Automated Inspection Method for Sheet Metal Parts Based on Multi-Perspective 3D Scanning

Abstract

The present invention discloses an automated inspection method for sheet metal parts based on multi-view 3D scanning. A 6-axis robotic arm drives a 3D camera to move, and is combined with the rotation of a turntable to obtain the point cloud data and image data of the outer surface of the sheet metal part. Through a structured light 3D reconstruction algorithm, a point cloud stitching algorithm based on coded points, a 3D hole center positioning algorithm based on multi-views, and an edge 3D reconstruction algorithm based on the part digital model, the accuracies of indexes such as the diameter of the high-precision sheet metal round hole, the center distance of the holes, the height, and the flatness reach within 0.1 mm. The system implemented by the present invention has good robustness, and solves the problem that current pure 2D vision measurement cannot measure indexes such as height, flatness, and hole diameters at different positions. It solves the problem of low measurement accuracy of workpieces caused by low stitching accuracy in pure 3D point cloud processing and low quality of edge point cloud reconstruction of sheet metal parts.

Description

Technical Field

[0001] The present invention relates to the field of industrial automation measurement technology, and in particular to an automated detection method for sheet metal parts based on multi-view 3D scanning. Background Art

[0002] With the development and progress of modern industry, especially in some high-precision processing industries, traditional detection methods are far from meeting the needs of production. Machine vision measurement technology is a non-contact measurement method based on optical imaging, digital image processing, and computer graphics, with a wider measurement range, higher measurement accuracy and efficiency.

[0003] At present, there are contact-type three-coordinate machines in the field of high-precision measurement. Although the three-coordinate machine has high accuracy, its structure is complex and expensive. It has high environmental requirements, time-consuming measurement path planning, and the probe is easily damaged. Non-contact measurement is divided into 2D visual measurement and 3D visual measurement. 2D visual measurement uses a single industrial camera and installs an industrial fixed-focus lens or a center lens. Generally, the workpiece needs to be back-lit. The use of industrial lenses can only be limited to measuring on a calibrated plane. If the workpiece indicators are not on the same plane, they cannot be measured. Generally, multiple cameras are required to measure different parameters of a workpiece. The system is highly complex, and workpiece switching is troublesome and difficult to achieve. In addition, it is impossible to measure parameters such as flatness and out-of-plane distance. With the installation of a center lens, the workpiece size measurement in the direction perpendicular to the camera has high accuracy, but it is also impossible to measure parameters such as flatness and out-of-plane distance.

[0004] 3D vision measurement uses line lasers in combination with motion platforms or handheld 3D scanners or structured light 3D cameras to perform 3D reconstruction on workpieces and process the acquired point clouds for measurement. When a line laser is combined with a motion platform, the measurement accuracy is high, but the field of view is small, and it is impossible to completely splice the entire outer surface of the workpiece. Generally, it is used to detect height and flatness. The hole pitch and round hole parameters depend on the quality of the point cloud at the hole edge, and the measurement accuracy is also limited. For a handheld 3D scanner, reflective target points need to be pasted on and around the workpiece, and the point cloud is processed for splicing. The scanner is carried by hand or a robot for scanning, and the point cloud of the inner wall of the round hole can be obtained from different angles. The measurement accuracy is relatively high, but for small holes or holes with shallow depths, the measurement accuracy also depends on the quality of the hole edge. Moreover, the reconstruction accuracy of the handheld scanner for the edge is not high. And it is difficult to achieve high-efficiency automation because stickers are required for each workpiece switch, and some workpieces that cannot be touched cannot have reflective points pasted. Manual post-processing of data calculation is also required, and online calculation is not possible. The cost of using a robotic arm with a 3D scanner is also very high, and automated measurement software needs to be written. A pure structured light 3D camera combined with a robot can reconstruct all the point clouds on the outer surface of the workpiece, but its reconstruction effect for the edge is also not good, with a difference of up to 1 to 2 pixel distances. Splicing is generally carried out through calibration with the robot, and the splicing accuracy depends on the absolute positioning accuracy of the robot, which is generally not high, about 0.5 mm. Robots with high absolute positioning accuracy are very costly. Or it depends on the characteristics of the workpiece itself for splicing, but for workpieces with unclear or symmetrical characteristics, the splicing effect is unpredictable and there is no good robustness. Summary of the Invention

[0005] The present invention provides an automated inspection method for sheet metal parts based on multi-view 3D scanning, which can achieve high-precision automated measurement of the aperture, hole pitch, height, and flatness of sheet metal processing parts.

