Anchor hole positioning method and system

Abstract

The invention discloses an anchor hole positioning method and system, and the method comprises the steps: carrying out the similarity matching of a preprocessed point cloud through employing a pre-scanned standard round end rectangular anchor point cloud; performing normal analysis on the preprocessed point cloud information to obtain a point cloud plane normal evaluation degree and plane parameters; weighting the point cloud matching degree and the point cloud normal evaluation degree to obtain a total cost, and processing the circular-end rectangular anchor point cloud according to the total cost to obtain a circular-end rectangular anchor point normal vector; performing coordinate transformation on the obtained round-end rectangular anchor point normal vector, and calculating to obtain a round-end rectangular center coordinate; and determining the center coordinate of the round-end rectangular anchor hole point according to whether the Gaussian process regression of the center point meets the standard or not. The problems that in the anchor protection operation process, the anchor point is manually positioned, the working efficiency is low, and danger exists are solved. The anchor point positioning accuracy is evaluated, the positioning precision is guaranteed, path planning is conducted through anchor hole position information, and the line control chassis is controlled to reach the anchor point position.

Description

Technical Field

[0001] The invention relates to image recognition processing and point cloud processing, and in particular to an anchor hole positioning method and system. Background Art

[0002] During the tunnel excavation process, it is necessary to carry out anchoring operations on the tunnel surface. Current mechanical equipment uses manual anchoring during the anchoring operation, and construction is carried out by manually positioning anchor points. The work efficiency is low, the working environment is harsh, and there is a certain degree of danger.

[0003] Patent 202211519958.2 provides a method, system, electronic device and medium for identifying drilling and anchor holes. Using a visual solution, only the center point position information is obtained, and no normal vector of the anchor hole is provided. Patent 202310239053.8 provides a method, system and electronic device for locating steel belt anchor holes of a drilling and anchoring robot. It is suitable for positioning a single target. When multiple anchor holes need to be processed at the same time, multiple operations are required, which is inefficient. The laser spot has poor effect when the surface texture is undulating. Patent 202211537950.9 A method, device, integrated excavation and anchoring machine and storage medium for locating anchor holes. Pre-processing is required to obtain the point cloud data of the anchor hole to be located, and only the center position information is obtained, without the normal vector information of the anchor hole. Patent 202211531596.9 A method, system, electronic device and medium for locating the center of a drilling and anchor hole. Binarization processing requires the selection of a suitable threshold, and the threshold may need to be adjusted according to the specific application scenario. Improper threshold selection affects the final center point positioning. At the same time, no normal vector is provided.

[0004] Therefore, there is an urgent need to develop a fused anchor hole positioning method and system to solve the problem of low work efficiency and danger in the process of anchoring and protection operations by manually positioning anchor points. Summary of the invention

[0005] The object of the present invention is to provide an anchor point positioning method and device for positioning a circular-end rectangular anchor point to determine the position and posture of the anchor point, and provide the control system with the center point position and normal vector of the anchor point.

[0006] The present invention is implemented according to the following technical solutions:

[0007] In a first aspect, the present application provides a method for locating an anchor hole point, the method comprising:

[0008] Obtain visual and point cloud information data that needs to be processed for target alignment;

[0009] Obtain the target frame coordinates through vision-based target detection and perform expansion processing on the target frame;

[0010] Obtain the point cloud of the round-end rectangular anchor hole points within the target frame after expansion processing, and pre-process the point cloud;

[0011] Use the pre-scanned standard round-end rectangular anchor point cloud to perform similarity matching on the pre-processed point cloud to obtain the point cloud matching degree;

[0012] Perform normal analysis on the processed point cloud to obtain the point cloud plane normal evaluation degree and plane parameters;

[0013] The point cloud matching degree and the point cloud normal evaluation degree are weighted to obtain the total cost, and the circular end rectangular anchor point point cloud is processed according to the total cost to obtain the circular end rectangular anchor point normal vector;

[0014] Perform coordinate transformation on the normal vector of the circular end rectangle anchor point to calculate the center coordinates of the circular end rectangle;

[0015] According to the plane parameters and the expanded target frame, points are randomly scattered in the plane, and the Gaussian process model is trained using the point cloud obtained by the random scattering.

[0016] The trained Gaussian process regression model is used to determine whether the center point meets the standard and determine the center coordinates of the rectangular anchor hole at the circular end.

