High-precision calibration positioning method based on image segmentation and three-dimensional point cloud

Abstract

According to the high-precision calibration positioning method based on image segmentation and three-dimensional point cloud, a calibration positioning device composed of a workbench, a one-dimensional sliding table, a six-degree-of-freedom mechanical arm, a 3D depth camera, a 2D camera and a zoom lens is adopted, and the sliding table is arranged on the workbench; the mechanical arm is connected with the sliding table; the 3D depth camera, the 2D camera and the zoom lens are all arranged at the tail end of the mechanical arm, and the zoom lens is located in front of the 2D camera; the calibration positioning method comprises the steps that the initial positions of the one-dimensional sliding table and the six-degree-of-freedom mechanical arm are set; shooting a reference object image and obtaining an image segmentation model through training; identifying a reference object by using the image segmentation model, and obtaining a position mask area of the reference object; calibrating a reference object; calibrating a measured object; and high-precision automatic tracking detection among a plurality of detected objects can be realized according to the calibrated detected objects. According to the invention, the problem that a picture with relatively high similarity cannot be obtained due to non-standard placement of a detected object and the like in the existing assembly quality intelligent detection is solved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent detection of assembly quality, and particularly to a high-precision calibration and positioning method based on image segmentation and three-dimensional point cloud. Background Art

[0002] In the intelligent detection of the quality of aerospace cylinder sections, in order to improve the accuracy of inspection, a picture with a high similarity is required. At present, to obtain a picture with a high similarity, it is necessary to fix the cylinder base in a specified area, rotate the cylinder to the specified picture during detection, and cooperate with the robotic arm and slide table to move to the specified position for detection. There are mainly the following problems in this way of obtaining images: 1) The moving range of the universal wheels of the cylinder base is limited, and it is easy to generate errors when fixing the base area, and it is easier to generate errors when rotating the cylinder to the specified picture, thus reducing the detection accuracy; 2) The rotation angle of the cylinder and the movement of the robotic arm are single, and it cannot adapt to the position of the cylinder module; 3) When detecting different processes, it is necessary to rotate the cylinder, and the operation is cumbersome, which affects the detection efficiency. Summary of the Invention

[0003] The purpose of the present invention is to provide a high-precision calibration and positioning method based on image segmentation and three-dimensional point cloud, so as to solve the problem that it is impossible to obtain a picture with a high similarity due to the irregular placement of the object to be measured and the like in the existing intelligent detection of assembly quality.

[0004] To achieve the above purpose, the present invention provides a high-precision calibration and positioning method based on image segmentation and three-dimensional point cloud, which adopts a calibration and positioning device composed of a one-dimensional slide table of a workbench, a six-degree-of-freedom robotic arm, a 3D depth camera, a 2D camera, and a zoom lens. The one-dimensional slide table is arranged on the workbench; the six-degree-of-freedom robotic arm is connected to the one-dimensional slide table; the D depth camera, the 2D camera, and the zoom lens are all arranged at the end of the six-degree-of-freedom robotic arm, and the zoom lens is located in front of the 2D camera; the calibration and positioning method includes:

[0005] 1) Set the initial positions of the one-dimensional slide table and the six-degree-of-freedom robotic arm;

[0006] 2) Take a reference object image and obtain an image segmentation model through training;

[0007] At the initial position, use the 3D depth camera to collect the two-dimensional RGB image of the cylinder section and label the reference object to obtain a data set; select part of the data from the data set, use the image segmentation algorithm for training to obtain a deep learning image segmentation model, and use the untrained data in the data set to verify the performance of the image segmentation model;

[0008] 3) Use the image segmentation model to identify the reference object and obtain the position mask area of the reference object;

[0009] Run the one-dimensional sliding table and the six-degree-of-freedom robotic arm to the initial position. Use a 3D depth camera to capture a 2D RGB image of the cylindrical cabin section. Use the image segmentation model obtained in step 2) to identify the reference object in the 2D RGB image captured by the 3D depth camera, and obtain the position mask area of the reference object formed by pixel points.

