under a specific application scenario, so the number of first attribute values is less than the number of characteristic attribute values;

[0073] determining a calibration ruler template set based on the representative attribute point of each calibration ruler template; since there are cases where the application scenarios are the same, that is, cases where X is the same, for example, X1Y1 and X2Y2 represent two calibration ruler templates with different attributes in different application scenarios, and X1Y1, X1Y2, and X1Y3 represent three calibration rulers with different attributes in the same application scenario;

[0074] acquiring the first attribute value of the to-be-designed calibration ruler, and performing screening and matching on the calibration ruler template set based on the first attribute value of the to-be-designed calibration ruler to obtain a plurality of calibration ruler templates under the first attribute value of the to-be-designed calibration ruler, which are recorded as to-be-determined calibration rulers;

[0075] Sort a plurality of to-be-determined calibration scales based on priority, and perform design based on the sorted plurality of to-be-determined calibration scales.

[0076] In an optional embodiment, the priority sorting is determined based on the manufacturing cost of the to-be-determined calibration scales, and the priority value is calculated based on a priority formula, where the priority formula is:

[0077] Wherein Q is the priority value, b1, b2 and b3 are proportional coefficients, 0<b1≤1, 0<b2≤1, 0<b3≤1, α is a price conversion coefficient, which is specifically determined based on actual application conditions, G represents the manufacturing cost of the to-be-determined calibration scale, S represents the adaptation value of the to-be-determined calibration scale in a specific application scenario, and L represents the adaptation value in the application scenario in the calibration scale template set;

[0078] By performing priority sorting on the plurality of to-be-determined calibration scales based on the priority formula, screening can be carried out according to actual design requirements, which facilitates the rapid and accurate screening of the range of the to-be-designed calibration scale from the calibration scale template set, and provides convenience for subsequent calibration work. The calibration scale manufactured according to the technical solution in the present application may be a rectangular square tube, which can be rotated and folded to change its posture; it can also be set as square tubes of other quantities as required, and a plurality of square tubes are connected by hinges, so that the calibration scale in the present application has high stability and flexibility, and enhances the accuracy and reliability of calibration results.

[0079] In an optional embodiment, the calibration layout unit comprises: determining a first calibration combination by performing combined target layout on the calibration scale designed based on a first parameter; the combined target comprises a precision target and an encoding target, the precision target is used for position extraction and recognition, and the encoding target is used for marker point tracking and orientation recognition; the precision target and the encoding target are pasted on the basis of back-glued alumina and the calibration scale; the alumina is marked by adopting a photolithographic combined target;

[0080] In an optional embodiment, the precision target may be circular, the shape of the encoding target may be adjusted as required to perform corresponding tracking and positioning, the encoding target includes but is not limited to a cross shape, a checkerboard shape, a rectangle, etc., the encoding target is located inside the precision target, the first parameter can be understood as performing the layout of the combined target according to actual needs in practical applications, for example, if in practical application, the layout needs to be performed according to the number of targets and the distance between each target, the first parameter can be understood as a comprehensive weighting coefficient of the number of targets and the target spacing, which reflects the distribution of the combined targets on the calibration scale, and when performing camera calibration by capturing images, each calibration parameter is calculated and analyzed based on the first parameter; performing precision calibration on the first calibration combination based on a high-precision motion mechanism;

[0081] In an optional implementation, by using a combination of targets, precision targets are used to calibrate the position, and directional targets are used to determine the orientation. This ensures that the calibration ruler can complete accuracy measurement and graphic matching even when it moves and rotates during the calibration process. The combination targets are attached to the calibration ruler using an adhesive-backed alumina bonding method. Existing combination targets use traditional printing methods, which have resolution issues and significant contour accuracy deviations. In this application, the alumina marking points are applied using photolithographic combination targets to improve resolution and contour accuracy. Furthermore, the adhesive used in this application is a highly stable double-sided adhesive that does not peel off or move in environments ranging from -30℃ to 60℃. The alumina thickness in this application is 0.25mm, which avoids wrinkles during processing and assembly, reducing errors in image processing.

