Assembly pose monocular vision measurement system-oriented uncertainty analysis method

Abstract

The invention discloses an uncertainty analysis method for an assembly pose monocular vision measurement system. The method comprises the following steps: S1, determining an uncertainty transmission route of an assembly pose combination system; s2, image pixel uncertainty A, image pixel uncertainty B and robot motion uncertainty are obtained and serve as input uncertainty; s3, sequentially calculating to obtain camera calibration uncertainty, pose calculation uncertainty, hand-eye calibration uncertainty and assembly pose uncertainty; and S4, analyzing the uncertainty sensitivity of hand-eye calibration and assembly poses in sequence. According to the method, after the input uncertainty is obtained, the sensitivity of the uncertainty and the input uncertainty of the assembly pose monocular vision measurement system is analyzed by using a Monte Carlo method. The method has the beneficial effects that the operation is simple and convenient, the feasibility is high, the obtained related conclusion has important significance for further improving the precision of the assembly pose, and a reference can be provided for uncertainty analysis of a complex system.

Description

Technical Field

[0001] The present invention relates to an uncertainty analysis method, in particular to an uncertainty analysis method for a monocular vision measurement system for assembly pose, belonging to the technical fields of machine vision measurement and robot automated assembly. Background Technique

[0002] Modern industry realizes the automation of production tasks through industrial robots. In the past, the automation of industrial robots mainly relied on offline programming and the repeatability of robots, which was relatively inflexible and required a lot of time and effort for programming and debugging. It was especially not suitable for the aerospace manufacturing field with extremely high manufacturing precision requirements. To solve this problem, measurement-assisted manufacturing emerged in the aerospace manufacturing field. Through various measurement systems, including laser trackers, lidar, iGPS, and vision measurement, etc., the movement of industrial robots is guided to flexibly complete various manufacturing tasks. Among the above measurement systems, monocular vision measurement has the advantages of small volume, high measurement accuracy, and strong environmental adaptability, and has been widely used in the aerospace manufacturing field, especially suitable for high-precision assembly tasks in narrow spaces. However, during the process of monocular vision measurement-guided assembly, there will be the transfer of various errors and uncertainties, affecting the overall assembly accuracy. Therefore, evaluating the uncertainty of the entire assembly pose measurement system is very important for error compensation to improve the overall assembly accuracy, and it can also provide guidance for subsequent related research.

[0003] Currently, no researchers have analyzed the uncertainty of the monocular vision measurement system for assembly pose, resulting in a lack of certain theoretical guidance when conducting research on the core technologies related to monocular vision measurement, which hinders the further improvement of assembly pose accuracy. Summary of the Invention

[0004] Technical Solution:

[0005] The purpose of the present invention is to provide an uncertainty analysis method for a monocular vision measurement system for assembly pose to solve the problems raised in the above background technique. To achieve the above purpose, the present invention provides the following technical solution: an uncertainty analysis method for a monocular vision measurement system for assembly pose, and the uncertainty analysis method for the monocular vision measurement system for assembly pose includes the following steps:

[0006] S1: Determine the uncertainty transfer route of the assembly pose combined system; the measurement model of the assembly pose combined system includes at least one of the following: camera imaging model, camera calibration model, vision measurement model, hand-eye calibration model;

[0007] S2: Obtain the image pixel uncertainty A, the image pixel uncertainty B, and the robot motion uncertainty as the input uncertainties. Among them, the image pixel uncertainty A represents the pixel uncertainty of the corner feature extraction of the checkerboard calibration board by the camera during camera calibration or hand-eye calibration. The image feature pixel uncertainty B represents the pixel uncertainty of the relevant feature extraction of the target image by the camera during pose calculation. The robot motion uncertainty is the positioning accuracy of the used robot.

[0008] S3: Calculate in sequence to obtain the camera calibration uncertainty, the pose calculation uncertainty, the hand-eye calibration uncertainty, and the assembly pose uncertainty. Among them, the uncertainties all independently include three position degrees of freedom X, Y, Z, and three attitude degrees of freedom A, B, C.

[0009] S4: Analyze the uncertainty sensitivities of hand-eye calibration and assembly pose in sequence.

[0010] As an improvement, the specific steps to obtain the image pixel uncertainty A in step S2 are as follows: Use an industrial camera to collect the feature of the checkerboard corner. During the image acquisition process, first set the camera frame rate to the highest threshold, repeat the image acquisition N times in the experiment, obtain the coordinate distributions of the same corner feature N times, record them, and calculate the standard deviations in the u direction and the v direction respectively as the uncertainty of the corner feature extraction, that is, the image pixel uncertainty A. Among them, the u direction and the v direction take the image collected by the camera as a plane rectangle, with the horizontal direction being the u direction and the vertical direction being the v direction. That is, for the coordinate distribution of the corner, the standard deviation in the u direction is the standard deviation of the abscissa of the corner, and the standard deviation in the v direction is the standard deviation of the ordinate of the corner. N is a positive integer.

[0011] As an improvement, in step S2, the steps to obtain the image pixel uncertainty B are as follows: Calculate the pose of the checkerboard calibration board in the camera coordinate system, select the PnP algorithm for the pose calculation method, and calculate the uncertainty of the checkerboard corner feature extraction, which is the image pixel uncertainty B.

[0012] As an improvement, the steps to obtain the camera calibration uncertainty in S3 are as follows: Add the previously obtained image pixel uncertainty A to the image features used for camera calibration to generate calibrated two-dimensional feature points and corresponding three-dimensional space points, then repeat running the camera calibration program, and statistically calculate the standard deviation of the output parameters, which is recorded as the camera calibration uncertainty.