[0006] The automated inspection method for sheet metal parts based on multi-view 3D scanning includes the following steps:

[0007] Step 1: Use computer software as the main control terminal and communicate and connect with the robot, 3D camera, and turntable.

[0008] Step 2: Manually place the sheet metal part at the center position of the turntable and select to start the system according to the type of the placed workpiece.

[0009] Step 3: The system controls the robotic arm to move the 3D camera to the planned specified position and stop. Cooperate with the rotating turntable to make the workpiece at a suitable angle, and then take 2D supplementary light pictures and 24 structured light pictures of the workpiece and save the data.

[0010] Step 4: Use the data of the 24 collected structured light pictures to reconstruct a dense point cloud through the structured light 3D reconstruction algorithm.

[0011] Step 5: Using a point cloud stitching algorithm based on coding points, stitch the dense point cloud into the turntable coding point coordinate system, save the matrix used for stitching calculation, and display the dense point cloud after reconstruction and stitching on the software interface;

[0012] Step 6: Repeat Steps 3 to 5 until the entire outer surface data of the workpiece is captured and calculated;

[0013] Step 7: Perform filtering on the stitched complete point cloud to remove miscellaneous points. Using a 3D hole center localization algorithm based on multi-view, calculate the 3D coordinates of the center of each hole in the picture through multiple 2D supplementary light pictures captured and saved in the previous step, perform point-to-point registration with the hole positions in the design digital model, obtain the rigid body transformation matrix for registration, and transform the point cloud into the digital model coordinate system;

[0014] Step 8: According to the captured 2D supplementary light pictures, use an edge 3D reconstruction algorithm based on the part digital model to calculate the actual discrete 3D coordinates of all contours;

[0015] Step 9: Perform least squares plane fitting on all discrete 3D coordinates of each individual circular contour. According to the fitted plane, correct the discrete 3D coordinates of each individual circular contour to a plane with a normal vector equal to (0, 0, 1). Then, use the X and Y coordinates of the corrected discrete 3D coordinates of the circular contour to perform least squares fitting on the circle to obtain the aperture value and the hole center coordinates;

[0016] Step 10: Restore the corrected hole center coordinates to obtain the hole center coordinates and hole center distance values in the digital model coordinate system;

[0017] Step 11: Through Step 7, transform the point cloud into the digital model coordinate system. Segment and intercept the point cloud according to the pre-planned height region to be obtained in the digital model, perform plane fitting on the intercepted region, and project the planned points onto the fitted plane to obtain the position coordinates in the digital model. Transform the position coordinates into the turntable plane coordinate system to obtain the height of this coordinate region on the turntable, and obtain the height information of the sheet metal part;

[0018] Step 12: Through Step 7, transform the point cloud into the digital model coordinate system. Segment and intercept the point cloud according to the pre-planned flatness region to be obtained in the digital model, and directly calculate the flatness of the segmented point cloud to obtain the flatness calculation result;

[0019] Step 13: Make all the required data into a table form, display it in the software, and save it as EXCEL to form a measurement report, and the system operation ends.

[0020] Further, in Step 4, when reconstructing the dense point cloud using the 24 captured structured light picture data through the structured light 3D reconstruction algorithm, the structured light 3D reconstruction algorithm is specifically:

[0021] Design the structured light encoding pattern according to the resolution of the camera to generate the corresponding encoding pattern;

[0022] Decode the collected structured light pattern to achieve pixel-level structured light image encoding;

[0023] According to the encoding result and the calibration parameters of the 3D camera, calculate the spatial coordinates corresponding to all encoded pixel points to form a dense 3D point cloud.

[0024] Further, step 5 is based on a point cloud stitching algorithm for encoded points, specifically:

[0025] Extract and identify the encoded points in the images collected at different poses to obtain the image coordinates and specific encodings of the centers of the encoded points;

[0026] Match the encoded points in the images at different poses according to the encoding result, solve the fundamental matrix between two adjacent images, and calculate the relative pose between different images;

[0027] Perform global pose optimization, calculate the spatial coordinates of all encoded points, and generate a point cloud stitching framework;

[0028] Stitch the point clouds obtained at different poses according to the framework to obtain the complete point cloud data.