[0082] In an optional implementation, a high-precision motion mechanism is used to detect the combined targets formed in this application. Based on the positioning and image acquisition of the high-precision motion mechanism, accuracy compensation is performed on the combined targets on the calibration ruler to obtain accurate relative position and other parameter relationships between the combined targets, thereby enhancing the calibration accuracy of the calibration ruler. The specific process includes: 5.1 specifically includes:

[0083] Based on the obtained initial combined target image, several sub-image units are divided. Based on the similarity of each sub-image unit within a preset range, similar sub-image units are determined, resulting in a set of similar sub-image units, denoted as the image unit set. This can be understood as follows: for several randomly divided sub-image units, for any sub-image unit, its corresponding similar sub-image units are determined within a preset range based on the similarity standard. A sub-image unit can have at least one similar sub-image unit. After traversing all sub-image units, several similar sub-image units can be determined, denoted as the image unit set.

[0084] An image unit matrix is ​​constructed based on the image unit set. The image unit matrix can be represented by E = E1 + E2, where E1 represents the optimized image unit matrix and E2 represents the noise matrix.

[0085] An optimal function is constructed for the image unit matrix based on integration and filtering constraints, and the pixel values ​​of the optimized sub-image units are determined analytically based on the optimal function through optimization iteration.

[0086] The optimized image unit is determined by aggregating the pixel values ​​of the optimized sub-image units, that is, by determining the combined target image; through the denoising processing of the initial combined target image in this application, the similarity between image information is fully utilized to further improve the accuracy of the combined target image information and reduce the interference of noise.

[0087] 5.2 A first candidate circle center is determined based on the grayscale information of the combined target image, a second candidate circle center is determined based on the edge information of the combined target image, and the positioning circle center of the combined target image is determined based on the first candidate circle center and the second candidate circle center; two candidate circle centers are determined by using two types of feature information, and the final positioning circle center is determined by analyzing the two candidate circle centers, which enhances the accuracy of circle center positioning;

[0088] 5.3 Determine the center error prediction value based on the center of the positioning circle in the combined target image, compensate the center of the positioning circle in the combined target image based on the center error prediction value, and determine the compensated center of the positioning circle; predict the error value of the positioning circle by combining the actual position of the positioning circle with the actual position of the combined target image, and compensate based on the error value to reduce the positioning error and improve the accuracy of the data processing results.

[0089] 5.4 Determine the positional parameter relationship between each combined target image based on the compensated positioning center, denoted as the first positional parameter. Perform coordinate transformation based on the first positional parameter to determine the positional parameter relationship between the actual combined targets, denoted as the second positional parameter. The second positional parameter is also known as the second parameter.

[0090] 5.5 Determine the second calibration combination based on the second parameter and the first calibration combination.

[0091] In an optional implementation, the positional parameter relationship between the center of each combined target image can be obtained based on the determined final compensated positioning center, i.e., the positional parameter relationship between each combined target image is determined, i.e., the first positional parameter. Based on the first positional parameter, the positional parameter relationship between the actual combined targets is determined through spatial transformation relationship, denoted as the second positional parameter, i.e., the second parameter. The second parameter is determined after compensating the first parameter using the method in this application, achieving a high-precision positioning effect. The second calibration combination is determined by combining the second parameter with the first calibration combination. High-precision accuracy compensation is performed on the first calibration combination, so that the finally determined calibration combination completes the calibration with high-precision parameters, improving the calibration accuracy of the calibration ruler. When calibrating the subsequent camera-captured image information, high-precision calibration parameters can be provided, thereby improving the accuracy of the image analysis results.

[0092] In an optional implementation, the pixel and grayscale information of the combined target image is obtained;

[0093] The combined target image is divided into N sub-image regions on average, and the center pixel information of the sub-image region is determined based on the grayscale information of the pixels in the sub-image region.

[0094] The center pixel of the combined target image, i.e., the first candidate circle center, is determined based on the center pixel information of the sub-image region;

[0095] In an optional implementation, the combined target image is divided into N sub-image regions, each of which can be considered as an arc segment region. The center pixel of the arc segment region is determined based on the grayscale information of the pixels within the arc segment region. Furthermore, the combined target image is composed of the center pixels of several arc segment regions. The center pixel of the combined target image is determined based on the N arc segment regions and their corresponding center pixels, thus obtaining the initial center of the combined target image, denoted as the first candidate center. In this application, by subdividing the combined target image into several arc segments, the computation speed is improved, and the extraction accuracy is further improved.

[0096] Edge extraction is performed on the combined target image to determine its edge information. This edge information is then categorized and filtered based on pixel continuity. Edge extraction includes, but is not limited to, extraction based on the Canny operator, star ray extraction, etc. After extracting the edge information of the combined target image, it is not directly fitted. The obtained edge information is further filtered and judged, which can be based on pixel continuity. Specifically, it can be understood that it is necessary to filter and retain smooth and continuous edge information features, with the retained edge width being a single pixel. By filtering the edge information, complete and continuous edge information is determined, inner edge information is removed, the number of edge information pixels is reduced, image processing efficiency is improved, and it provides convenience for fitting the center of the combined target image.

