Fuel assembly non-overlapping field of view multi-camera stitching calibration method

Abstract

The application discloses a fuel assembly non-overlapping field-of-view multi-camera splicing calibration method, which comprises the following steps: S1, arranging a plurality of camera groups in the vertical direction of the transmission channel of the fuel assembly; S2, arranging calibration pieces matched with the fuel assembly in the transmission channel; S3, acquiring actual measurement images of the opposite calibration pieces by each camera group, and combining the actual measurement images of the calibration pieces with standard images to calculate distortion parameters of each camera group; S4, according to the calibration pieces, constructing a projection matrix of the actual measurement images of the calibration pieces acquired by each camera group to a global panoramic image; and S5, according to the distortion parameters and the projection matrix, performing panoramic splicing on the actual measurement images of the fuel assembly acquired by the camera groups to generate a panoramic image. According to the calibration method, the cameras are arranged without overlapping, installation space is saved, the distortion parameters are solved with the aid of the calibration pieces, the projection matrix is constructed, the panoramic image is accurately spliced, and the detection precision and the detection efficiency of the nuclear power fuel assembly are improved.

Description

Technical Field

[0001] This invention relates to the field of non-destructive testing of nuclear fuel assemblies and computer vision, and in particular to a method for calibrating fuel assemblies using multi-camera stitching with non-overlapping fields of view. Background Technology

[0002] After long-term operation under high temperature and pressure, nuclear fuel assemblies may experience bending deformation or excessive growth, potentially affecting normal loading and unloading operations. Therefore, high-resolution imaging is necessary for dimensional inspection. However, existing visual measurement technologies have the following limitations: Single camera + mechanical rail: The traditional approach is to mount the camera on a mechanical rail and synthesize a panoramic image by scanning the surface of the components by moving the camera. However, this method is slow, has poor accuracy, and is bulky, and is increasingly unsuitable for the needs of nuclear power plants.

[0003] Single camera + video: A single camera is limited by its field of view and can only acquire partial images. The entire image needs to be captured by moving the components. Although a panoramic image can be synthesized by stitching together video streams, the surface of the fuel components has weak texture features (especially the smooth metallic surface of new components), resulting in the measurement accuracy of the synthesized panoramic image not meeting the requirements.

[0004] Multi-camera arrays with overlapping areas: Taking a fuel assembly deformation measurement system based on a camera array as an example, 16 cameras cover the assembly area, and a global image of the assembly is stitched together based on the features of overlapping areas. Although this approach achieves high precision, the excessive number of cameras results in a bulky device that is difficult to deploy in narrow fuel transport channels.

[0005] Existing technologies are either too slow to meet real-time requirements, have large splicing errors that do not meet accuracy requirements, or are too large to be used synchronously in the fuel transfer pool. None of these technologies can meet the requirements of a high-efficiency, high-precision component size measurement system in the fuel transfer channel during nuclear power plant loading and unloading and fuel transfer. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a method for calibrating fuel assembly with non-overlapping field of view multi-camera stitching.

[0007] The technical solution adopted by this invention to solve its technical problem is: to construct a multi-camera stitching calibration method for fuel assembly non-overlapping field of view, which includes the following steps: S1. Several camera groups are arranged in the vertical direction of the fuel assembly transmission channel, and the fields of view of the camera groups do not overlap. S2. A calibration component matching the fuel assembly is provided in the transmission channel; the calibration component is provided with a standard pattern and is positioned directly opposite the camera group; S3. By acquiring the measured image of the corresponding calibration component for each camera group, and combining the measured image of the calibration component with the standard graphic, the distortion parameters of each camera group are calculated. S4. Based on the calibration component, construct a projection matrix that maps the measured image of the calibration component acquired by each camera group to the global panoramic image; S5. Based on the distortion parameters and the projection matrix, perform panoramic stitching on the measured images of the fuel assembly collected by the camera group to generate a panoramic image.

[0008] In some embodiments, in step S1, the camera group consists of multiple cameras connected in series. The multiple cameras are arranged vertically at equal intervals along the vertical direction of the transmission channel. All cameras in the camera group achieve synchronous acquisition of the measured image through hardware synchronous triggering.

[0009] In some embodiments, in step S3, the standard pattern of the calibration component is a checkerboard pattern. The camera group acquires a measured image containing the checkerboard pattern, and the measured image is subjected to corner point extraction and distortion coefficient calculation. The corner point acquisition method is as follows: combining the local gray-level gradient features of the measured image with the Harris response operator, screening the checkerboard corner point candidate region, using the checkerboard geometric topological constraints to eliminate invalid candidate points, obtaining pixel-level initial corner points, and then fusing the least squares method and Newton's iteration algorithm to iteratively solve the sub-pixel coordinates corresponding to the initial corner points.

[0010] In some embodiments, in step S3, the radial distortion coefficients are solved by combining the measured images with standard graphics and using a hierarchical step-by-step iterative method combined with nonlinear joint optimization to obtain the distortion parameters corresponding to the camera group. The method is as follows: S31. Solve only k1, with k2=0 pre-set. With minimizing the image reprojection error as the optimization objective, use the Newton-Raphson iterative algorithm to solve for the optimal value of the dominant radial distortion coefficient k1. S32. Fix the optimal value of k1 obtained in the first step, keep the optimization objective of minimizing the reprojection error unchanged, and iteratively solve for the optimal value of the secondary radial distortion coefficient k2. S33. After completing the step-by-step iterative solution, the nonlinear least squares algorithm is used to jointly fine-tune and optimize k1 and k2.

