A Visually Guided Pose Calibration Method Based on Dimensionality Reduction of Multi-Source Coupling Error

Abstract

This invention discloses a vision-guided pose calibration method based on multi-source coupling error dimensionality reduction, belonging to the field of precision measurement technology. The method includes: constructing a transformation matrix from the part to the camera image plane coordinate system, deriving a comprehensive error analytical model from the 3D points of the part to the 2D points of the image; extracting high-sensitivity error parameters to obtain a sequence of estimated actual error parameters; constructing an error compensation transformation matrix for spatial inverse correction, generating a 2D projection point set after absorbing compensation for physical hardware defects, and removing interference points from the 2D projection point set; calling a nonlinear optimization solver to adjust the part pose, continuously iterating the part pose, and outputting the 3D pose information of the measured part. This invention achieves dynamic and accurate compensation of physical errors at the algorithm level by directly embedding the inverse error compensation matrix into the iterative projection rendering process of the underlying contour matching. Through continuous iterative convergence, it ultimately performs high-precision and robust 3D pose calibration of the part.

Description

Technical Field

[0001] This invention belongs to the field of precision measurement technology, specifically relating to a vision-guided pose calibration method based on multi-source coupling error dimensionality reduction. Background Technology

[0002] In the fields of high-end intelligent equipment manufacturing and precision industrial measurement, 3D reconstruction technology based on binocular stereo vision has become an important means of obtaining the spatial 3D shape of target objects.

[0003] However, in vision-guided scenarios for precision-machined rotating metal parts, existing technologies face severe physical and algorithmic challenges. The highly reflective and textureless physical characteristics of precision-machined metal surfaces result in uneven light reflection during machine vision forming, regardless of whether it's under natural ambient light or artificial fixed light sources. This leads to localized overexposure distortion in the image sensor, obscuring the true physical contours of the part with highlight patches and affecting measurement accuracy. Furthermore, existing binocular vision positioning systems lack a systematic error modeling framework. This results in a complex interplay of various error sources in actual physical measurement systems, including camera intrinsic residuals, lens distortion, extrinsic assembly deviations, and convergence deflection angle errors unique to converging structures. Ultimately, these factors limit the accuracy of 3D pose measurement, rendering existing binocular vision solutions unable to meet the dimensional measurement requirements of modern highly reflective and textureless metal parts. Summary of the Invention

[0004] In view of one or more of the above-mentioned defects or improvement needs of the existing technology, the present invention provides a vision-guided pose calibration method based on multi-source coupling error dimensionality reduction, which solves the problem of poor three-dimensional pose measurement accuracy caused by the intertwining of multiple error sources in the actual measurement process of the existing binocular vision positioning system.

[0005] To achieve the above objectives, this invention provides a vision-guided pose calibration method based on multi-source coupling error dimensionality reduction, which employs a binocular vision measurement system for calibration and includes the following steps: S1. Abstract each component in the binocular vision measurement system into a multi-rigid-body topology; wherein, the binocular vision measurement system includes a system base, a mounting bracket, a camera, a lens, a light source, and parts; construct a transformation matrix from the parts to the camera image plane coordinate system, and derive a comprehensive error analysis model from the three-dimensional points of the parts to the two-dimensional points of the image based on the transformation matrix; S2. Based on the Sobol global sensitivity analysis method of variance decomposition, the influence weight of the pose error of each component on the final pose accuracy is quantified, and the high-sensitivity error parameters are extracted to obtain the actual error parameter estimate sequence. S3. Construct an error compensation transformation matrix for spatial inverse correction based on the actual error parameter estimation sequence; place the error compensation transformation matrix before the current pose transformation matrix to correct the landing point trajectory of the three-dimensional contour on the virtual camera image plane and generate a two-dimensional projection point set after absorbing physical hardware defect compensation. S4. Eliminate interference points in the two-dimensional projection point set and establish a two-dimensional-three-dimensional point pair correspondence with high confidence. S5. Call the nonlinear optimization solver to adjust the part's pose; S6. Substitute the adjusted part pose into the pose transformation matrix in S3, repeat steps S3 to S5, and continuously iterate the part pose until the pose translation increment and rotation increment of the part pose before and after the update are both lower than the preset convergence limit threshold, and output the three-dimensional pose information of the tested part.

