Robot grabbing deviation correction method based on 3D camera

Abstract

The invention discloses a robot grabbing deviation correction method based on a 3D camera, and belongs to the technical field of industrial robot automatic control and vision measurement. Aiming at the problems of grabbing position offset and unstable round hole feature extraction caused by tray placement deviation, illumination variation and robot attitude error, the method comprises the following steps: collecting and positioning round hole point cloud at the tail end of a robot, realizing point cloud preprocessing by utilizing normal vector estimation, statistical filtering, clustering separation and edge extraction, and finishing coarse registration based on 4PCS; performing weighted fusion positioning on the circle center by combining GICP point cloud registration and RANSAC circle fitting; and then a perspective transformation matrix is constructed according to the final circle center and the theoretical circle center, and two-dimensional perspective correction is conducted on all grabbing point positions in the material disc. The center positioning stability and the overall grabbing precision can be improved under the complex working condition, and the manipulator is suitable for the fields of robot grabbing, material carrying, assembly positioning and the like in automatic production.

Description

Technical Field

[0001] This invention belongs to the field of industrial robot automation control and visual measurement technology, specifically relating to a method for correcting robot grasping points using stereo vision. Background Technology

[0002] With the continuous improvement of industrial automation, robots are increasingly used in material handling, assembly, and welding. Grasping and positioning accuracy has become a key factor affecting production efficiency and stability. To achieve accurate robot grasping in complex environments, visual measurement technology has gradually become the mainstream approach, including methods such as 2D image comparison, laser point / line feature extraction, and 3D point cloud matching. These methods can achieve high recognition accuracy in simple structural or controllable environments. However, in actual production lines, factors such as changes in tray posture, batch differences in workpieces, and robot posture errors can cumulatively affect grasping accuracy, making it difficult for existing methods to operate stably over long periods.

[0003] Existing image comparison methods based on 2D cameras rely on lighting conditions and target surface features. When the workpiece surface has reflections, oil stains, or insufficient texture, the recognition stability and correction accuracy will decrease significantly, resulting in a shift in the gripping position. Although laser point / line measurement methods have a certain degree of environmental adaptability, their measurement range is limited, making it difficult to adapt to complex situations such as overall tray offset and angle changes, and their overall correction efficiency is not high. While positioning methods based on 3D camera point cloud matching have advantages in terms of accuracy and spatial perception, they usually perform point cloud registration on a single workpiece, making it difficult to achieve rapid correction of the overall tray shape and failing to meet the needs of continuous feeding and batch gripping.

[0004] On the other hand, on automated production lines, the actual position and orientation of the material tray often deviate from their theoretical positions due to processing accuracy, handling methods, or placement errors, causing overall translation, rotation, or even dimensional distortion of the gripping points within the tray. The robot's own mechanical errors, installation errors, and posture changes further amplify the gripping deviation, making single-point positioning methods insufficient for overall error compensation. The lack of an effective means to extract key geometric features of the material tray and uniformly correct the entire tray has become a significant problem restricting the improvement of robot gripping accuracy and production efficiency.

[0005] In summary, current technologies cannot reliably and stably obtain the precise center position of the positioning hole under complex working conditions, nor can they correct the overall posture of the material tray through geometric relationships. Therefore, it is impossible to uniformly correct multiple gripping points within the material tray. There is an urgent need for a correction method that can robustly extract positioning features from noisy point clouds and establish a mapping relationship between the actual posture and the theoretical posture of the material tray, in order to improve the robot's gripping accuracy, reduce equipment downtime, and enhance the overall stability of automated production lines. Summary of the Invention

[0006] To address the problems in existing technologies, such as unstable gripping point recognition, inaccurate extraction of circular hole features, and difficulty in uniformly correcting the overall posture of the material tray due to factors like changes in lighting, differences in surface conditions, material tray placement deviations, and robot posture errors, which lead to robot gripping position offsets, difficulty in guaranteeing batch gripping accuracy, and susceptibility to interference in automated production processes, this invention proposes the following solution: A robot grasping and correction method based on a 3D camera includes the following steps: S1. Obtain the point cloud data of the positioning hole; S2. The point cloud data is preprocessed, and coarse registration is performed based on the preprocessing results to obtain the initial coarse registration pose of the positioning circular hole. S3. Using the preprocessed point cloud data and the initial pose of coarse registration as input, the center of the circle is located based on the GICP algorithm and the RANSAC circle fitting algorithm to obtain the final center coordinates of the positioning hole. S4. Based on the final center coordinates and the theoretical center coordinates of the material tray in the robot coordinate system, construct a perspective transformation matrix; S5. Input the theoretical coordinates of each gripping point in the material tray into the perspective transformation matrix to obtain the gripping points after correction.

