A robust calibration method for line laser robot vision system based on model reconstruction

Abstract

This application relates to the fields of robot calibration and computer vision, and discloses a robust calibration method for a line laser robot vision system based on model reconstruction. The method includes: fixing a line laser sensor and initializing a hand-eye matrix; installing a freeform surface calibration object on the robot's end effector to acquire surface point clouds and end effector pose data in multiple poses; transforming the point cloud to the robot's end effector coordinate system based on the initial hand-eye matrix to obtain a reconstructed point cloud; searching for the nearest point from the reconstructed point cloud to the calibration object surface and calculating the point-to-surface distance residual; deriving the Jacobian matrix of the residual with respect to the current hand-eye matrix using a Lie algebra right-multiplication perturbation model; establishing an optimization function with residual minimization as the objective and introducing adaptive weights; solving for the hand-eye matrix error vector; and performing manifold updates on the hand-eye matrix until convergence. This application eliminates the dependence on standard parts, removes intermediate fitting errors, and suppresses measurement posture interference through a robust weighting mechanism, thereby improving calibration accuracy and robustness.

Description

Technical Field

[0001] This application relates to the fields of robot calibration and computer vision, and in particular to a robust calibration method for a line laser robot vision system based on model reconstruction. Background Technology

[0002] In the field of industrial robot and machine vision integration applications, line laser sensors are widely used in high-precision operation scenarios such as part grasping, dimensional inspection, and weld seam tracking due to their advantages such as fast measurement speed, high accuracy, and strong anti-interference ability. Hand-eye calibration, as a core preliminary step for the collaborative work of line laser sensors and robots, aims to establish a precise transformation relationship (i.e., hand-eye matrix) between the line laser sensor coordinate system and the robot's relevant coordinate system (end-effector coordinate system or base coordinate system). This unifies the sensor measurement data with the robot's motion control coordinates, ensuring that the robot can accurately complete various tasks based on the spatial position information fed back by the sensor.

[0003] Currently, the hand-eye calibration method for traditional line laser sensors has formed a relatively mature technical path. Its core logic relies on the measurement and feature fitting of standard parts. The specific implementation process mainly includes three steps: First, a standard part with known precise geometric parameters (such as a standard sphere, a standard plane, etc.) is selected as the calibration reference. Second, the standard part is scanned by the line laser sensor to obtain the spatial point cloud data of the standard part in the sensor coordinate system. Then, a feature fitting algorithm is used to extract geometric feature points (such as the center of the standard sphere, feature points of the standard plane, etc.) from the massive point cloud data. Finally, a system of linear equations is constructed based on the extracted feature points. The system of equations is solved by a closed-form solution or iterative algorithm to finally obtain the hand-eye matrix and complete the calibration process.

[0004] However, the calibration accuracy of the aforementioned traditional calibration methods is highly dependent on the accuracy of the original measurement data and the reliability of the feature fitting algorithm. Under the complex measurement conditions of actual industrial sites, this method has the following limitations, which restrict further improvement in calibration accuracy and stability: First, measurement posture can easily cause sensor errors. The measurement process of a line laser sensor is inherently affected by laser diffraction and camera lens distortion. The measurement noise is not constant but increases significantly with the increase of the laser incident angle and the distance of the measurement point from the sensor's optimal focal plane. Especially when measuring large-curvature standard parts (such as large-curvature standard spheres), the accuracy distribution of the point cloud data obtained by line laser scanning shows obvious inconsistency in different regions. The point cloud data in some regions has high noise and low accuracy. Such erroneous raw measurement data will be directly introduced into the subsequent feature fitting and matrix solving process, thereby introducing hand-eye calibration errors and affecting calibration accuracy.