[0006] As a further improvement of the present invention, S1 includes: S101. Based on the binocular vision measurement system, establish the first motion chain from the system base to the first light source, the second motion chain from the system base to the second light source, and the measured object branch from the system base to the part. S102. Define the homogeneous transformation matrix between any two adjacent rigid bodies in the first kinematic chain, the second kinematic chain, and the branch of the object under test as follows: ; where the homogeneous transformation matrix For the ideal homogeneous transformation matrix Homogeneous transformation matrix with error The product; S103. Solve the error transformation matrix based on the first-order assumption of small error. ; S104. Solve for the comprehensive transformation matrix from the part to the camera image plane in the first motion chain. ; S105. Set the homogeneous coordinates of any three-dimensional feature point on the surface of the part in the world coordinate system as follows: Based on the perspective projection function, camera intrinsic parameter matrix and comprehensive transformation matrix, the pixel projection coordinates of any three-dimensional feature point on the surface of the part are obtained and mapped to the two-dimensional image plane.

[0007] As a further improvement of the present invention, in S103, the error transformation matrix The calculation method is as follows:

[0008] in, These represent the translation error components along the three orthogonal coordinate axes in space. These represent the minute rotational angular error components about the three orthogonal coordinate axes in space.

[0009] As a further improvement of the present invention, the synthesis transformation matrix in S104 Approximately expressed as:

[0010] in, The total number of all independent error parameters of the identified system. Representing the A specific physical error parameter For its tiny error value, For the first Each error parameter corresponds to the Jacobian incidence matrix of the ideal transformation matrix.

[0011] As a further improvement of the present invention, in S105, the pixel projection coordinates of any three-dimensional feature point on the surface of the part mapped to the two-dimensional image plane are... for:

[0012] in, The perspective projection function of this 3D point. This is the camera intrinsic parameter matrix in the first motion chain.

[0013] As a further improvement of the present invention, S2 includes: The comprehensive error index for 3D pose estimation is set as the output variable. ; The quasi-Monte Carlo method is used to sample within the probability density space of physical tolerance constraints, and the th error among multiple errors is calculated. The first-order sensitivity index of each error parameter With full-order sensitivity index ; A sensitivity threshold is set, and core parameters exceeding the threshold in the full-order sensitivity index are extracted to form a set of sensitive error sources. Offline equation fitting and calibration are performed on the parameters in the set using precision-machined standard parts to obtain a high-precision sequence of actual error parameter estimates. .

[0014] As a further improvement of the present invention, the error compensation transformation matrix for spatial inverse correction constructed based on the actual error parameter estimation sequence in S3 includes: Set the sequence of estimated actual error parameters as follows Jacobian correlation matrix extracted from multi-system systematic error analytical model Construct an error compensation transformation matrix for spatial inverse correction. :

[0015] in, It is a 4×4 identity matrix, which serves as the basis for the homogeneous transformation matrix; This is the linearized coefficient matrix obtained by performing a first-order Taylor expansion at each error parameter using the multi-system synthesis transformation matrix.

[0016] As a further improvement of the present invention, step S3, which involves placing the error compensation transformation matrix before the current pose transformation matrix to correct the trajectory of the three-dimensional contour on the virtual camera image plane and generating a two-dimensional projection point set after absorbing physical hardware defect compensation, includes: Within any iteration of the matching algorithm, the system maintains a currently estimated target six-DOF pose matrix. The three-dimensional model contour point set of the part is obtained through standard algorithms. Perspective projection is applied to a 2D image plane to generate a 2D projection point set after compensation for physical hardware defects:

[0017] in, This is the camera intrinsic parameter matrix.