[0007] Furthermore, S1 installs a 3D industrial camera at the end of the robot. The 3D camera is located 300 to 600 mm directly above the positioning hole. The 3D camera is triggered by TCP to take pictures and obtain the point cloud data of the positioning hole.

[0008] Furthermore, the method for obtaining the initial coarse registration pose of the positioning circular hole in S2 is as follows: S21. Perform normal vector estimation on the point cloud data to obtain the surface normal vector; S22. Noise points are removed based on statistical filtering; S23. Use Euclidean clustering to separate the point cloud data of non-positional circular hole regions; S24. Perform edge extraction on the remaining point cloud data to obtain the edge points of the positioning hole; S25. Match the edge points with the point cloud template of the standard part's circular hole edge contour, and obtain the initial coarse registration pose of the positioning circular hole through the 4PCS algorithm.

[0009] Furthermore, during edge extraction in S24, the edge points of the positioning circular hole are identified by comparing the changes in the angle between the point cloud normal vectors.

[0010] Furthermore, the method for obtaining the final center coordinates of the positioning hole in S3 is as follows: S31. Using the preprocessed point cloud data and the initial pose of coarse registration as input, perform point cloud registration through the GICP algorithm to obtain the coordinates of the first circle center. S32. Using the RANSAC circle fitting algorithm, select multiple point sets from the edge points, and perform plane fitting, point projection and circle fitting on each point set to obtain the coordinates of the second circle center. S33. Filter in-points based on error threshold and repeatedly fit to obtain the coordinates of the third circle center; S34. The first center coordinate, the second center coordinate, and the third center coordinate are combined in a weighted manner to obtain the final center coordinate.

[0011] Furthermore, in S31, point cloud registration is performed using the GICP algorithm, which includes calculating the registration score based on the root mean square error to determine the registration convergence.

[0012] Furthermore, when using the RANSAC algorithm for circle fitting in S32, each subset of the multiple point sets contains 3 points, and the spatial distance between any two of the three points is half the distance between the two furthest points.

[0013] Furthermore, in S4, a point correspondence is established between the final center coordinates of the positioning hole and the theoretical center coordinates of the material tray in the robot coordinate system, and the perspective transformation matrix is ​​constructed based on the point correspondence.

[0014] Furthermore, in S5, the theoretical coordinates of all gripping points in the material tray are input into the perspective transformation matrix in batch mode, and a two-dimensional perspective transformation is performed to obtain all corrected gripping points. The corrected gripping points are then sent to the robot control system for gripping operations.

[0015] Based on the same inventive concept, the present invention also proposes a computer storage medium for storing a computer program, wherein when the computer program is read by a computer, the computer executes the method described in the present invention.

[0016] Compared with the prior art, the present invention has the following beneficial effects: The robot grasping point correction method described in this invention achieves deep purification of the input point cloud by performing operations such as normal vector estimation, statistical filtering, clustering separation, and edge extraction on the original point cloud data. This processing ensures that the point cloud data used for subsequent registration has high continuity and low noise level, clearly expressing edge contour information. It effectively avoids the risks of noise interference and edge breakage that traditional methods are susceptible to under unstable point cloud quality conditions, and significantly improves the quality of acquired localization feature data.

[0017] The robot grasping point correction method described in this invention utilizes a point cloud template of the edge contour of a circular hole in a standard part, combined with a 4PCS coarse registration strategy, to establish a clear geometric reference. This template matching mechanism can accurately calculate the initial pose that conforms to the actual structure, successfully solving the problem of instability in coarse registration caused by the lack of significant local features due to the high symmetry of the circular hole. This provides optimal starting conditions for the subsequent GICP fine registration, effectively ensuring the rapid convergence of the registration process.

[0018] The robot grasping point correction method described in this invention employs a multi-source information weighted fusion mechanism that combines GICP point cloud registration with a RANSAC-based circle fitting strategy when determining the center of the positioning hole. Through steps such as random sampling, plane fitting, and interior point selection, this strategy effectively compensates for the fitting bias and local optima problems of single algorithms when facing uneven point clouds or local occlusion. Finally, by balancing and weighting the multiple circle center estimation results, this system can achieve high repeatability and high accuracy in circle center positioning even under complex field conditions.

