Flexible target and calibration method

Abstract

The invention discloses a flexible target and a calibration method. The flexible target comprises a plurality of target parts, and the target parts are movably connected in the mode that the shape and the relative position relation can be changed. Mark point patterns are arranged on the plurality of target parts; the spatial position relations of the mark point patterns on the multiple target parts can be mutually mapped according to shape and relative position relation change parameters when the shapes and the relative position relations of the multiple target parts are changed. According to the calibration method, the flexible target is adopted for calibration. When the target part of the flexible target is formed into a required form in a manner of changing the shape and the relative position relationship, the cameras in different directions can acquire required target images. The calibration method combines the structural characteristics of the flexible target and the spatial distortion coupling model, not only optimizes the geometrical relationship between the camera and the target, but also can dynamically correct the distortion along with the spatial position, thereby still keeping high precision in a multi-camera and multi-angle scene.

Description

Technical Field

[0001] This invention relates to the field of three-dimensional imaging technology, and in particular to a flexible target and calibration method. Background Technology

[0002] In applications such as 3D vision measurement, industrial inspection, and robot vision, multiple cameras often need to work simultaneously to acquire information about the target from different perspectives. To ensure that all cameras perform spatial reconstruction in a unified coordinate system, system calibration is essential.

[0003] Traditional calibration methods often use planar targets (such as checkerboard or dot arrays). However, in multi-camera systems, the camera installation angles vary, and some cameras may have large angles, obstructions, or large parallaxes, making it difficult for a single planar target to be effectively observed by all cameras simultaneously. Summary of the Invention

[0004] Therefore, it is necessary to provide a flexible target and calibration method to address the problem that multi-camera calibration cannot effectively observe targets due to varying orientations.

[0005] A flexible target, comprising:

[0006] Multiple target components, which are movably connected in a manner that allows for changes in shape and relative positional relationship;

[0007] The multiple target components are provided with marker patterns; the spatial positional relationship of the marker patterns on the multiple target components can be mapped to each other according to the shape and relative positional relationship change parameters when the multiple target components change shape and relative positional relationship.

[0008] In one embodiment, the plurality of target components are rectangular planar targets of the same shape, and the plurality of rectangular planar targets are flattened or folded to form a single planar target or a polyhedral column target.

[0009] In one embodiment, the plurality of target components are connected by at least one of slots, snaps, or hinges.

[0010] A calibration method, using the aforementioned flexible target to calibrate cameras located at multiple different orientations, includes:

[0011] Each camera captures an image of the pre-set target, obtaining the original pixel coordinates of the target point in the pixel coordinate system corresponding to each camera.

[0012] Initial estimates of the intrinsic and extrinsic parameters of each camera are made based on the original pixel coordinates of the marker points.

[0013] Spatial distortion coupling modeling is performed using initial intrinsic and extrinsic parameters, and the landmark points are projected onto the pixel coordinate system to obtain perspective pixel coordinates.

[0014] By utilizing the difference between the perspective pixel coordinates and the original pixel coordinates, a global reprojection error objective function is constructed;

[0015] Based on the aforementioned error objective function, the internal parameters, external parameters, and distortion parameters of each camera are iteratively optimized.

[0016] In one embodiment, it further includes:

[0017] The flexible target is unfolded or folded and fixed according to the position of the camera to obtain a target that can be captured by multiple cameras to capture the calibration pattern.

[0018] In one embodiment, the spatial distortion coupling modeling using initial intrinsic and extrinsic parameters includes:

[0019] Obtain the world coordinates of the marker points of each target component in the flexible target;

[0020] The world coordinates of the marker points are transformed to camera coordinates using rotation and translation parameters.

[0021] Transform camera coordinates to image coordinates using perspective parameters;

[0022] The corrected image coordinates are obtained by correcting the image coordinates using local distortion parameters;

[0023] Transform the corrected image coordinates to perspective pixel coordinates.

