Visual Recognition and Positioning Method for Bolting Holes of Rolling Mill

Abstract

Visual recognition and positioning method for bolt holes of rolling mills, comprising the following steps: Step 1, in the environment of the rolling mill production preparation workshop, use an eye-in-hand vision system to obtain the original three-dimensional point cloud of the bolt hole part of the rolling mill, and preprocess the obtained original point cloud of the bolt hole; Step 2: Use the random sample consensus method based on the plane model to fit and segment the optimal plane for the preprocessed point cloud to obtain the plane point cloud of the target bolt hole of the rolling mill; Step 3: Identify the target bolt hole; Step 4: Locate the center pose of the hole. According to the environment of the rolling mill production preparation workshop and the position characteristics of the bolt holes of the rolling mill, this method fits and segments the plane point cloud of the hole, uses a hole recognition method based on edge fitting to identify the target bolt hole, calculates the center pose of the hole, completes the recognition and positioning of the bolt holes of the rolling mill, and has a high recognition accuracy and a fast positioning speed.

Description

Technical Field

[0001] The present invention relates to a method for three-dimensional point cloud recognition and positioning, and particularly to a method for three-dimensional point cloud recognition and positioning of a rolling mill bolt hole, belonging to the technical field of image processing. Background Art

[0002] With the continuous development of the economic society, the demand for bar products in China is increasing day by day, and the quality requirements for bars are also getting higher and higher, which puts forward new requirements for the relevant equipment for bar production, especially for rolling mills. In bar production, the pass of the roll will be worn to varying degrees, and the roughness cannot meet the rolling requirements, resulting in the out-of-tolerance of the size of the rolled bar and affecting the quality of the bar. Therefore, quickly and timely changing the roll is of great significance for improving production efficiency and product quality. The time and workload of disassembling and assembling the bolts of the rolling mill can account for more than half of the roll-changing work. During the process of assembling a new roll, hexagon bolts for fixing the frame and the roll need to be finally installed.

[0003] At present, the assembly of rolling mill bolts worldwide mainly relies on semi-automatic mechanical equipment to assist manual operations, with long working hours, low efficiency, high labor intensity of workers, and low safety factor. It can be seen that the assembly of rolling mill bolts needs to be automated and intelligentized to replace manual operations. The automated assembly of rolling mill bolts requires the introduction of robots, and the intelligent recognition and positioning of bolt holes by robots requires the introduction of machine vision technology. Therefore, according to the industrial on-site production environment, a method for recognizing and positioning the image of a rolling mill bolt hole is proposed to lay a foundation for the industrial application of the automatic assembly of rolling mill bolts. Summary of the Invention

[0004] In order to solve the above technical problems, the present invention proposes a visual recognition and positioning method for rolling mill bolt holes, providing a technical basis for the automatic assembly of rolling mill bolts.

[0005] The visual recognition and positioning method for rolling mill bolt holes includes the following steps:

[0006] Step 1: In the environment of the rolling mill production preparation workshop, a hand-eye vision system is used to obtain the original three-dimensional point cloud of the rolling mill bolt hole part. The original three-dimensional point cloud includes the casing point cloud around the rolling mill bolt hole and some redundant points and adhesion points. The obtained original point cloud of the bolt hole is preprocessed. The specific steps are as follows: (1) Determine the effective coordinate range according to the relative position relationship between the rolling mill bolt hole and the vision system, and use a straight-through filter to filter out redundant points outside the range; (2) Use a voxel filter to downsample the point cloud; (3) Judge whether a point is an obvious outlier noise point according to the local point cloud density, and use a statistical filter to remove the adhesion points and outlier noise points on the hole plane;

[0007] Step 2: Apply the random sample consensus method based on the plane model to fit and segment the optimal plane for the preprocessed point cloud, and obtain the plane point cloud of the target bolt hole of the rolling mill.

[0008] Step 3: Identify the target bolt hole. The specific steps are as follows: (1) Use the least squares method to fit the local plane point by point, calculate the normal vector of each point, and use the included angle between the projection points of each point on the inner cutting plane of the neighborhood as the judgment basis for edge points to extract the edge point cloud of the hole plane; (2) According to the specification information of the rolling mill bolt hole, use the RANSAC space circle fitting algorithm based on the radius threshold to fit and identify the target bolt hole.