[0017] In some embodiments, the preprocessing of the point cloud includes:

[0018] Downsampling and removing outliers; the downsampling adopts a voxel grid filtering method, and the outlier removal adopts a statistical filtering method;

[0019] By rotating the matrix R p and the translation vector t p Perform conversion to transform the point cloud coordinate system into the initial global coordinate system;

[0020] Initialize the origin of the global coordinate system through the positioning system.

[0021] In some embodiments, the pre-scanned standard round-end rectangular anchor point cloud needs to be preprocessed, specifically: the point cloud coordinates are transformed to the coordinates of the center point of the round-end rectangle, the transformation matrix is ​​obtained by manual calibration method, the part is placed at a known coordinate from the origin of the laser radar coordinates for scanning, and the rotation matrix and translation vector are obtained.

[0022] In some embodiments, the use of the pre-scanned standard round-end rectangular anchor point cloud to perform similarity matching processing on the pre-processed point cloud to obtain the point cloud matching degree includes:

[0023] The ICP iterative closest point algorithm is used for registration to obtain the transformation matrix T, including the rotation matrix R and the translation vector t. The matching degree m is determined by calculating the registration error. The registration error is calculated using the root mean square error to obtain the matching degree. The root mean square error calculation method is:

[0024]

[0025] Where N is the number of point pairs, d i is the Euclidean distance between the i-th corresponding point pair.

[0026] In some embodiments, the performing normal analysis on the preprocessed point cloud information to obtain the point cloud plane normal evaluation degree and plane parameters includes:

[0027] The preprocessed point cloud information is subjected to PCA principal component analysis to obtain the plane normal vector n1. The PCA principal component analysis uses the SVD singular value decomposition method to obtain the point cloud normal vector n1. The preprocessed point cloud information is subjected to RANSAC random sampling consistency analysis to obtain the plane normal vector n2 and plane parameters. The plane parameter form is:

[0028] Ax+By+Cz+D=0

[0029] A, B, C are the plane normal vectors, and D is the distance from the plane to the origin;

[0030] The normal evaluation degree of the point cloud plane is the angle β between n1 and n2 n .

[0031] In some embodiments, a transformation matrix T may be obtained through ICP registration, where the transformation matrix T includes a rotation matrix R and a translation vector t:

[0032]

[0033] Among them, x, y, z are the coordinates of the center point of the pre-scanned standard circular end rectangular anchor point, and the coordinate value is (0, 0, 0). The point cloud coordinates are transformed into a coordinate system with the center point as the origin; x', y', z' are the coordinates of the anchor point center relative to the initial global coordinate system, which are used as the center point coordinates.

[0034] In some embodiments, randomly scattering points in the plane according to the plane parameters and the expanded target frame, and training the Gaussian process model using the point cloud obtained by the random scattering points, includes:

[0035] According to the given plane parameters and the expanded target box, randomly scatter points on the plane to generate a point cloud dataset. The generated point cloud dataset (X, Y, Z) is used to train the Gaussian process model, with X and Y as input and Z as output, to learn the distribution and characteristics of the points.

[0036] The Gaussian process model selects a radial basis function kernel, whose kernel function parameters include a length scale; a large length scale indicates that the kernel function is insensitive to changes in input data, that is, the model is relatively smooth; a small length scale indicates that the kernel function is sensitive to changes in input data, that is, the model may have large fluctuations; the kernel function is:

[0037]

[0038] Among them, l is the length dimension parameter, ||x i -x j || is the Euclidean distance between the input data points.

[0039] In some embodiments, whether the Gaussian process regression through the center point meets the standard includes:

[0040] The Gaussian process regression result is the predicted value point_pred and the uncertainty standard deviation sigma. If the center point coordinates (x', y', z') are already obtained, the Gaussian process regression model is used for prediction, where the input value is (x', y'), the predicted value is z, i.e., the height, and the corresponding uncertainty standard deviation sigma;

[0041] The 95% confidence interval (point_pred-1.96*sigma, point_pred+1.96) is determined by the output prediction value. If z' is within the 95% confidence interval, the center point coordinates (x', y', z') can be determined; if z' is not within the 95% confidence interval, the next target is processed.