[0010] 4) Reference object calibration;

[0011] Convert the RGB image of the position mask area of the reference object combined with the depth image formed by the 3D depth camera into 3D point cloud data; use the iterative algorithm to fit the 3D point cloud data of the reference object to the plane with the best degree. After obtaining the optimized plane, identify the edges and four corner points of the plane, and use the coordinates of the four corner points to obtain the pose (x, y, z, rx, ry, rz) of the center point of the reference object. Among them, x, y, and z are the components of the coordinates of the center point P of the reference object on the x-axis, y-axis, and z-axis of the 3D depth camera coordinate system, representing the straight-line distance of the center point P of the reference object relative to the origin O of the 3D depth camera coordinate system; rx, ry, and rz are the Euler angles, representing the rotation angle of P relative to O. rx is the rotation angle around the x-axis, ry is the rotation angle around the y-axis, and rz is the rotation angle around the z-axis; transmit the pose information of the center point of the reference object to the six-degree-of-freedom robotic arm.

[0012] 5) Calibration of the object under test;

[0013] Calibrating the object under test means calibrating the pose of the object under test relative to the reference object, specifically as follows:

[0014] a) Define the pose matrix of the reference object;

[0015] Define the pose matrix of the reference object through the pose information of the center point of the reference object:

[0016] T A =[R A |t A 4×4

[0017] Where R A is the rotation matrix of the reference object, and t A is the position vector of the reference object;

[0018] Calculate the rotation matrix R A :

[0019] 3D matrix rotated around the z-axis:

[0020] 3D matrix rotated around the x-axis:

[0021] 3D matrix rotated around the y-axis:​

[0022] Therefore, the rotation matrix R of the reference object A is as follows:

[0023]

[0024] b) Define the fixed distance vector between the object under test and the reference object;

[0025] Assume that the fixed distance between the object under test and the reference object is represented in the reference object coordinate system as:

[0026] l A = [l x , l y , l z T

[0027] where l x , l y , l z represent the components of the coordinates of the center point of the object under test on the X-axis, Y-axis, and Z-axis of the reference object coordinate system respectively, and the origin of the reference object coordinate system is the center point of the reference object;

[0028] c) Calculate the position vector t of the object under test in the reference object coordinate system B ;

[0029] t B = t A + R A ·l A

[0030] d) Calculate the rotation matrix of the object under test

[0031] There is a relative rotation angle θ between the object under test and the reference object. Calculate the relative rotation matrix R rel of the object under test:

[0032]

[0033] According to the relative rotation matrix R rel calculate the rotation matrix R B of the object under test as follows:

[0034] R B = R A ·R rel

[0035] e) Calculate the pose matrix T of the object under test B

[0036]

[0037] ​f) Extract the Euler angles of the object under test from the pose matrix of the object under test;

[0038] After obtaining the six-dimensional pose information of the center point of the object under test through the above calculations, it is transmitted to the six-degree-of-freedom robotic arm, and the six-dimensional pose information of the center point of the object under test is saved.

[0039] The above high-precision calibration and positioning method based on image segmentation and three-dimensional point cloud, wherein, during detection, the cylindrical cabin section is randomly placed within the specified area; first, perform steps 1) to 4) to complete the calibration of the reference object; calculate the pose of the six-degree-of-freedom robotic arm according to the six-dimensional pose information of the center point of the object under test in the reference object coordinate system, and adjust the pose of the six-degree-of-freedom robotic arm according to the calculated pose of the six-degree-of-freedom robotic arm; after the pose is adjusted, use a 2D camera to collect the image of the object under test for quality inspection. According to the six-dimensional pose information of the center point of the object under test in the reference object coordinate system, high-precision automatic tracking detection between multiple objects under test can be achieved; when detecting the next object under test after detecting one object under test, directly go from the previous object under test to the next object under test.

[0040] Compared with the prior art, the beneficial technical effects of the present invention are:

[0041] 1) The high-precision calibration and positioning method based on image segmentation and three-dimensional point cloud of the present invention can ensure a high image similarity without strictly specifying the specific placement positions of the base and the cylindrical cabin section, solves the problem in the existing intelligent detection of assembly quality that it is impossible to obtain a picture with a relatively high similarity due to the irregular placement of the object under test, etc., and at the same time realizes the automation of the recognition and positioning of the object under test, realizes the automatic movement of the robotic arm when detecting multiple objects under test, greatly improves the efficiency and accuracy of the assembly quality inspection work, reduces manual interference, can effectively avoid the influence of human factors, and further improves the accuracy of image intelligent detection;

[0042] 2) Since the relative pose between the object under test and the reference object remains unchanged, when the reference object is accurately calibrated, the displacement of the robotic arm is not affected by rotation, base tilt, etc., and high-precision automatic displacement between multiple objects under test can be achieved, laying a solid foundation for subsequent recognition and detection based on computer vision; the high-precision calibration and positioning method based on image segmentation and three-dimensional point cloud of the present invention greatly reduces the image presentation error caused by humans, thereby improving the accuracy of image recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The high-precision calibration and positioning method based on image segmentation and three-dimensional point cloud of the present invention is given by the following embodiments and drawings.