[0024] In one embodiment, obtaining the world coordinates of the marker points of each target component in the flexible target includes:

[0025] Obtain the world coordinates of a marker point on one of the target components in a flexible target;

[0026] Other target components in the flexible target obtain the world coordinates of their marker points based on their spatial positional relationship with the target component that determines the world coordinates.

[0027] In one embodiment, it further includes:

[0028] For the optimized parameters, the marker points are back-projected onto the image plane of each camera;

[0029] The back-projected points on the camera image plane are compared with the original pixels. If the back-projected points deviate from the original pixels by more than a threshold, the original pixels are removed so that they do not participate in the optimization of the error objective function.

[0030] In one embodiment, if the number of original pixels removed exceeds a set ratio of the total number of pixels in the image, the entire image is removed so that it does not participate in the optimization of the error objective function.

[0031] In one embodiment, the error objective function is iteratively optimized using the Levenberg-Marquardt algorithm.

[0032] When the target components of the aforementioned flexible target are shaped and their relative positions changed to achieve the desired form, cameras positioned at different locations can acquire the required target images. Through folding and interlocking, it can adapt to camera systems with varying numbers and angles. The multi-angle, multi-faceted structure ensures simultaneous observation by more cameras, improving calibration robustness. More modules can be assembled as needed, supporting large-scale N-camera calibration. It is foldable for storage and can be deployed on-site for convenient engineering applications. This calibration method combines the structural characteristics of the flexible target with a spatial distortion coupling model. The flexible target can adapt to different camera installation angles and workspaces through folding and interlocking, ensuring sufficient effective calibration points for each camera even in complex measurement environments. Furthermore, spatial distortion coupling allows the calibration process to not only optimize the geometric relationship between the camera and the target but also dynamically correct distortion based on spatial position, thus maintaining high accuracy in multi-camera, multi-angle scenarios. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of a flexible target structure according to one embodiment;

[0034] Figure 2A This is a schematic diagram of a flexible target unfolded into a planar target according to one embodiment;

[0035] Figure 2B A schematic diagram of a flexible target folded into a triangular prism shape according to one embodiment;

[0036] Figure 2C A schematic diagram of a flexible target folded into a tetrahedral cylindrical target according to one embodiment;

[0037] Figure 2D A schematic diagram of a flexible target folded into a hexahedral cylindrical target according to one embodiment;

[0038] Figure 3 This is a schematic diagram of a flexible target formed by splicing triangular target components to create a polyhedral target, as shown in one embodiment.

[0039] Figure 4 Here is a flowchart of a calibration method according to one embodiment;

[0040] Figure 5 This is a flowchart of step S104 in a calibration method according to one embodiment. Detailed Implementation

[0041] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Preferred embodiments of the invention are shown in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0042] Calibration of multi-camera systems is used to determine the intrinsic and extrinsic parameters of the cameras and the spatial relationships between them. The goal of calibration is to eliminate imaging errors and provide a precise geometric basis for applications such as multi-view image fusion and 3D reconstruction.

[0043] The calibration process mainly includes intrinsic parameter calibration and extrinsic parameter calibration. Intrinsic parameter calibration determines the inherent properties of each camera, including focal length, principal point coordinates, and distortion coefficients. These parameters describe the mathematical model by which the camera projects points in three-dimensional space onto a two-dimensional image plane. Intrinsic parameters can be obtained by capturing checkerboard images in different poses and calculating them. Extrinsic parameter calibration is used to solve for the relative poses between cameras, i.e., translation vectors and rotation matrices. Extrinsic parameters define the positional relationships of multiple cameras in a unified world coordinate system.

[0044] The checkerboard image described above is called a target. A target is a reference object used in multi-camera calibration to provide known spatial coordinates and establish the relationship between the world coordinate system and the camera image coordinate system. Besides checkerboard targets, there are also dot array targets, etc. Checkerboard targets are more commonly used because their corner points are easily identifiable and their accuracy is high. The size of each square on a checkerboard target is fixed, and the target surface must be flat and non-reflective. The camera captures images of the target and identifies its feature points as markers.