[0009] Step 4: Locate the center pose of the hole. The specific steps are as follows: According to the relative pose between the robot and the vision system, establish the mutual relationship between the camera coordinate system and the robot tool coordinate system. Use the least squares method to fit the center pose of the target bolt hole in the camera coordinate system, and multiply it by the transformation matrix between the camera coordinate system and the robot tool coordinate system to obtain the center pose of the hole in the robot base coordinate system.

[0010] In the RANSAC space circle fitting algorithm based on the radius threshold in Step 3, on the basis of the traditional RANSAC space circle fitting algorithm, the radius threshold judgment is added. When the result is the target circle, the result is directly output; when the result is not the target circle, the current point set is cleared, and the last outlier point set is used as the input point cloud to perform the RANSAC space circle fitting again until the target circle is fitted.

[0011] Beneficial effects of the method of the present invention:

[0012] According to the environment of the rolling mill production preparation workshop and the position characteristics of the rolling mill bolt holes, the three-dimensional point cloud of the bolt hole part is collected through the robot hand-eye vision system, the hole plane point cloud is fitted and segmented, a hole recognition method based on edge fitting is used to identify the target bolt hole, the center pose of the hole is calculated, and the recognition and positioning of the rolling mill bolt hole are completed, with high recognition accuracy and fast positioning speed. Description of the drawings

[0013] Figure 1 is a schematic structural diagram of the robot hand-eye vision system adopted by the method of the present invention;

[0014] Figure 2 is a flow chart of the visual recognition and positioning method of the rolling mill bolt hole of the present invention;

[0015] Figure 3 is a flow chart of the RANSAC space circle fitting algorithm based on the radius threshold of the method of the present invention. Detailed implementation manners

[0016] Combined with the attached Figures 1-3 Explain the system structure and operation steps of the present invention.

[0017] Visual recognition and positioning method for bolt holes of rolling mills, comprising the following steps:

[0018] Step 1: In the environment of a rolling mill production preparation workshop, use an eye-in-hand vision system to obtain the original three-dimensional point cloud of the bolt hole part of the rolling mill. The original three-dimensional point cloud includes the housing point cloud around the bolt hole of the rolling mill and some redundant points and adhesion points. Preprocess the obtained original point cloud of the bolt hole. The specific steps are as follows: (1) Determine the effective coordinate range according to the relative position relationship between the bolt hole of the rolling mill and the vision system, and use straight-through filtering to filter out redundant points outside the range; (2) Use voxel filtering to downsample the point cloud; (3) Judge whether a point is an obvious outlier noise point according to the local point cloud density, and use statistical filtering to remove the adhesion points and outlier noise points on the hole plane;

[0019] Step 2: Use the random sample consensus method based on the plane model to fit and segment the optimal plane for the preprocessed point cloud to obtain the plane point cloud of the target bolt hole of the rolling mill;

[0020] Step 3: Identify the target bolt hole. The specific steps are as follows: (1) Use the least squares method to fit the local plane point by point, calculate the normal vector of each point, and use the included angle between the projection points on the inscribed plane of the neighborhood of each point as the judgment basis for edge points to extract the edge point cloud of the hole plane; (2) According to the specification information of the bolt hole of the rolling mill, use the RANSAC space circle fitting algorithm based on the radius threshold to fit and identify the target bolt hole;

[0021] Step 4: Locate the center pose of the hole. The specific steps are as follows: According to the relative pose between the robot and the vision system, establish the mutual relationship between the camera coordinate system and the robot tool coordinate system, use the least squares method to fit the center pose of the target bolt hole in the camera coordinate system, and multiply it by the transformation matrix between the camera coordinate system and the robot tool coordinate system to obtain the center pose of the hole in the robot base coordinate system.

