A non-co-visibility multi-camera online calibration method

Abstract

The present application relates to a kind of non-look at under multi-camera online calibration method, belong to camera calibration technical field.The present application first obtains the real-time pose of each camera under respective calibration board system by means of calibration board, then the real-time pose of rigid body fixed with camera is obtained using visual capture system, after the synchronous alignment of the pose data obtained, the pose transformation matrix between each camera and rigid body is calculated, and then the pose transformation matrix between each camera is calculated.The present application does not need to exist look at between camera, does not need to use complex auxiliary equipment, just using calibration board and visual capture system, the external parameter of multi-camera can be obtained in real time, the automatic high-precision online real-time calibration of multi-camera is realized, the problem of low efficiency and low accuracy of multi-camera calibration under non-look at is solved, and the efficiency and precision of multi-camera calibration under non-look at are improved.

Description

Technical Field

[0001] This invention relates to an online calibration method for multiple cameras under non-common-view conditions, belonging to the field of camera calibration technology. Background Technology

[0002] Currently, with the rapid development of industries such as autonomous driving and unmanned systems, research and applications using multi-camera sensors for vehicle collaboration, target tracking, and environmental perception have received widespread attention. The extrinsic parameter calibration results of multi-camera systems directly affect their positioning accuracy and overall performance; therefore, accurate extrinsic parameter calibration is crucial. Traditional multi-camera extrinsic parameter calibration methods often rely on the premise of shared viewing areas between cameras. They extract feature points from the shared viewing area using each camera and calibrate the extrinsic parameters based on epipolar geometry principles. However, traditional methods fail to calibrate multi-camera systems without shared viewing areas. Existing calibration methods for multi-camera systems without shared viewing areas include: Yang Tao's method using a camera array optimization calibration with non-overlapping fields of view, but this requires cameras to be able to see common 3D point coordinates, meaning a shared viewing area still exists; Zhuang Chungang's method using auxiliary cameras and auxiliary calibration boards, but this involves numerous auxiliary devices and is an offline calibration scheme; and Huang's method using the position information of a mobile robot to calibrate the cameras, but the robot's own positioning accuracy affects the calibration results.

[0003] Therefore, current online calibration methods for multiple cameras under non-common-view conditions still require a common-view area or a lot of auxiliary equipment for calibration, resulting in problems such as low calibration efficiency, low accuracy, and inability to calibrate online. Summary of the Invention

[0004] The purpose of this invention is to provide an online calibration method for multiple cameras under non-common-view conditions, so as to solve the problems of low calibration efficiency and low calibration accuracy of multiple cameras under non-common-view conditions.

[0005] To solve the above-mentioned technical problems, this invention provides a method for online calibration of multiple cameras under non-common-view conditions, the method comprising the following steps:

[0006] 1) Each camera takes real-time pictures of its own calibration board, and obtains the real-time pose of each camera in its own calibration board coordinate system with the help of the calibration board. The real-time pose of each camera in its own calibration board coordinate system is the pose transformation matrix from the camera coordinate system to the calibration board coordinate system.

[0007] 2) Use a vision capture system to acquire the real-time pose of a rigid body fixed to multiple cameras;

[0008] 3) Synchronize and align the acquired camera pose data and rigid body pose data in time;

[0009] 4) Based on the synchronized pose data of each camera and the pose data of the rigid body, calculate the pose transformation matrix from the camera coordinate system to the rigid body coordinate system.

[0010] 5) Calculate the pose transformation matrix between cameras based on the pose transformation matrix from each camera coordinate system to the rigid body coordinate system, and realize online calibration of multiple cameras under non-common-view conditions.