[0011] In some embodiments, the cost function of the reprojection error is: Where k1 and k2 are radial distortion coefficients, and E(k1,k2) is the cost function of reprojection error, (u i ,v i ) is the actual coordinate of the pixel in the original distorted image, (ui model (k1,k2),v i model (k1,k2) are the ideal coordinates of the pixel after distortion model correction; The expression for solving the dominant radial distortion coefficient k1 using the Newton-Raphson iterative algorithm is as follows: Among them, k1 (n+1) This is the updated value of the radial distortion coefficient k1 after the (n+1)th iteration, k1 (n) It is the current value of the radial distortion coefficient k1 at the nth iteration, E'(k1) (n) ) is the cost function E in k1 (n) The first derivative at point E''(k1) (n) ) is the cost function E in k1 (n) The second derivative at point .

[0012] In some embodiments, in step S4, a projection matrix is ​​constructed based on the calibration component. The standard graphic of the calibration component is divided into multiple local calibration regions, each local calibration region corresponding to a group of cameras. A rigid transformation relationship is established between the coordinate system of the local calibration region and the coordinate system of the global panoramic image. The process of establishing the rigid transformation relationship is as follows: S41. Measure the four edge corners of each chessboard calibration board using a coordinate measuring machine to obtain the global three-dimensional coordinate data corresponding to all corners, and use this as the reference data from the multi-local coordinate system to the global coordinate system. S42. Calculate the geometric centroid based on the global 3D coordinates of the four edge corner points. Use the geometric centroid as the origin of the local coordinate system. Construct initial local plate surface basis vectors based on the two adjacent edges of the calibration plate plane. Perform orthogonalization processing on the initial local plate surface basis vectors to construct local orthogonal bases and global orthogonal bases respectively, and generate corresponding local basis matrices. With the global basis matrix Through the global basis matrix With the local basis matrix Solving for the rotation matrix R using orthogonal mapping operations; S43. Based on the obtained rotation matrix R, combined with the local centroid coordinates of the calibration plate in the local coordinate system... Global centroid coordinates in the global coordinate system obtained by coordinate measuring machine (CMM) measurement The translation vector t used to compensate for the position deviation is calculated. S44. Integrate the rotation matrix R and the translation vector t to construct a rigid transformation homogeneous matrix.

[0013] In some embodiments, the rotation matrix R is expressed as: The expression for the translation vector t used to compensate for the position deviation is as follows: The expression for constructing the rigid transformation homogeneous matrix is ​​as follows: in, It is a homogeneous matrix that maps points in the local coordinate system of the local calibration plate to the global coordinate system.

[0014] In some embodiments, a homography matrix is ​​used to characterize the projection mapping relationship between the camera image plane and the calibration board world plane. The homography matrix is ​​a 3×3 composite matrix that fuses the poses involved in the camera, and is solved by least squares method or singular value decomposition. The homography matrix satisfies the following: Where X i Y i It is the two-dimensional coordinate of the corner point in world coordinates, x i y i It is the pixel position of the corner point in the image coordinate system, s i H is the scale factor, and H is the homography matrix of the composite effects of poses within a 3×3 camera.

[0015] In some embodiments, in step S4, using the first camera as a reference, the ratio s1 of the checkerboard side length to the ideal geometric side length in its image is calculated, and the homography matrix scale of other cameras is unified accordingly, using H iG Backproject each image onto the global coordinate plane, H iG The solution expression is: Among them, H iG H is the global projection matrix. iB For each camera, the homography matrix is... T is a rigid transformation homogeneous matrix GB A 3×3 submatrix.

[0016] In some embodiments, in step S5, after obtaining a panoramic image by completing the panoramic stitching of the measured image based on the distortion parameters and projection matrix, an accuracy verification is carried out. The accuracy verification method is as follows: select multiple checkerboard or feature point regions of known size in the generated panoramic image, calculate the ratio difference between their pixel distance in the image space and the global physical distance, and verify the geometric consistency and overall measurement accuracy of the stitched image by comparing the consistency between the pixel scale and the real physical scale.

[0017] The present invention offers the following advantages: It deploys multiple sets of non-overlapping camera groups in the vertical direction of the fuel assembly transmission channel, providing full coverage imaging of the fuel assembly's vertical height without blind spots. This meets the panoramic inspection requirements of slender fuel assemblies. Furthermore, the non-overlapping field-of-view layout fundamentally avoids the image data redundancy and repetitive computation problems caused by traditional multi-camera overlapping acquisition, effectively reducing the computational load of image acquisition, transmission, and stitching algorithms, and improving online imaging processing efficiency. A calibration component is configured within the transmission channel, positioned directly opposite each camera group. This accurately simulates the actual imaging conditions of the fuel assembly, avoiding calibration failures caused by inconsistencies between the calibration and detection scenarios, and ensuring that each camera group can independently and completely acquire standard graphic information. During the calibration phase, the present invention uses only the measured images from the calibration component combined with standard graphics to solve for the distortion parameters of each camera group. Relying on the precise geometric dimensions and topological features of the standard graphics of the calibration component, the distortion parameters of each camera can be accurately solved, achieving high-precision distortion correction for a single camera. Meanwhile, this invention constructs a global projection matrix corresponding to each camera group based on the measured images of the calibration component, and establishes a rigid mapping relationship between each independent non-overlapping field of view and the global panoramic coordinate system using the standard geometric features of the calibration component, so as to achieve global unification of the scale, position and angle of the multi-camera field of view and ensure the seamless stitching accuracy of the non-overlapping field of view images. Attached Figure Description