[0013] The camera calibration method is based on MATLAB. First, a number of calibration plate images taken in different postures are read in advance, for example, 10 to 20 images. Second, the detectCheckerboardPoints operator provided by MATLAB is used to detect the corner points of the calibration plate. For example, the coordinates of the corner points on the calibration plate are obtained by inputting the set calibration plate image and the actual size of the calibration plate. Then, the pre-obtained image pixel uncertainty A value is determined.

[0014] Again, perform several cycles, for example, 10,000 to 15,000 times. In each cycle, superimpose the pre-obtained image pixel uncertainty A on the detected calibration plate corner coordinates, and use the estimateCameraParameters operator provided by MATLAB to complete the calibration of the camera internal parameters.

[0015] Finally, the standard deviation of all the internal parameters of the camera after the last cycle is calculated.

[0016] As an improvement, the steps for obtaining the uncertainty of pose calculation in S3 are: adding the pre-acquired image pixel uncertainty B to a certain image feature used for pose calculation, adding the pre-acquired camera uncertainty to the camera intrinsic parameter, and then repeatedly running the pose calculation method, and counting the standard deviation of the output parameters, recorded as the uncertainty of pose calculation, where the pose includes three position degrees of freedom X, Y, Z, and three attitude degrees of freedom A, B, C.

[0017] The specific steps of the pose calculation method are:

[0018] (1)1 Read the pre-collected images of the target to be measured taken in different postures, such as a calibration plate;

[0019] (2) Use the detectCheckerboardPoints operator provided by MATLAB to detect the corner points of the calibration board. However, it is not limited to corner point detection. Other detections can also be performed. Corner point detection is only one of the items in the calculation. The others are mature and conventional detection items.

[0020] (3) Determine the camera calibration uncertainty value obtained in advance;

[0021] (4) Perform 5000 to 15000 cycles. In each cycle, superimpose the pre-obtained camera calibration uncertainty onto the internal parameters of the camera, and use the extrinsics operator built in MATLAB to complete the calculation of the 6-degree-of-freedom pose of the calibration board in the camera coordinate system; that is, complete the internal parameters of the camera after superimposing the uncertainty, the corner coordinates detected on the calibration board; output: the process of the 6-degree-of-freedom pose of the target in the camera coordinate system.

[0022] (5) Calculate the standard deviation of the 5000 to 15000 pose calculations; that is, complete the input: all 6-degree-of-freedom poses of 10000 pose calculations; output: the standard deviation of the 6-degree-of-freedom poses of the pose calculations.

[0023] As an improvement, the steps to obtain the hand-eye calibration uncertainty in S3 are: Add the pre-obtained image pixel uncertainty A to the image features for hand-eye calibration, add the pre-obtained camera uncertainty to the camera internal parameters to obtain the pose of the calibration board in the camera coordinate system, then add the pre-obtained robot motion uncertainty to the robot pose for hand-eye calibration, and then use the Horaud hand-eye calibration method, repeat running the hand-eye calibration method, and statistically output the standard deviation of the parameters, denoted as the uncertainty of the hand-eye calibration.

[0024] The specific steps of the hand-eye calibration method are: (1) First, read the poses of the calibration board in the camera coordinate system at 5 to 15 different poses pre-calculated, and the corresponding robot poses one by one for robot hand-eye calibration;

[0025] (2) Then determine the pre-obtained pose calculation uncertainty value and the robot motion uncertainty value;

[0026] (3) Then perform 5000 to 15000 cycles. In each large cycle, there are 8 to 15 small cycles. Superimpose the pre-obtained pose calculation uncertainty onto the pose of the calibration board in the camera coordinate system, and superimpose the pre-obtained robot motion uncertainty onto the corresponding robot pose one by one; after superimposition, use the calibrateHandEye operator built in the OpenCV vision library and the HORAUD method to complete the hand-eye calibration;

[0027] For example, 10 sets of poses of the calibration board in the camera coordinate system after superimposing the uncertainty and 10 sets of corresponding robot motion poses can be input; the 6-degree-of-freedom hand-eye pose relationship is output.

[0028] (4) Finally, calculate the standard deviation of 5000 to 15000 hand-eye calibrations;

[0029] For example, the 6-DOF poses of 10,000 hand-eye calibrations can be input; the standard deviation of the 6-DOF poses obtained from the hand-eye calibration is output.

[0030] As an improvement, the steps to obtain the uncertainty of the assembly pose in S3 are as follows: Add the pre-obtained pose calculation uncertainty to the pose of the target in the camera coordinate system, add the pre-obtained hand-eye calibration uncertainty to the hand-eye relationship matrix, then repeatedly run the assembly pose calculation method, and statistically calculate the standard deviation of the output parameters, which is denoted as the uncertainty of the assembly pose.

[0031] The specific steps of the assembly pose calculation method are as follows: (1) Determine a pose calculated without superimposed uncertainty and a hand-eye relationship matrix without superimposed uncertainty (including all 6-DOF poses of the hand-eye calibration);

[0032] (2) Determine the pre-obtained pose calculation uncertainty value and the hand-eye calibration uncertainty value;

[0033] (3) Perform 5000 - 15000 loops. In each large loop, add the pre-obtained pose calculation uncertainty to the pose calculated by the pose, and add the pre-obtained hand-eye calibration uncertainty to the pose of the hand-eye calibration. After the superposition, perform matrix multiplication on the two to complete the calculation of the 6-DOF assembly pose;

[0034] For example, the input is the pose calculated by the pose after superimposing the uncertainty and the pose of the hand-eye calibration after superimposing the uncertainty; the 6-DOF assembly pose is output;

[0035] (4) Calculate the standard deviation of 5000 - 15000 assembly poses.