[0011] This invention first uses a calibration board to obtain the real-time pose of each camera within its respective calibration board system. Then, it uses a vision capture system to acquire the real-time pose of a rigid body fixed to the camera. After synchronizing and aligning the acquired pose data, it calculates the pose transformation matrix from each camera coordinate system to the rigid body coordinate system, and then calculates the pose transformation matrix between the cameras. This invention does not require a shared field of view between cameras or complex auxiliary equipment. It only requires a calibration board and a vision capture system to acquire the extrinsic parameters of multiple cameras in real time, achieving automated, high-precision online real-time calibration of multiple cameras. This solves the problems of low efficiency and low accuracy in multi-camera calibration under non-shared-view conditions, improving the efficiency and accuracy of multi-camera calibration under non-shared-view conditions.

[0012] Furthermore, in step 4), the pose transformation matrices from each camera coordinate system to the rigid body coordinate system are optimized and adjusted. The process is as follows: Based on the synchronized pose data of each camera and the rigid body pose data, the initial pose transformation matrix from each camera coordinate system to the rigid body coordinate system is calculated respectively. The error equation of the initial pose transformation matrix is ​​constructed, the Jacobian matrix of the error equation is derived, and the initial pose transformation matrix is ​​optimized through LM iteration to obtain the pose transformation matrix from each camera coordinate system to the rigid body coordinate system.

[0013] This invention establishes error equations for the initial pose transformation matrices from each camera coordinate system to the rigid body coordinate system. Using the knowledge of Lie groups and Lie algebras, it derives the Jacobian matrix of the error function through a perturbation model. The initial pose transformation matrices are then optimized using the LM method to obtain the pose transformation matrices from each camera coordinate system to the rigid body coordinate system. This improves the accuracy of the pose transformation matrices from each camera coordinate system to the rigid body coordinate system, enhances the accuracy of subsequent calculations, and effectively improves the accuracy of online multi-camera calibration.

[0014] Furthermore, the error equation is as follows:

[0015]

[0016] Among them, P i(j-1) Let P be the 3D coordinates of the i-th calibration board corner point in the camera coordinate system at frame j-1. ij Let be the three-dimensional coordinates of the i-th calibration board corner point in the camera coordinate system at frame j. Let be the pose transformation matrix from the camera coordinate system to the rigid body coordinate system. Let be the relative pose of the rigid body at frame j relative to frame j-1.

[0017] This invention constructs an error equation by using the three-dimensional coordinates of the calibration plate corner points in the camera coordinate system in two adjacent frames, the pose transformation matrix from the camera coordinate system to the rigid body coordinate system, and the relative pose of the rigid body in two adjacent frames. This can effectively reduce the error of the initial pose transformation matrix from the camera coordinate system to the rigid body coordinate system.

[0018] Furthermore, the method for deriving the Jacobian matrix is ​​as follows: using the knowledge of Lie groups and Lie algebras, the Jacobian matrix of the error equation is derived using the perturbation model.

[0019] This invention utilizes the knowledge of Lie groups and Lie algebras to solve the problem of difficulty in solving the error equation caused by the fact that the actual physical meaning of the rotation matrix does not satisfy the addition operation by using the perturbation model to derive the Jacobian matrix.

[0020] Furthermore, the method for calculating the initial pose transformation matrix from each camera coordinate system to the rigid body coordinate system is as follows: based on the pose transformation relationship between two adjacent frames of the camera and two adjacent frames of the rigid body, and taking advantage of the fact that the poses of the camera and the rigid body remain unchanged, the pose transformation matrix is ​​obtained. The pose transformation matrix is ​​decomposed into a rotation matrix and a translation vector. The rotation matrix and the translation vector are solved, and the pose transformation matrix from each camera coordinate system to the rigid body coordinate system is calculated based on the obtained rotation matrix and translation vector.

[0021] The pose transformation relationship between two adjacent frames of the camera and two adjacent frames of the rigid body is as follows:

[0022]

[0023]

[0024] The pose transformation matrix is:

[0025]

[0026] The decomposed pose transformation matrix is:

[0027]

[0028] in, Let be the pose transformation matrix between two adjacent frames of the camera. Let be the pose transformation matrix of the rigid body between two adjacent frames. Let k be the real-time pose of the camera in the calibration board coordinate system. The real-time pose of the camera in the calibration board coordinate system at time k+1 is given. Let k be the real-time pose of the rigid body in the coordinate system of the vision capture system at time k. This represents the real-time pose of the rigid body in the coordinate system of the visual capture system at time k+1. Let be the pose transformation matrix from the camera coordinate system to the rigid body coordinate system. For the rotation matrix of the decomposition, is the translation vector of the decomposition.