[0018] To more clearly illustrate the technical solution of the present invention, the present invention will be further described below in conjunction with the accompanying drawings and embodiments. It should be understood that the following drawings only show some embodiments of the present invention and therefore should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort. In the drawings: Figure 1 This is a flowchart illustrating the multi-camera stitching calibration method for non-overlapping field of view of fuel assemblies in some embodiments of the present invention; Figure 2 This is a schematic diagram illustrating the application of the multi-camera stitching calibration method for non-overlapping field of view of fuel assemblies in some embodiments of the present invention; Figure 3 This is a schematic diagram of the installation of calibration components for the multi-camera stitching calibration method for non-overlapping field of view of fuel assemblies in some embodiments of the present invention. Detailed Implementation

[0019] To provide a clearer understanding of the technical features, objectives, and effects of this invention, specific embodiments are now described in detail with reference to the accompanying drawings. In the following description, it should be understood that the orientations or positional relationships indicated by terms such as "front," "rear," "upper," "lower," "left," "right," "longitudinal," "horizontal," "vertical," "horizontal," "top," "bottom," "inner," "outer," "head," and "tail" are based on the orientations or positional relationships shown in the accompanying drawings, and are constructed and operated in a specific orientation. They are only for the convenience of describing this technical solution and do not indicate that the device or element referred to must have a specific orientation; therefore, they should not be construed as limitations on this invention.

[0020] It should also be noted that, unless otherwise explicitly specified and limited, terms such as "installation," "connection," "linking," "fixing," and "setting" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. When an component is referred to as being "on" or "below" another component, the component can be located "directly" or "indirectly" on the other component, or there may be one or more intermediary components. The terms "first," "second," "third," etc., are only for the convenience of describing this technical solution and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, features defined with "first," "second," "third," etc., may explicitly or implicitly include one or more of that feature. For those skilled in the art, the specific meaning of the above terms in this invention can be understood according to the specific circumstances.

[0021] Please see Figure 1 This invention relates to a method for calibrating fuel assemblies using multi-camera stitching with non-overlapping fields of view, as described in some embodiments of the present invention. Specifically, it is a calibration method for high-precision stitching of images from a multi-camera array with non-overlapping fields of view for fuel assemblies. This technology is mainly applied to the calibration of high-precision measurements of parameters such as the external dimensions, deformation, and defects of fuel assemblies in nuclear power plants. It enables rapid and accurate detection in confined spaces such as transfer pools without affecting the efficiency of the loading and unloading process. The method includes the following steps: S1. Several camera groups are arranged in the vertical direction of the fuel assembly transmission channel, and the fields of view of the camera groups do not overlap. S2. Set a calibration piece in the transmission channel that matches the fuel assembly; the calibration piece has a standard graphic and is set facing the camera group; S3. Collect the measured images of the corresponding calibration components for each camera group, and calculate the distortion parameters of each camera group by combining the measured images of the calibration components with the standard graphics. S4. Based on the calibration components, construct a projection matrix that maps the measured images of the calibration components collected by each camera group to the global panoramic image; S5. Based on the distortion parameters and projection matrix, perform panoramic stitching on the measured images of the fuel assembly collected by the camera group to generate a panoramic image.

[0022] Understandably, this invention deploys multiple sets of non-overlapping camera groups in the vertical direction of the fuel assembly transmission channel, enabling full coverage imaging of the fuel assembly's vertical height without blind spots. This meets the panoramic inspection requirements of slender fuel assemblies. Simultaneously, the non-overlapping field-of-view layout fundamentally avoids the image data redundancy and repetitive computation problems caused by traditional multi-camera overlapping acquisition, effectively reducing the computational load of image acquisition, transmission, and stitching algorithms, and improving online imaging processing efficiency. Furthermore, a calibration component is configured within the transmission channel, positioned directly opposite each camera group. This accurately simulates the actual imaging conditions of the fuel assembly, avoiding calibration failures caused by inconsistencies between the calibration and detection scenarios, and ensuring that each camera group can independently and completely acquire standard graphic information. During the calibration phase, this invention uses only the measured images from the calibration component combined with standard graphics to solve for the distortion parameters of each camera group. Relying on the precise geometric dimensions and topological features of the standard graphics of the calibration component, the distortion parameters of each camera can be accurately solved, achieving high-precision distortion correction for a single camera. Meanwhile, this invention constructs a global projection matrix corresponding to each camera group based on the measured images of the calibration component, and establishes a rigid mapping relationship between each independent non-overlapping field of view and the global panoramic coordinate system using the standard geometric features of the calibration component, so as to achieve global unification of the scale, position and angle of the multi-camera field of view and ensure the seamless stitching accuracy of the non-overlapping field of view images.

[0023] In step S1, the camera group consists of multiple cameras connected in series. These cameras are arranged vertically at equal intervals along the transmission channel. All cameras in the camera group achieve synchronous acquisition of the measured images through hardware synchronization triggering. In a specific embodiment, the camera group consists of five cameras connected in series. The cameras are arranged vertically at intervals with no overlap in their fields of view. Frame synchronization is maintained through hardware triggering. Specifically, during nuclear power plant unloading, fuel assemblies are transferred from the reactor building to the fuel building. During nuclear power plant loading, fuel assemblies are transferred from the fuel building to the reactor building. During both loading and unloading, the fuel assemblies pass through the fuel transmission channel and are removed from the fuel tank at a non-uniform speed of approximately 12 m / s. Due to the confined environment and overhaul requirements, the equipment can only complete all data acquisition during this process. Figure 2 and Figure 3As shown, this method uses five cameras 2 connected in series for image acquisition. Before implementation, equipment needs to be deployed. Camera group 3 is arranged in the transmission channel of fuel assembly 1. Each camera group can acquire an image of one side of fuel assembly 1. Depending on the requirements, up to four camera groups 3 can be deployed to complete the simultaneous measurement of four sides. The cameras 2 are arranged vertically at intervals, with no overlap in the field of view. Frame synchronization is maintained by hardware triggering. They are clamped onto the four sides of the cuboid fuel assembly 1. To calibrate the series camera group 3, this method uses a custom-made rectangular calibration component 4 with the same width and length as fuel assembly 1. A checkerboard calibration plate 5 is installed at the position directly opposite each camera 2. The position of each checkerboard calibration plate 5 is precisely calibrated by a coordinate measuring machine at the factory. During use, the series camera group 3 and the special calibration component 4 are directly opposite each other, and the series camera group 3 completes the image acquisition work through synchronous triggering.