[0019] The robot grasping point correction method described in this invention constructs a perspective transformation matrix describing the overall posture changes of the material tray based on the correspondence between the actual and theoretical center coordinates of the positioning holes. By establishing a unified mapping using the geometric relationship of the four key positioning holes of the material tray, this invention can simultaneously correct the translation, rotation, and nonlinear scale differences of the material tray. This mechanism overcomes the limitation that relying solely on local registration cannot reflect the overall posture changes of the material tray, achieving global compensation for the posture deviation of the material tray.

[0020] The robot gripping point correction method described in this invention inputs the theoretical coordinates of all gripping points in the material tray into a perspective transformation matrix in batches, and performs a uniform two-dimensional perspective transformation. This scheme ensures that all gripping positions are corrected within the same geometric mapping framework, completely eliminating error propagation and accumulation caused by multiple independent corrections. This enables the robot to maintain extremely high consistency and operational stability in batch and continuous production scenarios, significantly optimizing the cycle time efficiency of the production line.

[0021] This invention has the ability to stably acquire the features of the positioning circular holes, accurately restore the overall posture of the material tray, and batch correct the gripping points under complex working conditions. It can maintain high gripping accuracy and consistency in environments with large material tray placement errors, complex lighting conditions, and frequent changes in robot posture. It is applicable to technical fields such as industrial robot automatic gripping, intelligent logistics handling, workpiece assembly positioning, and automatic welding loading that require unified correction of batch gripping points. Attached Figure Description

[0022] Figure 1 This is a flowchart of the correction method described in the implementation method; Figure 2 This is a schematic diagram of point cloud filtering using the correction method described in the implementation method; Figure 3 This is a schematic diagram illustrating the extraction of point cloud edge information using the correction method described in the implementation method. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Implementation Method 1 like Figure 1 As shown, a robot grasping and correction method based on a 3D camera includes the following steps: S1. Obtain the point cloud data of the positioning hole; S2. The point cloud data is preprocessed, and coarse registration is performed based on the preprocessing results to obtain the initial coarse registration pose of the positioning circular hole. S3. Using the preprocessed point cloud data and the initial pose of coarse registration as input, the center of the circle is located based on the GICP algorithm and the RANSAC circle fitting algorithm to obtain the final center coordinates of the positioning hole. S4. Based on the final center coordinates and the theoretical center coordinates of the material tray in the robot coordinate system, construct a perspective transformation matrix; S5. Input the theoretical coordinates of each gripping point in the material tray into the perspective transformation matrix to obtain the gripping points after correction.

[0025] By acquiring the point cloud of the positioning holes, performing preprocessing and coarse registration, combining GICP registration and RANSAC circle fitting to determine the center position, constructing the tray posture perspective transformation matrix, and batch calculating the correction coordinates of the gripping points, a unified correction of the overall tray posture deviation and the robot gripping deviation is achieved. This method can maintain high gripping consistency in scenarios with high noise, unstable tray placement, or large robot repetitive positioning errors, enabling the robot to have stable batch gripping capabilities in automated production lines.

[0026] Furthermore, S1 installs a 3D industrial camera at the end of the robot. The 3D camera is located 300 to 600 mm directly above the positioning hole. The 3D camera is triggered by TCP to take pictures and obtain the point cloud data of the positioning hole.

[0027] By mounting a 3D industrial camera on the robot's end effector and limiting the shooting distance, the camera can acquire point cloud data of the positioning circular holes with a fixed posture and stable perspective, reducing the impact of ambient light variations, perspective changes, and installation errors on point cloud quality. Utilizing a TCP signal-triggered image capture mechanism also ensures synchronization between the robot and the camera's movements, improving the temporal consistency of the point cloud data.

[0028] Furthermore, the method for obtaining the initial coarse registration pose of the positioning circular hole in S2 is as follows: S21. Perform normal vector estimation on the point cloud data to obtain the surface normal vector; S22. Noise points are removed based on statistical filtering; S23. Use Euclidean clustering to separate the point cloud data of non-positional circular hole regions; S24. Perform edge extraction on the remaining point cloud data to obtain the edge points of the positioning hole; S25. Match the edge points with the point cloud template of the standard part's circular hole edge contour, and obtain the initial coarse registration pose of the positioning circular hole through the 4PCS algorithm.