[0029] Furthermore, the rotation matrix is ​​solved using the quaternion method, and the translation vector is solved using the least squares method.

[0030] Furthermore, in step 4), the formula for calculating the pose transformation matrix between the cameras is:

[0031]

[0032] in, Let be the pose transformation matrix between the i-th camera and the j-th camera. Let be the pose transformation matrix from the j-th camera coordinate system to the rigid body coordinate system. Let be the pose transformation matrix from the i-th camera coordinate system to the rigid body coordinate system.

[0033] Further, in step 1), the method for obtaining the real-time pose of each camera in its respective calibration board coordinate system using a calibration board is as follows: the coordinates of the corner points of the calibration board in the calibration board coordinate system are known; the camera obtains the coordinates of the corner points on the calibration board in the pixel coordinate system in real time; and based on the camera intrinsic parameters and distortion coefficients, the pose transformation matrix from the calibration board coordinate system to the camera coordinate system is calculated. Then, the pose transformation matrix from the camera coordinate system to the calibration board coordinate system is calculated. The formula for calculating the pose transformation matrix from the camera coordinate system to the calibration board coordinate system is as follows:

[0034]

[0035]

[0036] in, and These are the rotation matrix and translation vector from the camera coordinate system to the calibration board coordinate system, respectively.

[0037] This invention does not require complex auxiliary equipment; it only requires a calibration board to obtain the real-time pose of each camera in the calibration board coordinate system online, enabling subsequent real-time online calibration of multiple cameras. Attached Figure Description

[0038] Figure 1 This is a flowchart of the online calibration method for multiple cameras under non-common-view conditions according to the present invention. Detailed Implementation

[0039] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings. The non-common-view multi-camera online calibration method of the present invention first obtains the real-time pose of each camera in its respective calibration board coordinate system using a calibration board. Then, it uses a visual capture system to acquire the real-time pose of the rigid body fixed to the camera. After synchronizing and aligning the acquired pose data, it calculates the pose transformation matrix from each camera coordinate system to the rigid body coordinate system, and then calculates the pose transformation matrix between cameras. The implementation flow of this method is as follows: Figure 1 As shown below, a detailed explanation will be provided with specific examples.

[0040] 1. Each camera takes real-time pictures of its own calibration board, and obtains the real-time pose of each camera in its own calibration board coordinate system with the help of the calibration board. The real-time pose is the pose transformation matrix from the camera coordinate system to the calibration board coordinate system.

[0041] Taking the Aruco calibration board as an example, the camera captures real-time images of the calibration board and obtains the coordinates of the corner points of the calibration board in the pixel coordinate system. The homogeneous coordinates of the corner points are denoted as P. c =[uv 1] T The 3D coordinates of the corner points in the camera coordinate system are calculated based on the camera intrinsic parameters and distortion coefficients. Camera intrinsic parameter calibration methods are relatively mature and simple; therefore, it is assumed that the camera's intrinsic parameter matrix and distortion coefficients are known. If camera intrinsic parameter calibration is required, the Zhang Zhengyou calibration method is used. The calibration principle of the Zhang Zhengyou calibration method is as follows:

[0042] Assume the homogeneous coordinates of the top corner of the chessboard in the pixel coordinate system and the world coordinate system (i.e., the chessboard coordinate system) are respectively: [uv 1] T and [X] w Y w Z w 1] T From the relationship between the world coordinate system, camera coordinate system, image coordinate system, and pixel coordinate system, we can see that:

[0043]

[0044] Where u0 and v0 are the coordinates of the origin of the image coordinate system in the pixel coordinate system, respectively, and Z is the coordinate of the origin of the image coordinate system in the pixel coordinate system. c Let f be the Z-axis coordinate of the checkerboard corner point in the camera coordinate system, a and b be the number of pixels per unit length on the x-axis and y-axis of the image coordinate system corresponding to the u-axis and v-axis of the pixel coordinate system, respectively, f be the effective focal length of the camera, and R and t be the rotation matrix and translation vector from the world coordinate system to the camera coordinate system, respectively.