[0044] Figure 1 It is a schematic diagram of the high-precision calibration and positioning device in the embodiment of the present invention.

[0045] Figure 2 This is the flowchart for calibrating the reference object in the embodiments of the present invention. Detailed implementation manners

[0046] The following will further describe in detail Figures 1 to 2 the high-precision calibration and positioning method based on image segmentation and three-dimensional point cloud of the present invention.

[0047] Figure 1 This is the schematic diagram of the high-precision calibration and positioning device in the embodiments of the present invention.

[0048] As Figure 1 , the device includes a workbench 1, a one-dimensional sliding table 2, a six-degree-of-freedom robotic arm 3, a 3D depth camera 4, a 2D camera, and a zoom lens (5);

[0049] The workbench 1 is used to place other components, and its design needs to consider stability and load-bearing capacity. In this embodiment, the material of the workbench 1 is selected as high-strength stainless steel to ensure the minimum deformation during long-term operation;

[0050] The one-dimensional sliding table 2 is arranged on the workbench 1; the cylindrical cabin section 6 is placed horizontally, the workbench 1 is close to the cylindrical cabin section 6, and the one-dimensional sliding table 2 is parallel to the cylindrical cabin section 6;

[0051] The six-degree-of-freedom robotic arm 3 is connected to the one-dimensional sliding table 2 and can move on the one-dimensional sliding table 2. To ensure the stability of the six-degree-of-freedom robotic arm 3 when it stops at a certain position on the one-dimensional sliding table 2 for shooting, the one-dimensional sliding table 2 uses a servo motor to achieve the precise stop of the six-degree-of-freedom robotic arm 3 on the one-dimensional sliding table 2; the six-degree-of-freedom robotic arm 3 adopts a six-axis design, and each joint is driven by a high-precision servo motor and is equipped with a position sensor to ensure the position accuracy of each joint;

[0052] The 3D depth camera 4, the 2D camera, and the zoom lens are all arranged at the end of the six-degree-of-freedom robotic arm 3, and the zoom lens is located in front of the 2D camera; the 2D camera can obtain high-definition RGB (Red, Green, Blue) images of the object to be measured. In this embodiment, the 2D camera uses a 2D camera with a high resolution of 12 million pixels, which can ensure image clarity and detail capture ability. The zoom lens can adjust the focal length according to needs to obtain images with different perspectives and different magnification factors, enhancing the adaptability of the device; the 3D depth camera 4 is used to obtain the three-dimensional information of the object to be measured, and the RGB-D (RGB-Depth) image obtained by the 3D depth camera 4 is an important device for achieving precise recognition and positioning.

[0053] Figure 2 This is the flowchart for calibrating the reference object in the embodiments of the present invention.

[0054] Combined with Figure 1 and Figure 2, the high-precision calibration and positioning method based on image segmentation and 3D point cloud in this embodiment includes:

[0055] 1) Set the initial positions of the one-dimensional sliding table 2 and the six-degree-of-freedom robotic arm 3;

[0056] Select a rigid module as a reference object on the end face of the cylindrical cabin section 6 near the six-degree-of-freedom robotic arm. Preferably, the rigid module is in the shape of a cuboid;

[0057] The initial positions of the one-dimensional sliding table 2 and the six-degree-of-freedom robotic arm 3 should ensure that both the 3D depth camera 4 and the 2D camera can capture the entire view of the end of the cylindrical cabin section 6 near the six-degree-of-freedom robotic arm;

[0058] Set the distance between the one-dimensional sliding table 2 and the cylindrical cabin section 6 and the initial pose of the six-degree-of-freedom robotic arm 3. The initial pose of the six-degree-of-freedom robotic arm 3 includes the position of the six-degree-of-freedom robotic arm 3 on the one-dimensional sliding table 2 and the attitude of the six-degree-of-freedom robotic arm 3;

[0059] 2) Take reference object images and obtain an image segmentation model through training;