[0029] Further, step 7 is based on a 3D positioning algorithm for hole centers in multiple views, and the specific processing process is as follows:

[0030] Extract the hole contours in the images collected at different poses to obtain the image coordinates of the hole centers;

[0031] Calculate the epipolar lines according to the obtained fundamental matrix for image matching of the hole centers to achieve matching of the hole centers in all collected images;

[0032] Merge the matching results of the hole centers into the matching results of the encoded points, and then perform global pose optimization again to calculate the 3D coordinates of all hole centers.

[0033] Further, step 8 is specifically:

[0034] Extract the edge contours in the images collected at different poses, discretize the contours at the pixel level, and obtain the discrete image points of the contours;

[0035] Register the hole center coordinates obtained in the hole center positioning algorithm module with the hole center coordinates in the part digital model to obtain the spatial pose of the digital model relative to all shooting poses;

[0036] Extract all the contours that need to be reconstructed in the digital model, and perform discretization to obtain the spatial discrete points of the contours;

[0037] Back-project the spatial discrete points back into the corresponding images according to the spatial pose, find the image contour discrete points closest to the back-projected discrete points, and perform encoding;

[0038] Calculate the actual discrete 3D coordinates of all the contours according to the obtained pose parameters.

[0039] The beneficial effects of the present invention are:

[0040] The present invention solves the problems that 2D vision in the field of non-contact measurement cannot measure flatness and non-coplanar dimensions, and that 3D vision cannot perform high-precision automated measurement of aperture and hole pitch. Through encoding, point cloud is stitched with high precision, and then through the idea of 2D and 3D fusion, the hole edges are reconstructed with high precision to obtain high-precision sheet metal workpiece size information. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The drawings described herein are used to provide a further understanding of the present invention, and constitute a part of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention, and do not constitute an improper limitation to the present invention. In the drawings:

[0042] Figure 1 is the system flowchart design diagram of the present invention;

[0043] Figure 2 is the standard plate stp model diagram of the present invention;

[0044] Figure 3 is the digital model diagram of the sheet metal part 1 tested by the present invention;

[0045] Figure 4 is the detection result of the sheet metal part 2 by the system of the present invention;

[0046] Figure 5 is the detection result of the sheet metal part 3 by the system of the present invention;

[0047] Figure 6 is the detection result of the sheet metal part 4 by the system of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0048] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the specific embodiments of the present invention and the corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0049] The technical solutions provided by the embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings.

[0050] The method used in the present invention is based on a camera clamping motion module, a 3D camera module, and a turntable module. The camera clamping motion module includes a 6-axis robotic arm and a camera fixing tooling. The 3D camera module includes a camera, a projector, and an annular light source. The turntable module mainly includes a PLC, a stepper motor, and a disc with two circles of coding points on it.

[0051] As Figure 1 shown, the automated inspection method for sheet metal parts based on multi-view 3D scanning includes the following steps:

[0052] Step 1: Use computer software as the master control terminal to establish communication connections with the robot, 3D camera, and turntable.

[0053] Step 2: Manually place the sheet metal part at the center position of the turntable, select according to the type of the placed workpiece, and start the system.

[0054] Step 3: The system controls the robotic arm to move the 3D camera to the planned designated position and stop, cooperate with the rotating turntable to make the workpiece at a suitable angle, then take 2D supplementary light pictures and 24 structured light pictures of the workpiece, and save the data.

[0055] Step 4: Use the collected data to reconstruct the point cloud through the structured light 3D reconstruction algorithm. The specific process of the structured light 3D reconstruction algorithm is as follows:

[0056] ① Design the structured light coding pattern according to the resolution of the camera, generate the corresponding coding pattern for the projection of structured light;

[0057] ② Decode the collected structured light pattern to achieve pixel-level structured light image coding;

[0058] ③ According to the coding result and the calibration parameters of the 3D camera, calculate the spatial coordinates corresponding to all coding pixel points to form a dense 3D point cloud.