[0029] Through operations such as normal vector estimation, noise point cloud removal, clustering separation and edge extraction, the point cloud information related to the positioning of the circular hole is made more concentrated and clear, avoiding noise, occlusion or background structure interference with the shape of the circular hole. At the same time, by matching with the standard part template and using the 4PCS algorithm to obtain the initial pose of coarse registration, the initial pose has a clear geometric reference, which is conducive to improving the stability of subsequent registration and fitting.

[0030] Furthermore, during edge extraction in S24, the edge points of the positioning circular hole are identified by comparing the changes in the angle between the point cloud normal vectors.

[0031] By utilizing the change in the angle between the normal vectors to identify and locate the edge points of the circular hole, edge extraction can be judged based on the local geometric changes of the point cloud, avoiding reliance on simple spatial distribution or point cloud density. This allows for the effective identification of the circular hole contour even when there are local missing points in the point cloud or measurement noise, thus improving the accuracy of edge point extraction.

[0032] Furthermore, the method for obtaining the final center coordinates of the positioning hole in S3 is as follows: S31. Using the preprocessed point cloud data and the initial pose of coarse registration as input, perform point cloud registration through the GICP algorithm to obtain the coordinates of the first circle center. S32. Using the RANSAC circle fitting algorithm, select multiple point sets from the edge points, and perform plane fitting, point projection and circle fitting on each point set to obtain the coordinates of the second circle center. S33. Filter in-points based on error threshold and repeatedly fit to obtain the coordinates of the third circle center; S34. The first center coordinate, the second center coordinate, and the third center coordinate are combined in a weighted manner to obtain the final center coordinate.

[0033] The initial center position is obtained by GICP point cloud registration, and the center obtained by RANSAC circle fitting is used as a supplement. By fusing multiple center estimation results, the final center calculation does not depend on a single algorithm. In cases of uneven point cloud density, noise interference at edge points, or local defects in the positioning hole, this combined strategy can reduce bias and improve the robustness and repeatability of center positioning.

[0034] Furthermore, in S31, point cloud registration is performed using the GICP algorithm, which includes calculating the registration score based on the root mean square error to determine the registration convergence.

[0035] The convergence of GICP registration is evaluated based on the root mean square error, enabling real-time assessment of registration quality and avoiding center estimation shifts caused by insufficient convergence. When the number of point clouds is insufficient or the edge shapes are irregular, this mechanism can promptly adjust or terminate unreliable registration processes, improving overall registration accuracy.

[0036] Furthermore, when using the RANSAC algorithm for circle fitting in S32, each subset of the multiple point sets contains 3 points, and the spatial distance between any two of the three points is half the distance between the two furthest points.

[0037] By limiting the spatial distance between three points in the RANSAC subset, sufficient geometric dispersion among the sampling points is ensured, reducing the instability of circle fitting caused by overly concentrated point sets. When there is significant noise or defects at the edge of the circular hole, this constraint can improve the reliability of the fitting, making the circle center estimation more consistent with the actual structure.

[0038] Furthermore, in S4, a point correspondence is established between the final center coordinates of the positioning hole and the theoretical center coordinates of the material tray in the robot coordinate system, and the perspective transformation matrix is ​​constructed based on the point correspondence.

[0039] A perspective transformation matrix is ​​constructed based on the correspondence between the final center coordinates and the theoretical center coordinates, enabling the material tray posture to be represented through a unified mapping. This method can simultaneously reflect translation, rotation, and scale differences, avoiding the limitations of traditional single-point correction methods that cannot express the overall posture changes of the material tray, and allowing subsequent multi-point corrections to be executed uniformly.

[0040] Furthermore, in S5, the theoretical coordinates of all gripping points in the material tray are input into the perspective transformation matrix in batch mode, and a two-dimensional perspective transformation is performed to obtain all corrected gripping points. The corrected gripping points are then sent to the robot control system for gripping operations.

[0041] By batch inputting the theoretical coordinates of the gripping points on the material tray and performing perspective transformation, multiple gripping points can complete the correction within the same transformation framework, avoiding the accumulation of errors caused by separate calculations. When the production cycle is fast or there are many gripping points, this solution can maintain the consistency and stability of the gripping action, which is beneficial to improving the continuous operation capability of automated production lines.