[0022] In the RANSAC space circle fitting algorithm based on the radius threshold in Step 3, on the basis of the traditional RANSAC space circle fitting algorithm, add the radius threshold judgment. When the result is the target circle, directly output the result; when the result is not the target circle, clear the current point set, take the last outlier point set as the input point cloud and perform RANSAC space circle fitting again until the target circle is fitted.

[0023] The method of the present invention uses a robot hand-eye vision system, which includes a robot floor rail 1, an industrial robot 2, a line laser stereo camera 3, a wrench 4, and a rolling mill bolt library 5. The robot floor rail 1 is horizontally arranged on one side of the rolling mill to be disassembled and assembled. The industrial robot 2 is placed on the robot floor rail 1. A line laser stereo camera 3 and a wrench 4 are arranged at the end of the industrial robot 2. The rolling mill bolt library 5 is arranged within the working space of the industrial robot 2 for storing rolling mill bolts. The robot floor rail adds an additional degree of freedom to the six-degree-of-freedom industrial robot, expanding the working range of the robot. A line laser stereo camera is installed at the end of the robot, adopting the Eye-in-Hand working mode. After visual segmentation and positioning are completed, the wrench at the end of the robot is guided to achieve bolt disassembly.

[0024] For the bolts that need to be disassembled during the roll change of the RRCA445-HS type finishing mill, the present invention φ takes the 22 bolt holes as samples. To verify the feasibility of the visual recognition and positioning method, a steel mill production environment is simulated in the laboratory. Multiple three-dimensional point cloud images of the bolt hole parts of the rolling mill are obtained through the robot hand-eye vision system. In order to locate the pose of the target bolt hole, processes such as preprocessing of the original point cloud, hole plane segmentation, target hole recognition, and hole center positioning are carried out on the original point cloud, and finally the visual positioning of the rolling mill bolts is completed.

[0025] 1. Preprocessing of the original point cloud

[0026] Due to the complex industrial site environment, the original point cloud collected by the line laser stereo camera contains some irrelevant noise points and adhesion points between planes. The present invention preprocesses the original point cloud successively by means of direct filtering, voxel filtering, and statistical filtering to obtain a point cloud with clear boundaries, easy to segment, and no adhesion between planes. The specific steps are as follows:

[0027] (1) Direct filtering

[0028] Since there are a large number of fixed point clouds in the point cloud background that are irrelevant to the main structure and will affect the target extraction, direct filtering is usually the simplest method to remove the fixed point cloud. Set the upper and lower limit thresholds for the points ([[]] x m , y m , z m ) on the x axis ([[]] X 1, X 2), y axis ([[]] Y 1, Y 2), z axis ([[]] Z 1, Z 2). Retain the points within the range. Then the point coordinates after direct filtering are:

[0029]

[0030] The present invention needs to filter out irrelevant point clouds in the background and adopts z axial filtering, and the upper and lower limit thresholds for filtering are taken as (200, 800).

[0031] (2) Voxel filtering

[0032] The amount of point cloud data directly collected by the stereo camera is relatively large, seriously affecting the processing efficiency, and there will be redundant points when describing the shape features. Voxel filtering is used to streamline the original point cloud, and the downsampling operation of the point cloud is completed without destroying the geometric structure and shape features of the target.

[0033] The point cloud range is divided into cubes of the same size, the centroid position of each cube is calculated, and the point cloud is streamlined by replacing all points in the cube with the centroid point. The side length of the cube in voxel filtering d is extremely important. d If the value is too large, some important points including the edge features of the target will be filtered out, resulting in excessive streamlining of the point cloud. d If the value is too small, the number of point clouds in the cube is too small to play the role of streamlining the point cloud. Therefore, the present invention determines through experiments d = 0.75 mm.

[0034] (3) Statistical filtering

[0035] The line laser measurement method often introduces outlier measurement noise when extracting the light strip. Its characteristic is that it is sparsely distributed in the three-dimensional space and scattered outside the main structure of the point cloud. This kind of noise can be removed through statistical filtering. Statistical filtering is to statistically analyze each point in the point cloud and remove the points that do not meet the criteria based on a certain standard. Statistically analyze each point in the point cloud. Theoretically, the distances of all points form a Gaussian distribution, with an expected value of and a standard deviation of. Let a point P i ([[]] x i , y i , z i ) to any point P m ([[]] x m , y m , z m ) is:

[0036]

[0037] And Pi The point cloud composed of the k nearest k points is the k neighborhood, and the average distance from this point to all points within the D avg-n neighborhood is:

[0038]

[0039] where, , .