[0036] For example, the input is all 6-DOF poses of 10,000 assembly poses; the standard deviation of the 6-DOF poses of the assembly pose is output.

[0037] As an improvement, in S4, the steps to analyze the uncertainty sensitivity of the hand-eye calibration are as follows: First, increase the X uncertainty in the pose calculation uncertainty from 0 by m times in sequence and analyze the change of the hand-eye calibration uncertainty; then increase the A uncertainty in the pose calculation uncertainty from 0 by n times in sequence and analyze the change of the hand-eye calibration uncertainty; subsequently, increase the position uncertainties X, Y, Z in the robot motion uncertainty from 0 by s times in sequence and analyze the change of the hand-eye calibration uncertainty; finally, increase the attitude uncertainties A, B, C in the robot motion uncertainty from 0 by t times in sequence and analyze the change of the hand-eye calibration uncertainty; where m, n, t, s are positive integers.

[0038] As an improvement, in S4, the steps for analyzing the uncertainty sensitivity of the assembly pose are as follows: First, increase the X uncertainty in the pose calculation uncertainty from 0 by h several times in sequence, and analyze the change in the assembly pose uncertainty; then increase the A uncertainty in the pose calculation uncertainty from 0 by i times in sequence, and analyze the change in the assembly pose uncertainty; subsequently, increase the X uncertainty in the hand-eye calibration uncertainty from 0 by j times in sequence, and analyze the change in the assembly pose uncertainty; finally, increase the A uncertainty in the hand-eye calibration uncertainty from 0 by k times in sequence, and analyze the change in the assembly pose uncertainty; where h, i, j, and k are positive integers.

[0039] Beneficial effects:

[0040] Compared with the prior art, the beneficial effects of the present invention are: A method for analyzing the uncertainty of a monocular vision measurement system for assembly pose is proposed, including the determination of the system uncertainty transfer route, the uncertainty analysis of the monocular vision measurement system for assembly pose, and the sensitivity analysis of the input uncertainty. The method of the present invention can be applied to any form of monocular vision system for assembly pose and can effectively analyze their uncertainties.

[0041] (1) A series of conclusions obtained by the method of the present invention have important guiding effects on the accuracy analysis, error compensation, and further improvement of the assembly pose accuracy of the monocular vision measurement system for assembly pose.

[0042] (2) The method of the present invention can be extended to a binocular vision system and can provide a reference for the uncertainty analysis of other complex systems. Description of the drawings

[0043] Figure 1 It is a schematic flow chart of the method of the present invention.

[0044] Figure 2 It is a schematic diagram of the composition of the monocular vision measurement system for the object assembly pose of the method of the present invention.

[0045] Figure 3 It is an uncertainty transfer route diagram of the monocular vision measurement system for the object assembly pose of the method of the present invention.

[0046] Figure 4 It is the change of the hand-eye calibration uncertainty with the pose calculation X uncertainty in Example 1 of the method of the present invention.

[0047] Figure 5 It is the change of the hand-eye calibration uncertainty with the pose calculation A uncertainty in Example 1 of the method of the present invention.

[0048] Figure 6 It is the change of the hand-eye calibration uncertainty with the robot motion position uncertainty in Example 1 of the method of the present invention.

[0049] Figure 7 For the change of the hand-eye calibration uncertainty with the robot motion pose uncertainty in Embodiment 1 of the method of the present invention.

[0050] Figure 8 For the change of the assembly pose uncertainty with the pose calculation X uncertainty in Embodiment 1 of the method of the present invention.

[0051] Figure 9 For the change of the assembly pose uncertainty with the pose calculation A uncertainty in Embodiment 1 of the method of the present invention.

[0052] Figure 10 For the change of the assembly pose uncertainty with the hand-eye calibration X uncertainty in Embodiment 1 of the method of the present invention.

[0053] Figure 11 For the change of the assembly pose uncertainty with the hand-eye calibration A uncertainty in Embodiment 1 of the method of the present invention.

[0054] In the figure: six-degree-of-freedom robot 1, object with pose to be measured 2, industrial camera for visual measurement 3, robot base coordinate system 4, robot end coordinate system 5, object coordinate system 6, camera coordinate system 7, hand-eye calibration process 8, robot motion process 9, visual measurement process 10. Detailed implementation manners

[0055] The technical solutions in the embodiments of the present invention will be clearly and completely described below, so that those skilled in the art can better understand the advantages and features of the present invention, and thus make a clearer definition of the protection scope of the present invention. The embodiments described in the present invention are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative work belong to the protection scope of the present invention.

[0056] Please refer to Figures 1-3 , the present invention provides an uncertainty analysis method for a monocular vision measurement system for assembly pose, and the uncertainty analysis method for the monocular vision measurement system for assembly pose includes the following steps:

[0057] S1: Determine the uncertainty transfer route of the assembly pose combination system;

[0058] S2: Obtain the image pixel uncertainty A, the image pixel uncertainty B and the robot motion uncertainty as the input uncertainties;

[0059] S3: Calculate and obtain the camera calibration uncertainty, the pose calculation uncertainty, the hand-eye calibration uncertainty, and the assembly pose uncertainty in sequence;

[0060] S4: Analyze the uncertainty sensitivity of hand-eye calibration and assembly pose in sequence.

[0061] After obtaining the structure of S4 in the present invention, it is to explore whether the final assembly pose can meet the requirements of relevant measurement and assembly tasks. Analyzing the uncertainty sensitivity of hand-eye calibration and assembly pose is to, when the uncertainty of the final assembly pose does not meet the task requirements, adjust and optimize the input uncertainty to improve the accuracy of the final assembly pose from the perspective of the highest efficiency.