[0060] At the initial position, use the 3D depth camera 4 to collect the two-dimensional RGB image of the cylindrical cabin section 6 (mainly the two-dimensional RGB image of the end face of the cylindrical cabin section 6) and label the reference object to obtain a dataset; select some data from the dataset and use an image segmentation algorithm for training to obtain a deep learning image segmentation model, and use the untrained data in the dataset to verify the performance of the image segmentation model;

[0061] When using the 3D depth camera 4 to collect the two-dimensional RGB image of the cylindrical cabin section 6, roll the cylindrical cabin section 6 to collect the two-dimensional RGB images of the reference object from different perspectives;

[0062] 3) Use the image segmentation model to identify the reference object and obtain the position mask area of the reference object;

[0063] Run the one-dimensional sliding table 2 and the six-degree-of-freedom robotic arm 3 to the initial position, use the 3D depth camera 4 to take the two-dimensional RGB image of the cylindrical cabin section 6, and use the image segmentation model obtained in step 2) to identify the reference object in the two-dimensional RGB image captured by the 3D depth camera 4 to obtain the position mask area of the reference object formed by pixel points;

[0064] 4) Reference object calibration;

[0065] Convert the RGB image of the position mask area of the reference object and the depth image formed by the 3D depth camera 4 into three-dimensional point cloud data; use an iterative algorithm to fit the three-dimensional point cloud data of the reference object to the plane with the best degree. After obtaining the optimized plane, identify the edges and four corner points of the plane, and use the coordinates of the four corner points to obtain the pose of the center point of the reference object. The obtained pose of the center point of the reference object is based on the x, y, z, rx, ry, rz values in the 3D depth camera coordinate system. Among them, x, y, z are the components of the coordinates of the center point P of the reference object on the x-axis, y-axis, and z-axis of the 3D depth camera coordinate system, representing the straight-line distance of the center point P of the reference object relative to the origin O of the 3D depth camera coordinate system; rx, ry, rz are Euler angles, representing the rotation angle of P relative to O. rx is the angle of rotation around the x-axis, ry is the angle of rotation around the y-axis, and rz is the angle of rotation around the z-axis; transmit the pose information of the center point of the reference object to the six-degree-of-freedom robotic arm 3;

[0066] Obtaining the pose information of the center point of the reference object completes the calibration of the reference object, and save the pose information of the center point of the reference object;

[0067] 5) Calibration of the object to be measured;

[0068] The object to be measured is a module on the cylindrical cabin section 6, and the relative pose between the object to be measured and the reference object remains unchanged; calibrating the object to be measured is to calibrate the pose of the object to be measured relative to the reference object;

[0069] The calibration of the object to be measured is specifically as follows:

[0070] g) Define the pose matrix of the reference object;

[0071] Define the pose matrix of the reference object through the pose information (6D pose information) of the center point of the reference object:

[0072] T A =[R A |t A 4×4

[0073] Among them, R A is the rotation matrix of the reference object, and t A is the position vector of the reference object;

[0074] Calculate the rotation matrix R A of the reference object:

[0075] 3D matrix rotated around the z-axis:

[0076] 3D matrix rotated around the x-axis:

[0077] 3D matrix rotated around the y-axis:​

[0078] Therefore, the rotation matrix R of the reference object A is as follows:

[0079]

[0080] h) Define the fixed distance vector between the object under test and the reference object;

[0081] Assume that the fixed distance between the object under test and the reference object is represented in the reference object coordinate system as:

[0082] l A =[l x , l y , l z T

[0083] where l x , l y , l z represent the components of the coordinates of the center point of the object under test on the X-axis, Y-axis, and Z-axis of the reference object coordinate system respectively, and the origin of the reference object coordinate system is the center point of the reference object;

[0084] i) Calculate the position vector of the object under test in the reference object coordinate system;

[0085] The position vector t of the object under test B can be obtained by rotating and translating the position vector of the reference object:

[0086] t B = t A + R A · l A

[0087] j) Calculate the rotation matrix of the object under test

[0088] Since there is a relative rotation angle θ between the object under test and the reference object, the rotation matrix of the object under test is not the same as that of the reference object, and the relative rotation matrix R of the object under test needs to be calculated rel :

[0089]

[0090] According to the relative rotation matrix R rel calculate the rotation matrix R of the object under test again B as follows:

[0091] R B = R A · R rel

[0092] k) Calculate the pose matrix of the object under test​

[0093] After obtaining the rotation matrix and position vector of the object to be measured, the pose matrix of the object to be measured can be obtained as follows:

[0094]