[0040] Set the coefficient of standard deviation s , and calculate the standard distance threshold d std-n as the basis for judging whether it is an outlier:

[0041]

[0042] Compare the selected data point P n average distance D avg-n with the standard distance threshold d std-n . When D avg-n is outside the d std-n range, it is defined as an outlier and deleted. When D avg-n is within the d std-n range, the point is retained. Remove all outliers and retain the remaining points as the point cloud after statistical filtering.

[0043] 2. Segmentation of the point cloud of the bolt hole plane

[0044] The orifice plane contains complete hole features. Therefore, for the segmentation of the hole point cloud, finally, the hole plane needs to be extracted. The plane can be completely segmented by using the method of fitting the plane with the random sample consensus algorithm to extract the inliers.

[0045] Random Sample Consensus (RANSAC) estimates the parameters of a mathematical model iteratively from a set of observed data containing outliers. The data that conforms to the assumed model parameters is called inliers, and the remaining points are called outliers. This algorithm requires that, with a certain confidence probability, the minimum number of samples N of the basic subset and the probability P of obtaining at least one good subset P > t satisfy the following relationship:

[0046]

[0047] in, t is the probability of an interior point in the data set, that is , N The minimum amount of data required to calculate the model parameters, m is the number of iterations.

[0048] According to the plane characteristics of the casing around the bolt hole, the plane model is selected as the mathematical model for estimation:

[0049] Ax + By + Cz = D

[0050] Calculate plane parameters by picking three points A , B , C , D , fit the optimal plane, and use the inner points after fitting as the reserved points to segment the bolt hole plane of the rolling mill. Error threshold selected in the RANSAC fitting process T and iteration threshold m 0 directly affects the optimal plane estimation effect. Error threshold T It mainly limits the range of the up and down jumps of the midpoint of the estimated plane. The value will affect the flatness of the fitted plane. The maximum number of iterations varies depending on the point cloud, so when setting the iteration threshold m If 0 is too small, the iteration will often stop before the optimal effect is achieved, thus outputting an erroneous result. T 2mm, m 0 is 50.

[0051] 3. Identify the target bolt hole

[0052] The hole plane point cloud after the point cloud segmentation operation contains the complete information of the hole. The plane outside the hole is composed of points arranged in space, and there is no point cloud arrangement inside the hole. Therefore, locating the center point of the hole that does not actually exist in the point cloud becomes the key to the problem.

[0053] The present invention adopts a hole recognition method based on edge fitting, extracts all edge points of the hole plane point cloud, performs target circle fitting of the edge point cloud, and thus identifies the target bolt hole. The specific steps are as follows:

[0054] (1) Point cloud edge extraction based on normal estimation method

[0055] The outermost points in a point cloud are usually called edge points. The characteristic of an edge point is that most of the points around it are on the same side of the point. Any point in a point cloud has a normal vector, and whether the point is an edge point can be determined based on the angle between the normal vectors of adjacent points and the tangent plane.

[0056] For the point cloud P ={ p i | p i R 3 , i = 1, 2, …, m}}, set the search radius r The inner neighborhood points are . Use principal component analysis to calculate the normal vector of the point cloud P . Given that P passes through its k The centroid of the neighborhood points is p 0, and the normal vector n satisfies , calculate the covariance matrix M :

[0057]

[0058] Perform eigenvalue decomposition on M . The eigenvector corresponding to the smallest eigenvalue is the normal vector. The normal vector of this plane is the normal vector of the point p n .

[0059] According to p i and its normal vector n to construct the tangent plane of this point . Project the points in N ( p i ) onto the tangent plane , denoted as . Select a point in . Take as the u axis, n as the w axis, as the v axis. Take p i as the coordinate center to establish a local coordinate system, denoted as .