[0045] Since the chessboard is a two-dimensional plane, its z-axis coordinate is 0, i.e., Z. w =0. Therefore, the formula can be simplified to:

[0046]

[0047] Where s is a non-zero scale factor, r1 and r2 are the first and second columns of the rotation matrix R, respectively, which is a 3×1 matrix, and K is the camera's intrinsic parameter matrix, satisfying:

[0048]

[0049] Where u0 and v0 are the coordinates of the origin of the image coordinate system in the pixel coordinate system, respectively, and f x f is the focal length of the camera along the x-axis. y This represents the focal length of the camera along the y-axis.

[0050] The camera's intrinsic parameters can be calculated using three or more images and the homography matrix.

[0051] When radial distortion is taken into account, we get:

[0052]

[0053] Among them, (u,v) and Let (u0, v0) be the pixel coordinates without distortion and the pixel coordinates with distortion, respectively. (x, y) are the coordinates of the principal image point; (x, y) are the image coordinates without distortion; and k1 and k2 are the distortion coefficients.

[0054] The results were:

[0055]

[0056] Since the positions of the calibration plate corner points in the calibration plate coordinate system are known, the homogeneous coordinates of the corner points in the calibration plate coordinate system are denoted as P. w =[x w y w z w 1] T Based on the corresponding positions of multiple corner points and the spatial relationship between the pixel coordinate system and the calibration board coordinate system, an equation is established to obtain the pose transformation matrix from the calibration board coordinate system to the camera coordinate system. The established equation is as follows:

[0057]

[0058]

[0059] Where K is the intrinsic parameter matrix of the camera, and is a known quantity. Let be the pose transformation matrix from the calibration board coordinate system to the camera coordinate system. These are the rotation matrix and translation vector, respectively. Based on the PnP principle, constraint equations are established, and the pose transformation matrix of the camera coordinate system relative to the Aruco calibration board coordinate system is obtained through direct linear transformation (DLT). The pose transformation matrix from the camera coordinate system to the calibration board coordinate system is calculated according to formula (6). The real-time pose of each camera in its respective calibration plate coordinate system is obtained.

[0060]

[0061] in, and These are the rotation matrix and translation vector from the camera coordinate system to the calibration board coordinate system, respectively.

[0062] 2. Use a vision capture system to obtain the real-time pose of a rigid body fixed to multiple cameras.

[0063] A rigid body is established, fixed to multiple cameras. Since the rigid body is fixed to the cameras, the positional relationship between the rigid body and the cameras is constant but unknown. To accurately obtain the real-time pose of the rigid body, a commonly used high-precision measurement device—a visual motion capture system—is employed. The visual motion capture system is a 6-DOF dynamic tracking motion system with millimeter or even sub-millimeter accuracy and a sampling frequency exceeding several hundred hertz, enabling real-time acquisition of the rigid body's high-precision position and orientation. Therefore, by fixing multiple marker spheres, a marker sphere rigid body is established under the visual capture system. The visual capture system tracks the rigid body in real time and acquires its real-time pose, denoted as [the real-time pose of the rigid body is denoted as ].

[0064] 3. Synchronize and align the acquired camera pose data and rigid body pose data in time.

[0065] The real-time poses of the multiple cameras and the rigid body are known, but they are not synchronized. To obtain time-synchronized pose data of the multiple cameras and the rigid body, the pose data of the multiple cameras and the rigid body are first published to the robot operating system (ROS) via wireless communication. The time synchronization function of ROS is called to perform soft synchronization alignment of the pose data of the multiple cameras and the rigid body in time, so as to realize the real-time synchronization of the pose data of the multiple cameras and the rigid body. The synchronized pose data of the multiple cameras and the rigid body is then received for subsequent calculations.