[0042] Implementation Method 2 This embodiment integrates multiple technical solutions from the aforementioned embodiments, and combines them with the application requirements of robots in welding and material loading scenarios. It fully implements the robot grasping point correction process based on stereo vision, and verifies the performance of the method through specific experiments.

[0043] 1. Acquisition and preprocessing of point cloud data for positioning circular holes In the point cloud acquisition stage, this embodiment mounts a 3D industrial camera on the flange of a six-axis robot. The extrinsic parameter matrix of the camera relative to the robot coordinate system is determined using a robot hand-eye calibration method. In practical applications, the robot moves the camera to a height range of 300mm to 600mm directly above the positioning holes. A TCP signal triggers the 3D industrial camera to capture point cloud data, and the point cloud data including the area of ​​the positioning holes is stored. Positioning holes are set at the four right angles of the rectangular tray. The acquired point cloud data simultaneously includes the tray's shape, the area of ​​the holes, and the point cloud of the surrounding environment.

[0044] Because raw point clouds contain a large amount of redundant data, noise points, and outliers, directly using them for point cloud registration can easily lead to the registration process getting stuck in local optima, registration errors, or slow convergence. This implementation method preprocesses the point cloud before registration, specifically including the following steps: 1.1 The Normal Estimation algorithm is used to estimate the point cloud normal vector, and surface normal vector information is assigned to each point to provide geometric constraints for point cloud registration; 1.2, such as Figure 2 As shown, point cloud filtering is performed based on statistical analysis methods, and isolated noise points are removed by utilizing neighborhood statistical characteristics, thereby reducing the impact of noise on the registration results. 1.3 The Euclidean point cloud clustering method is used to group the point cloud. The point cloud clusters that are obviously far away from the positioning hole area are removed, and only the local point cloud related to the positioning hole is retained to improve the accuracy of the point cloud correspondence. 1.4, such as Figure 3 As shown, the point cloud edge information is extracted by comparing the angle of change of the normal vector angle. While retaining the outline features of the circular hole, the internal redundant data is removed. The figure shows the overall outline, the outline of the circular hole and the local magnified view from left to right. 1.5. A standard part circular hole edge contour point cloud template is used, which is a circular hole workpiece obtained by machining. The point cloud data of the standard part circular hole edge contour point cloud template is prepared by the known PCL method. The preprocessed point cloud and the template are coarsely registered using the 4PCS (Four-PointsConcurrentSets) algorithm. The initial pose of the circular hole in the robot coordinate system is calculated, thereby reducing the search space of the subsequent fine registration and guiding the subsequent registration to converge to the global optimal solution.

[0045] 2. Registration of point cloud data for positioning circular holes In the center localization stage, this implementation method introduces the RANSAC circle fitting algorithm to correct the deviation of the registered center data based on the traditional GICP algorithm, thereby improving the robustness and accuracy of center localization. The specific process is as follows.

[0046] 2.1. Using the template point cloud of the standard part's circular hole and the initial pose for coarse registration together as input, the GICP algorithm is used for point cloud registration to obtain the registration center. This implementation considers that the point cloud does not have obvious extreme distribution characteristics such as discreteness, sparsity, and non-uniformity, and calculates the registration score through the root mean square error (RMSE). , is used to characterize the registration quality, and its calculation expression is shown in equation (1):

[0047] in, For the points in the template point cloud, For the corresponding points in the actual point cloud, This represents the number of points involved in the registration. The smaller the RMSE value, the higher the registration score, indicating that the GICP registration result is more reliable.

[0048] 2.2 Edge point set obtained from preprocessing Selected from Subset Each subset The sample contains three points, and the spatial distance between any two of these three points must be no less than half the distance between the two largest points in the edge point set. The edge contour of the circular hole is a closed circular ring structure in space. By limiting the dispersion of the three points in the subset on the ring, the problem of unstable circle fitting caused by overly concentrated sampling points can be avoided, thus improving the fitting accuracy.

[0049] 2.3. Use the SVD decomposition method to decompose each subset. By fitting the spatial plane in which it lies, the parameters of the plane equation can be obtained. The equation of the plane is Subset Points within the projection plane are projected onto this plane, and the equation of the spatial circle is fitted using the least squares method within the projection plane to calculate the coordinates of the center of the corresponding subset. and radius To evaluate the fitting accuracy, this implementation calculates the difference between the distance from the subset points to the fitting circle and the radius thereof. Its expression is shown in equation (2):

[0050] in, For subset The first in One point, It is a plane normal vector. If Below a given error threshold If the value is true, then the point is considered an interior point of the fitted circle; otherwise, it is discarded.