[0005] Secondly, fitting errors lead to unstable calibration results. Existing traditional calibration methods often obtain key feature points by fitting geometric features such as the center of a standard sphere and a standard plane, and then correct the robot's kinematic errors by solving linear equations to complete hand-eye calibration. However, in actual measurement, the raw point cloud data inevitably contains measurement noise. This noise interference will inevitably produce fitting errors in the feature fitting process, and the noise distribution differs between different measurements, resulting in slight deviations in the coordinates of the geometric feature points obtained from each fitting. This deviation will affect the solution results of subsequent linear equations through data transmission, causing the final hand-eye matrix to exhibit fluctuations, i.e., the calibration results are unstable and it is difficult to guarantee consistency across multiple calibrations.

[0006] Third, calibration algorithms are prone to getting trapped in local optima. Traditional calibration methods often use standard spheres or planes as constraint references. Such constraints often suffer from imbalance: either insufficient number of extracted feature points leads to missing constraints, resulting in non-unique solutions to the linear equations; or excessive number of feature points leads to over-constraints, causing contradictions within the equations. When using iterative algorithms to solve the hand-eye matrix, this constraint imbalance can cause the iterative optimization process to easily get trapped in local optima instead of approximating the true global optimum. Ultimately, the resulting hand-eye matrix deviates significantly from the true transformation relationship, failing to meet the requirements of high-precision robot operations. Summary of the Invention

[0007] To address the issues of traditional line laser hand-eye calibration heavily relying on high-precision standard parts and struggling to overcome measurement posture-induced errors, this application provides a robust calibration method for line laser robot vision systems based on model reconstruction.

[0008] In a first aspect, this application provides a robust calibration method for a line laser robot vision system based on model reconstruction, comprising the following steps: Using a line laser sensor and a calibration plate fixed to the end of the robot, the initial hand-eye matrix of the robot is obtained through a hand-eye calibration algorithm; A freeform surface calibration object is installed at the end of a robot, and the robot end is controlled to move to multiple poses. The surface point cloud data of the freeform surface calibration object and the corresponding robot end pose data are obtained in each pose. Based on the initial hand-eye matrix, the surface point cloud data is transformed to the robot end effector coordinate system to obtain the reconstructed point cloud; Search for the nearest point in the reconstructed point cloud to the surface of the freeform calibration object, and calculate the point-to-surface distance residual; The Jacobian matrix of the point-to-surface distance residual with respect to the current hand-eye matrix is ​​derived using the Lie algebra right-multiplication perturbation model. An optimization function is established based on the Jacobian matrix, with the objective of minimizing the residual distance between points and surfaces and introducing adaptive weights. Solve the optimization function to obtain the current hand-eye matrix error vector. Update the current hand-eye matrix using an exponential mapping and use the updated hand-eye matrix as the new current hand-eye matrix. Repeat the iteration until the convergence condition is met.

[0009] Furthermore, the steps for calculating the point-to-surface distance residual include: The three-dimensional model of the freeform surface calibration object is discretized into a dense model point cloud, and the base coordinate system of the three-dimensional model coincides with the coordinate system of the robot end flange; The nearest neighbor search algorithm is used to search for the nearest point in the dense model point cloud for each point in the reconstructed point cloud, forming a nearest point pair; Calculate the local surface normal vector at the nearest point in the dense model point cloud; The normal distance residual for each pair of nearest points is determined based on the distance metric from the point to the tangent plane, which is the point-to-plane distance residual.

[0010] Furthermore, the step of deriving the Jacobian matrix of the point-to-surface distance residual with respect to the current hand-eye matrix through the Lie algebra right-multiplication perturbation model includes: introducing a right-multiplication perturbation variable located in the Lie algebra space onto the current hand-eye matrix, the right-multiplication perturbation variable containing the rotation error component and translation error component of the hand-eye matrix; applying the right-multiplication perturbation variable to the current hand-eye matrix through exponential mapping; substituting it into the transformation process of transforming the surface point cloud data to the robot end-effector coordinate system based on the initial hand-eye matrix; and linearly expanding it to obtain the Jacobian matrix of the point-to-surface distance residual with respect to the current hand-eye matrix.