[0045] To address the issue of ineffective target observation in multi-camera systems due to varying camera orientations, this invention provides a flexible target. This flexible target comprises multiple target components. These components are movably connected in a manner that allows for changes in shape and relative position. Each target component is provided with a marker pattern. The spatial positional relationship of the marker patterns on the multiple target components can be mapped to each other based on changes in shape and relative positional parameters when the multiple target components change their shape and relative positional relationship.

[0046] When the target components are shaped into the desired form by changing their shape and relative positional relationships, they can be used by cameras in different positions to capture the desired target images.

[0047] When the target components change shape or relative position (e.g., folding, rotating), the spatial positions of the marker points on each component will change accordingly. The change in the spatial position of the marker points can be correlated using parameters related to changes in shape and position (such as rotation angle, folding amplitude, etc.). Specifically, if target component A and target component B are known to be rotatably connected, when target component A rotates 30° relative to target component B, the spatial coordinate transformation relationship of all marker points on target component A relative to the marker points on target component B can be calculated using the geometric parameters of the rotation axis (such as axis length, connection point coordinates). The target components should be made of a solid, rigid material, ensuring that the relative positions of their own marker points remain unchanged, with only the overall orientation changing as the target components move.

[0048] In some of these embodiments, such as Figure 1 As shown, the multiple target components are rectangular planar targets of the same shape. These rectangular planar targets are flattened or folded to form a single planar target or a polyhedral cylindrical target. Depending on different needs, the flexible target can be unfolded into a planar target. Figure 2A ), folded into a triangular prism target ( Figure 2B ), folded into a tetrahedral cylindrical target ( Figure 2C ) and hexahedral cylindrical targets ( Figure 2D ), etc. Other desired target shapes can also be formed, but the embodiments of this invention are not limited thereto. For example... Figure 3 As shown, the target component can also be triangular, and the surface of the target component module is printed with marking patterns. By splicing the modules together, a polyhedral target can be formed for omnidirectional camera calibration.

[0049] In some embodiments, the plurality of target components are connected by at least one of slots, snap-fits, or hinges. Slot connections allow for detachable assembly of components through the engagement of flanges and grooves, suitable for scenarios requiring rapid configuration changes (such as frequent switching of calibration modes in a laboratory). Snap-fit ​​connections utilize elastic structures to achieve rapid fixing and separation of components, balancing stability and convenience, and are commonly used in handheld portable flexible targets. Hinge connections achieve rotational movement of components through shaft structures, allowing precise control of the rotation angle, suitable for calibration tasks requiring quantitative adjustment of component orientation. In practical applications, combinations of various connection methods can also be used. For example, the main structure of a flexible target can be rotated using hinges, while edge components can be quickly assembled and disassembled using snap-fits, ensuring precise control of the main configuration while improving flexibility of use.

[0050] The aforementioned flexible target, through folding and interlocking, can be adapted to camera systems with varying numbers and angles. Its multi-angle, multi-faceted structure ensures simultaneous observation by more cameras, improving calibration robustness. More modules can be assembled as needed, supporting large-scale N-camera calibration. It is foldable for storage and can be deployed in the field for convenient engineering applications.

[0051] Based on the aforementioned flexible target, embodiments of the present invention also provide a calibration method. For example... Figure 4 As shown, the method includes the following steps:

[0052] S102: Each camera captures an image of the pre-set target, obtaining the original pixel coordinates of the marker points in the corresponding pixel coordinate system for each camera. Each camera simultaneously captures an image of the preset target, obtaining the original coordinates of the marker points in the pixel coordinate system. The pixel coordinate system has its origin at the top left corner of the image, with the horizontal axis being the u-axis and the vertical axis being the v-axis, and the coordinate unit being pixels. For example, when using a checkerboard target, the computer identifies each intersection of black and white squares using a corner detection algorithm (such as Harris corner detection) and outputs its (u, v) pixel coordinates.