[0051] traversing all Subset After repeating the plane fitting, circle fitting, and interior point selection steps described above, this embodiment obtains the set of circle centers for each subset. and the set of all interior points Based on this, the center set of the circles is calculated using a probability-weighted average method. , , Expectations in direction And combine them to form a subset to fit the center coordinates of the circle. Then, using the interior point set... Construct a new point set Repeat the plane fitting and circle fitting process to obtain the center of the fitted circle. The center of this circle reflects, to some extent, the geometric center of the high-confidence edge point that has been verified multiple times.

[0052] 2.4. In this embodiment, the registration center will be... Subset fitting center Fitting the center of the circle to the interior point By performing weighted fusion, the final center coordinates of the positioning hole in the robot coordinate system are obtained. Its expression is shown in equation (3):

[0053] in, Reflecting the reliability of GICP registration, when the registration quality is high, the registration result has a greater weight in the final circle center calculation; when the registration quality decreases, the circle center obtained by RANSAC fitting has a higher weight in the final result, thus achieving a more balanced circle center positioning effect under different working conditions.

[0054] 3. Material tray perspective correction During the tray perspective correction stage, this embodiment uses a tooling tray and a cylinder clamping device to fix the tray in the Euler angle of the robot coordinate system. The basic consistency is maintained to ensure the overall horizontality of the material tray and the four positioning holes. The robot sequentially moves the 3D industrial camera above the four positioning holes, repeating the point cloud acquisition, preprocessing, and center positioning process described above. Finally, the actual center coordinates of the four positioning holes are obtained in the robot coordinate system. Meanwhile, based on the tooling design, the theoretical center coordinates of the four positioning holes under ideal conditions are set. The theoretical coordinates are then projected onto the XOY plane of the robot coordinate system.

[0055] Due to factors such as machining errors, clamping errors, and placement deviations, a quadrilateral whose actual center is on the XOY plane usually cannot be aligned with a theoretical quadrilateral through simple translation, rotation, and scaling transformations. Therefore, this embodiment constructs a two-dimensional perspective transformation matrix. This describes the mapping relationship between the theoretical center and the actual center of a circle. Matrix The general form is shown in equation (4):

[0056] In the XOY plane of the robot coordinate system, with the coordinates of the theoretical center of the circle... Coordinates of the actual center of the circle The matrix is ​​solved by establishing four sets of point correspondences and using the following perspective transformation relationship. The parameters are shown in equation (5):

[0057] By establishing and solving a system of linear equations for the four pairs of points, a matrix satisfying the above constraints can be obtained. After obtaining the perspective transformation matrix, this embodiment calculates the theoretical coordinates of each gripping point in the material tray. As input, the corrected capture points are calculated using equation (6). :

[0058] in, In order to capture the height coordinates of the point in the theoretical coordinate system, this embodiment maintains that the Z-axis height in the welding loading station is determined by the tooling and workpiece structure and does not participate in the planar perspective transformation.

[0059] As shown in Tables 1-3, to verify the performance of this embodiment in practical applications, multiple sets of material trays were selected as samples for testing at different time periods during three shifts of production: 8:00–11:00, 13:00–16:00, and 18:00–21:00. The center deviations of the four positioning holes and the deviations of the gripping points inside the material trays were recorded. Without the correction strategy, the maximum deviation of the positioning hole center was distributed in the range of 1.5mm to 2.8mm; after applying this embodiment, the deviation was reduced to the range of 1.2mm to 1.6mm. In the tests of five typical gripping points (upper left, lower left, upper right, lower right, and middle), the maximum deviation reached 3.1mm without correction, while the maximum deviation was reduced to 1.1mm after using this embodiment. These results indicate that this embodiment has high engineering adaptability in terms of overall material tray error compensation, stable extraction of positioning hole features, and batch correction of gripping points, and can effectively reduce abnormal downtime and fixture damage caused by inaccurate gripping.

[0060] Table 1 - Maximum center deviation for each group before correction (mm)

[0061] Table 2 - Maximum center deviation for each group after correction (mm)

[0062] Table 3 - Comparison Data (mm)

[0063] The technical process described in this embodiment can be deployed to the robot controller via computer program, enabling point cloud acquisition, preprocessing, center positioning, perspective transformation calculation, and grasping instruction generation to form an automated closed-loop process. It is suitable for continuous, multi-batch production scenarios and combines software deployment flexibility with hardware compatibility.