[0011] Furthermore, the adaptive weights are determined based on the ratio of the point-to-surface distance residual of the current reconstruction point to the mean residual of the current frame point cloud: a first weight is assigned to reconstruction points whose residuals are not greater than the mean; a second weight is assigned to reconstruction points whose residuals are between the mean and a preset multiple of the mean, which decreases as the residual increases; and a zero weight is assigned to reconstruction points whose residuals are greater than the preset multiple of the mean; wherein, the first weight is greater than the maximum value of the second weight.

[0012] Furthermore, the adaptive weight is calculated as follows: the ratio of the residual distance between the current reconstructed point and the mean residual of the current frame point cloud is calculated; when the ratio is not greater than a first threshold, the weight is 1; when the ratio is between the first threshold and a second threshold, the weight decreases linearly as the ratio increases; when the ratio is greater than the second threshold, the weight is 0.

[0013] Furthermore, the steps for solving the optimization function include: constructing a hand-eye matrix error vector identification objective function based on the optimization function, and obtaining the hand-eye matrix error vector by iteratively solving the Newton method with the introduction of damping coefficient.

[0014] Furthermore, the step of updating the current hand-eye matrix through exponential mapping includes: converting the solved hand-eye matrix error vector into a transformation matrix of the Lie group space through exponential mapping, compensating the current hand-eye matrix by right multiplication to obtain the updated hand-eye matrix, and repeating the iteration until the change in the optimization function value between the two iterations is less than a preset threshold.

[0015] Secondly, this application provides a robust calibration device for a line laser robot vision system based on model reconstruction, comprising: The data acquisition and reconstruction module is used to acquire the initial hand-eye matrix and control the robot end effector to move to multiple poses, acquire the surface point cloud data of the freeform surface calibration object and the corresponding robot end effector pose data in each pose; and transform the surface point cloud data to the robot end effector coordinate system based on the initial hand-eye matrix to obtain the reconstructed point cloud. An optimization module is used to search for the nearest point from the reconstructed point cloud to the surface of the freeform calibration model and to calculate the point-to-surface distance residual. The Jacobian matrix of the point-to-surface distance residual with respect to the current hand-eye matrix is ​​derived by using the Lie algebra right-multiplication perturbation model. Based on the Jacobian matrix, an optimization function is established with the goal of minimizing the point-to-surface distance residual and introducing adaptive weights. The iterative update module is used to solve the optimization function, obtain the current hand-eye matrix error vector, perform manifold update on the current hand-eye matrix through exponential mapping, and use the updated hand-eye matrix as the new current hand-eye matrix. The iteration is repeated until the convergence condition is met.

[0016] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-mentioned robust calibration method for a line laser robot vision system based on model reconstruction.

[0017] Fourthly, this application provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-mentioned robust calibration method for a line laser robot vision system based on model reconstruction.

[0018] In summary, this application includes at least one of the following beneficial technical effects: 1. This application eliminates the reliance on high-precision standard parts, requiring only arbitrary free-form surfaces as calibration references to achieve a refined solution for the hand-eye matrix. Because free-form surfaces possess rich geometric features, they can provide sufficient and non-degenerate constraints for the optimization process, effectively avoiding the technical trap of falling into local optima due to insufficient or excessive constraints in traditional methods, thus enhancing the global convergence capability and stability of the calibration algorithm.

[0019] 2. This application directly transforms the multi-view point cloud acquired by the line laser sensor to the robot's end-effector coordinate system using a hand-eye matrix transformation. The residual normal distance between the reconstructed point cloud and the freeform surface calibration model is used as the optimization objective, avoiding the intermediate feature fitting step in traditional methods and thus eliminating the cumulative error introduced by fitting. Theoretical analysis and experimental verification show that this method can improve the overall calibration accuracy of the system by more than 10% compared to traditional methods.