[0042] In a second aspect, the present invention provides an anchor hole positioning device, the device comprising:

[0043] An acquisition module is used to acquire visual and point cloud information data that needs to be processed for target alignment;

[0044] An expansion module is used to obtain the target frame coordinates through vision-based target detection and to expand the target frame;

[0045] A preprocessing module is used to obtain a point cloud of round-end rectangular anchor holes within the target frame after expansion processing, and to preprocess the point cloud;

[0046] A matching degree module is used to use the pre-scanned standard round-end rectangular anchor point cloud to perform similarity matching processing on the pre-processed point cloud to obtain the point cloud matching degree;

[0047] The normal evaluation module is used to perform normal analysis on the pre-processed point cloud information to obtain the point cloud plane normal evaluation degree and plane parameters;

[0048] The comprehensive weighted evaluation module is used to weight the point cloud matching degree and the point cloud normal evaluation degree to obtain the total cost, and process the circular end rectangular anchor point point cloud according to the total cost to obtain the circular end rectangular anchor point normal vector;

[0049] The center coordinate calculation module is used to perform coordinate transformation on the normal vector of the circular end rectangle anchor point to calculate the center coordinate of the circular end rectangle;

[0050] The training module is used to randomly scatter points in the plane according to the plane parameters and the expanded target box, and train the Gaussian process model using the point cloud obtained by the random scattering;

[0051] The judgment module is used to determine whether the Gaussian process regression of the center point meets the standard and determine the center coordinates of the rectangular anchor hole point at the circular end.

[0052] In the third aspect, the present invention provides an anchor hole positioning system, which includes a sensor module, a computing platform and a wire-controlled chassis; wherein the sensor module includes a depth camera and a binocular vision positioning camera, and the computing platform uses the ubuntu ROS system to run the fusion anchor hole positioning method described in any one of claims 1-8 to determine the center coordinates of the anchor point; the computing platform plans a path according to the center coordinates of the anchor point, uses the MPC controller to output control signals including speed and steering angular velocity, and realizes movement through communication with the wire-controlled chassis via the CAN bus.

[0053] Beneficial effects of the present invention

[0054] The present invention directly obtains point cloud information within the target frame through the processing of visual and point cloud information, reduces the amount of point cloud information processing, and reduces the amount of calculation. The present invention combines 2D and 3D information to obtain the position and normal vector information of the anchor hole, and evaluates the accuracy of anchor point positioning to ensure positioning accuracy, uses the anchor hole position information for path planning, and controls the wire-controlled chassis to reach the anchor point position. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The accompanying drawings are part of the present invention and are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but do not constitute an improper limitation of the present invention. Obviously, the accompanying drawings described below are only some embodiments. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0056] In the attached picture:

[0057] Figure 1 A flow chart of an anchor hole positioning method provided by an embodiment of the present invention;

[0058] Figure 2 A schematic diagram of the structure of an anchor hole positioning system provided in one embodiment of the present invention.

[0059] It should be noted that these drawings and textual descriptions are not intended to limit the conceptual scope of the present invention in any way, but are intended to illustrate the concept of the present invention for those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The following embodiments are used to illustrate the present invention but are not used to limit the scope of the present invention.

[0061] The following is combined with Figure 1 The specific implementation method of the present invention is further described. Figure 1 As shown, an embodiment of the present disclosure provides an anchor hole positioning method, the method comprising the following steps:

[0062] Step S100: Obtaining visual and point cloud information data to be processed for target alignment.

[0063] Specifically, an RGB depth camera is used to obtain aligned visual and point cloud information. If point cloud information cannot be obtained directly, lidar and visual sensors can be used, but joint calibration is required.

[0064] Step S200: Obtain the target frame coordinates through vision-based target detection, and perform expansion processing on the target frame.

[0065] Furthermore, the target detection algorithm uses yolov5 to process the target frame and expand the target frame to obtain the processed target frame.

[0066] The method of expanding the target frame is to expand along the long side and the short side of the rectangular target frame, and the side lengths are multiplied by parameters α1 and α2, where α1>1, α2>1. The purpose of expanding the target frame is to facilitate obtaining the normal vector.

[0067] Step S300: obtaining a point cloud of round-end rectangular anchor points within the target frame after the dilation process, and preprocessing the point cloud to obtain a preprocessed point cloud.

[0068] Using aligned visual and point cloud information to directly obtain point cloud information within the target box can reduce the amount of point cloud information processing and reduce the amount of calculation.