[0018] As a further improvement of the present invention, S4 includes: Using discrete pixels in the two-dimensional projection point set after absorbing physical hardware defect compensation as anchor points, the system expands outward along the two-dimensional normal direction of the point on the rendering contour to perform sub-pixel level real physical edge search. A gradient direction consistency check is introduced to eliminate interference points with excessively large gradient angles and establish a high-confidence 2D-3D point pair correspondence.

[0019] The aforementioned improved technical features can be combined with each other as long as they do not conflict with each other.

[0020] In summary, the beneficial effects of the above-described technical solutions conceived by this invention compared with the prior art include: (1) The vision-guided pose calibration method based on multi-source coupling error dimensionality reduction of the present invention derives the homogeneous transformation matrix of error under all six degrees of freedom small perturbations by abstracting the rigid body topology and motion transmission chain of the system, and constructs the analytical solution from the physical bottom error source to the pixel-level projection deviation; then, the core high-sensitivity error parameters are extracted by global sensitivity analysis dimensionality reduction, and multi-source coupling error dimensionality reduction is performed. Compared with the conventional error post-processing method, the present invention directly embeds the inverse error compensation matrix into the iterative projection rendering process of the bottom contour matching, realizes the dynamic and accurate compensation of physical error at the algorithm level, and finally obtains high-precision and high-robust three-dimensional pose calibration of parts in the face of complex optical scenes with high reflectivity and no texture through continuous iterative convergence. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the vision-guided pose calibration method based on multi-source coupling error dimensionality reduction in an embodiment of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0023] In the description of this invention, it should be understood that, unless otherwise stated, the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this invention and simplifying the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention.

[0024] Furthermore, unless otherwise stated, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0025] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., 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, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0026] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can mean that the first feature is in direct contact with the second feature, or that the first feature is in indirect contact with the second feature through an intermediate medium. Furthermore, "above," "over," and "on top" of the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply that the first feature is at a lower horizontal level than the second feature.

[0027] Example: Please see Figure 1 The preferred embodiment of the present invention uses a binocular vision measurement system for calibration based on multi-source coupling error dimensionality reduction and includes the following steps: S1. The components in the binocular vision measurement system are abstracted into a multi-rigid-body topology; wherein, the binocular vision measurement system includes a system base, a mounting bracket, a camera, a lens, a light source, and other parts; A transformation matrix from the part to the camera image plane coordinate system is constructed, and a comprehensive error analysis model from the three-dimensional points of the part to the two-dimensional points of the image is derived based on the transformation matrix. S2. Based on the variance decomposition-based Sobol global sensitivity analysis method, the influence weight of each component's pose error on the final pose accuracy is quantified, and the high-sensitivity error parameters are extracted to obtain the actual error parameter estimate sequence; here, the high-sensitivity error parameters are the set of error parameter sources that exceed the sensitivity threshold. S3. Construct an error compensation transformation matrix for spatial inverse correction based on the actual error parameter estimation sequence; place the error compensation transformation matrix before the current pose transformation matrix to correct the landing point trajectory of the three-dimensional contour on the virtual camera image plane and generate a two-dimensional projection point set after absorbing physical hardware defect compensation. S4. Eliminate interference points in the two-dimensional projection point set and establish a two-dimensional-three-dimensional point pair correspondence with high confidence. S5. Call the nonlinear optimization solver to adjust the part's pose; S6. Substitute the adjusted part pose into the pose transformation matrix in S3, repeat steps S3 to S5, and continuously iterate the part pose until the pose translation increment and rotation increment of the part pose before and after the update are both lower than the preset convergence limit threshold, and output the three-dimensional pose information of the tested part.

[0028] The vision-guided pose calibration method based on multi-source coupling error dimensionality reduction in this invention derives a homogeneous error transformation matrix containing all six degrees of freedom small perturbations by abstracting the rigid body topology and motion transmission chain of the system, and constructs an analytical solution from the physical underlying error source to the pixel-level projection deviation. Then, it uses global sensitivity analysis to extract the core high-sensitivity error parameters and performs multi-source coupling error dimensionality reduction. Compared with conventional error post-processing methods, this invention directly embeds the inverse error compensation matrix into the iterative projection rendering process of the underlying contour matching, realizing dynamic and accurate compensation of physical errors at the algorithm level. Through continuous iterative convergence, it finally obtains a high-precision and robust three-dimensional pose calibration of parts in complex optical scenes with high reflectivity and no texture.