[0024] It is worth mentioning that commonly used calibration methods, such as the Zhang Zhengyou calibration method, establish a projective transformation (homography matrix) between image coordinates and world plane coordinates using multiple planar calibration plate images (usually a checkerboard pattern) in different poses, and then solve for the camera's intrinsic and extrinsic parameters from these relationships. However, this approach is not suitable due to the nonlinear distortion underwater and the high-temperature, high-radiation environment of nuclear power plant pools, which makes it impossible to obtain planar calibration plate images in different poses.

[0025] Although underwater imaging environments require light to pass through multiple media interfaces such as water, protective glass, air, and lens, resulting in multiple nonlinear refractions and thus violating the ideal pinhole imaging model, the imaging conditions in this system design are characterized by "static distance and fixed posture": the camera position is fixed, the fuel assembly is rigidly constrained in the fuel tank, and its posture and distance only change slightly. In underwater refraction environments, the traditional Brown-Conrady model is still applicable, but it requires actual calibration points to compensate for errors caused by interface refraction. This paper proposes a simpler and more stable approximate calibration method. This method, specifically designed for fixed distance, fixed posture, and underwater use, focuses only on correcting image distortion and establishing the projection matrix from the image to the panoramic view.

[0026] It is known that underwater imaging by a camera is mainly caused by geometric distortion due to refraction. According to Snell's law: Where n w Let n be the refractive index of water. a Let θ be the refractive index of air. w and θ a This corresponds to the incident angle underwater and in air. Therefore, it can be seen that refraction primarily causes radial distortion in underwater imaging. For the classic camera distortion model Brown-Conrady: in, Let represent the radial position of a pixel in an ideal, distortion-free imaging model, and 'r' represent the radial position of a pixel with distortion. k1, k2, and k3 are the radial distortion coefficients. For cases of non-severe distortion, the higher-order distortion represented by k3 can usually be ignored, and only k1 and k2 need to be considered, with k1 playing the dominant role and k2 being a secondary optimization. To prevent conventional methods from easily getting trapped in local optima and to improve the solution speed, this method proposes a step-by-step iterative strategy for solving distortion, incorporating Newton's iteration.

[0027] In step S3, the standard graphic of the calibration piece is a checkerboard pattern. A camera group acquires measured images containing the checkerboard pattern, and corner points are extracted and distortion coefficients are calculated from these images. The corner point acquisition method involves combining local gray-level gradient features of the measured image with the Harris response operator to filter candidate checkerboard corner point regions. Invalid candidate points are eliminated using the checkerboard's geometric topological constraints to obtain initial pixel-level corner points. Then, the least squares method and Newton's iteration algorithm are fused to iteratively solve for the sub-pixel coordinates corresponding to the initial corner points. This step can be implemented using built-in functions in OpenCV or MATLAB. Next, using the known grid spacing of the calibration board, an ideal coordinate set for each corner point in an ideal planar coordinate system is generated. The ideal coordinates use the center of the calibration board or a fixed angle as the origin, and the planar coordinates of each corner point are calculated in units of the grid side length for subsequent distortion calculation. Understandably, by employing a multi-level corner extraction strategy that combines grayscale gradient and Harris operator initial selection, geometric constraint screening, and sub-pixel iterative optimization, pseudo-corner points caused by underwater image noise and refraction interference are effectively eliminated. At the same time, the corner point positioning accuracy is improved to the sub-pixel level, providing high-precision raw data support for subsequent distortion coefficient calculation, homography matrix calculation, and projection matrix construction, thereby reducing calibration errors from the source.

[0028] In step S3, combining the measured images and standard graphics, the radial distortion coefficients are solved using a hierarchical step-by-step iterative approach combined with nonlinear joint optimization to obtain the distortion parameters corresponding to the camera group. The method is as follows: S31. Solve only k1, with k2=0 pre-set. With minimizing the image reprojection error as the optimization objective, use the Newton-Raphson iterative algorithm to solve for the optimal value of the dominant radial distortion coefficient k1. S32. Fix the optimal value of k1 obtained in the first step, keep the optimization objective of minimizing the reprojection error unchanged, and iteratively solve for the optimal value of the secondary radial distortion coefficient k2. S33. After completing the step-by-step iterative solution, the nonlinear least squares algorithm is used to jointly fine-tune and optimize k1 and k2.

[0029] A hierarchical step-by-step iterative optimization strategy is adopted to solve the biradial distortion coefficients. First, the dominant distortion is coarsely calculated, then the secondary distortion is finely adjusted, and finally the global optimization is combined. This avoids the defects of traditional synchronous iterative solutions, such as easy divergence, easy getting trapped in local optima, and inaccurate parameter solutions. The hierarchical constraints improve the stability and convergence accuracy of underwater nonlinear distortion solutions and significantly improve the underwater image distortion correction effect.