[0060] Calculate the clockwise angle between the vector from the other points p i in to u and the coordinate axis . Take the pairwise differences of the neighborhood angles to obtain the angle set ​, where:

[0061]

[0062] At this time, is the included angle between two adjacent vectors. Arrange the elements in in descending order, and find the maximum included angle . When is greater than the threshold , this point is considered an edge point.

[0063] The angle threshold during edge extraction is an important indicator affecting the judgment of edge points. Through a large number of experiments, the present invention determines that is when the extraction effect is the best.

[0064] (2) Fitting the target hole using the RANSAC spatial circle fitting algorithm based on the radius threshold

[0065] All the edge points of the hole plane point cloud include not only the hole edge but also the plate edge. It is necessary to find and extract the edge points of the target circular hole among all the edges. The pose of the hole edge in the point cloud space is uncertain, and a spatial circle needs to be fitted. When a spherical surface intersects a plane, the intersection line is a circle. Therefore, the rectangular coordinate equation of the spatial circle can be expressed as:

[0066]

[0067] When the traditional RANSAC algorithm fits a spatial circle, the output is the optimal spatial circle. Usually, a curve with a larger number of points and a smaller curvature is more likely to be output as the result compared to another curve with a smaller number of points and a curvature closer to a circle. However, usually, the target circular hole in the hole plane edge is not the optimal spatial circle, resulting in the fact that the fitted circle is not the target circle, and the phenomenon of fitting error occurs.

[0068] To solve the problem that the spatial circle fitted by the traditional RANSAC algorithm is not necessarily the target circle, combined with the condition of the known hole diameter information during the installation of the rolling mill bolts, the present invention adds a radius constraint of the fitted circle during the fitting process, and limits the output circle radius within a certain range:

[0069]

[0070] In the formula, r 0 represents the known target hole radius; represents the tolerance error of the radius.

[0071] Input the edge point cloud of the hole plane. According to the principle of the RANSAC algorithm, first randomly select 3 points, and calculate the spatial circle parameters determined by the three points, including the center coordinates ( x , y , z), normal vector ( k x , k y , k z ) and circle radius r ; then calculate the distance from each point in the point cloud to the center of the circle d , and judge whether the difference between the distance and the radius is within a certain threshold range. The points within the range are determined as inliers, otherwise they are put into the outlier set; then count the number of inliers M whether it is greater than the minimum inlier threshold. If so, use the least squares method to calculate the circle parameters fitted by the inliers. If not, go to the next step; iterate the above process until the number of iterations k exceeds the maximum iteration threshold k max , and obtain the inlier point set of the fitted circle and the circle parameters; at this time, judge whether the circle radius r is within the set radius range . If it is within the range, it means that the fitted circle is the target hole. If not, it means that the optimal fit is not the required data; at this time, clear the current point set, take the outlier point set of the last time as the input point cloud and perform RANSAC spatial circle fitting again until the edge data of the required target hole is obtained, output the inlier point set and the fitted circle parameters, and complete the recognition and positioning of the target hole.

[0072] The present invention conducts multiple experiments on a bolt hole with a diameter of 22 mm, sets the target hole radius r 0 to be 11 mm, and the tolerance error of the radius is 1 mm.

[0073] 4. Center pose of the positioning hole

[0074] The present invention adopts the visual and robot position relationship of Eye-in-Hand. The positions of the camera coordinate system and the robot end tool coordinate system are fixed. Therefore, it is necessary to determine the pose transformation relationship between the robot end and the camera.

[0075] Establish the following coordinate systems for the system: robot base coordinate system C base , robot tool coordinate system C tcp , camera coordinate system C camera and calibration plate coordinate system C board . Among them, base H tcp represents the conversion relationship from the robot tool coordinate system to the robot base coordinate system; camera H boardRepresents the transformation relationship from the calibration board coordinate system to the camera coordinate system; base H board Represents the transformation relationship from the calibration board coordinate system to the robot base coordinate system; tcp H camera Represents the transformation relationship from the camera coordinate system to the robot tool coordinate system.