[0069] Furthermore, the point cloud information within the target frame can be directly obtained using the RGB depth camera, or it can be calculated based on the internal parameters. The calculation formula is:

[0070] X=(u-cx)×Z / f x

[0071] Y=(u-cy)×Z / f y

[0072] Z = depth_value × depth_scale

[0073] f x ,f y is the focal length; cx, cy are the coordinates of the center point of the main point image pixel; depth_value is the depth information of the pixel in the image, and depth_scale is the scale factor. X, Y, Z are the converted point cloud information. Using the ROS system and PCL library, it is very convenient to process lidar and point cloud data. For processing using the PCL library, the point cloud can be converted into the ROS standard point cloud format. Using visual sensors and lidars, point cloud information within the target frame can be directly obtained.

[0074] In an embodiment of the present application, preprocessing the point cloud includes:

[0075] Step S310: downsampling and removing outliers.

[0076] To downsample using voxel grid filtering, you need to set the voxel grid size; to remove outliers using statistical filtering, you need to set the number of neighborhood points and the standard deviation multiplier threshold from the mean.

[0077] Step S320: Using the rotation matrix R p and the translation vector t p Perform a transformation to convert the point cloud coordinate system to the initial global coordinate system.

[0078] Step S330: Initialize the origin of the global coordinate system through the positioning system.

[0079] The origin of the initial global coordinate system can be initialized by the positioning system, which can use the SLAM algorithm based on LiDAR, the positioning algorithm based on vision or the IMU fusion algorithm. Currently, the T265 positioning camera is used for positioning, combined with the inertial measurement unit IMU and the visual inertial odometer for positioning.

[0080] Step S400: using the pre-scanned standard round-end rectangular anchor point cloud, perform similarity matching processing on the pre-processed point cloud to obtain the point cloud matching degree.

[0081] In the embodiment of the present application, the standard circular end rectangular anchor point cloud is pre-scanned, and the point cloud needs to be pre-processed to transform the point cloud coordinates into the circular end rectangular center point coordinates.

[0082] The specific method is: use the manual calibration method to obtain the transformation matrix, place the part at the known coordinates from the laser radar coordinate origin for scanning, and obtain the rotation matrix and translation vector.

[0083] For the pre-scanned standard circular end rectangular anchor point cloud, it is necessary to prepare the standard circular end rectangular anchor point cloud as a standard for similarity matching. The pre-processed point cloud is processed for similarity matching, and the ICP iterative closest point algorithm is used for registration to obtain the transformation matrix T, including the rotation matrix R and the translation vector t. The matching degree m is determined by calculating the registration error, and the registration error is calculated using the root mean square error to obtain the matching degree.

[0084] Root mean square error calculation method:

[0085]

[0086] Where N is the number of point pairs, d i is the Euclidean distance between the i-th corresponding point pair.

[0087] Step S500: Perform normal analysis on the processed point cloud to obtain the point cloud plane normal evaluation degree and plane parameters.

[0088] Perform normal analysis on the processed point cloud, perform PCA principal component analysis to obtain the plane normal vector n1, and use the SVD singular value decomposition method to obtain the point cloud normal vector n1. Perform RANSAC random sampling consistency analysis on the preprocessed point cloud information to obtain the plane normal vector n2 and plane parameters.

[0089] Plane parameter form:

[0090] Ax+By+Cz+D=0

[0091] A, B, C are the plane normal vectors, and D is the distance from the plane to the origin.

[0092] Obtain the point cloud plane normal evaluation and plane parameters. The point cloud plane normal evaluation is the angle β between n1 and n2. n .

[0093] n1 obtained by PCA principal component analysis represents the normal vector of the entire point cloud, and n2 is the normal vector of a plane fitted in the point cloud. The angle between the two can be used to judge the quality of the normal vector obtained after point cloud preprocessing.

[0094] Step S600: The point cloud matching degree m and the point cloud normal evaluation degree β n The total cost is obtained by weighting, and the point cloud of the circular end rectangular anchor point is processed according to the total cost to obtain the normal vector of the circular end rectangular anchor point.

[0095] f cost =γ1m+γ2β n

[0096] Among them, f costis the total cost, γ1 and γ2 are the weighted values ​​of the point cloud matching degree and the point cloud normal evaluation degree, and the sum of the weighted values ​​is 1, so γ1=0.5 and γ2=0.5 can be selected.

[0097] If it is less than the threshold, the point cloud of the circular end rectangle anchor point is processed to obtain the normal vector of the circular end rectangle anchor point. If it is greater than the threshold, the process returns to step S100 to select other targets and start again.