[0066] 4. Based on the synchronized pose data of each camera and the pose data of the rigid body, calculate the pose transformation matrix from the camera coordinate system to the rigid body coordinate system.

[0067] Given the pose data of each camera after synchronization with the rigid body for each frame, the pose transformation relationship between two adjacent frames of camera A and two adjacent frames of the rigid body at time k is:

[0068]

[0069] in, Let be the pose transformation matrix between two adjacent frames of the camera. Let be the pose transformation matrix of the rigid body between two adjacent frames. Let k be the real-time pose of camera A in the calibration board coordinate system. Let k+1 be the real-time pose of camera A in the calibration board coordinate system. Let k be the real-time pose of the rigid body in the coordinate system of the vision capture system at time k. The real-time pose of the rigid body in the coordinate system of the visual capture system at time k+1.

[0070] Wherein, the pose transformation matrix from the camera coordinate system to the rigid body coordinate system of camera A is set as This is the quantity to be determined. Since the camera is fixed to the rigid body, the pose of the camera and the rigid body remains unchanged, that is:

[0071]

[0072] According to the matrix propagation rules:

[0073]

[0074] Then we have:

[0075]

[0076] Decomposing the pose transformation matrix into rotation and translation matrices, we have:

[0077]

[0078] in, For the rotation matrix of the decomposition, is the translation vector of the decomposition.

[0079] Expanding, we get:

[0080]

[0081] For rotation matrix Solve for the relationship between quaternions and rotation matrices:

[0082]

[0083] make Based on the properties of quaternion multiplication, we can simplify to:

[0084]

[0085] in, They represent and The conjugate imaginary part, express The antisymmetric matrix, where I is the identity matrix.

[0086] For multi-frame poses, equation (15) can be derived from the equations, and the SVD decomposition calculation can be performed.

[0087]

[0088] For the calculation of the translation vector, according to formula (12), let For multi-frame poses, the equations are as follows:

[0089]

[0090] The translation vectors between each camera coordinate system and the rigid body coordinate system can be obtained by solving the least squares problem.

[0091] The initial pose transformation matrix T from each camera coordinate system to the rigid body coordinate system is obtained using the above method. c r .

[0092] The initial value of the pose transformation matrix from the camera coordinate system to the rigid body coordinate system is known. To further optimize the pose from the camera coordinate system to the rigid body coordinate system, the error equation of the pose transformation matrix is ​​established as follows:

[0093]

[0094] Among them, P i(j-1) Let P be the 3D coordinates of the i-th calibration board corner point in the camera coordinate system at frame j-1. ij Let be the three-dimensional coordinates of the i-th calibration board corner point in the camera coordinate system at frame j. Let be the pose transformation matrix from the camera coordinate system to the rigid body coordinate system. Let be the relative pose of the rigid body at frame j relative to frame j-1.

[0095] Because visual capture systems can achieve sub-millimeter accuracy, the pose of rigid bodies is considered... There is no error. Based on the actual physical meaning of the rotation matrix, it does not satisfy addition. Therefore, using the knowledge of Lie groups and Lie algebras, the Jacobian matrix of the pose error equation is derived using the perturbation model. The specific process is as follows:

[0096] Based on the relevant knowledge of Lie groups and Lie algebras, the Lie algebra corresponding to the Lie group SE(3) is se(3), and se(3) is located at... In space, This represents a six-dimensional vector. Its relationships are:

[0097]

[0098] Where ξ is an element of the Lie algebra se(3), ρ is the translation vector, and φ is the rotation vector. It is a three-dimensional vector. Let ξ be the Lie algebra of a three-dimensional vector. ∧ φ ∧ Let be the antisymmetric matrices corresponding to ξ and φ, respectively. It is a four-dimensional vector matrix.

[0099] Assume the quantity to be determined The disturbance is The Lie algebra of the disturbance term is δξ=[δρ δφ] T , and The corresponding Lie algebras are ξ and δξ, respectively.