[0097] The center pixel of the combined target image, i.e., the second candidate circle center, is determined by fitting the filtered edge information;

[0098] The pixel association relationship between the first candidate circle center and the second candidate circle center is determined based on the first candidate circle center and the second candidate circle center.

[0099] If the pixel association relationship meets the preset conditions, the positioning center of the combined target image is determined based on the first candidate center and the second candidate center.

[0100] In an optional implementation, the center pixel of the combined target image, i.e. the second candidate circle center, is determined by fitting the filtered edge information; this can be understood as including, but not limited to, fitting the edge information based on Euclidean distance, fitting the edge information based on 8 random edge points combined with the optimal function, etc.

[0101] In an optional implementation, the correlation between the obtained first candidate circle center and the second candidate circle center is determined. The correlation can be understood as the difference in the number of pixels between the first candidate circle center and the second candidate circle center. If the difference in the number of pixels does not meet a preset condition, the above steps are repeated. If the difference in the number of pixels meets the preset condition, and the first candidate circle center and the second candidate circle center coincide, then the point is the circle center of the combined target image. If the first candidate circle center and the second candidate circle center do not highly coincide, then the final circle center is determined by high-fitting based on the two candidate circle centers, that is, the positioning circle center of the combined target image. Through the method for determining the circle center of the combined target image in this application, the first candidate circle center is determined based on grayscale information to enhance the noise resistance capability, and the second candidate circle center is determined based on edge information to enhance the fitting accuracy. Through the method for determining the positioning circle center of the combined target image in this application, the computational load of image processing is reduced, the image processing speed is increased, the specific pixel information of the circle center is accurately determined, and the accuracy of circle center positioning is improved.

[0102] In an optional implementation, the combined target images are rotated to determine a combined target image set, and a center error set is determined based on the combined target image set and the positioning center of the combined target images. The rotation can be based on the X-axis or the Y-axis. By rotating the combined target images, several combined target images, i.e., a combined target image set, can be obtained. The center error is determined by comparing the positioning center of the combined target images with the actual position of the combined target images, and the center error set is further determined based on the center error.

[0103] The prediction model is trained based on the circle center error set to determine the trained prediction model, which is denoted as the circle center error prediction model. By constructing the prediction model, the prediction effect of the circle center error is achieved, reducing the amount of computation and improving the stability of the error prediction result.

[0104] The predicted center error value of the positioning circle of the combined target image is determined based on the center error prediction model.

[0105] When the predicted center error value is greater than zero, the first compensation method is performed based on the predicted center error value to determine the center of the compensated combined target image.

[0106] When the predicted center error value is less than zero, a second compensation method is performed based on the predicted center error value to determine the center of the compensated combined target image.

[0107] In an optional implementation, error compensation is performed on the positioning center based on the predicted center error value to achieve accurate positioning. By comparing the predicted center error value with zero, i.e., by determining whether the error generated by the positioning center is a positive or negative error, different forms of compensation are performed. If the predicted center error value is greater than zero, the first compensation method is executed, i.e., the pixel coordinates of the positioning center are added to the predicted center error value to obtain the compensated center. If the predicted center error value is less than zero, the second compensation method is executed, i.e., the pixel coordinates of the positioning center are subtracted from the predicted center error value to obtain the compensated center. By judging different error relationships of the positioning center and performing different compensation methods, the robustness of the compensation result is enhanced, so that error compensation of the positioning center can be achieved when dealing with different error situations, and the compensated center information can be obtained, which facilitates the determination of the positional parameter relationship between various combined target images.

[0108] In an optional implementation, the obtained center error set is randomly divided into a training set and a test set, wherein the training set accounts for 80% of the center error set and the test set accounts for 20% of the center error set, and the training set and the test set are normalized.

[0109] The input of the prediction model is determined based on the pixel information of the positioning center and the rotation information, and the output layer of the prediction model is determined based on the error prediction value of the positioning center.

[0110] An optimization function is constructed to optimize the initial parameters of the prediction model and determine the optimal parameters.

[0111] The prediction model is trained based on the optimal parameters and the training set, and the training results of the prediction model are compared with the preset error based on the test set.

[0112] Iterative calculations are performed until the training results meet the preset error to obtain the trained prediction model, i.e., the center error prediction model.