[0053] S104: Initially estimate the intrinsic and extrinsic parameters of each camera based on the pixel coordinates of the marker points. Intrinsic parameters describe the inherent properties of the camera, including focal length (fx, fy), principal point coordinates (cx, cy), etc., determining the projection relationship from 3D points to 2D pixels. Extrinsic parameters include the rotation matrix R and translation vector t, describing the relative pose of the camera to the world coordinate system. Using the target plane as a plane of the world coordinate system (z=0), solve for the intrinsic and extrinsic parameters using perspective transformation equations based on the known world coordinates and pixel coordinates of the marker points under multiple poses. In a multi-camera system, the relative extrinsic parameters between cameras also need to be initially estimated using marker points in the common field of view.

[0054] S106: Spatial distortion coupling modeling is performed using initial intrinsic and extrinsic parameters to project the marker point onto the pixel coordinate system, obtaining the perspective pixel coordinates. In actual camera imaging, lens distortion (such as radial distortion causing straight-line imaging to bend) will cause the projected point to deviate from the ideal position. First, the projected coordinates of the marker point under the ideal pinhole model are calculated based on the initially estimated intrinsic and extrinsic parameters. Then, distortion parameters are introduced to correct them, ultimately obtaining the perspective pixel coordinates that include the distortion effect. For example, radial distortion can be expressed as a function of the distance from the pixel to the principal point; the greater the distance, the more pronounced the distortion. The coupling model can more realistically reflect the imaging process.

[0055] S108: Construct a global reprojection error objective function using the difference between the perspective pixel coordinates and the original pixel coordinates. Reprojection error refers to the deviation between the original pixel coordinates (actual measured values) and the perspective pixel coordinates (model predicted values) of the marker points, usually measured using Euclidean distance. The global reprojection error objective function can be the sum of squares of the reprojection errors of all marker points. This function is essentially a multivariate function of parameters; the more reasonable the parameter values, the smaller the function value.

[0056] S110: Based on the aforementioned error objective function, iteratively optimize the intrinsic, extrinsic, and distortion parameters of each camera. The Levenberg-Marquardt (LM) algorithm is commonly used. Its core principle is to find the direction that causes the error function to decrease most rapidly by taking the derivative, and then gradually adjust the intrinsic, extrinsic, and distortion parameters. For example, if in a certain iteration it is found that a parameter is too small, resulting in a large reprojection error, the algorithm will increase the value of that parameter according to the error gradient, iterating repeatedly until the error converges. In a multi-camera system, the optimization process must simultaneously consider the parameter correlations of each camera to ensure consistency in the global coordinate system.

[0057] In some embodiments, prior to step S102, the method further includes: flattening or folding the flexible target according to the camera's orientation and fixing it, thereby obtaining a target that can be captured by multiple cameras to obtain the calibration pattern. Figure 1 As shown, the multiple target components are rectangular planar targets of the same shape. These rectangular planar targets are flattened or folded to form a single planar target or a polyhedral cylindrical target. Depending on different needs, the flexible target can be unfolded into a planar target. Figure 2A ), folded into a triangular prism target ( Figure 2B ), folded into a tetrahedral cylindrical target ( Figure 2C ) and hexahedral cylindrical targets ( Figure 2D ) etc. For example Figure 3 As shown, the target component can also be triangular, and the surface of the target component module is printed with marking patterns. By splicing the modules together, a polyhedral target can be formed for omnidirectional camera calibration. Other desired target shapes can also be formed, but this embodiment of the invention is not limited to these.

[0058] In some of these embodiments, such as Figure 5 As shown, step S104 may specifically include:

[0059] S202: Transforming the world coordinates of the marker point to camera coordinates using rotation and translation parameters: The transformation from world coordinates to camera coordinates is dominated by extrinsic parameters (rotation matrix and translation vector). World coordinates are a user-defined three-dimensional reference frame (e.g., the plane where the target is located is Z=0), while camera coordinates have the camera's optical center as the origin. For example, the rotation matrix describes the change in direction, and the translation vector records the position offset. Combining the two can transform the world coordinates (X, Y, Z) of the marker point on the target into coordinates (Xc, Yc, Zc) in the camera coordinate system. This step solves the problem of the relative pose between the camera and the target.