[0100] make Establish a left perturbation model for the Lie group, for Multiplying by a perturbation gives:

[0101]

[0102] Where J is the Jacobian matrix and δξ ​​is the perturbation amount. Lie algebras.

[0103] Optimization using the Levenberg-Marquardt (LM) method yields the following linear equation for the increment:

[0104] (H+λD T D)Δx=g (19)

[0105] Where H = J T J, g = -J T e and λ are the damping factors, and Δx is the quantity to be optimized.

[0106] To simplify the equation, D takes the identity matrix I, resulting in:

[0107] (H+λI)Δx=g (20)

[0108] The initial pose transformation matrices of the camera and rigid body are optimized through LM iteration to obtain the pose transformation matrices from each camera coordinate system to the rigid body coordinate system, denoted as . n is the number of cameras.

[0109] 5. Based on the pose transformation matrix from each camera coordinate system to the rigid body coordinate system, calculate the pose transformation matrix between each camera to achieve online calibration of multiple cameras under non-common-view conditions.

[0110] Given the pose transformation matrices between each camera and the rigid body. The pose transformation matrix between each camera can be obtained according to the formula, which is the extrinsic parameter of each camera. At this point, the extrinsic parameter calibration of the camera is completed.

[0111]

[0112] in, Let be the pose transformation matrix between the i-th camera and the j-th camera. Let be the pose transformation matrix from the j-th camera coordinate system to the rigid body coordinate system. Let be the pose transformation matrix from the i-th camera coordinate system to the rigid body coordinate system.

[0113] This invention proposes an automated online multi-camera calibration method for non-common-view conditions, a real-time online calibration technology for multiple cameras. Addressing the issue of multiple cameras lacking a common-view region, a rigid body is created and fixed to the multiple cameras. Using a calibration board and a vision capture system, the real-time poses of the cameras and the rigid body are acquired online. First, the pose transformation matrix from the multi-camera coordinate system to the rigid body coordinate system is obtained. Then, the pose transformation matrix between each camera is calculated, enabling real-time online acquisition of the extrinsic parameters of the multiple cameras. This allows for automated online calibration of multiple cameras without the need for robots or other complex equipment, improving the efficiency and accuracy of multi-camera calibration under non-common-view conditions. It can be widely applied to various complex and challenging multi-camera calibration scenarios, such as production workshops and target tracking. Furthermore, this invention establishes the initial pose transformation matrix equation from the camera coordinate system to the rigid body coordinate system, derives the Jacobian matrix of the error equation using Lie groups and Lie algebras, and performs pose optimization through iterative LM method, further improving the accuracy of online multi-camera calibration and effectively solving the problems of difficulty and low accuracy in online calibration of multiple cameras under non-common-view conditions.

Claims

1. A method for online calibration of multiple cameras under non-common-view conditions, characterized in that, The calibration method includes the following steps: 1) Each camera takes real-time pictures of its own calibration board, and obtains the real-time pose of each camera in its own calibration board coordinate system with the help of the calibration board. The real-time pose of each camera in its own calibration board coordinate system is the pose transformation matrix from the camera coordinate system to the calibration board coordinate system. 2) Use a vision capture system to acquire the real-time pose of a rigid body fixed to multiple cameras; 3) Synchronize and align the acquired camera pose data and rigid body pose data in time; 4) Based on the synchronized pose data of each camera and the pose data of the rigid body, calculate the initial pose transformation matrix from each camera coordinate system to the rigid body coordinate system, construct the error equation of the initial pose transformation matrix, derive the Jacobian matrix of the error equation, and optimize the initial pose transformation matrix through LM iteration to obtain the pose transformation matrix from each camera coordinate system to the rigid body coordinate system. 5) Calculate the pose transformation matrix between cameras based on the pose transformation matrix from each camera coordinate system to the rigid body coordinate system, and realize online calibration of multiple cameras under non-common-view conditions.