[0098] The matching degree m is determined by calculating the registration error, and the registration error is calculated using the root mean square error to obtain the matching degree. If m is larger, it means that the point cloud is farther away from the pre-scanned standard circular end rectangular anchor point cloud, that is, the matching degree is lower.

[0099] The point cloud plane normal evaluation and plane parameters are obtained. The point cloud plane normal evaluation is the angle βn between n1 and n2. The larger the angle, the greater the deviation between the two. Therefore, the larger the weighted cost, the lower the matching degree and the mismatch of the normal.

[0100] Step S700: performing coordinate transformation on the normal vector of the round end rectangle anchor point to calculate the center coordinates of the round end rectangle.

[0101] The translation vector t obtained by ICP registration in step S400 is the coordinate of the center point relative to the initial global coordinate system. Through ICP registration in S400, a transformation matrix T can be obtained, and the transformation matrix T includes a rotation matrix R and a translation vector t.

[0102]

[0103] Where x, y, and z are the coordinates of the center point of the pre-scanned standard circular end rectangular anchor point, and the coordinate value is (0, 0, 0). In step S400, the point cloud coordinates have been transformed into a coordinate system with the center point as the origin using the calibration method. x', y', and z' are the coordinates of the center of the anchor point relative to the initial global coordinate system, and the coordinates of the center point are obtained.

[0104] Step S800: randomly scatter points on the plane according to the plane parameters and the expanded target frame, and train the Gaussian process regression model using the point cloud obtained by the random scattering.

[0105] According to the given plane parameters and the expanded target box, points are randomly scattered on the plane to generate a point cloud dataset. The generated point cloud dataset (X, Y, Z) is used to train the Gaussian process model, with X and Y as input and Z as output, to learn the distribution and characteristics of the points.

[0106] The Gaussian process model selects the radial basis function kernel (RBF Kernel). The parameters include the length scale, which is the sensitivity of the kernel function to the input data. A smaller length scale means that the kernel function is more sensitive to input changes and can capture smaller changes. A larger length scale means that the kernel function is less sensitive to input changes and is more suitable for capturing global trends.

[0107] Furthermore, the kernel function is:

[0108]

[0109] Where l is the length dimension parameter, ||x i -x j || is the Euclidean distance between the input data points.

[0110] Step S900: using the trained Gaussian process regression model to determine whether the center point meets the standard, so as to determine the center coordinates of the circular end rectangular anchor hole point.

[0111] The credibility of the center point is determined by the trained Gaussian process regression model. The credibility of the center point is evaluated by the model's predicted value point_pred and uncertainty (standard deviation) sigma.

[0112] The Gaussian process regression result is the predicted value point_pred and the uncertainty (standard deviation) sigma. If the center point coordinates (x', y', z') have been obtained in S7, the prediction is made through the Gaussian process regression model, where the input value is (x', y'), the predicted value is z, i.e., the height, and the corresponding uncertainty (standard deviation) sigma.

[0113] The 95% confidence interval (point_pred-1.96*sigma, point_pred+1.96) is determined by the output prediction value. If z' is within the 95% confidence interval, the center point coordinates (x', y', z') are determined, and the center point coordinates and normal vector of the round end rectangle anchor point are obtained and output. Otherwise, return to step S100 and start again to process the next target.

[0114] The following is an embodiment of the anchor hole positioning device of the present invention, which can be used to perform the embodiment of the anchor hole positioning method of the present invention. For details not disclosed in the embodiment of the anchor hole positioning device of the present invention, please refer to the embodiment of the anchor hole positioning method of the present invention.

[0115] In one embodiment, an anchor hole positioning device is provided, the device comprising:

[0116] An acquisition module is used to acquire visual and point cloud information data that needs to be processed for target alignment;

[0117] An expansion module is used to obtain the target frame coordinates through vision-based target detection and to expand the target frame;

[0118] A preprocessing module is used to obtain a point cloud of round-end rectangular anchor holes within the target frame after expansion processing, and to preprocess the point cloud;

[0119] A matching degree module is used to use the pre-scanned standard round-end rectangular anchor point cloud to perform similarity matching processing on the pre-processed point cloud to obtain the point cloud matching degree;

[0120] The normal evaluation module is used to perform normal analysis on the pre-processed point cloud information to obtain the point cloud plane normal evaluation degree and plane parameters;