[0059] Step 5: Through the point cloud stitching algorithm based on coding points, stitch the point cloud to the turntable target point coordinate system with high precision, save the matrix used for stitching calculation, and display the reconstructed and stitched point cloud on the software interface. The specific process of the point cloud stitching algorithm based on coding points is as follows:

[0060] ① Extract and identify the coding points in the images collected at different poses, and obtain the image coordinates and specific coding of the centers of the coding points;

[0061] ②Match the encoded points in the images with different poses according to the encoding results, so as to solve the fundamental matrix between two adjacent images, and further calculate and obtain the relative pose between different images;

[0062] ③Perform global pose optimization, calculate the spatial coordinates of all encoded points, and generate a point cloud stitching skeleton;

[0063] ④Stitch the point clouds obtained at different poses according to the skeleton to obtain the complete point cloud data.

[0064] Step 6, repeat Step 3 to Step 5 until the entire outer surface data of the workpiece is completed by shooting and calculation.

[0065] Step 7, perform filtering on the stitched complete point cloud to remove noise points. Through the obtained stitching matrix and multiple 2D supplementary light pictures saved in the previous step, using the multi-view based 3D hole center localization algorithm, calculate the center coordinates of each hole in the picture. Perform point-to-point registration with the hole positions in the design digital model to obtain the rigid body transformation matrix for registration, and then the point cloud can be transformed to the digital model coordinate system. This algorithm module is used for high-precision 3D localization of the hole centers on the part surface. The specific process is as follows:

[0066] ①Extract the hole contours in the images collected at different poses to obtain the image coordinates of the hole centers;

[0067] ②According to the fundamental matrix obtained in the stitching algorithm module, the epipolar lines can be calculated for image matching of the hole centers, and the matching of the hole centers in all collected images is realized;

[0068] ③Merge the matching results of the hole centers into the matching results of the encoded points, and then perform another global pose optimization to calculate the 3D coordinates of all hole centers.

[0069] Step 8, according to the captured 2D supplementary light pictures, use the edge 3D reconstruction algorithm based on the part digital model to calculate the actual discrete 3D coordinates of all contours. This algorithm module is used for 3D reconstruction of the part edge contours (including hole contours). The specific process is as follows:

[0070] ①Extract the edge contours in the images collected at different poses, discretize the contours at the pixel level, and obtain the image discrete points of the contours;

[0071] ②Register the hole center coordinates obtained in the hole center localization algorithm module with the hole center coordinates in the part digital model to obtain the spatial pose of the digital model relative to all shooting poses;

[0072] ③Extract all the contours that need to be reconstructed in the digital model, and perform discretization to obtain the spatial discrete points of the contours;

[0073] ④Back-project the spatial discrete points in ③ back to the corresponding images according to the pose in ②, find the discrete points on the image contour closest to the back-projected discrete points, and perform encoding.

[0074] ⑤Calculate the actual discrete 3D coordinates of all contours according to the pose parameters obtained in Algorithm Module 3.

[0075] Step 9: Perform least squares plane fitting on all discrete 3D coordinates of each individual circular contour. According to the fitted plane, correct the discrete 3D coordinates of each individual circular contour to a plane with a normal vector equal to (0, 0, 1). Then, only use the X and Y coordinates of the corrected discrete 3D coordinates of the circular contour to perform least squares fitting on the circle to obtain high-precision aperture values and hole center coordinates.

[0076] Step 10: Restore the corrected hole center coordinates back to obtain the hole center coordinates in the digital model coordinate system. After knowing the hole center coordinates of each hole in the digital model coordinate system, obtain high-precision hole center distance values.

[0077] Step 11: Transform the point cloud to the digital model coordinate system through Step 7. Segment and intercept the point cloud according to the pre-planned height region to be measured in the digital model. Perform plane fitting on the intercepted region and project the planned points onto the fitted plane to obtain high-precision position coordinates in the digital model. Transform these coordinates to the turntable plane coordinate system to obtain the height of this coordinate region on the turntable and obtain the height information of the sheet metal part.

[0078] Step 12: Transform the point cloud to the digital model coordinate system through Step 7. Segment and intercept the point cloud according to the pre-planned flatness region to be measured in the digital model. Directly calculate the flatness of the segmented point cloud to obtain the flatness calculation result.