2. The online calibration method for multiple cameras under non-common-view conditions according to claim 1, characterized in that, The synchronization alignment involves publishing the pose data of each camera and the pose data of the rigid body to the robot operating system ROS via wireless communication, and calling the time synchronization function of ROS to perform soft synchronization alignment of the pose data of each camera and the pose data of the rigid body in time.

3. The online calibration method for multiple cameras under non-common-view conditions according to claim 1, characterized in that, The error equation is: ； in, When it is the (j-1)th frame The three-dimensional coordinates of the corner points of the calibration plate in the camera coordinate system When it is the j-th frame, the first... The three-dimensional coordinates of the corner points of the calibration plate in the camera coordinate system Let be the pose transformation matrix from the camera coordinate system to the rigid body coordinate system. Let be the relative pose of the rigid body at frame j relative to frame j-1.

4. The online calibration method for multiple cameras under non-common-view conditions according to claim 1, characterized in that, The method for deriving the Jacobian matrix is ​​as follows: using the knowledge of Lie groups and Lie algebras, the Jacobian matrix of the error equation is derived using the perturbation model.

5. The online calibration method for multiple cameras under non-common-view conditions according to claim 1, characterized in that, The method for calculating the initial pose transformation matrix from each camera coordinate system to the rigid body coordinate system is as follows: Based on the pose transformation relationship between two adjacent frames of the camera and two adjacent frames of the rigid body, and taking advantage of the fact that the poses of the camera and the rigid body remain unchanged, the pose transformation matrix is ​​obtained. The pose transformation matrix is ​​decomposed into a rotation matrix and a translation vector. The rotation matrix and the translation vector are solved, and the pose transformation matrix from each camera coordinate system to the rigid body coordinate system is calculated based on the obtained rotation matrix and translation vector. The pose transformation relationship between two adjacent frames of the camera and two adjacent frames of the rigid body is as follows: ； The pose transformation matrix is: ； The decomposed pose transformation matrix is: ； in, Let be the pose transformation matrix between two adjacent frames of the camera. Let be the pose transformation matrix of the rigid body between two adjacent frames. Let k be the real-time pose of the camera in the calibration board coordinate system. The real-time pose of the camera in the calibration board coordinate system at time k+1 is given. Let k be the real-time pose of the rigid body in the coordinate system of the vision capture system at time k. This represents the real-time pose of the rigid body in the coordinate system of the visual capture system at time k+1. Let be the pose transformation matrix from the camera coordinate system to the rigid body coordinate system. , , For the rotation matrix of the decomposition, , , is the translation vector of the decomposition.

6. The online calibration method for multiple cameras under non-common-view conditions according to claim 5, characterized in that, The rotation matrix is ​​solved using the quaternion method, and the translation vector is solved using the least squares method.

7. The online calibration method for multiple cameras under non-common-view conditions according to claim 1, characterized in that, In step 4), the formula for calculating the pose transformation matrix between the cameras is as follows: ； in, Let be the pose transformation matrix between the i-th camera and the j-th camera. Let be the pose transformation matrix from the j-th camera coordinate system to the rigid body coordinate system. Let be the pose transformation matrix from the i-th camera coordinate system to the rigid body coordinate system.

8. The online calibration method for multiple cameras under non-common-view conditions according to claim 1, characterized in that, In step 1), the method for obtaining the real-time pose of each camera in its respective calibration board coordinate system using a calibration board is as follows: the coordinates of the corner points of the calibration board in the calibration board coordinate system are known; the camera obtains the coordinates of the corner points on the calibration board in the pixel coordinate system in real time; and based on the camera intrinsic parameters and distortion coefficients, the pose transformation matrix from the calibration board coordinate system to the camera coordinate system is calculated. Then, the pose transformation matrix from the camera coordinate system to the calibration board coordinate system is calculated. The formula for calculating the pose transformation matrix from the camera coordinate system to the calibration board coordinate system is as follows: ； in, and These are the rotation matrix and translation vector from the camera coordinate system to the calibration board coordinate system, respectively.

Images