[0076] From the transformation relationship, it can be known that:

[0077] base H tcp tcp H camera camera H board = base H board

[0078] Since tcp H base base H board = tcp H board , finally, it can be deduced that:

[0079]

[0080] Therefore, solving the hand-eye calibration problem of Eye-in-Hand can be simplified to solving the transformation relationship between board H tcp and camera H board The transformation relationship between the two coordinate systems. The calibration board coordinate system is established by the three-point teaching method of the robot. The transformation relationship from the robot tool coordinate system to the calibration board coordinate system corresponding to the moment when the camera starts to collect data is read by the robot controller as board H tcp .

[0081] Use the camera to collect an image of the calibration board, and record the three-dimensional coordinates of the points a , b , c in the camera coordinate system in the grayscale image ([[]] x a1 , y a1 , z a1 ), ([[]] xb1 , y b1 , z b1 ),( x c1 , y c1 , z c1 ), it can be known by measuring the calibration plate that ab The distance is x 2. ac The distance is y 2, then in the calibration plate coordinate system a , b , c The coordinates of the three points are (0, 0, 0), ([[]] x 2, 0, 0), (0, [[[]] y 2, 0). From the three pairs of point coordinates in the two coordinate systems, the rigid body transformation matrix of the two coordinate systems can be solved. The centroids of the three points on the two coordinate systems are obtained respectively O camera , O board , and the matrix H is established:

[0082]

[0083] Among them, n represents the matrix dimension, p camera and p board represent the coordinates of the points in the camera coordinate system and the calibration plate coordinate system respectively. Perform SVD decomposition on this matrix SVD( H ) = [[[[]] U , S , V , and the rotation matrix R and the translation vector T can be obtained:

[0084]

[0085] From this, camera H board can be calculated, and finally the conversion relationship from the camera coordinate system to the robot tool coordinate system tcp H camera is obtained.

[0086] The positioning hole needs to obtain the position and attitude of the hole center, and establish a hole center coordinate system. The origin of the coordinate system O is the hole center, Z the X axis is perpendicular to the hole plane and faces inside the hole, and theY The plane formed by the axis XOY is parallel to the hole plane. The center coordinates of the spatial circle can be obtained by fitting the target spatial circle O ( x , y , z ) and the normal vector K ( kx , ky , kz ). In the camera coordinate system, the spatial center coordinates are the hole center position. Therefore, it is necessary to solve the hole center attitude according to the center normal vector.

[0087] Establish the coordinate system of the hole in the camera coordinate system X 0 Y 0 Z 0. The origin XYZ is the hole center. The unit vectors O ( a ( a 1, a 2, a 3) and b ( b 1, b 2, b 3) are in the same direction as the Z axis and the Z 0 axis respectively. At this time, the calculated center normal vector K is coaxial with the Z axis and in the opposite direction. In theory, the Z axis should be perpendicular to the hole plane. At this time, the O point coordinates are the hole center position.

[0088] In the spatial positioning of the hole, it is necessary to ensure the perpendicularity of the hole, that is, the Z axis is perpendicular to the hole plane. For the X axis and the Y axis, there are no strict requirements for the direction. Therefore, the problem of solving the hole attitude can be simplified to obtaining the spatial position relationship between the unit vector a and the unit vector b . Since the vector K is coaxial and in the opposite direction to the unit vector a , this problem is to solve the spatial position relationship between the vector - K and the vector b . By solving the corresponding rotation and translation matrix from two known vectors, the pose of the hole in the camera coordinate system can be obtained camera H object .

[0089] Therefore, the pose of the hole in the robot coordinate system base H object is:

[0090] base H object = base H tcp tcp H camera camera H object

[0091] Among them, base H tcp represents the pose of the end - effector tool coordinate system relative to the base coordinate system at the moment when the camera starts to capture images, which is directly read on the robot teaching pendant.