[0020] 3. In the iterative optimization process, this application dynamically calculates adaptive weights based on the ratio of the current residual of each measurement point to the mean residual of the frame in which it is located. Low-quality outliers are assigned lower weights or even eliminated, while high-quality measurement points are assigned higher weights, thereby eliminating or suppressing abnormal data interference. This mechanism effectively suppresses the problem of inconsistent data accuracy caused by changes in measurement posture (such as fluctuations in incident angle or object distance), significantly improves the robustness of calibration results, and effectively ensures the high accuracy and reliability of hand-eye calibration. Attached Figure Description

[0021] Figure 1 This is a flowchart of a robust calibration method for a line laser robot vision system based on model reconstruction, according to an embodiment of this application. Figure 2 This is a schematic diagram of the linear laser scanning calibration object in this embodiment; Figure 3 This is a schematic diagram of the discrete point cloud of the object calibrated by the freeform surface in this embodiment; Figure 4 This is a schematic diagram illustrating sensor errors caused by attitude measurement. Figure 5 This is a schematic diagram of the robust weight change curve in this embodiment; Figure 6 This is the convergence curve of this embodiment. Detailed Implementation

[0022] The following is in conjunction with the appendix Figure 1-6 This application will be described in further detail.

[0023] This application discloses a robust calibration method for a line laser robot vision system based on model reconstruction. The hardware system includes an industrial robot, a line laser sensor fixedly installed within the robot's workspace, and a freeform surface calibration object mounted on the robot's end effector. The line laser sensor integrates a laser projector and an industrial camera, capable of acquiring three-dimensional point cloud data of the surface of the object being measured. The freeform surface calibration object is an arbitrary irregular surface object with continuous curvature, its centroid located inside the calibration object, and the three-dimensional model of the freeform surface calibration object is known, with its base coordinate system coinciding with the coordinate system of the robot's end effector flange.

[0024] Reference Figure 1The robust calibration method for a line laser robot vision system based on model reconstruction includes the following steps: Step 1: Using a line laser sensor and a calibration plate fixed to the end of the robot, obtain the initial hand-eye matrix of the robot through a hand-eye calibration algorithm; Step 2: Install a freeform surface calibration object on the robot end effector, control the robot end effector to move to multiple poses, and obtain the surface point cloud data of the freeform surface calibration object and the corresponding robot end effector pose data in each pose; Step 3: Based on the initial hand-eye matrix, transform the surface point cloud data to the robot end effector coordinate system to obtain the reconstructed point cloud; Step 4: Search for the nearest point from the reconstructed point cloud to the surface of the freeform calibration object, and calculate the point-to-surface distance residual; Step 5: Derive the Jacobian matrix of the point-to-surface distance residual with respect to the current hand-eye matrix by right multiplying the perturbation model using Lie algebra; Step 6: Based on the Jacobian matrix, establish an optimization function with the objective of minimizing the residual distance between points and surfaces and introducing adaptive weights; Step 7: Solve the optimization function to obtain the current hand-eye matrix error vector. Update the current hand-eye matrix using the exponential mapping and use the updated hand-eye matrix as the new current hand-eye matrix. Repeat the iteration until the convergence condition is met.

[0025] This application proposes a robust calibration method for line laser robot vision systems based on model reconstruction. It eliminates the reliance on high-precision standard parts or customized calibration objects used in traditional methods, requiring only an arbitrary free-form surface as a reference. The hand-eye matrix is ​​simultaneously corrected by minimizing reconstruction errors, and an adaptive weight based on relative normal distance is introduced during the iterative calibration process. This effectively suppresses data accuracy inconsistencies caused by changes in measurement posture (such as fluctuations in incident angle or object distance). Compared to traditional methods that rely on fitting standard spheres or planar features, are susceptible to fitting quality issues, and easily get trapped in local optima at large curvatures or non-ideal angles, this application fundamentally eliminates the dependence on specific geometric features, improves upon the inherent defects of measurement posture-induced errors, and achieves low-cost, highly flexible in-situ calibration. This significantly improves the accuracy and robustness of the calibration results, making it particularly suitable for rapid calibration of long-term in-service systems in industrial settings.