[0062] Among them, the step S1 includes the following steps:

[0063] A1: According to the specific hardware settings of the assembly pose combination system, sort out the involved measurement models, including but not limited to: camera imaging model, camera calibration model, visual measurement model, hand-eye calibration model, and determine the system uncertainty transfer route;

[0064] A2: According to the system uncertainty transfer route, analyze the interaction of uncertainties in each part.

[0065] Among them, the step S2 includes the following steps:

[0066] A1: Obtain the image pixel uncertainty A;

[0067] A2: Obtain the image pixel uncertainty B; where the image pixel uncertainty A and the image feature pixel uncertainty B are essentially non-interfering and uncorrelated due to different feature extraction objects.

[0068] A3: Obtain the robot motion uncertainty.

[0069] Furthermore, in the step A1, the image pixel uncertainty A is related to calibration, including camera parameter calibration and robot hand-eye calibration. During the calibration process, a calibration board with higher feature extraction accuracy is usually used. Preferably, taking the checkerboard calibration board as an example as the object to be measured.

[0070] In the present invention, the pre-obtained image pixel uncertainty A is added to the image features for camera calibration to generate calibrated two-dimensional feature points and corresponding three-dimensional space points, and then the camera calibration program is repeatedly run, and the standard deviation of the output parameters is statistically recorded as the uncertainty of camera calibration.

[0071] Furthermore, in the step A2, the image pixel uncertainty B is related to specific visual measurement tasks. In visual measurement tasks, relevant features of the object to be measured need to be extracted, and the relevant features vary according to different objects to be measured.

[0072] Since the uncertainty B of the image pixels affects the pose calculation uncertainty under its combined action with the pose calculation method. For the simplicity and operability of the method, in the present invention, the pose of the checkerboard calibration board in the camera coordinate system is directly calculated, and the PnP algorithm is selected as the pose calculation method. The corresponding image pixel uncertainty B is the image pixel uncertainty A, that is, the uncertainty of the checkerboard corner feature extraction.

[0073] Among them, the step S3 includes the following steps:

[0074] A1: Obtain the camera calibration uncertainty;

[0075] A2: Pose calculation uncertainty;

[0076] A3: Obtain the hand-eye calibration uncertainty;

[0077] A4: Obtain the assembly pose uncertainty.

[0078] Further, in the step A1, the pre-obtained image pixel uncertainty A is added to the image features for camera calibration to generate calibrated two-dimensional feature points and corresponding three-dimensional space points. Then, the camera calibration program is repeatedly run, and the standard deviation of the output parameters is statistically calculated and recorded as the uncertainty of camera calibration.

[0079] Further, in the step A2, the pre-obtained image pixel uncertainty B is added to a certain determined image feature for pose calculation, and the pre-obtained camera uncertainty is added to the camera internal parameters. Then, the pose calculation program is repeatedly run, and the standard deviation of the output parameters is statistically calculated and recorded as the uncertainty of pose calculation.

[0080] Further, in the step A3, the pre-obtained image pixel uncertainty A is added to the image features for hand-eye calibration, and the pre-obtained camera uncertainty is added to the camera internal parameters to obtain the pose of the calibration board in the camera coordinate system. Then, the pre-obtained robot motion uncertainty is added to the robot pose for hand-eye calibration. Then, the Horaud hand-eye calibration method is used, and the hand-eye calibration program is repeatedly run, and the standard deviation of the output parameters is statistically calculated and recorded as the uncertainty of hand-eye calibration.

[0081] Further, in the step A4, the pre-obtained pose calculation uncertainty is added to the pose of the target in the camera coordinate system, and the pre-obtained hand-eye calibration uncertainty is added to the hand-eye relationship matrix. Then, the assembly pose calculation program is repeatedly run, and the standard deviation of the output parameters is statistically calculated and recorded as the uncertainty of assembly pose.

[0082] Among them, the step S4 includes the following steps:

[0083] A1: Analyze the uncertainty sensitivity of hand-eye calibration;

[0084] A2: Analyze the uncertainty sensitivity of the assembly pose.

[0085] Further, in the step A1, when analyzing the uncertainty sensitivity of hand-eye calibration, first increase the X uncertainty in the pose calculation uncertainty from 0 several times in sequence, and analyze the change of hand-eye calibration uncertainty; then increase the A uncertainty in the pose calculation uncertainty from 0 several times in sequence, and analyze the change of hand-eye calibration uncertainty; subsequently, increase the position uncertainty (XYZ) in the robot motion uncertainty from 0 several times in sequence, and analyze the change of hand-eye calibration uncertainty; finally, increase the attitude uncertainty (ABC) in the robot motion uncertainty from 0 several times in sequence, and analyze the change of hand-eye calibration uncertainty.

[0086] Further, in the step A2, when analyzing the uncertainty sensitivity of the assembly pose, first increase the X uncertainty in the pose calculation uncertainty from 0 several times in sequence, and analyze the change of the assembly pose uncertainty; then increase the A uncertainty in the pose calculation uncertainty from 0 several times in sequence, and analyze the change of the assembly pose uncertainty; subsequently, increase the X uncertainty in the hand-eye calibration uncertainty from 0 several times in sequence, and analyze the change of the assembly pose uncertainty; finally, increase the A uncertainty in the hand-eye calibration uncertainty from 0 several times in sequence, and analyze the change of the assembly pose uncertainty.