[0060] S204: Transforming Camera Coordinates to Image Coordinates via Perspective Parameters: The transformation from camera coordinates to image coordinates depends on perspective parameters (focal length and principal point in the intrinsic parameters). Based on the pinhole model, a 3D point (Xc, Yc, Zc) is projected onto the imaging plane through the optical center, forming 2D image coordinates (x, y). This process simulates the perspective effect of 3D space shrinking to a 2D plane, such as distant objects appearing smaller.

[0061] S206: Obtaining corrected image coordinates by using local distortion parameters: Local distortion parameter correction addresses deviations caused by lens defects. Radial distortion causes edge points to deviate (such as barrel distortion in fisheye lenses), and tangential distortion causes offset due to the lens and sensor not being parallel. By correcting (x,y) with distortion coefficients, more accurate image coordinates (x',y') are obtained.

[0062] S208: Transform the corrected image coordinates to perspective pixel coordinates: The transformation from image coordinates to perspective pixel coordinates is a unit conversion, converting image coordinates in physical units (such as millimeters) to pixel units (u,v) according to pixel size, completing the full mapping from the three-dimensional world to a two-dimensional image. This chain is crucial in fields such as binocular vision and 3D reconstruction; for example, in industrial inspection, this process enables the precise measurement of part dimensions.

[0063] In some embodiments, step S202 further includes: obtaining the world coordinates of a marker point on one of the target components in the flexible target; and mapping the world coordinates of marker points on other target components in the flexible target to the target components whose world coordinates are determined, based on their spatial positional relationship with the target component whose world coordinates are determined. Changes in the spatial position of the marker points can be correlated using shape and positional relationship parameters (such as rotation angle, folding amplitude, etc.). Specifically, if target component A and target component B are known to be rotatably connected, when target component A rotates 30° relative to target component B, the spatial coordinate transformation relationship of all marker points on target component A relative to the marker points on target component B can be calculated using the geometric parameters of the rotation axis (such as axis length, connection point coordinates).

[0064] In some embodiments, after parameter optimization in step S110, the calibration method may further include: back-projecting the marker points onto the image planes of each camera based on the optimized parameters; comparing the back-projected points on the image planes with the original pixels; if the back-projected points deviate from the original pixels by more than a threshold, then the original pixels are removed so that they do not participate in the optimization of the error objective function. That is, based on the optimized parameters, the observed pixel coordinates and predicted pixel coordinates are compared; if a small number of predicted pixel coordinates deviate too much from the observed pixel coordinates, they can be removed to avoid affecting the optimization. Further, if the number of removed original pixels accounts for more than a set proportion of the total number of pixels in the image, then the entire image is removed so that it does not participate in the optimization of the error objective function.

[0065] The calibration process will be described in detail below:

[0066] 1. Flexible target imaging and data acquisition.

[0067] Foldable and modular flexible targets are deployed within the measurement volume to cover different viewpoints of multiple cameras. The measurement volume refers to the three-dimensional spatial area where a multi-camera system can simultaneously acquire clear images and ultimately measure three-dimensional information (such as size, position, and shape). It is the effective workspace determined by the field of view of the multiple cameras, their installation location, and the measurement task.

[0068] Each camera acquires images of the target, extracts binary-coded feature points, and obtains the coordinates of the observed pixels. Here, i represents the camera number, and j represents the target point number. The observed pixel coordinates are the original pixel coordinates from the steps described above.

[0069] 2. Initial parameter estimation.

[0070] Camera extrinsic parameters were analyzed using a conventional model. Camera internal parameters and initial distortion parameters An estimate is made. Among them... These are rotation and translation parameters. It is focal length, These are the principal point coordinates, also known as the optical center of the camera. 'i' is the camera identifier, indicating which camera it is.