[0030] The cost function for reprojection error is: Where k1 and k2 are radial distortion coefficients, and E(k1,k2) is the cost function of reprojection error, (u i ,v i ) is the actual coordinate of the pixel in the original distorted image, (u i model (k1,k2),v i model (k1,k2) is the ideal coordinate of a pixel after distortion model correction.

[0031] The expression for solving the dominant radial distortion coefficient k1 using Newton's iterative algorithm is as follows: Among them, k1 (n+1) This is the updated value of the radial distortion coefficient k1 after the (n+1)th iteration, k1 (n) It is the current value of the radial distortion coefficient k1 at the nth iteration, E'(k1) (n) ) is the cost function E in k1 (n) The first derivative at point E''(k1) (n) ) is the cost function E in k1 (n) The second derivative at point .

[0032] Specifically, the construction of the cost function for reprojection error needs to be based on the checkerboard image corner data extracted in step S2, where (u i ,v i (u) represents the actual pixel coordinates of corner points extracted from the checkerboard image in the original distorted image. This data is obtained by extracting corner point candidates through local gray-level gradient and Harris response function, followed by checkerboard geometric constraint screening and optimization using least squares and Newton iteration. It represents the actual imaging position of the corner points under the influence of underwater distortion; i model (k1,k2),v i model(k1,k2) represents the ideal coordinates of the corner point after distortion model correction. This data is based on the ideal coordinates of the corner point generated by the known grid spacing of the checkerboard calibration board, and is calculated by combining the radial distortion coefficients k1 and k2 obtained by step-by-step iterative solution. It represents the theoretical position of the corner point under the distortion-free ideal imaging model. The core function of the cost function is to quantify the deviation between the actual imaging position and the ideal imaging position of the corner point, provide an optimization target for the iterative solution of the radial distortion coefficients k1 and k2, ensure the accuracy of underwater distortion correction, and lay the foundation for subsequent image stitching and size measurement.

[0033] The solution to the iterative formula for Newton's iteration for k1 requires the derivative data of the reprojection error cost function E, as well as the initial and current iteration values ​​of the radial distortion coefficient k1; where k1 (n) E'(k1) represents the current value of the radial distortion coefficient k1 in the nth iteration. This data is an intermediate parameter in the iteration process, representing the approximate value of the radial distortion coefficient obtained in the current iteration step. The initial value can be set based on experience with the underwater imaging environment or obtained through rough calculation. (n) ) is the cost function E in k1 (n) The first derivative at point E''(k1) (n) ) is the cost function E in k1 (n) The second derivative at k1, both are calculated based on the expression of the cost function E, combined with the actual and ideal pixel coordinates of the corner point, and represent the cost function at the current k1. (n) The rate of change and curvature at the point are determined. This iterative formula minimizes the cost function E by gradually updating the value of k1, ultimately obtaining the accurate radial distortion coefficient k1. This avoids the drawback of conventional solution methods easily getting trapped in local optima, while improving the efficiency of solving the distortion coefficient.

[0034] In step S4, a projection matrix is ​​constructed based on the calibration component. The standard graphic of the calibration component is divided into multiple local calibration regions, each corresponding to a group of cameras. A rigid transformation relationship is established between the coordinate system of the local calibration region and the coordinate system of the global panoramic image. The process of establishing the rigid transformation relationship is as follows: S41. Measure the four edge corners of each chessboard calibration board using a coordinate measuring machine to obtain the global three-dimensional coordinate data corresponding to all corners, and use this as the reference data from the multi-local coordinate system to the global coordinate system. S42. Calculate the geometric centroid based on the global 3D coordinates of the four edge corners. Use the geometric centroid as the origin of the local coordinate system. Construct initial local plate surface basis vectors based on the two adjacent edges of the calibration plate plane. Perform orthogonalization processing on the initial local plate surface basis vectors to construct local orthogonal bases and global orthogonal bases respectively, and generate corresponding local basis matrices. With the global basis matrix Through the global basis matrix With local basis matrix Solving for the rotation matrix R using orthogonal mapping operations; S43. Based on the obtained rotation matrix R, combined with the local centroid coordinates of the calibration plate in the local coordinate system. Global centroid coordinates in the global coordinate system obtained by coordinate measuring machine (CMM) measurement The translation vector t used to compensate for the position deviation is calculated. S44. Integrate the rotation matrix R and the translation vector t to construct a rigid transformation homogeneous matrix.

[0035] The R expression is: The expression for the translation vector t used to compensate for positional deviation is calculated as follows: The expression for constructing the homogeneous matrix of a rigid transformation is: Among them, T GB It is a rigid transformation homogeneous matrix that maps the midpoint of the local calibration plate coordinate system to the global coordinate system.

[0036] Specifically, The global three-dimensional coordinates of the four edge corner points of each chessboard calibration board measured by a coordinate measuring machine are constructed. This data represents the precise spatial position of each corner point in the global coordinate system and is the core basis for establishing the global coordinate system benchmark. Based on the position of the corner point in its own local coordinate system, this data represents the relative position of the corner point in the local coordinate system of the corresponding chessboard calibration board, and is used to reflect the attitude of the local coordinate system. The essence is to unify the pose of the local coordinate system to the global coordinate system through orthogonal mapping between the global and local bases, thus solving the problem of inconsistent poses between the local and global coordinate systems of different cameras. The rigid transformation homogeneous matrix integrates the rotation matrix R and the translation vector t, and this matrix is ​​the subsequent global projection matrix H. iG The solution provides key attitude and position parameters, which is the core link between the precise measurement of the coordinate measuring machine and the global mapping of the image, ensuring that multi-camera images with non-overlapping fields of view can be stitched together based on a unified global coordinate system.