[0113] In optional implementations, optimization based on constructing an optimization function can be understood as including, but not limited to, optimization based on a genetic algorithm. This involves encoding the initial parameters of the prediction model, constructing an optimization function to filter parameters that meet certain conditions, and determining the optimization function based on the difference between the output values ​​after training and the output values ​​in the dataset. The training of the prediction model in this application includes, but is not limited to, using a training function. The initial parameters of the prediction model are optimized to obtain the optimal parameters, and the model is trained based on the optimal parameters to improve the convergence speed of the model, reduce the influence of extreme values, and enhance the stability of the prediction model.

[0114] In an optional implementation, the positional relationship of the combined target images in the first coordinate system is determined based on the compensated positioning center, and is denoted as the first positional relationship. Based on the first positional relationship, the positional relationship between each combined target image in the first coordinate system is determined, and is denoted as the second positional relationship.

[0115] The second positional relationship is used to determine the actual positional relationship of the combined target based on a preset transformation relationship, and is denoted as the third positional relationship.

[0116] The second parameter is determined by updating and compensating the first parameter based on the third positional relationship.

[0117] In an optional implementation, the positional relationship of the combined target images in the first coordinate system is determined based on the compensated positioning center. In this application, the first coordinate system can be understood as determining the positional relationship of the combined target images in the high-precision motion mechanism, that is, determining the positional relationship of the combined target images in the high-precision motion mechanism through the updated and compensated positioning center, thereby determining the positional relationship between each combined target image in the high-precision motion mechanism, i.e., the second positional relationship. Further, the second positional relationship is used to determine the actual positional relationship of the combined targets in the world coordinate system based on a preset transformation relationship. The preset transformation relationship is determined based on the rotation and translation of the image coordinate system and the world coordinate system. The actual positional relationship of each combined target is determined through the preset transformation relationship, i.e., the third positional relationship. The third positional relationship is the actual positional relationship between the combined targets determined by the high-precision motion mechanism after high-precision compensation. Based on the third positional relationship, the first parameter is updated to determine the second parameter. The second parameter has overcome the error of the first parameter being determined manually, and the accuracy compensation of the first parameter has been achieved. By determining the high-precision parameters between the combined targets, a high-precision calibration ruler is formed, thereby improving the accuracy of the calibration device.

[0118] In an optional implementation, the 3D calibration device performs pose transformations based on a preset number of transformations M, combined with a binocular camera, to complete the calibration. Specifically, during binocular camera calibration, the calibration ruler of the second calibration combination is mounted on the corresponding tripod platform to form a 3D calibration device. During calibration measurements, the 3D calibration device is placed in the space to be measured, and pose transformations are performed. The number of pose transformations is set to M, where M≥15. After each pose transformation, a photograph is taken using the binocular camera to record the data. The acquired data is then used for binocular camera calibration. By transforming the pose of the calibration combination, multi-angle and multi-directional shooting is achieved, obtaining multi-directional calibration data. Experimental verification shows that at least 15 pose transformations are sufficient to determine comprehensive calibration data, improving the accuracy and reliability of the calibration results, reducing the impact of incomplete calibration data on the calibration results, and thus enhancing the completeness of the calibration data.

[0119] This invention is not limited to the specific embodiments described above. Any modifications made by those skilled in the art based on the above concept without creative effort are within the scope of protection of this invention.

Claims

1. A three-dimensional calibration device, characterized in that, The device includes: The calibration ruler manufacturing unit includes a calibration ruler material design unit and a calibration ruler design unit. The calibration ruler material design unit includes a method for designing zero-expansion materials, comprising: a preset stacking method; preset laying angles, wherein the laying angles are -40°, 40°, -40°, 40°, 0°, -40°, 40°, -40°, 40°, 90°, 40°, -40°, 40°, -40°, 0°, 40°, -40°, 40°, -40°; obtaining carbon fiber synthetic fibers, and laying the carbon fiber synthetic fibers based on the preset laying angles and the preset stacking method to determine the zero-expansion material; the calibration ruler design unit includes designing the calibration ruler to be designed based on a calibration ruler template set. Calibration layout unit: Based on the first parameter, the designed calibration ruler is combined with the target layout to determine the first calibration combination; the first calibration combination is calibrated for accuracy based on the high-precision motion mechanism; Compensation Unit: Determines the positioning center of the combined target image based on the grayscale information and edge information of the combined target image; performs error compensation on the positioning center based on the predicted error value of the positioning center of the combined target image; analyzes and determines a second parameter based on the compensated positioning center; and determines a second calibration combination based on the second parameter and the first calibration combination. Specifically, it includes: 5.1 Obtaining an initial combined target image; determining an optimized initial combined target image based on image structure similarity processing of the initial combined target image, denoted as the combined target image; 5.2 Determining a first candidate center based on the grayscale information of the combined target image; determining a second candidate center based on the edge information of the combined target image; and determining the positioning center of the combined target image based on the first candidate center and the second candidate center. Specifically, it includes: obtaining the pixel points and grayscale information of the combined target image; dividing the combined target image into N sub-image regions on an average basis; determining the center pixel information of the sub-image regions based on the grayscale information of the pixels in the sub-image regions; and determining the center pixel of the combined target image, i.e., the first... 5.1 Candidate center; ... Calibration unit: The binocular camera and the second calibration combination are combined to form a three-dimensional calibration device, and the image information captured by the binocular camera is calibrated based on the three-dimensional calibration device and the second parameter.