[0087] Example 1

[0088] Purpose of Example 1: Completely analyze the uncertainty of the specific assembly pose barrage vision measurement system

[0089] Step 1: Determine the uncertainty transfer route of the assembly pose combination system, specifically as follows:

[0090] According to Figure 3 , the image feature pixel uncertainty A represents the pixel uncertainty of the camera's extraction of corner features in the checkerboard calibration board during camera calibration or hand-eye calibration, while the image feature pixel uncertainty B represents the pixel uncertainty of the camera's extraction of relevant features in the target image during pose calculation. The assembly pose uncertainty is jointly determined by the pose calculation uncertainty and the hand-eye calibration uncertainty. Figure 3 The uncertainty of each part in

[0091] (1) Pose calculation uncertainty: The original data of the pose calculation process is the target image data. By combining the relevant features in the target image with the camera internal parameter data obtained through calibration, the pose of the target in the camera coordinate system is output. Therefore, the sources of pose calculation uncertainty are the image feature pixel uncertainty B and the camera calibration uncertainty.

[0092] (2) Uncertainty of hand-eye calibration: The Horaud hand-eye calibration method is adopted. The original data required for hand-eye calibration are several calibration board images collected by an industrial camera and the corresponding end-effector poses of the robot. First, the corner features in the calibration board images are extracted, and the pose of the calibration board in the camera coordinate system is calculated based on the camera internal parameters. Then, the hand-eye calibration is completed in combination with the corresponding end-effector poses of the robot, and the hand-eye relationship matrix is output. Therefore, the sources of uncertainty in hand-eye calibration are the uncertainty A of image feature pixels, camera calibration uncertainty, and robot motion uncertainty.

[0093] (3) Camera calibration uncertainty: The Zhang Zhengyou calibration method is adopted. The original data required for camera calibration are several calibration board images collected by an industrial camera. The corner features in the calibration board images are extracted, and the camera internal parameters are output. Therefore, the source of camera calibration uncertainty is the uncertainty A of image feature pixels.

[0094] Step 2: Obtain the calculation uncertainties such as image pixel uncertainty, which are specifically as follows:

[0095] (1) For the image pixel uncertainty A, information such as the model parameters of the industrial camera used in the experiment is shown in Table 1. The collected feature is the checkerboard corner feature. During the image acquisition process, in order to reduce the influence caused by environmental changes, the camera frame rate is set to the highest. In the experiment, the image acquisition is repeated 500 times, and the coordinate distributions of the same corner feature are obtained 500 times and recorded. The standard deviations in the u direction and v direction are calculated respectively as the uncertainty of corner feature extraction, that is, the image pixel uncertainty A, as shown in Table 2.

[0096] Table 1 Camera model parameters for repeated experiments

[0097]

[0098] Table 2 Uncertainty of corner feature extraction

[0099]

[0100] (2) For the image pixel uncertainty B, under the combined action of it and the pose calculation method, it affects the pose calculation uncertainty. Here, the pose of the checkerboard calibration board in the camera coordinate system is directly calculated, and the PnP algorithm is selected as the pose calculation method. The corresponding image pixel uncertainty B is the image pixel uncertainty A, that is, the uncertainty of checkerboard corner feature extraction.

[0101] (3) For the robot motion uncertainty, the positioning accuracy of the robot used in the present invention is taken as its motion uncertainty. The robot position positioning accuracy is 0.03 mm, and the attitude positioning accuracy is 5×10 -5rad. Therefore, the position motion uncertainty of the robot is set to 0.03 mm, and the attitude motion uncertainty is 5×10 -5 rad.

[0102] Step 3: Calculate the camera calibration uncertainty, pose calculation uncertainty, hand-eye calibration uncertainty, and assembly pose uncertainty in sequence, as follows:

[0103] (1) For the camera calibration uncertainty, the number of iterations in the present invention is determined to be 10,000 times. First, the corner feature extraction uncertainty in Table 2 is used as the input uncertainty to calculate the camera calibration uncertainty, as shown in Table 3.

[0104] Table 3 Camera Calibration Uncertainty

[0105]

[0106] (2) For the pose calculation uncertainty, subsequently, the corner feature extraction uncertainty in Table 2 and the camera calibration uncertainty in Table 3 are used as the input uncertainties to calculate the pose calculation uncertainty, as shown in Table 4. Since the radian values are relatively small, 7 decimal places are uniformly retained. The same applies hereinafter.

[0107] Table 4 Pose Calculation Uncertainty

[0108]

[0109] (3) For the hand-eye calibration uncertainty. Since the pose calculation uncertainty in the present invention is to calculate the pose of the checkerboard calibration board in the camera coordinate system, the robot motion uncertainty and the pose calculation uncertainty in Table 4 are directly used as the input uncertainties here. The Horaud hand-eye calibration method is selected to calculate the hand-eye calibration uncertainty, as shown in Table 5.

[0110] Table 5 Hand-Eye Calibration Uncertainty

[0111]

[0112] (4) For the assembly pose uncertainty. Finally, the pose calculation uncertainty in Table 4 and the hand-eye calibration uncertainty in Table 5 are used as the input uncertainties to calculate the assembly pose uncertainty, as shown in Table 6.

[0113] Table 6 Assembly Pose Uncertainty

[0114]

[0115] (5) First, observe the hand-eye calibration uncertainty in Table 5. Although the input uncertainty, robot motion uncertainty, and pose calculation uncertainty are all around 0.03 mm (XYZ), after the hand-eye calibration process, the hand-eye calibration uncertainty in Table 5 reaches 1.5 mm (XYZ). Then, observe the assembly pose uncertainty in Table 6. Although the pose calculation uncertainty is around 0.03 mm (XYZ), due to the relatively large hand-eye calibration uncertainty, the final assembly pose uncertainty is further increased based on the hand-eye calibration uncertainty. To explore the influence of the change in input uncertainty on the hand-eye calibration uncertainty and assembly pose uncertainty, the sensitivity analysis of the hand-eye calibration uncertainty and assembly pose uncertainty will be carried out separately below.