[0029] As a further improvement of the present invention, S1 specifically includes: S101. Based on the binocular vision measurement system, establish the first motion chain from the system base to the first light source, the second motion chain from the system base to the second light source, and the measured object branch from the system base to the part. S102. Define the homogeneous transformation matrix between any two adjacent rigid bodies in the first kinematic chain, the second kinematic chain, and the branch of the object under test as follows: ; where the homogeneous transformation matrix For the ideal homogeneous transformation matrix Homogeneous transformation matrix with error The product; S103. Solve the error transformation matrix based on the first-order assumption of small error. ; S104. Solve for the comprehensive transformation matrix from the part to the camera image plane in the first motion chain. ; S105. Set the homogeneous coordinates of any three-dimensional feature point on the surface of the part in the world coordinate system as follows: Based on the perspective projection function, camera intrinsic parameter matrix and comprehensive transformation matrix, the pixel projection coordinates of any three-dimensional feature point on the surface of the part are obtained and mapped to the two-dimensional image plane.

[0030] Furthermore, as a specific embodiment of the present invention, the binocular vision measurement system of the present invention is a convergent binocular vision measurement system. When abstracting each component into a multi-rigid-body topology, each component is defined as a rigid body (Body). Body0 (B0) is the system base (global reference coordinate system); Body1 (B1) is the left camera mounting bracket (including angle adjustment mechanism); Body2 (B2) is the left camera (including image sensor); Body3 (B3) is the left lens (including optical system); Body4 (B4) is the right camera mounting bracket (including angle adjustment mechanism); Body5 (B5) is the right camera (including image sensor); Body6 (B6) is the right lens (including optical system); Body7 (B7) is the coaxial ring shadowless light source (left); Body8 (B8) is the coaxial ring shadowless light source (right); and Body9 (B9) is the rotating part being measured. This binocular vision measurement system includes two main motion chains and one branch chain of the measured object. The first kinematic chain is B0→B1→B2→B3→B7; the second kinematic chain is B0→B4→B5→B6→B8; and the branch chain of the tested part is B0→B9.

[0031] Further, step S102 of the present invention includes: for any two adjacent rigid bodies in the above-mentioned conduction chain... and The homogeneous transformation matrix of the ideal and error is derived, and the transfer relationship between spatial position and attitude is expressed using a 4×4 homogeneous transformation matrix. A description is provided. Furthermore, considering the unavoidable tolerances in actual physical assembly and manufacturing processes, the true homogeneous transformation matrix is ​​defined as the ideal homogeneous transformation matrix. Homogeneous transformation matrix with error The product of:

[0032] Further, step S103 includes: based on the first-order approximation assumption of small error, the error transformation matrix... It can be linearly expanded into a fourth-order identity matrix. With the small error matrix sum:

[0033] in, These represent the translation error components along the three orthogonal coordinate axes in space. These represent the minute rotational angular error components about the three orthogonal coordinate axes in space.

[0034] As an optional embodiment of the present invention, step S104 includes: based on this, comprehensive error modeling is completed, and a global comprehensive transformation relationship from the coordinate system of the measured part to the coordinate systems of each camera image plane is established using the low-order volume array method. Taking the first kinematic chain as an example, the comprehensive transformation matrix is ​​transferred from the rigid body B9 of the measured part to the rigid body B3 of the left camera image plane. Expressed as the concatenated product of adjacent matrices along the path:

[0035] Substitute the specific matrix containing the error term and expand it:

[0036] Ignoring the higher-order minima resulting from the product of error parameters, a first-order Taylor expansion of the above formula yields a comprehensive transformation matrix that can be approximated as the product of the ideal transformation matrix and the total deviation matrix accumulated across the entire system error chain:

[0037] Where N is the total number of identified independent error parameters of the system. Representing the A specific physical error parameter For its tiny error value, For the first Each error parameter corresponds to the Jacobian incidence matrix of the ideal transformation matrix.