[0095] l) Extract Euler angles

[0096] After obtaining the pose matrix of the object to be measured, the Euler angles of the object to be measured can be extracted therefrom;

[0097] After obtaining the six-dimensional pose information of the center point of the object to be measured through the above calculations, it is transmitted to the six-degree-of-freedom robotic arm 3, and the six-dimensional pose information of the center point of the object to be measured is saved;

[0098] Repeating step 5) can complete the calibration of multiple objects to be measured, that is, obtain the six-dimensional pose information of the center points of multiple objects to be measured in the reference object coordinate system;

[0099] The calibration of the object to be measured is used for subsequent automatic detection;

[0100] 6) Run detection

[0101] The cylindrical cabin section 6 is randomly placed within the specified area; steps 1) to 4) are executed to complete the calibration of the reference object; according to the six-dimensional pose information of the center point of the object to be measured in the reference object coordinate system that has been obtained, the pose of the six-degree-of-freedom robotic arm 3 is calculated (that is, the displacement amounts of the joints of the six-degree-of-freedom robotic arm 3 are calculated), and the pose of the six-degree-of-freedom robotic arm 3 is adjusted according to the calculated pose of the six-degree-of-freedom robotic arm 3; after the pose is adjusted, a 2D camera is used to collect an image of the object to be measured for quality detection;

[0102] After detecting one object to be measured and proceeding to detect the next object to be measured, directly go from the previous object to be measured to the next object to be measured, without having the six-degree-of-freedom robotic arm 3 return to the initial position and take a calibration shot again. This can improve the time efficiency and avoid errors caused by re-calibration, thereby improving the detection accuracy.

[0103] Since the relative pose between the object to be measured and the reference object remains unchanged, and when the reference object is calibrated accurately, the displacement of the robotic arm is not affected by rotation, base tilt, etc., high-precision automatic tracking detection between multiple objects to be measured can be achieved, laying a solid foundation for subsequent recognition detection based on computer vision; the high-precision calibration and positioning method based on image segmentation and three-dimensional point cloud of the present invention greatly reduces the image presentation error caused by humans, thereby improving the accuracy of image recognition.

Claims

1. A high-precision calibration and positioning method based on image segmentation and 3D point cloud, characterized in that A calibration and positioning device consists of a one-dimensional sliding table on the workbench, a six-degree-of-freedom robotic arm, a 3D depth camera, a 2D camera, and a zoom lens. The one-dimensional sliding table is set on the workbench; the six-degree-of-freedom robotic arm is connected to the one-dimensional sliding table; The 3D depth camera, 2D camera, and zoom lens are all set at the end of the six-degree-of-freedom robotic arm, and the zoom lens is in front of the 2D camera; The calibration and positioning method includes: 1) Set the initial positions of the one-dimensional sliding table and the six-degree-of-freedom robotic arm; 2) Take reference object images and obtain an image segmentation model through training; 3) Use the image segmentation model to identify the reference object and obtain the position mask area of the reference object; 4) Calibrate the reference object; 5) Calibrate the object to be measured.

2. The high-precision calibration and positioning method based on image segmentation and 3D point cloud according to claim 1, wherein Select a rigid module as the reference object on the end face of the cylinder section near the six-degree-of-freedom robotic arm; the initial positions of the one-dimensional sliding table and the six-degree-of-freedom robotic arm should ensure that both the 3D depth camera and the 2D camera can capture the entire view of the end of the cylinder section near the six-degree-of-freedom robotic arm.

3. A high-precision calibration and positioning method based on image segmentation and 3D point cloud according to claim 1, characterized in that, In step 1), set the distance between the one-dimensional sliding table and the cylinder section and the initial pose of the six-degree-of-freedom robotic arm. The initial pose of the six-degree-of-freedom robotic arm includes its position on the one-dimensional sliding table and its attitude.

4. The high-precision calibration and positioning method based on image segmentation and three-dimensional point cloud according to claim 1, characterized in that, Step 2) includes: At the initial position, use the 3D depth camera to collect the two-dimensional RGB image of the cylinder section and annotate the reference object to obtain a data set; select part of the data from the data set and use an image segmentation algorithm for training to obtain a deep learning image segmentation model, and use the untrained data in the data set to verify the performance of the image segmentation model.

5. A high-precision calibration and positioning method based on image segmentation and 3D point cloud as claimed in claim 1, characterized in that, In step 2), when using the 3D depth camera to collect the two-dimensional RGB image of the cylinder section, roll the cylinder section to collect the two-dimensional RGB images of the reference object from different perspectives.