[0116] Step 4: Analyze the uncertainty sensitivity of hand-eye calibration and assembly pose in turn, as follows:

[0117] (1) First, analyze the sensitivity of the hand-eye calibration uncertainty. First, increase the X uncertainty in the pose calculation uncertainty from 0 to 0.01, 0.02, 0.03, 0.04 mm in turn. The change in the hand-eye calibration uncertainty is as Figure 4 shown.

[0118] Analyze Figure 4 the results in. Along with the increase in the pose calculation X uncertainty, the A, B, and C uncertainties in the hand-eye calibration uncertainty do not change significantly, but the X, Y, and Z uncertainties increase to a certain extent respectively. Among them, the increase in the X uncertainty is relatively small, while the increase in the Y and Z uncertainties is relatively large and consistent.

[0119] Then, increase the A uncertainty in the pose calculation uncertainty from 0 to 0.0001, 0.0002, 0.0003, 0.0004 rad in turn. The change in the hand-eye calibration uncertainty is as Figure 5 shown.

[0120] Analyze Figure 5 the results in. Along with the increase in the pose calculation A uncertainty, the X, Y, Z, A, B, and C uncertainties in the hand-eye calibration uncertainty all increase to a certain extent. In terms of position uncertainty, the increase in the X uncertainty is the largest, the increase in the Z uncertainty is the second, and the increase in the Y uncertainty is the smallest; in terms of attitude uncertainty, the increase in the B and C uncertainties is relatively large and consistent, and the increase in the A uncertainty is relatively small.

[0121] Subsequently, increase the position uncertainty (XYZ) in the robot motion uncertainty from 0 to 0.01, 0.02, 0.03, 0.04 mm in turn. The change in the hand-eye calibration uncertainty is as Figure 6 shown.

[0122] AnalyzeFigure 6 In the results of , with the increase of the uncertainty of the robot's motion position, the uncertainties A, B, and C in the hand-eye calibration uncertainty do not change significantly, but the uncertainties X, Y, and Z increase to a certain extent respectively. Among them, the increase range of the X uncertainty is relatively small, while the increase ranges of the Y and Z uncertainties are relatively large and consistent. It should be noted that since the XYZ uncertainties of the robot's motion increase simultaneously, the increase ranges of the XYZ in the hand-eye calibration uncertainty are all larger than those in . Figure 7 in .

[0123] Finally, the attitude uncertainty (ABC) in the robot's motion uncertainty is increased from 0 to 0.00005, 0.0001, 0.00015, and 0.0002 mm in sequence, and the changes in the hand-eye calibration uncertainty are as shown in . Figure 7 shown.

[0124] Analysis Figure 7 In the results of , with the increase of the robot's motion attitude uncertainty, the uncertainties X, Y, Z, A, B, and C in the hand-eye calibration uncertainty all increase to a certain extent. In terms of the position uncertainty, the increase range of the X uncertainty is relatively large, while the increase ranges of the Y and Z uncertainties are relatively small and basically consistent; in terms of the attitude uncertainty, the increase ranges of the A and B uncertainties are relatively large and consistent, and the increase range of the C uncertainty is relatively small. Here, the ABC uncertainties of the robot's motion also increase simultaneously.

[0125] Summary Figures 4 to 7 Based on the results of , the following conclusions can be drawn: For the hand-eye calibration uncertainty, the position uncertainty in the input uncertainty (including the pose calculation uncertainty and the robot's motion uncertainty) has a relatively limited impact on the final hand-eye calibration uncertainty, only having a certain impact on the position uncertainty among them. On the contrary, the attitude uncertainty in the input uncertainty has a greater impact on the final hand-eye calibration uncertainty, and at the same time affects both the position uncertainty and the attitude uncertainty among them. For the attitude uncertainty, there is even a certain amplification effect. As shown in , when the input uncertainty increases from 0.0001 to 0.0002 rad, only increasing by 0.0001 rad, but it causes the X uncertainty in the final hand-eye calibration to increase by approximately 1 mm and the B uncertainty to increase by approximately 0.001 rad. This phenomenon also explains why the input position uncertainty is about 0.03 mm, but the final hand-eye calibration position uncertainty exceeds 1 mm. Figure 8 shown in , when the input uncertainty increases from 0.0001 to 0.0002 rad, only increasing by 0.0001 rad, but it causes the X uncertainty in the final hand-eye calibration to increase by approximately 1 mm and the B uncertainty to increase by approximately 0.001 rad. This phenomenon also explains why the input position uncertainty is about 0.03 mm, but the final hand-eye calibration position uncertainty exceeds 1 mm.

[0126] (2) Next, analyze the sensitivity of the assembly pose uncertainty. First, increase the X uncertainty in the pose calculation uncertainty from 0 to 0.01, 0.02, 0.03, and 0.04 mm in sequence, and the changes in the assembly pose uncertainty are as shown in . Figure 8 shown.

[0127] Analysis Figure 8 In the results, with the increase of the uncertainty of pose calculation in X, the uncertainties of A, B, and C in the hand-eye calibration uncertainty do not change significantly. However, the uncertainties of X, Y, and Z increase to a certain extent, and the increase amplitudes are all small. Analyzing the reason, it should be that the uncertainty of pose calculation position is much smaller than the uncertainty of hand-eye calibration position. Therefore, when the uncertainty of pose calculation position increases slightly, the impact on the uncertainty of the final assembly pose is not obvious.

[0128] Then, the uncertainty of A in the pose calculation uncertainty is increased from 0 to 0.0001, 0.0002, 0.0003, 0.0004 rad in sequence, and the change of the assembly pose uncertainty is as Figure 9 shown.