2. The three-dimensional calibration device according to claim 1, characterized in that, The calibration ruler design unit specifically includes: Obtain application scenarios for numerous calibration ruler templates and assign values ​​to them to obtain the first attribute value of the calibration ruler template; obtain the feature attribute values ​​of numerous calibration ruler templates. The representative attribute points of each calibration ruler template are determined by combining the first attribute value and the feature attribute value, denoted as X1Y1, X2Y2, X3Y3, ... XiYj, where X represents the first attribute value, Y represents the feature attribute value, i represents the number of existing first attribute values, and j represents the number of existing feature attribute values, and i... <j; The calibration ruler template set is determined based on the representative attribute points of each calibration ruler template; Obtain the first attribute value of the calibration ruler to be designed, and filter and match the calibration ruler template set based on the first attribute value of the calibration ruler to be designed to obtain several calibration ruler templates under the first attribute value of the calibration ruler to be designed, which are denoted as the calibration ruler to be designed. Several undetermined calibration scales are sorted according to priority, and the design is carried out based on the sorted undetermined calibration scales.

3. The three-dimensional calibration device according to claim 1, characterized in that, The calibration layout unit further includes a combined target comprising a precision target and a coded target, wherein the precision target and the coded target are bonded together based on adhesive-backed alumina and the calibration ruler; the alumina is marked using photolithography combined targets.

4. A three-dimensional calibration device according to claim 1, characterized in that, Specifically, 5.3 includes: The combined target image is rotated to determine the combined target image set, and the center error set is determined based on the combined target image set and the positioning center of the combined target image; The prediction model is trained based on the circle center error set to determine the trained prediction model, which is denoted as the circle center error prediction model. The predicted center error value of the positioning circle of the combined target image is determined based on the center error prediction model. When the predicted center error value is greater than zero, the first compensation method is performed based on the predicted center error value to determine the center of the compensated combined target image. When the predicted center error value is less than zero, a second compensation method is performed based on the predicted center error value to determine the center of the compensated combined target image.

5. A three-dimensional calibration device according to claim 1, characterized in that, Building and training a prediction model specifically includes: The obtained center error set is randomly divided into a training set and a test set, with the training set accounting for 80% of the center error set and the test set accounting for 20% of the center error set. The training set and the test set are then normalized. The input of the prediction model is determined based on the pixel information and rotation information of the positioning center, and the output layer of the prediction model is determined based on the error prediction value of the positioning center. An optimization function is constructed to optimize the initial parameters of the prediction model and determine the optimal parameters. The prediction model is trained based on the optimal parameters and the training set, and the training results of the prediction model are compared with the preset error based on the test set. Iterative calculations are performed until the training results meet the preset error to obtain the trained prediction model, i.e., the center error prediction model.

6. A three-dimensional calibration device according to claim 1, characterized in that, Specifically, 5.4 includes: The positional relationship of the combined target images in the first coordinate system is determined based on the compensated positioning center, and is denoted as the first positional relationship. The positional relationship between each combined target image in the first coordinate system is determined based on the first positional relationship, and is denoted as the second positional relationship. The second positional relationship is used to determine the actual positional relationship of the combined target based on a preset transformation relationship, and is denoted as the third positional relationship. The second parameter is determined by updating and compensating the first parameter based on the third positional relationship.

7. A three-dimensional calibration device according to claim 1, characterized in that, The calibration unit further includes: The three-dimensional calibration device performs posture transformation based on a preset number of transformations M and combines it with a binocular camera to complete the calibration, where M≥15.

Images