[0037] Understandably, for multiple cameras arranged in a series with no overlapping fields of view, this method uses a unified global calibration board to unify the coordinates of each camera. Each global calibration board contains several local calibration boards, each corresponding to a camera. After each local calibration board is assembled onto the global calibration board, several control points are measured using a coordinate measuring machine (CMM) to obtain the precise positional relationship of the board in the global coordinate system. For each local calibration board, the global 3D coordinates of its four corner points are measured using a CMM. With the centroids of the four corner points as the center, local board surface basis vectors are constructed using two adjacent edges, and orthogonalization is performed on them to obtain orthogonal bases for both local and global coordinates. The rotation matrix is ​​obtained by multiplying the global basis matrix by the local basis matrix. Based on the measured global coordinates from the CMM, an orthogonal basis matrix is ​​constructed. The rotation matrix is ​​solved through orthogonal mapping of the basis matrix, and the translation vector is solved by combining the centroid alignment mechanism. A standardized rigid homogeneous transformation matrix is ​​constructed to achieve precise unification of all local calibration board coordinate systems to the global coordinate system. This method addresses the stitching misalignment issue caused by varying poses and positional offsets among multiple cameras in a non-overlapping field of view, providing a precise pose and position reference for globally unified mapping of multiple cameras. All parameters are based on measured physical data, exhibiting significantly higher robustness and accuracy than pure image calibration methods. This approach avoids local optima, has fewer iterative variables, and improves solution efficiency. It also facilitates future expansion should higher-order distortion coefficients (k3) be required.

[0038] In step S4, a homography matrix is ​​used to characterize the projection mapping relationship between the camera image plane and the calibration board world plane. The homography matrix is ​​a 3×3 composite matrix that fuses the poses of the camera. It is solved by least squares method or singular value decomposition. The homography matrix satisfies the following: Where X i Y i It is the two-dimensional coordinate of the corner point in world coordinates, x i y i It is the pixel position of the corner point in the image coordinate system, s i H is the scale factor, and H is the homography matrix of the composite effects of poses within a 3×3 camera.

[0039] Understandably, by combining the measured physical coordinates of the three coordinate measuring machine to solve the homography matrix, the projection mapping relationship between the image plane of each camera and the physical calibration plane can be accurately established. Noise interference can be suppressed by least squares or singular value decomposition to ensure the stability and accuracy of the homography matrix solution, and to provide core mapping parameters for subsequent scale unification and projection matrix construction.

[0040] In step S4, using the first camera as a reference, the ratio s1 of the checkerboard side length to the ideal geometric side length in its image is calculated. Based on this, the homography matrix scale of other cameras is unified, using H... iG Backproject each image onto the global coordinate plane, H iG The solution expression is: Among them, H iG H is the global projection matrix. iB For each camera, the homography matrix is... T is a rigid transformation homogeneous matrix GB A 3×3 submatrix.

[0041] Understandably, the scaling of the homography matrix needs to be based on the first camera. Using the precisely calibrated checkerboard dimensions from a coordinate measuring machine, the ratio s1 of the checkerboard side length to the ideal geometric side length in the first camera's image is calculated. This ratio is then used as a benchmark to unify the homography matrix scale of other cameras, ensuring that all cameras have the same imaging scale. This provides a data foundation for the subsequent construction of the projection matrix and coordinate unification mapping. Based on this scale-unified homography matrix, combined with the rigid transformation homogeneous matrix T... GB The inverse of the 3×3 submatrix is ​​used to construct the global projection matrix H from the image to the panoramic image. iG Homography matrix H iB The rigid transformation homogeneous matrix T of . GB The product of the 3×3 submatrices essentially maps the scale-unified image coordinates of each camera to the global coordinate system through a rigid transformation, thereby achieving unified stitching of multi-camera images with non-overlapping fields of view. The scale-unified homography matrix H... iB This ensures scale consistency when images from each camera are mapped to the global coordinate system. This ensures the consistency of pose when the images from each camera are mapped to the global coordinate system, and the two work together to form the global projection matrix H. iG Through the global projection matrix H iG Complete the back-projection transformation of images from each camera to the global coordinate plane, providing coordinate mapping support for high-precision panoramic image stitching.

[0042] In step S5, after obtaining the panoramic image by completing the panoramic stitching of the measured image based on the distortion parameters and projection matrix, accuracy verification is carried out. The accuracy verification method is as follows: select multiple checkerboard or feature point regions of known size in the generated panoramic image, calculate the ratio difference between their pixel distance in the image space and the global physical distance, and verify the geometric consistency and overall measurement accuracy of the stitched image by comparing the consistency between the pixel scale and the real physical scale.

[0043] Specifically, in step S5, after completing distortion correction for each camera, unifying the coordinates of the local calibration board, and solving the geometric relationships of the cameras, panoramic image stitching and accuracy verification are further performed. Based on the homography matrix of each camera and its corresponding local calibration board, the pixel coordinates of each camera image can be mapped to the corresponding local calibration board plane coordinate system. Then, using the homogeneous transformation matrix from the local calibration board to the global calibration board obtained from the three coordinate measuring machines, the transformation from the local plane to the global unified coordinate system is achieved. After two levels of coordinate mapping, the pixels of any camera can be uniformly expressed as points in the global calibration board coordinate system. After generating the panoramic image, to verify geometric consistency and measurement accuracy, multiple checkerboard or feature point regions of known size are selected in the panoramic image, and the ratio difference between their pixel distance in image space and global physical distance is calculated. This method achieves unified mapping of multi-camera images through geometric transformation chain fusion, and can still construct a high-precision panoramic image under non-overlapping field of view conditions, while taking into account imaging consistency and physical size consistency, providing a high-precision two-dimensional mapping foundation for subsequent target recognition and size measurement.