6. The high-precision calibration and positioning method based on image segmentation and 3D point cloud according to claim 1, characterized in that Step 3) includes: Run the one-dimensional sliding table and the six-degree-of-freedom robotic arm to the initial position, use the 3D depth camera to take the two-dimensional RGB image of the cylinder section, and use the image segmentation model obtained in step 2) to identify the reference object in the two-dimensional RGB image taken by the 3D depth camera to obtain the position mask area of the reference object formed by pixel points.

7. A high-precision calibration and positioning method based on image segmentation and three-dimensional point cloud according to claim 1, characterized in that, Step 4) includes: Convert the RGB image of the position mask area of the reference object and the depth image formed by the 3D depth camera into three-dimensional point cloud data; use an iterative algorithm to fit the three-dimensional point cloud data of the reference object to the plane with the best fit. After obtaining the optimized plane, identify the edges and four corner points of the plane, and use the coordinates of the four corner points to obtain the pose (x, y, z, rx, ry, rz) of the center point of the reference object. Among them, x, y, and z are the components of the coordinates of the center point P of the reference object on the x-axis, y-axis, and z-axis of the 3D depth camera coordinate system, representing the straight-line distance of the center point P of the reference object relative to the origin O of the 3D depth camera coordinate system; rx, ry, and rz are Euler angles, representing the rotation angle of P relative to O. rx is the rotation angle around the x-axis, ry is the rotation angle around the y-axis, and rz is the rotation angle around the z-axis; Transmit the pose information of the center point of the reference object to the six-degree-of-freedom robotic arm.

8. A high-precision calibration and positioning method based on image segmentation and 3D point cloud as described in claim 7, characterized in that, In step 5), calibrating the object under test refers to calibrating the pose of the object under test relative to the reference object, specifically as follows: a) Define the pose matrix of the reference object; Define the pose matrix of the reference object through the pose information of the center point of the reference object: T A = [R A |t A 4×4 ​ where R A is the rotation matrix of the reference object, and t A is the position vector of the reference object; Calculate the rotation matrix R of the reference object A : 3D matrix rotated about the z-axis: 3D matrix rotated about the x-axis: 3D matrix rotated around the y-axis: Therefore, the rotation matrix R of the reference object A is as follows: b) Define the fixed distance vector between the object under test and the reference object; Assume that the fixed distance between the object under test and the reference object is expressed in the coordinate system of the reference object as: l A ＝[l x ,l y ,l z ] T where l x , l y , l z respectively represent the components of the coordinates of the center point of the object under test on the X-axis, Y-axis, and Z-axis of the reference object coordinate system, and the origin of the reference object coordinate system is the center point of the reference object; c) Calculate the position vector t of the object to be measured in the reference object coordinate system B ; t B = t A + R A · l A d) Calculate the rotation matrix of the object under test There is a relative rotation angle θ between the object to be measured and the reference object, and the relative rotation matrix R of the object to be measured is calculated rel : According to the relative rotation matrix R rel Calculate the rotation matrix R of the object under test B as follows: R B = R A ·R rel e) Calculate the pose matrix T of the object under test B f) Extract the Euler angles of the object under test from the pose matrix of the object under test; After obtaining the six-dimensional pose information of the center point of the object under test through the above calculations, transmit it to the six-degree-of-freedom robotic arm and save the six-dimensional pose information of the center point of the object under test.

9. A high-precision calibration and positioning method based on image segmentation and 3D point cloud as claimed in claim 8, characterized in that During detection, the cylindrical cabin section is randomly placed within the specified area; first, perform steps 1) to 4) to complete the calibration of the reference object; calculate the pose of the six-degree-of-freedom robotic arm based on the six-dimensional pose information of the center point of the object under test in the coordinate system of the reference object that has been obtained, and adjust the pose of the six-degree-of-freedom robotic arm according to the calculated pose; After the pose is adjusted, use a 2D camera to collect an image of the object under test for quality detection.

10. A high-precision calibration and positioning method based on image segmentation and three-dimensional point cloud according to claim 9, characterized in that, Based on the six-dimensional pose information of the center point of the object under test in the coordinate system of the reference object that has been obtained, high-precision automatic tracking detection between multiple objects under test can be achieved; when detecting the next object under test after detecting one object under test, directly move from the previous object under test to the next object under test.

Images