[0038] Based on this, step S105 includes: setting the homogeneous coordinates of any three-dimensional feature point on the surface of the measured part in the world coordinate system as follows: The three-dimensional point is projected using the perspective projection function. and the intrinsic parameter matrix of the left camera Pixel projection coordinates mapped onto the two-dimensional image plane for:

[0039] By combining the display replacement with the expansion of the comprehensive error model, the pixel projection position can be decoupled to the ideal projection position. with by Image projection deviation jointly triggered by several underlying physical error parameters The sum of these factors establishes an analytical mapping path from physical multibody error sources to two-dimensional image distortion.

[0040] Furthermore, the full error chain of a binocular positioning system includes dozens of minute perturbation parameters such as convergence deflection angle error, installation tilt error, and lens distortion residual. Full compensation for all parameters would lead to ill-conditioned solutions and overfitting. Therefore, this invention introduces a Sobol global sensitivity analysis method based on variance decomposition to quantify the impact weights of various physical errors on the final spatial pose estimation accuracy.

[0041] Specifically, as an optional embodiment of the present invention, step S2 of the present invention specifically includes: The comprehensive error index for 3D pose estimation is set as the output variable. ; The quasi-Monte Carlo method is used to sample within the probability density space of physical tolerance constraints, and the th error among multiple errors is calculated. The first-order sensitivity index of each error parameter With full-order sensitivity index ; A sensitivity threshold is set, and core parameters exceeding the threshold in the full-order sensitivity index are extracted to form a set of sensitive error sources. Offline equation fitting and calibration are performed on the parameters in the set using precision-machined standard parts to obtain a high-precision sequence of actual error parameter estimates. .

[0042] Specifically, the comprehensive error index for 3D pose estimation is defined as the output variable. The quasi-Monte Carlo method is used to sample within the probability density space of the physical tolerance constraints, and the first... The first-order sensitivity index of each error parameter With full-order sensitivity index The first-order sensitivity index represents the proportion of output variance caused independently by a single parameter, while the full-order sensitivity index represents the proportion of total variance caused by the parameter itself and its interaction with other parameters, as detailed below:

[0043]

[0044] Based on this, a sensitivity threshold is set, and a set of sensitive error sources consisting of core parameters whose full-order sensitivity exponent exceeds the set threshold is extracted. Then, offline equation fitting and calibration are performed on the parameters in the set using precision-machined standard parts to obtain a high-precision sequence of actual error parameter estimates. .

[0045] Furthermore, the core of this invention lies in directly embedding the comprehensive error model, analyzed and calibrated using multibody theory, as a low-level constraint into the iterative solution engine for refined contour matching. By applying pre-emptive physical correction at each 3D-to-2D dimensionality reduction rendering stage, high-precision robust matching under highly reflective, textureless conditions can be achieved. The specific embedding mechanism is as follows: S3. Dynamic construction of the inverse error compensation transformation matrix: based on the pre-calibrated and extracted sequence of actual error parameter estimates. Jacobian correlation matrix extracted from multi-system systematic error analytical model Construct an error compensation transformation matrix for spatial inverse correction. :

[0046] This error compensation transformation matrix can mathematically cancel out systematic geometric transmission deviations caused by hardware assembly and camera optical path structure. Here... The identity matrix is ​​a 4×4 matrix, serving as the benchmark for the homogeneous transformation matrix and representing the ideal state without any error perturbation; the Jacobian correlation matrix is ​​extracted from the multi-system error analytical model. The linearized coefficient matrix is ​​obtained by performing a first-order Taylor expansion of the multi-system integrated transformation matrix at each error parameter. It is used to quantify the local influence of a single physical error on the final pose transformation and serves as a mathematical bridge connecting physical errors and algorithm compensation.