[0129] Analysis Figure 9 In the results, with the increase of the uncertainty of A in the pose calculation, the uncertainties of X, Y, Z, A, B, and C in the assembly pose uncertainty all increase to a certain extent, but the increase amplitudes are not obvious. The reason is also that the uncertainty of pose calculation attitude is much smaller than the uncertainty of hand-eye calibration attitude.

[0130] Subsequently, the uncertainty of X in the hand-eye calibration uncertainty is increased from 0 to 1, 2, 3, 4 mm in sequence, and the change of the assembly pose uncertainty is as Figure 10 shown.

[0131] Analysis Figure 10 In the results, with the increase of the uncertainty of X in the hand-eye calibration, the uncertainties of Y, Z, A, B, and C in the assembly pose uncertainty do not change significantly, but the uncertainty of X increases to a large extent, and the increasing trend is gradually accelerating. However, due to the uncertainty of pose calculation being much smaller than the uncertainty of hand-eye calibration, there is a certain dilution effect, and the increase amplitude of the uncertainty of X in the assembly pose is smaller than the increase amplitude of the uncertainty of X in the hand-eye calibration.

[0132] Finally, the uncertainty of A in the hand-eye calibration uncertainty is increased from 0 to 0.0015, 0.003, 0.0045, 0.006 mm in sequence, and the change of the assembly pose uncertainty is as Figure 11 shown.

[0133] Analysis Figure 11 In the results, with the increase of the uncertainty of A in the hand-eye calibration, the uncertainties of X, Z, B, and C in the assembly pose uncertainty do not change significantly, but the uncertainties of Y and A increase to a large extent. Due to the specific configuration of the robot hand-eye relationship in this study, the uncertainty of A in the hand-eye calibration affects the uncertainty of Y in the assembly pose.

[0134] Summary Figures 8 to 11From the results above, the following conclusions can be drawn: for the assembly pose uncertainty, if there is a difference of more than one order of magnitude between the pose calculation uncertainty and the hand-eye calibration uncertainty in the input uncertainty, then the change in the smaller uncertainty has a negligible impact on the uncertainty of the final result and can be ignored; while the change in the larger position uncertainty has a greater impact on the position uncertainty of the final result, and the change in the larger attitude uncertainty has a greater impact on the position and attitude uncertainties of the final result.

[0135] (3) Further summarize the above conclusions: to obtain a high hand-eye calibration accuracy, that is, a low hand-eye calibration uncertainty, low pose calculation and robot motion uncertainties are required, especially the attitude uncertainty among them; and to obtain a high assembly pose accuracy, it is necessary to control the pose calculation and hand-eye calibration uncertainties within the same order of magnitude as much as possible. If one of the uncertainties is too high, then no matter how precise the other is, it will be of no avail. Therefore, to a certain extent, to obtain a high assembly pose accuracy, improving the hand-eye calibration accuracy and the pose calculation accuracy are equally important.

[0136] Example 2

[0137] Step 1: After completing the construction of the hardware platform, in order to preliminarily evaluate the accuracy of the entire assembly system, the relevant input uncertainties were obtained using the method of this article, and a preliminary analysis of the system uncertainty was carried out. The uncertainty analysis results are shown in Table 7.

[0138] Table 7 Uncertainty of the assembly system before

[0139]

[0140] Step 2: As shown in Table 7, the initial accuracy of the assembly system is far from the actual assembly accuracy requirements. Especially the hand-eye calibration uncertainty is relatively large, which greatly affects the uncertainty of the assembly pose. Guided by this conclusion, relevant accuracy optimization work was carried out to improve the hand-eye calibration accuracy and the pose measurement accuracy, and finally these two uncertainties were reduced to within the same order of magnitude. After completing the above work, the method of the present invention was continued to re-obtain the relevant input uncertainties, and the uncertainty of the assembly system was further analyzed. The results are shown in Table 8.

[0141] Table 8 Uncertainty of the current assembly system

[0142]

[0143] Step 3: The data in Table 8 show that the accuracy of the current assembly system has basically reached the assembly accuracy requirements, that is, the assembly system already has the assemblability. With the support of the data in Table 8, a certain system was successfully assembled.

[0144] Through the above embodiments, on the one hand, it proves the feasibility of the method of the present invention for the uncertainty analysis of complex systems, especially applicable to large assembly systems with high precision requirements; on the other hand, it proves the accuracy of the simulation data obtained by the method of the present invention. The simulation results are close to the final assembly results, indicating that the method of the present invention can better characterize the accuracy of the assembly pose of the entire system.

[0145] The conclusion of this Embodiment 2 is that when the method of the present invention is applied to actual assembly production, the following are achieved: 1. Preliminarily evaluate the accuracy of the entire assembly system before assembly and find that the existing accuracy cannot meet the requirements; 2. After precision optimization, re-evaluate the accuracy of the entire assembly system and confirm that the existing accuracy already meets the requirements.

[0146] The above embodiments only represent one implementation manner of the present invention, and the description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention patent shall be subject to the appended claims.

Claims

1. An uncertainty analysis method for a monocular vision measurement system for assembly pose, characterized in that The uncertainty analysis method for the monocular vision measurement system for assembly pose includes the following steps: S1: Determine the uncertainty transfer route of the assembly pose combination system; the measurement models of the assembly pose combination system include at least one of the following: camera imaging model, camera calibration model, vision measurement model, hand-eye calibration model; S2: Obtain the image pixel uncertainty A, the image pixel uncertainty B, and the robot motion uncertainty as input uncertainties; among them, the image pixel uncertainty A represents the pixel uncertainty of the camera's corner feature extraction in the checkerboard calibration board during camera calibration or hand-eye calibration; the image feature pixel uncertainty B represents the pixel uncertainty of the camera's relevant feature extraction in the target image during pose calculation; the robot motion uncertainty is the positioning accuracy of the used robot; S3: Calculate in sequence to obtain the camera calibration uncertainty, the pose calculation uncertainty, the hand-eye calibration uncertainty, and the assembly pose uncertainty; among them, the uncertainties independently include three position degrees of freedom X, Y, Z, and three attitude degrees of freedom A, B, C; S4: Analyze the uncertainty sensitivity of hand-eye calibration and assembly pose in sequence.