[0026] In step 1, using a line laser sensor and a calibration plate fixed to the robot's end effector, the initial hand-eye matrix of the robot is obtained through a hand-eye calibration algorithm. Specific steps include: The calibration board is fixed to the robot's end effector. Images of the calibration board in multiple poses are acquired using a vision camera integrated with a line laser sensor. The classic Tsai-Lenz algorithm is used to calculate the hand-eye relationship to obtain the initial hand-eye matrix. .

[0027] In step 2, a freeform surface calibration object is installed on the robot's end effector. The robot's end effector is controlled to move to multiple poses, and the surface point cloud data of the freeform surface calibration object and the corresponding robot end effector pose data are acquired at each pose. The specific steps include: Step 2.1: Adjust the robot's end effector pose until the freeform surface calibration object is completely within the measurement field of view of the line laser sensor, such as... Figure 2 As shown.

[0028] Step 2.2: Collect single-frame point cloud data of the freeform calibration surface and simultaneously record the current robot end-effector pose.

[0029] Step 2.3: Keep the freeform surface calibration object within the measurement field of view, change the robot end effector pose, and repeat step 2.2.

[0030] Step 2.4: After collecting 5-10 sets of data in a single local area, repeat steps 2.1 to 2.3 to readjust the robot pose and observe the freeform surface calibration object from different perspectives to obtain no less than 30 sets of point cloud data and robot end-effector pose data pairs.

[0031] In step 3, based on the initial hand-eye matrix, the surface point cloud data is transformed to the robot's end effector coordinate system to obtain the reconstructed point cloud. The specific calculation formula for constructing the reconstructed point cloud is as follows: Formula (1) In formula (1), Let represent the homogeneous coordinates of the i-th frame of the point cloud in the line laser sensor coordinate system. This represents the homogeneous coordinates of the point cloud in the i-th frame in the robot's end effector coordinate system. This represents the initial hand-eye matrix obtained in step 1. This represents the robot end-effector pose matrix corresponding to the i-th image capture, and all of them... Convert to This allows for the reconstruction of the point cloud of the freeform surface calibration object.

[0032] In step 4, the nearest point from the reconstructed point cloud to the surface of the freeform calibration object is searched, and the point-to-surface distance residual is calculated. Specific steps include: Step 4.1: Discretize the 3D model of the freeform surface calibration object into a dense model point cloud, such as... Figure 3 As shown. Since the base coordinate system of the 3D model coincides with the coordinate system of the robot's end flange, theoretically, the reconstructed point cloud should be precisely fitted to the surface of the 3D model.

[0033] Step 4.2: Select the i-th frame of the reconstructed point cloud according to the acquisition order. The KD-tree nearest neighbor search algorithm is used to reconstruct the point cloud. Each point in Search for the nearest point in the dense model point cloud. Form the nearest point pair .

[0034] Step 4.3: Calculate the nearest points in the dense model point cloud using principal component analysis. Local surface normal vector at the location The subscript k represents the index of the nearest point in the dense model point cloud. To eliminate directional ambiguity, the normal vector direction is uniformly redirected to point to the centroid inside the freeform surface calibration object.

[0035] Step 4.4: Based on the distance metric from a point to the tangent plane, determine the normal distance residual for each pair of closest points, i.e., the point-to-plane distance residual. : Formula (2) In formula (2), for transpose, superscript i For point cloud frame count, subscript j This refers to the index of a point in a single frame point cloud.

[0036] In step 5, the Jacobian matrix of the point-to-surface distance residual with respect to the current hand-eye matrix is ​​derived by right-multiplying the perturbation model using Lie algebra. Specific steps include: Hand-eye matrix error Represented as: Formula (3) In formula (3), Multiply by the disturbance variable on the right. These are the rotation error vector and translation error vector of the hand-eye matrix, respectively. Represents three-dimensional real space.