[0079] Step 13: Make all the data to be measured into a table form, display it in the software, and save it as EXCEL to form a measurement report, and the system operation ends.

[0080] The specific embodiments of the present invention are as follows:

[0081] First, place a special metrology Invar tetrahedron. The actual data of the Invar tetrahedron has been measured by professional equipment. The standard distance accuracy between the small balls of the Invar tetrahedron is less than 0.001 mm. The robot with a 3D camera performs point cloud reconstruction and stitching on the standard balls on the Invar tetrahedron, fits the small balls in the point cloud and performs distance measurement to obtain the following data, proving that the point cloud reconstruction plus stitching accuracy is less than 0.1 mm. The point cloud reconstruction stitching data is shown in Table 1:

[0082] Table 1

[0083]

[0084] As shown Figure 2 in the figure, a standard plate is prepared, and coding points are pasted on the standard plate. The aperture, hole pitch, height, and flatness of the inspection plate are all measured by a three-coordinate measuring machine with an accuracy of 0.001 mm.

[0085] The standard plate is placed on the turntable, sampled through this system, and solved using the system algorithm. The obtained result proves that the data obtained by this system is less than 0.1. The reference value represents the actual value measured by the three-coordinate measuring machine, the tolerance represents the set tolerance, and the actual value represents the value calculated by the system software algorithm for the workpiece. If the reference value minus the actual value is within the tolerance range, the data in this row is displayed as qualified; otherwise, it is unqualified.

[0086] Four different sheet metal parts are prepared. Through the system steps of the present invention, the system is started to automatically sample and calculate the sheet metal parts respectively, and the results are displayed. Figure 3 is the digital model of sheet metal part 1, Figure 4 is the software calculation result of sheet metal part 2, Figure 5 is the software calculation result of sheet metal part 3, Figure 6 is the software calculation result of sheet metal part 4. In the calculation result diagram, the reference value is the drawing design value of the sheet metal part, the upper tolerance and the lower tolerance represent the deviations that the machining in the drawing cannot exceed, and the actual value represents the value obtained by system calculation. If the reference value minus the actual value does not exceed the designed deviation, it is displayed as qualified; otherwise, it is displayed as unqualified. All the above workpieces have obtained qualified results. Moreover, the present invention can not only measure aperture, hole pitch, flatness, height information, but also measure the distance between edges in space, the distance between an edge and a surface, the distance between surfaces, the distance between different holes, etc.

[0087] In summary, the present invention solves the problems that 2D vision in the non-contact measurement field cannot measure flatness and non-planar dimensions, and that 3D vision cannot perform high-precision automated measurement of aperture and hole pitch. Through high-precision stitching of coding points to point clouds, and then through the idea of 2D and 3D fusion, high-precision reconstruction of the hole edge is carried out to obtain high-precision size information of sheet metal workpieces.

[0088] The above description is only for the embodiments of the present invention and is not intended to limit the present invention. For those skilled in the art, various changes and modifications can be made to the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the scope of the claims of the present invention.