[0047] Forward rendering projection of the 3D contour with physical error compensation: In any iteration of the matching algorithm, the system maintains a currently estimated target six-DOF pose matrix. The standard algorithm directly uses the point set of the 3D CAD model contour of the part. Perspective projection onto a two-dimensional image plane. This invention replaces this with an error-compensated projection model:

[0048] By The matrix is ​​placed before the current pose transformation matrix, and the rendering engine forcibly corrects the trajectory of the 3D contour on the virtual camera's image plane, generating a compensated 2D projection point set that fully absorbs the physical hardware defects. .

[0049] S4. Searching for the true edge using the gradient domain method and multi-dimensional filtering: After obtaining a high-precision compensated projection contour, the system calculates the current actual acquired image... Gradient field data. Using discrete pixels as anchor points, a one-dimensional sub-pixel-level real physical edge search is performed outwards, strictly following the two-dimensional normal direction of that point on the rendered contour. Since the rendering starting position has been... Matrix correction ensures the normal search path accurately traverses the actual edge region of the target object, significantly reducing the probability of false matches caused by highlight patches or initial projection offsets. Simultaneously, a gradient direction consistency check is introduced to eliminate interfering points with excessively large gradient angles, establishing a high-confidence 2D-3D point pair correspondence. The selection of interfering points with excessively large gradient angles is adjusted based on actual computational requirements.

[0050] S5. Nonlinear Pose Optimization Update and Convergence Judgment: Based on the high-purity matching point pairs established in the above steps, the system calls the nonlinear pose estimation solver to solve for the updated pose that minimizes the reprojection error. .

[0051] S6. The algorithm determines whether the translation and rotation increments of the pose before and after the update are lower than the preset convergence limit threshold. If the convergence requirement is not met, the new pose is substituted into S3 to perform forward rendering and matching optimization with error compensation, and steps S3 to S5 are repeated; if convergence has been achieved, the iteration process is exited, and the three-dimensional pose information of the measured part with extremely high absolute spatial accuracy is output.

[0052] Furthermore, the present invention can send the three-dimensional pose information of the part to be tested obtained in step S6 to the robotic arm, and then the robotic arm can perform the subsequent part grasping task.

[0053] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-source coupling error dimension reduction visual guidance pose calibration method based on a binocular visual measurement system, characterized in that, Includes the following steps: S1. Abstract each component in the binocular vision measurement system into a multi-rigid-body topology; wherein, the binocular vision measurement system includes a system base, a mounting bracket, a camera, a lens, a light source, and parts; construct a transformation matrix from the parts to the camera image plane coordinate system, and derive a comprehensive error analysis model from the three-dimensional points of the parts to the two-dimensional points of the image based on the transformation matrix; S1 includes: S101. Based on the binocular vision measurement system, establish the first motion chain from the system base to the first light source, the second motion chain from the system base to the second light source, and the measured object branch from the system base to the part. S102、Set the homogeneous transformation matrix between any two adjacent rigid bodies in the first kinematic chain, the second kinematic chain and the measured object branch as ; wherein the homogeneous transformation matrix is the product of an ideal homogeneous transformation matrix and an error homogeneous transformation matrix . S103. Solve the error transformation matrix based on the first-order assumption of small error. ; S104. Solve for the comprehensive transformation matrix from the part to the camera image plane in the first motion chain. ; The comprehensive transformation matrix Approximately expressed as: in, The total number of all independent error parameters of the identified system. Representing the A specific physical error parameter For its tiny error value, For the first Each error parameter corresponds to the Jacobian incidence matrix of the ideal transformation matrix; S105. Set the homogeneous coordinates of any three-dimensional feature point on the surface of the part in the world coordinate system as follows: The pixel projection coordinates of any three-dimensional feature point on the surface of the part are obtained by mapping it to the two-dimensional image plane based on the perspective projection function, the camera intrinsic parameter matrix and the comprehensive transformation matrix. S2. Based on the Sobol global sensitivity analysis method of variance decomposition, the influence weight of the pose error of each component on the final pose accuracy is quantified, and the high-sensitivity error parameters are extracted to obtain the actual error parameter estimate sequence. S3. Construct an error compensation transformation matrix for spatial inverse correction based on the actual error parameter estimation sequence; place the error compensation transformation matrix before the current pose transformation matrix to correct the landing point trajectory of the three-dimensional contour on the virtual camera image plane and generate a two-dimensional projection point set after absorbing physical hardware defect compensation. S4. Eliminate interference points in the two-dimensional projection point set and establish a two-dimensional-three-dimensional point pair correspondence with high confidence. S5. Call the nonlinear optimization solver to adjust the part's pose; S6. Substitute the adjusted part pose into the pose transformation matrix in S3, repeat steps S3 to S5, and continuously iterate the part pose until the pose translation increment and rotation increment of the part pose before and after the update are both lower than the preset convergence limit threshold, and output the three-dimensional pose information of the tested part.