[0064] The above detailed description of the technical solution provided by the present invention is intended to highlight the advantages and benefits of the technical solution provided by the present invention. However, the above detailed embodiments are not intended to limit the scope of protection of the present invention. Any reasonable modifications and improvements to the present invention, recombination of embodiments, and equivalent substitutions based on the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0065] Those skilled in the art will understand that the above description is merely a preferred embodiment of the present invention, and the features described in the various embodiments and / or claims disclosed in the present invention can be combined or combined in various ways, even if such combinations or combinations are not explicitly described in the disclosure of the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle scope of the present invention should be considered to fall within the protection scope of the present invention.

[0066] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and modifications.

Claims

1. A robot grasping and correction method based on a 3D camera, characterized in that, Includes the following steps: S1. Obtain the point cloud data of the positioning hole; S2. The point cloud data is preprocessed, and coarse registration is performed based on the preprocessing results to obtain the initial coarse registration pose of the positioning circular hole. S3. Using the preprocessed point cloud data and the initial pose of coarse registration as input, the center of the circle is located based on the GICP algorithm and the RANSAC circle fitting algorithm to obtain the final center coordinates of the positioning hole. S4. Based on the final center coordinates and the theoretical center coordinates of the material tray in the robot coordinate system, construct a perspective transformation matrix; S5. Input the theoretical coordinates of each gripping point in the material tray into the perspective transformation matrix to obtain the gripping points after correction.

2. The method according to claim 1, characterized in that, S1 installs a 3D industrial camera at the end of the robot. The 3D camera is located 300 to 600 mm directly above the positioning hole. The 3D camera is triggered by TCP to take pictures and obtain the point cloud data of the positioning hole.

3. The method according to claim 1, characterized in that, The method for obtaining the initial coarse registration pose of the positioning hole in S2 is as follows: S21. Perform normal vector estimation on the point cloud data to obtain the surface normal vector; S22. Noise points are removed based on statistical filtering; S23. Use Euclidean clustering to separate the point cloud data of non-positional circular hole regions; S24. Perform edge extraction on the remaining point cloud data to obtain the edge points of the positioning hole; S25. Match the edge points with the point cloud template of the standard part's circular hole edge contour, and obtain the initial coarse registration pose of the positioning circular hole through the 4PCS algorithm.

4. The method according to claim 3, characterized in that, When performing edge extraction in S24, the edge points of the positioning circular hole are identified by comparing the changes in the angle between the point cloud normal vectors.

5. The method according to claim 3, characterized in that, The method for obtaining the final center coordinates of the positioning hole in S3 is as follows: S31. Using the preprocessed point cloud data and the initial pose of coarse registration as input, perform point cloud registration through the GICP algorithm to obtain the coordinates of the first circle center. S32. Using the RANSAC circle fitting algorithm, select multiple point sets from the edge points, and perform plane fitting, point projection and circle fitting on each point set to obtain the coordinates of the second circle center. S33. Filter in-points based on error threshold and repeatedly fit to obtain the coordinates of the third circle center; S34. The first center coordinate, the second center coordinate, and the third center coordinate are combined in a weighted manner to obtain the final center coordinate.

6. The method according to claim 5, characterized in that, In S31, point cloud registration is performed using the GICP algorithm, which includes calculating the registration score based on the root mean square error to determine the registration convergence.

7. The method according to claim 5, characterized in that, When using the RANSAC algorithm for circle fitting in S32, each subset of the multiple point sets contains 3 points, and the spatial distance between any two of the three points is half the distance between the two furthest points.

8. The method according to claim 1, characterized in that, In S4, a point correspondence is established between the final center coordinates of the positioning hole and the theoretical center coordinates of the material tray in the robot coordinate system, and the perspective transformation matrix is ​​constructed based on the point correspondence.

9. The method according to claim 1, characterized in that, In S5, the theoretical coordinates of all gripping points in the material tray are input into the perspective transformation matrix in batch mode, and a two-dimensional perspective transformation is performed to obtain all corrected gripping points. The corrected gripping points are then sent to the robot control system for gripping operations.

10. A computer storage medium for storing computer programs, characterized in that, When the computer program is read by the computer, the computer executes the method according to any one of claims 1-9.

Images