Claims

1. The automatic detection method of sheet metal parts based on multi-view 3D scanning is characterized by: The following steps are involved: Step 1: Use the computer software as the main control terminal to communicate with the robot, 3D camera, and turntable; Step 2: manually place the sheet metal part at the center of the turntable and select and start the system according to the type of workpiece to be placed; Step 3: The system controls the robot arm to move with the 3D camera to the planned designated position and stops, cooperates with the rotating turntable to make the workpiece at a suitable angle, and then takes a 2D fill light picture of the workpiece and 24 structured light pictures, and saves the data; Step 4: Use the collected 24 structured light image data to reconstruct a dense point cloud through a structured light 3D reconstruction algorithm; Step 5, stitching the dense point cloud to the turntable coding point coordinate system through the coding point based point cloud stitching algorithm, saving the matrix used for stitching calculation, and displaying the dense point cloud after the reconstruction and stitching in the software interface; Step 6, repeating steps 3 to 5 until the entire outer surface data of the workpiece is photographed and calculated; Step 7: filter the spliced ​​point cloud to remove noise, use the multi-view based hole center 3D positioning algorithm, and calculate the 3D coordinates of the center of each hole in the picture through multiple 2D fill light pictures taken and saved in the previous step, and perform point-to-point registration with the hole positions of the designed digital model to obtain the registered rigid body transformation matrix, and transform the point cloud to the digital model coordinate system; Step 8, according to the captured 2D fill light picture, using the edge 3D reconstruction algorithm based on the part digital model, calculate the actual discrete 3D coordinates of all contours; Step 9, perform least squares plane fitting on all discrete 3D coordinates of each individual circular contour, and correct each discrete 3D coordinate of the circular contour to a plane whose normal vector is equal to (0, 0, 1) according to the fitted plane, and then use the X and Y coordinates of the corrected discrete 3D coordinates of the circular contour to perform least squares fitting on the circle to obtain the aperture value and the coordinates of the hole center; Step 10, restore the corrected hole center coordinates to obtain the hole center coordinates and hole center distance values ​​in the digital model coordinate system; Step 11, transforming the point cloud to the digital model coordinate system through step 7, segmenting and intercepting the point cloud according to the required height area pre-planned under the digital model, performing plane fitting on the intercepted area, and projecting the planned points to the fitted plane to obtain the position coordinates under the digital model, transforming the position coordinates to the turntable plane coordinate system, obtaining the height of this coordinate area on the turntable, and obtaining the height information of the sheet metal part; Step 12, transforming the point cloud into the digital model coordinate system through step 7, segmenting and intercepting the point cloud according to the desired flatness area pre-planned under the digital model, and directly performing flatness calculation on the segmented point cloud to obtain the flatness calculation result; Step 13, put all the required data into a table format, display it in the software, and save it in EXCEL to form a measurement report, and the system operation ends.

2. The automatic detection method for sheet metal parts based on multi-view 3D scanning according to claim 1 is characterized in that: The step 4 uses the collected 24 structured light image data to reconstruct a dense point cloud through a structured light 3D reconstruction algorithm. The structured light 3D reconstruction algorithm is specifically as follows: Design the mechanism light coding pattern according to the resolution of the camera and generate the corresponding coding pattern; Decode the collected structured light patterns to achieve pixel-level structured light image encoding; According to the encoding results and the calibration parameters of the 3D camera, the spatial coordinates corresponding to all encoded pixels are calculated to form a dense 3D point cloud.

3. The automatic detection method for sheet metal parts based on multi-view 3D scanning according to claim 1 is characterized in that: The point cloud stitching algorithm based on the coded points in step 5 is specifically as follows: Extract and identify the code points in the images collected at different positions and obtain the image coordinates and specific codes of the code point center; According to the encoding results, the encoding points in the images with different postures are matched, the basic matrix between two adjacent images is solved, and the relative postures between different images are calculated; Perform global pose optimization, calculate the spatial coordinates of all coded points, and generate a point cloud splicing skeleton; The point clouds obtained at different positions are spliced ​​according to the skeleton to obtain complete point cloud data.

4. The automatic detection method for sheet metal parts based on multi-view 3D scanning according to claim 1 is characterized in that: The step 7 is based on the multi-view hole center 3D positioning algorithm, and the specific processing process is as follows: Extract the hole contour from the images collected at different positions and obtain the image coordinates of the hole center; According to the acquired basic matrix, the epipolar lines are calculated for image matching of the hole centers, so as to achieve matching of the hole centers in all acquired images; The matching results of the hole centers are merged with the matching results of the encoding points, and then a global pose optimization is performed to calculate the 3D coordinates of all hole centers.

5. The automatic detection method for sheet metal parts based on multi-view 3D scanning according to claim 1 is characterized in that: The step 8 is specifically as follows: Extract edge contours from images collected at different positions, discretize the contours at the pixel level, and obtain image discrete points of the contours; According to the hole center coordinates obtained in the hole center positioning algorithm module, the hole center coordinates in the part digital model are aligned to obtain the spatial pose of the digital model relative to all shooting poses; Extract all contours that need to be reconstructed in the digital model, and discretize them to obtain spatial discrete points of the contours; Back-projecting the spatial discrete points back to the corresponding image according to the spatial pose, finding the image contour discrete points closest to the back-projected discrete points, and encoding them; The actual discrete 3D coordinates of all contours are calculated based on the acquired pose parameters.

Images