[0037] Substituting formula (3) into formula (1), we get: Formula (4) In formula (4), For point The error is minute. This is the robot's end-effector pose matrix. For hand-eye rotation matrix, Let the hand-eye translation vector be... For visual measurement points.

[0038] Expanding further, we get: Formula (5) In formula (5), This is the pose transformation matrix from the visual measurement point to the robot's end effector. It is a 3×3 identity matrix. It is a 3×6 coefficient matrix. This is the error vector of the hand-eye matrix.

[0039] In step 6, an optimization function is established based on the Jacobian matrix, with the objective of minimizing the residual distance between points and surfaces and incorporating adaptive weights. The specific steps include: According to the reconstruction principle, if the hand-eye matrix has no error, the reconstructed point cloud should coincide with the dense model point cloud. Therefore, this method corrects the hand-eye error through iterative compensation. Assuming the reconstructed points after single-step compensation are... ,in In the i-th scan, the first... j One reconstruction point, If we represent the compensation difference, then the residual distance from the point to the surface after compensation can be written as: Formula (6) In formula (6), For point Time normal distance, It is a 3×6 coefficient matrix, that is, from the first... i In the point cloud of the first frame j Calculated from points M matrix.

[0040] Generally, by minimizing the iteration Ideal results can be obtained Matrix, where n The total number of frames in the point cloud. m i This represents the total number of points in a single frame of the point cloud; however, as... Figure 4 As shown, when scanning a surface, the inconsistency of the normals of each point on the surface and the different scanning depths cause the points at the edges of the line to shift. In this case, directly using distance minimization will lead to errors in the reconstructed point cloud correction. Therefore, a weighted distance minimization objective function is proposed, i.e., the optimization function is as follows: Formula (7) In formula (7), The adaptive weights, determined based on the residual distances from the reconstructed points to the surface, are calculated as follows: Formula (8) In formula (8), λ For the rate of descent, e It is a natural constant; It is the ratio of the point-to-surface distance residual of the current reconstructed point to the mean residual of the point cloud in the current frame, defined as: ,in, for The 2-norm.

[0041] The curve is as follows Figure 5 As shown, during the correction process, if the residual distance from the current reconstructed point to the surface is not greater than the mean residual of the point cloud in the current frame, the point is considered a normal reconstruction point, and its weight is set. If the residual distance from the current reconstructed point to the surface is 1-2 times the mean residual, the point is considered a reliable outlier. Its weight is set to decrease as the residual increases, and a descent rate is set. ,when When it approaches 2, This allows for the connection of subsequent weights; if the residual distance from the current reconstructed point to the surface is greater than twice the mean, the point is considered an outlier and its weight is set to 0. The weights of each point change with the point cloud position during the iteration process, reducing the uncertainty caused by the measurement attitude and ultimately enabling the point cloud to reach the ideal position.

[0042] In step 7, the optimization function is solved to obtain the current hand-eye matrix error vector. The current hand-eye matrix is ​​then updated using an exponential mapping, and the updated hand-eye matrix is ​​used as the new current hand-eye matrix. This process is repeated iteratively until the convergence condition is met. Specifically, this includes two steps: solving for the error vector and error compensation. Step 7.1, Error Vector Solution: Based on the optimization function constructed in Step 6 (Formula (7)), expand it to obtain the hand-eye matrix error vector identification target function: Formula (9) The hand-eye matrix error vector is obtained by solving formula (9) using the damped Newton method: Formula (10) In formula (10), I It is the identity matrix. The damping coefficient is taken as in this method. The damped Newton method can make the hand-eye matrix error vector identification objective function converge more stably.