[0044] In summary, this invention discloses a calibration method for high-precision image stitching using a multi-camera array with non-overlapping fields of view, applicable to fuel assemblies. This method is similar to the "multi-camera array with overlapping areas" scheme, but the improved calibration method reduces the number of cameras to five, thus minimizing the equipment size for deployment in narrow fuel transport channels. It enables automated calibration and stitching, allowing simultaneous measurement during loading and unloading, significantly improving work efficiency. It solves the following problems: underwater distortion correction is a technical challenge; this technology achieves highly robust distortion correction in single-image scenarios. Due to the reduced number of cameras, there is no shared field of view among them; this method designs a non-overlapping area calibration method to achieve rapid underwater calibration of the equipment. Furthermore, this method performs intelligent appearance inspection and dimensional measurement simultaneously with unloading, intelligently identifying fuel surface anomalies to ensure core fuel safety. It also measures the assemblies' external dimensions, enabling timely detection of assembly deformation and anomalies, and timely detection of post-irradiation fuel characteristics exceeding expectations, ensuring fuel safety. Simultaneously, it accumulates a large amount of post-irradiation performance data for fuel assemblies.

[0045] In summary, this multi-camera stitching calibration method for fuel assemblies with non-overlapping fields of view effectively solves the technical problems of inconsistent calibration benchmarks, low underwater distortion correction accuracy, difficulty in stitching non-overlapping fields of view, and poor measurement reliability in existing technologies, demonstrating significant technical advantages. This invention employs calibration components matched to the fuel assembly and combines a coordinate measuring machine (CMM) to complete high-precision positioning and calibration of the checkerboard calibration plate. Synchronous triggering of image acquisition by the cameras ensures the consistency and accuracy of data acquired by multiple cameras, making it suitable for full-domain detection scenarios of large-size, long cylindrical fuel assemblies. Simultaneously, this invention relies on checkerboard standard geometric parameters combined with a layered, step-by-step iterative optimization strategy to solve for the underwater radial distortion coefficient, specifically correcting nonlinear distortion caused by medium refraction and lens distortion in underwater imaging, effectively improving image quality. Furthermore, this invention uses the measured global three-dimensional coordinates of the CMM as a rigid benchmark, establishing a rigid transformation relationship between the local coordinate system and the global coordinate system of the multiple cameras. Through homography matrix solving and unified calibration of the scale of the multiple cameras, a global projection matrix is ​​constructed, effectively eliminating stitching misalignment, geometric distortion, and image fragmentation problems caused by differences in the attitude and scale of the multiple cameras. Furthermore, this invention verifies accuracy through scale comparison of standard physical size feature regions, achieving a closed-loop detection process encompassing calibration, image mapping and stitching, and accuracy verification, effectively ensuring the geometric consistency and physical measurement accuracy of the panoramic image. In summary, this solution improves the calibration stability and stitching accuracy of underwater panoramic imaging of fuel assemblies, meeting the requirements for high-precision appearance inspection, dimensional measurement, and defect identification of nuclear power fuel assemblies, demonstrating strong engineering practicality and scalability.

[0046] It is understood that the above embodiments only illustrate preferred embodiments of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can freely combine the above technical features without departing from the concept of the present invention, and can also make several modifications and improvements, all of which fall within the protection scope of the present invention. Therefore, all equivalent transformations and modifications made with respect to the scope of the claims of the present invention should fall within the scope of the claims of the present invention.

Claims

1. A method for calibrating a fuel assembly with non-overlapping field of view using multiple cameras, characterized in that, Includes the following steps: S1. Several camera groups are arranged in the vertical direction of the fuel assembly transmission channel, and the fields of view of the camera groups do not overlap. S2. A calibration component matching the fuel assembly is provided in the transmission channel; the calibration component is provided with a standard pattern and is positioned directly opposite the camera group; S3. By acquiring the measured image of the corresponding calibration component for each camera group, and combining the measured image of the calibration component with the standard graphic, the distortion parameters of each camera group are calculated. S4. Based on the calibration component, construct a projection matrix that maps the measured image of the calibration component acquired by each camera group to the global panoramic image; S5. Based on the distortion parameters and the projection matrix, perform panoramic stitching on the measured images of the fuel assembly collected by the camera group to generate a panoramic image.

2. The method for non-overlapping field-of-view multi-camera stitching calibration of fuel assembly according to claim 1, characterized in that, In step S1, the camera group consists of multiple cameras connected in series. The multiple cameras are arranged vertically at equal intervals along the vertical direction of the transmission channel. All cameras in the camera group achieve synchronous acquisition of the measured images through hardware synchronous triggering.

3. The method for non-overlapping field-of-view multi-camera stitching calibration of fuel assembly according to claim 1, characterized in that, In step S3, the standard pattern of the calibration component is a checkerboard pattern. The camera group acquires a measured image containing the checkerboard pattern, and the measured image is used to extract corner points and solve for distortion coefficients. The corner point acquisition method is as follows: combining the local gray-level gradient features of the measured image with the Harris response operator, screening the checkerboard corner point candidate region, using the checkerboard geometric topological constraints to eliminate invalid candidate points, obtaining pixel-level initial corner points, and then fusing the least squares method and Newton's iteration algorithm to iteratively solve the sub-pixel coordinates corresponding to the initial corner points.