Claims

1. A method for visually identifying and locating bolt holes of a rolling mill, comprising the following steps: Step 1: In the steel rolling production preparation workshop environment, the robot hand-eye vision system is used to obtain the original three-dimensional point cloud of the bolt hole of the rolling mill. The original three-dimensional point cloud includes the casing point cloud around the bolt hole of the rolling mill and redundant points and adhesion points. The obtained original point cloud of the bolt hole is preprocessed. The specific steps are as follows: (1) Determine the effective coordinate range based on the relative position relationship between the bolt hole of the rolling mill and the visual system, and use straight-through filtering to filter out redundant points outside the range; (2) Use voxel filtering to downsample the point cloud; (3) Determine whether the point is an obvious outlier noise point based on the local point cloud density, and use statistical filtering to remove hole plane adhesion points and outlier noise points; Step 2: The preprocessed point cloud is fitted and segmented using a random sampling consistency method based on a plane model to obtain a plane point cloud of the target bolt hole of the rolling mill; Step 3: Identify the target bolt hole. The specific steps are as follows: (1) The least squares method is used to fit the local plane point by point, and the normal vector of each point is calculated. The angle between the normal vector of the adjacent point and the tangent plane is used as the basis for edge point judgment. If the angle exceeds the threshold, it is identified as an edge point, and the hole plane edge point cloud is extracted; (2)According to the specification information of the rolling mill bolt holes, the RANSAC spatial circle fitting algorithm based on the radius threshold is used to fit and identify the target bolt holes. A radius constraint of the fitting circle is added during the traditional RANSAC spatial circle fitting process, and the output circle radius is limited within a certain range: , where r 0 represents the known target hole radius; represents the tolerance error of the radius. Input the edge point cloud of the hole plane. According to the principle of the RANSAC algorithm, first randomly select 3 points, and calculate the spatial circle parameters determined by the three points, including the center coordinates ( x , y , z ), the normal vector ( k x , k y , k z ) and the circle radius r ; then calculate the distance d from each point in the point cloud to the center of the circle, and judge whether the difference between the distance and the radius is within a certain threshold range. The points within the range are determined as inliers, otherwise they are put into the outlier set; then count the number of inliers M and check if it is greater than the minimum inlier threshold. If it is, the circle parameters fitted by the inliers are calculated using the least squares method. If not, go to the next step; iterate the above process until the iteration number k exceeds the maximum iteration threshold k max , and obtain the inlier point set and circle parameters of the fitted circle; at this time, judge whether the circle radius r is within the set radius range . If it is within the range, it means that the fitted circle is the target hole. If not, it means that the optimal fit is not the required data. At this time, clear the current point set, take the last outlier point set as the input point cloud and perform RANSAC spatial circle fitting again until the edge data of the required target hole is obtained, and output the inlier point set and the fitted circle parameters to complete the identification of the target hole; Step 4: Locate the center pose of the hole. The specific steps are as follows: According to the relative pose between the robot and the visual system, establish the relationship between the camera coordinate system and the robot tool coordinate system, fit the center pose of the target bolt hole in the camera coordinate system, and obtain the center pose of the hole in the robot base coordinate system by multiplying it with the transformation matrix between the camera coordinate system and the robot tool coordinate system.

2. The visual recognition and positioning method of the rolling mill bolt hole according to claim 1, characterized in that: The robot hand-eye vision system used in step 1 of the method comprises a robot ground track (1), an industrial robot (2), a line laser stereo camera (3), a wrench (4) and a rolling mill bolt library (5). The robot ground track (1) is horizontally arranged on one side of the rolling mill to be disassembled and assembled, the industrial robot (2) is placed on the robot ground track (1), the line laser stereo camera (3) and the wrench (4) are arranged at the end of the industrial robot (2), and the rolling mill bolt library (5) is arranged in the working space of the industrial robot (2) for storing rolling mill bolts.

3. The visual recognition and positioning method of the rolling mill bolt holes according to claim 1, characterized in that: Pose of the hole in the robot base coordinate system base H object It is as follows: base H object = base H tcp tcp H camera camera H object Among them, base H tcp represents the pose of the end - effector tool coordinate system relative to the base coordinate system at the moment when the camera starts to capture images, which is directly read on the robot teaching pendant. tcp H camera is the conversion relationship from the camera coordinate system to the robot tool coordinate system. camera H object is the position of the hole in the camera coordinate system.

Images