[0043] Step 7.2, Error Compensation: Solve Then, it is first transformed into a transformation matrix in Lie group space through exponential mapping, and then the current hand-eye matrix is ​​compensated by right multiplication of the perturbation variable to obtain the updated hand-eye matrix: Formula (11) In formula (11), This represents the hand-eye matrix after a single update. This step is repeated until the difference between the optimization function values ​​of two consecutive iterations is less than a preset threshold of 10. -5 Stop iteration, the convergence curve is as follows Figure 6 As shown.

[0044] This embodiment also provides a robust calibration device for a line laser robot vision system based on model reconstruction, characterized in that it includes: The data acquisition and reconstruction module is used to acquire the initial hand-eye matrix and control the robot end effector to move to multiple poses, acquire the surface point cloud data of the freeform surface calibration object and the corresponding robot end effector pose data in each pose; and transform the surface point cloud data to the robot end effector coordinate system based on the initial hand-eye matrix to obtain the reconstructed point cloud. An optimization module is used to search for the nearest point from the reconstructed point cloud to the surface of the freeform calibration model and to calculate the point-to-surface distance residual. The Jacobian matrix of the point-to-surface distance residual with respect to the current hand-eye matrix is ​​derived by using the Lie algebra right-multiplication perturbation model. Based on the Jacobian matrix, an optimization function is established with the goal of minimizing the point-to-surface distance residual and introducing adaptive weights. The iterative update module is used to solve the optimization function, obtain the current hand-eye matrix error vector, perform manifold update on the current hand-eye matrix through exponential mapping, and use the updated hand-eye matrix as the new current hand-eye matrix. The iteration is repeated until the convergence condition is met.

[0045] This embodiment also provides an electronic device, which may include a processor, a communication interface, a memory, and a communication bus, wherein the processor, communication interface, and memory communicate with each other through the communication bus. The processor can call logical instructions in the memory to execute the aforementioned robust calibration method for a line laser robot vision system based on model reconstruction.

[0046] Furthermore, when the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0047] This embodiment also provides a non-transitory computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it implements the above-described robust calibration method for a line laser robot vision system based on model reconstruction.

[0048] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A robust calibration method for a line laser robot vision system based on model reconstruction, characterized in that: Includes the following steps: Using a line laser sensor and a calibration plate fixed to the end of the robot, the initial hand-eye matrix of the robot is obtained through a hand-eye calibration algorithm; A freeform surface calibration object is installed at the end of a robot, and the robot end is controlled to move to multiple poses. The surface point cloud data of the freeform surface calibration object and the corresponding robot end pose data are obtained in each pose. Based on the initial hand-eye matrix, the surface point cloud data is transformed to the robot end effector coordinate system to obtain the reconstructed point cloud; Search for the nearest point in the reconstructed point cloud to the surface of the freeform calibration object, and calculate the point-to-surface distance residual; The Jacobian matrix of the point-to-surface distance residual with respect to the current hand-eye matrix is ​​derived using the Lie algebra right-multiplication perturbation model. An optimization function is established based on the Jacobian matrix, with the objective of minimizing the residual distance between points and surfaces and introducing adaptive weights. Solve the optimization function to obtain the current hand-eye matrix error vector. Update the current hand-eye matrix using an exponential mapping and use the updated hand-eye matrix as the new current hand-eye matrix. Repeat the iteration until the convergence condition is met.

2. The robust calibration method for a line laser robot vision system based on model reconstruction according to claim 1, characterized in that: The steps for calculating the point-to-surface distance residual include: The three-dimensional model of the freeform surface calibration object is discretized into a dense model point cloud, and the base coordinate system of the three-dimensional model coincides with the coordinate system of the robot end flange; The nearest neighbor search algorithm is used to search for the nearest point in the dense model point cloud for each point in the reconstructed point cloud, forming a nearest point pair; Calculate the local surface normal vector at the nearest point in the dense model point cloud; The normal distance residual for each pair of nearest points is determined based on the distance metric from the point to the tangent plane, which is the point-to-plane distance residual.