4. The method for non-overlapping field-of-view multi-camera stitching calibration of fuel assembly according to claim 1, characterized in that, In step S3, combining the measured images and standard graphics, the radial distortion coefficients are solved using a hierarchical step-by-step iterative approach combined with nonlinear joint optimization to obtain the distortion parameters corresponding to the camera group. The method is as follows: S31. Solve only k1, with k2=0 pre-set. With minimizing the image reprojection error as the optimization objective, use the Newton-Raphson iterative algorithm to solve for the optimal value of the dominant radial distortion coefficient k1. S32. Fix the optimal value of k1 obtained in the first step, keep the optimization objective of minimizing the reprojection error unchanged, and iteratively solve for the optimal value of the secondary radial distortion coefficient k2. S33. After completing the step-by-step iterative solution, the nonlinear least squares algorithm is used to jointly fine-tune and optimize k1 and k2.

5. The method for non-overlapping field-of-view multi-camera stitching calibration of fuel assembly according to claim 4, characterized in that, The cost function for the reprojection error is: Where k1 and k2 are radial distortion coefficients, and E(k1,k2) is the cost function of reprojection error, (u i ,v i ) is the actual coordinate of the pixel in the original distorted image, (u i model (k1,k2),v i model (k1,k2) are the ideal coordinates of the pixel after distortion model correction; The expression for solving the dominant radial distortion coefficient k1 using the Newton-Raphson iterative algorithm is as follows: Among them, k1 (n+1) This is the updated value of the radial distortion coefficient k1 after the (n+1)th iteration, k1 (n) It is the current value of the radial distortion coefficient k1 at the nth iteration, E'(k1) (n) ) is the cost function E in k1 (n) The first derivative at point E''(k1) (n) ) is the cost function E in k1 (n) The second derivative at point .

6. The method for non-overlapping field-of-view multi-camera stitching calibration of fuel assembly according to claim 1, characterized in that, In step S4, a projection matrix is ​​constructed based on the calibration component. The standard graphic of the calibration component is divided into multiple local calibration regions, each corresponding to a group of cameras. A rigid transformation relationship is established between the coordinate system of the local calibration region and the coordinate system of the global panoramic image. The process of establishing the rigid transformation relationship is as follows: S41. Measure the four edge corners of each chessboard calibration board using a coordinate measuring machine to obtain the global three-dimensional coordinate data corresponding to all corners, and use this as the reference data from the multi-local coordinate system to the global coordinate system. S42. Calculate the geometric centroid based on the global 3D coordinates of the four edge corner points. Use the geometric centroid as the origin of the local coordinate system. Construct initial local plate surface basis vectors based on the two adjacent edges of the calibration plate plane. Perform orthogonalization processing on the initial local plate surface basis vectors to construct local orthogonal bases and global orthogonal bases respectively, and generate corresponding local basis matrices. With the global basis matrix Through the global basis matrix With the local basis matrix Solving for the rotation matrix R using orthogonal mapping operations; S43. Based on the obtained rotation matrix R, combined with the local centroid coordinates of the calibration plate in the local coordinate system... Global centroid coordinates in the global coordinate system obtained by coordinate measuring machine (CMM) measurement The translation vector t used to compensate for the position deviation is calculated. S44. Integrate the rotation matrix R and the translation vector t to construct a rigid transformation homogeneous matrix.

7. The method for non-overlapping field-of-view multi-camera stitching calibration of fuel assembly according to claim 6, characterized in that, The expression for the rotation matrix R is: The expression for the translation vector t used to compensate for the position deviation is as follows: The expression for constructing the rigid transformation homogeneous matrix is ​​as follows: in, It is a homogeneous matrix that maps points in the local coordinate system of the local calibration plate to the global coordinate system.

8. The method for non-overlapping field-of-view multi-camera stitching calibration of fuel assembly according to claim 1, characterized in that, In step S4, a homography matrix is ​​used to characterize the projection mapping relationship between the camera image plane and the calibration board world plane. The homography matrix is ​​a 3×3 composite matrix that fuses the poses of the camera. It is solved by least squares method or singular value decomposition. The homography matrix satisfies the following: Where X i Y i It is the two-dimensional coordinate of the corner point in world coordinates, x i y i It is the pixel position of the corner point in the image coordinate system, s i H is the scale factor, and H is the homography matrix of the composite effects of poses within a 3×3 camera.

9. The method for non-overlapping field-of-view multi-camera stitching calibration of fuel assembly according to claim 1, characterized in that, In step S4, using the first camera as a reference, the ratio s1 of the checkerboard side length to the ideal geometric side length in its image is calculated. Based on this, the homography matrix scale of other cameras is unified, using H... iG Backproject each image onto the global coordinate plane, H iG The solution expression is: Among them, H iG H is the global projection matrix. iB For each camera, the homography matrix is... T is a rigid transformation homogeneous matrix GB A 3×3 submatrix.

10. The method for non-overlapping field-of-view multi-camera stitching calibration of fuel assembly according to claim 1, characterized in that, In step S5, after obtaining the panoramic image by completing the panoramic stitching of the measured image based on the distortion parameters and projection matrix, accuracy verification is carried out. The accuracy verification method is as follows: select multiple checkerboard or feature point regions of known size in the generated panoramic image, calculate the ratio difference between their pixel distance in the image space and the global physical distance, and verify the geometric consistency and overall measurement accuracy of the stitched image by comparing the consistency between the pixel scale and the real physical scale.

Images