[0071] By combining the target template, a point set in the world coordinate system is established. It mainly refers to the set of marker points on the target. For the flexible target of this invention, after being unfolded or folded, the world coordinates of the marker points on each target component can be mapped to each other according to their relative positional relationships.

[0072] Based on the camera projection model:

[0073] ;

[0074] , ;

[0075] 'c' stands for camera. According to... The world coordinates of the marker can be converted to camera coordinates.

[0076] , It refers to the horizontal and vertical positions of the marker point in the camera coordinate system; It is the depth from the marker point to the camera, that is, the Z coordinate of the marker point in the camera coordinate system. These are the coordinates of the marker point on the normalized image plane.

[0077] pass , This yields the coordinates on the normalized image plane.

[0078] 3. Spatial distortion coupling modeling.

[0079] The measurement volume is divided into a three-dimensional mesh, and local distortion parameters are assigned to each mesh. .

[0080] any point in space Local distortion correction under camera i:

[0081] ;

[0082] Corrected normalized coordinates:

[0083] ;

[0084] ;

[0085] in, ;

[0086] The pixel coordinates of the perspective are:

[0087] ;

[0088] 4. Optimize the objective function.

[0089] Construct the global reprojection error objective function:

[0090] ;

[0091] in:

[0092] : A projection function that includes spatial distortion correction; that is .

[0093] Robust weights are used to reduce the impact of outliers.

[0094] ;

[0095] in:

[0096] The residual of a point, The set threshold;

[0097] This represents a set of distortion parameters for each camera.

[0098] 5. Iterative optimization.

[0099] Optimization parameters include:

[0100] Camera internal parameters: External reference: Target point coordinates: Local distortion parameters: .

[0101] Update using the LM algorithm:

[0102] ;

[0103] in:

[0104] : All parameters to be optimized; r: Residual vector; W: Weight matrix; J: Jacobian matrix; Damping factor.

[0105] 6. Back projection error elimination strategy.

[0106] For the optimized parameters, the 3D points are back-projected onto the image planes of each camera:

[0107] ;

[0108] Comparison with actual observed values:

[0109] ;

[0110] like If the proportion of removed points in an image exceeds a set threshold, the entire image is removed. If the proportion of removed points in an image exceeds a set threshold, the overall quality of the image is considered poor (e.g., motion blur, exposure problems, or severe target occlusion). Therefore, the entire image will be removed from the optimization dataset, meaning that the camera's observation data for that frame will no longer participate in any optimization rounds.

[0111] 7. Convergence and Result Output.

[0112] Repeat steps 4-6 until the error function E converges or decreases to the threshold.

[0113] Output: Optimized camera intrinsic and extrinsic parameters; 3D point coordinates of the flexible target; local spatial distortion parameters.

[0114] Experimental results show that within a measurement range of 10m×4m×3m, the reprojection error decreased from 0.28 pixels to 0.09 pixels, and the 3D intersection error decreased from 0.41mm to 0.12mm.

[0115] The calibration method described above divides the measurement volume into a three-dimensional mesh, with each mesh corresponding to a set of local distortion coefficients. When any spatial point is projected onto the image plane, in addition to the inherent distortion of the camera, further corrections are made based on the local coefficients of its corresponding mesh, achieving "spatial position-distortion coupling" modeling. This method can effectively eliminate errors caused by spatially non-uniform distortion in scenes with large differences in the perspectives of multiple cameras and large volumes, significantly improving the accuracy of reprojection and 3D intersection.

[0116] For this nonlinear least squares optimization problem, this invention employs the LM algorithm for iterative solution. The LM algorithm can adaptively switch between gradient descent and the Gauss-Newton method, ensuring both stability and accelerating convergence, thereby ensuring that the optimization process is efficient and reliable.

[0117] This invention combines the structural characteristics of a flexible target with a spatial distortion coupling model. The flexible target can adapt to different camera mounting angles and workspaces through folding and splicing, ensuring that it can still provide sufficient effective calibration points for each camera in complex measurement environments. Furthermore, spatial distortion coupling enables the calibration process to not only optimize the geometric relationship between the camera and the target, but also dynamically correct distortion according to spatial position, thus maintaining high accuracy in multi-camera, multi-angle scenarios.