3. The robust calibration method for a line laser robot vision system based on model reconstruction according to claim 1, characterized in that: The steps for deriving the Jacobian matrix of the point-to-surface distance residual with respect to the current hand-eye matrix using the Lie algebra right-multiplication perturbation model include: introducing a right-multiplication perturbation variable located in the Lie algebra space onto the current hand-eye matrix, the right-multiplication perturbation variable containing the rotation error component and translation error component of the hand-eye matrix; applying the right-multiplication perturbation variable to the current hand-eye matrix through exponential mapping; substituting it into the transformation process of transforming the surface point cloud data to the robot end-effector coordinate system based on the initial hand-eye matrix; and linearizing and expanding it to obtain the Jacobian matrix of the point-to-surface distance residual with respect to the current hand-eye matrix.

4. The robust calibration method for a line laser robot vision system based on model reconstruction according to claim 1, characterized in that: The adaptive weights are determined based on the ratio of the residual distance between the current reconstructed point and the mean residual of the current frame point cloud: a first weight is assigned to reconstructed points whose residuals are not greater than the mean; a second weight is assigned to reconstructed points whose residuals are between the mean and a preset multiple of the mean; and a zero weight is assigned to reconstructed points whose residuals are greater than the preset multiple of the mean. The first weight is greater than the maximum value of the second weight.

5. A robust calibration method for a line laser robot vision system based on model reconstruction according to claim 4, characterized in that: The adaptive weight is calculated as follows: the ratio of the residual distance between the current reconstructed point and the mean residual of the current frame point cloud is calculated. When the ratio is not greater than the first threshold, the weight is 1; when the ratio is between the first threshold and the second threshold, the weight decreases linearly as the ratio increases; when the ratio is greater than the second threshold, the weight is 0.

6. The robust calibration method for a line laser robot vision system based on model reconstruction according to claim 1, characterized in that: The steps for solving the optimization function include: constructing a hand-eye matrix error vector identification objective function based on the optimization function, and obtaining the hand-eye matrix error vector by iteratively solving the Newton method with the introduction of damping coefficient.

7. The robust calibration method for a line laser robot vision system based on model reconstruction according to claim 1, characterized in that: The steps of updating the current hand-eye matrix using exponential mapping include: converting the solved hand-eye matrix error vector into a transformation matrix in the Lie group space through exponential mapping; compensating the current hand-eye matrix using right multiplication to obtain the updated hand-eye matrix; and repeating the iteration until the change in the optimization function value between two iterations is less than a preset threshold.

8. A robust calibration device for a line laser robot vision system based on model reconstruction, characterized in that: include: The data acquisition and reconstruction module is used to acquire the initial hand-eye matrix and control the robot end effector to move to multiple poses, acquire the surface point cloud data of the freeform surface calibration object and the corresponding robot end effector pose data in each pose; and transform the surface point cloud data to the robot end effector coordinate system based on the initial hand-eye matrix to obtain the reconstructed point cloud. The optimization module is used to search for the nearest point from the reconstructed point cloud to the surface of the freeform calibration model and to calculate the point-to-surface distance residual. The Jacobian matrix of the point-to-surface distance residual with respect to the current hand-eye matrix is ​​derived by right-multiplying the perturbation model using Lie algebra. An optimization function based on the Jacobian matrix is ​​established with the objective of minimizing the residual distance between points and surfaces and introducing adaptive weights. The iterative update module is used to solve the optimization function, obtain the current hand-eye matrix error vector, perform manifold update on the current hand-eye matrix through exponential mapping, and use the updated hand-eye matrix as the new current hand-eye matrix. The iteration is repeated until the convergence condition is met.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, it implements a robust calibration method for a line laser robot vision system based on model reconstruction as described in any one of claims 1-7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements a robust calibration method for a line laser robot vision system based on model reconstruction as described in any one of claims 1-7.

Images