2. The vision-guided pose calibration method based on multi-source coupling error dimensionality reduction according to claim 1, characterized in that, In S103, the error transformation matrix The calculation method is as follows: in, These represent the translation error components along the three orthogonal coordinate axes in space. These represent the minute rotational angular error components about the three orthogonal coordinate axes in space.

3. The vision-guided pose calibration method based on multi-source coupling error dimensionality reduction according to claim 1, characterized in that, In S105, the pixel projection coordinates of any three-dimensional feature point on the surface of the part mapped onto the two-dimensional image plane are... for: in, The perspective projection function of this 3D point. This is the camera intrinsic parameter matrix in the first motion chain.

4. The vision-guided pose calibration method based on multi-source coupling error dimensionality reduction according to claim 1, characterized in that, S2 includes: The comprehensive error index for 3D pose estimation is set as the output variable. ; The quasi-Monte Carlo method is used to sample within the probability density space of physical tolerance constraints, and the th error among multiple errors is calculated. The first-order sensitivity index of each error parameter With full-order sensitivity index ; A sensitivity threshold is set, and core parameters exceeding the threshold in the full-order sensitivity index are extracted to form a set of sensitive error sources. Offline equation fitting and calibration are performed on the parameters in the set using precision-machined standard parts to obtain a high-precision sequence of actual error parameter estimates. .

5. The vision-guided pose calibration method based on multi-source coupling error dimensionality reduction according to claim 1, characterized in that, The error compensation transformation matrix for spatial inverse correction constructed in S3 based on the actual error parameter estimation sequence includes: Set the sequence of estimated actual error parameters as follows Jacobian correlation matrix extracted from multi-system systematic error analytical model Construct an error compensation transformation matrix for spatial inverse correction. : in, It is a 4×4 identity matrix, which serves as the basis for the homogeneous transformation matrix; This is the linearized coefficient matrix obtained by performing a first-order Taylor expansion at each error parameter using the multi-system synthesis transformation matrix.

6. The vision-guided pose calibration method based on multi-source coupling error dimensionality reduction according to claim 5, characterized in that, In step S3, the error compensation transformation matrix is ​​placed before the current pose transformation matrix to correct the trajectory of the three-dimensional contour on the virtual camera image plane, generating a two-dimensional projection point set after absorbing physical hardware defect compensation, including: Within any iteration of the matching algorithm, the system maintains a currently estimated target six-DOF pose matrix. The three-dimensional model contour point set of the part is obtained through standard algorithms. Perspective projection is applied to a 2D image plane to generate a 2D projection point set after compensation for physical hardware defects: in, This is the camera intrinsic parameter matrix.

7. The vision-guided pose calibration method based on multi-source coupling error dimensionality reduction according to claim 1, characterized in that, S4 includes: Using discrete pixels in the two-dimensional projection point set after absorbing physical hardware defect compensation as anchor points, the system expands outward along the two-dimensional normal direction of the point on the rendering contour to perform sub-pixel level real physical edge search. A gradient direction consistency check is introduced to eliminate interference points with excessively large gradient angles and establish a high-confidence 2D-3D point pair correspondence.

Images