[0118] Furthermore, the method of this invention introduces a robust weighting mechanism and a backprojection error elimination strategy in the iterative optimization: when certain calibration points or images have large errors, their weights are automatically reduced or they are eliminated to avoid abnormal data affecting the overall accuracy. This mechanism significantly enhances the robustness and stability of the method, making it particularly suitable for complex industrial environments such as assembly line production.

[0119] Through the above process, the present invention can output high-precision camera intrinsic and extrinsic parameters, spatial coordinates of flexible target points, and local spatial distortion parameters, enabling rapid and consistent calibration of multi-camera systems, and significantly improving the accuracy of three-dimensional measurement and on-site calibration efficiency in large-volume, multi-view scenarios.

[0120] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0121] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A flexible target, characterized in that, include: Multiple target components, which are movably connected in a manner that allows for changes in shape and relative positional relationship; The multiple target components are provided with marker dot patterns; The spatial positional relationship of the marker patterns on multiple target components can be mapped to each other based on the shape and relative positional relationship change parameters when the multiple target components change shape and relative positional relationship.

2. The flexible target according to claim 1, characterized in that, The multiple target components are rectangular planar targets of the same shape. Multiple rectangular planar targets are flattened or folded to form a single planar target or a polyhedral cylindrical target.

3. The flexible target according to claim 1, characterized in that, The plurality of target components are connected to each other by at least one of slots, snaps, or hinges.

4. A calibration method, characterized in that, Using the flexible target according to any one of claims 1 to 3 to calibrate cameras located in multiple different orientations includes: Each camera captures an image of the pre-set target, obtaining the original pixel coordinates of the target point in the pixel coordinate system corresponding to each camera. Initial estimates of the intrinsic and extrinsic parameters of each camera are made based on the original pixel coordinates of the marker points. Spatial distortion coupling modeling is performed using initial intrinsic and extrinsic parameters, and the landmark points are projected onto the pixel coordinate system to obtain perspective pixel coordinates. By utilizing the difference between the perspective pixel coordinates and the original pixel coordinates, a global reprojection error objective function is constructed; Based on the aforementioned error objective function, the internal parameters, external parameters, and distortion parameters of each camera are iteratively optimized.

5. The calibration method according to claim 4, characterized in that, Also includes: The flexible target is unfolded or folded and fixed according to the position of the camera to obtain a target that can be captured by multiple cameras to capture the calibration pattern.

6. The calibration method according to claim 4, characterized in that, The spatial distortion coupling modeling using initial intrinsic and extrinsic parameters includes: Obtain the world coordinates of the marker points of each target component in the flexible target; The world coordinates of the marker points are transformed to camera coordinates using rotation and translation parameters. Transform camera coordinates to image coordinates using perspective parameters; The corrected image coordinates are obtained by correcting the image coordinates using local distortion parameters; Transform the corrected image coordinates to perspective pixel coordinates.

7. The calibration method according to claim 6, characterized in that, The process of obtaining the world coordinates of the marker points of each target component in the flexible target includes: Obtain the world coordinates of a marker point on one of the target components in a flexible target; Other target components in the flexible target obtain the world coordinates of their marker points based on their spatial positional relationship with the target component that determines the world coordinates.

8. The calibration method according to claim 4, characterized in that, Also includes: For the optimized parameters, the marker points are back-projected onto the image plane of each camera; The back-projected points on the camera image plane are compared with the original pixels. If the back-projected points deviate from the original pixels by more than a threshold, the original pixels are removed so that they do not participate in the optimization of the error objective function.

9. The calibration method according to claim 8, characterized in that, If the number of original pixels removed exceeds a set proportion of the total number of pixels in the image, the entire image will be removed so that it does not participate in the optimization of the error objective function.

10. The calibration method according to claim 4, characterized in that, The error objective function is iteratively optimized using the Levenberg-Marquardt algorithm.

Images