2. The uncertainty analysis method for a monocular vision measurement system for assembly pose according to claim 1, characterized in that: The specific steps to obtain the image pixel uncertainty A in step S2 are as follows: Use an industrial camera to collect the feature of the checkerboard corner. During the image acquisition process, first set the camera frame rate to the highest threshold. In the experiment, take pictures N times repeatedly, obtain the coordinate distributions of the same corner feature N times, record them, and calculate the standard deviations in the u direction and the v direction respectively as the uncertainty of corner feature extraction, that is, the image pixel uncertainty A, where the u direction and the v direction are based on the image collected by the camera as a plane rectangle, with the horizontal direction being the u direction and the vertical direction being the v direction. That is, for the coordinate distribution of the corner, the standard deviation in the u direction is the standard deviation of the corner abscissa, and the standard deviation in the v direction is the standard deviation of the corner ordinate, and N is a positive integer.

3. The uncertainty analysis method for a monocular vision measurement system for assembly pose according to claim 2, characterized in that: In step S2, the steps to obtain the image pixel uncertainty B are as follows: Calculate the pose of the checkerboard calibration board in the camera coordinate system, select the PnP algorithm for the pose calculation method, and calculate the uncertainty of checkerboard corner feature extraction, which is the image pixel uncertainty B.

4. The uncertainty analysis method for a monocular vision measurement system for assembly pose according to claim 2, wherein: The steps to obtain the camera calibration uncertainty in S3 are as follows: Add the previously obtained image pixel uncertainty A to the image features used for camera calibration to generate calibrated two-dimensional feature points and corresponding three-dimensional space points, then repeatedly run the camera calibration method, and statistically calculate the standard deviation of the output parameters, which is recorded as the camera calibration uncertainty.

5. The uncertainty analysis method for a monocular vision measurement system for assembly pose according to claim 3, wherein: The steps to obtain the pose calculation uncertainty in S3 are as follows: Add the previously obtained image pixel uncertainty B to a certain determined image feature used for pose calculation, add the previously obtained camera uncertainty to the camera internal parameters, then repeatedly run the pose calculation method, and statistically calculate the standard deviation of the output parameters, which is recorded as the pose calculation uncertainty, where the pose includes three position degrees of freedom X, Y, Z, and three attitude degrees of freedom A, B, C.

6. The uncertainty analysis method for a monocular vision measurement system for assembly pose according to claim 4, wherein: The steps to obtain the hand-eye calibration uncertainty in S3 are as follows: Add the pre-obtained image pixel uncertainty A to the image features for hand-eye calibration, add the pre-obtained camera uncertainty to the camera internal parameters to obtain the pose of the calibration board in the camera coordinate system, then add the pre-obtained robot motion uncertainty to the robot pose for hand-eye calibration, and then use the Horaud hand-eye calibration method. Repeat running the hand-eye calibration method and statistically calculate the standard deviation of the output parameters, which is denoted as the hand-eye calibration uncertainty.

7. An uncertainty analysis method for a monocular vision measurement system for assembly pose according to claim 5, characterized in that: The steps to obtain the uncertainty of the assembly pose in S3 are as follows: Add the pre-obtained pose calculation uncertainty to the pose of the target in the camera coordinate system, add the pre-obtained hand-eye calibration uncertainty to the hand-eye relationship matrix, and then repeat running the assembly pose calculation method and statistically calculate the standard deviation of the output parameters, which is denoted as the uncertainty of the assembly pose.

8. The uncertainty analysis method for a monocular vision measurement system for assembly pose according to claim 1, characterized in that: In S4, the steps to analyze the uncertainty sensitivity of hand-eye calibration are as follows: First, increase the X uncertainty in the pose calculation uncertainty from 0 by m times in sequence and analyze the change in hand-eye calibration uncertainty; then increase the A uncertainty in the pose calculation uncertainty from 0 by n times in sequence and analyze the change in hand-eye calibration uncertainty; subsequently, increase the position uncertainties X, Y, and Z in the robot motion uncertainty from 0 by s times in sequence and analyze the change in hand-eye calibration uncertainty; finally, increase the attitude uncertainties A, B, and C in the robot motion uncertainty from 0 by t times in sequence and analyze the change in hand-eye calibration uncertainty; where m, n, t, and s are positive integers.

9. The uncertainty analysis method for a monocular vision measurement system for assembly pose according to claim 1, characterized in that: In S4, the steps to analyze the uncertainty sensitivity of the assembly pose are as follows: First, increase the X uncertainty in the pose calculation uncertainty from 0 by h times (a certain number of times) in sequence and analyze the change in the assembly pose uncertainty; then increase the A uncertainty in the pose calculation uncertainty from 0 by i times in sequence and analyze the change in the assembly pose uncertainty; subsequently, increase the X uncertainty in the hand-eye calibration uncertainty from 0 by j times in sequence and analyze the change in the assembly pose uncertainty; finally, increase the A uncertainty in the hand-eye calibration uncertainty from 0 by k times in sequence and analyze the change in the assembly pose uncertainty; where h, i, j, and k are positive integers.

Images