[0121] The comprehensive weighted evaluation module is used to weight the point cloud matching degree and the point cloud normal evaluation degree to obtain the total cost, and judge whether it meets the requirements according to the total cost threshold; if it meets the requirements, continue to process the circular end rectangular anchor point point cloud to obtain the circular end rectangular anchor point normal vector; if it does not meet the requirements, return to process the next target;

[0122] The center coordinate calculation module is used to perform coordinate transformation on the normal vector of the circular end rectangle anchor point to calculate the center coordinate of the circular end rectangle;

[0123] The training module is used to randomly scatter points in the plane according to the plane parameters and the expanded target box, and train the Gaussian process model using the point cloud obtained by the random scattering;

[0124] The judgment module is used to judge whether the Gaussian process regression of the center point meets the standard; if it meets the standard, the center coordinates of the rectangular anchor hole at the circular end are obtained; if it does not meet the standard, the next target is processed.

[0125] It should be noted that the anchor hole positioning device provided in the above embodiment only uses the division of the above functional modules as an example when executing the anchor hole positioning method. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the anchor hole positioning device provided in the above embodiment belongs to the same concept as the anchor hole positioning method embodiment. The implementation process thereof is detailed in the anchor hole positioning method embodiment, which will not be repeated here.

[0126] like Figure 2As shown, in one embodiment, an anchor hole positioning system is proposed, which includes a sensor, a computing platform and a mobile chassis. The sensor includes a depth camera and a binocular vision positioning camera. The computing platform uses the Ubuntu system and the ROS system to run a 2D-3D fusion anchor hole positioning method to determine the center coordinates of the anchor point. The computing platform plans a path according to the center coordinates of the anchor point, uses an MPC controller to output control signals including speed and steering angular velocity, and realizes movement through CAN bus and wire-controlled chassis communication.

[0127] In the description provided herein, a large number of specific details are described. However, it is understood that embodiments of the present invention can be practiced without these specific details. In some instances, well-known methods, structures and techniques are not shown in detail so as not to obscure the understanding of this description.

[0128] In addition, those skilled in the art will understand that, although some embodiments described herein include certain features included in other embodiments but not other features, the combination of features of different embodiments also means being within the scope of protection of the present invention and forming different embodiments. For example, in the above embodiments, those skilled in the art can use them in combination according to the known technical solutions and the technical problems to be solved by this application.

[0129] The above is only a preferred embodiment of the present invention, and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technician familiar with this patent can make some changes or modify the technical contents suggested above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention without departing from the content of the technical solution of the present invention still fall within the scope of the solution of the present invention.

Claims

1. An anchor hole positioning method, characterized in that: The method comprises: Obtain visual and point cloud information data that needs to be processed for target alignment; Obtain the target frame coordinates through vision-based target detection and perform expansion processing on the target frame; Obtain the point cloud of the round-end rectangular anchor hole points within the target frame after expansion processing, and pre-process the point cloud; Use the pre-scanned standard round-end rectangular anchor point cloud to perform similarity matching on the pre-processed point cloud to obtain the point cloud matching degree; Perform normal analysis on the processed point cloud to obtain the point cloud plane normal evaluation degree and plane parameters; The point cloud matching degree and the point cloud normal evaluation degree are weighted to obtain the total cost, and the circular end rectangular anchor point point cloud is processed according to the total cost to obtain the circular end rectangular anchor point normal vector; Perform coordinate transformation on the normal vector of the circular end rectangle anchor point to calculate the center coordinates of the circular end rectangle; According to the plane parameters and the expanded target frame, points are randomly scattered in the plane, and the Gaussian process model is trained using the point cloud obtained by the random scattering. The trained Gaussian process regression model is used to determine whether the center point meets the standard and determine the center coordinates of the rectangular anchor hole at the circular end.

2. The anchor hole positioning method according to claim 1, characterized in that: The preprocessing of the point cloud comprises: Downsampling and removing outliers; the downsampling adopts a voxel grid filtering method, and the outlier removal adopts a statistical filtering method; By rotating the matrix R p and the translation vector t p Perform conversion to transform the point cloud coordinate system into the initial global coordinate system; Initialize the origin of the global coordinate system through the positioning system.

3. The anchor hole positioning method according to claim 1, characterized in that: The pre-scanned standard circular end rectangular anchor point cloud needs to be pre-processed to transform the point cloud coordinates to the circular end rectangle center point coordinates; the specific method is: use a manual calibration method to obtain a transformation matrix, place the part at a known coordinate from the laser radar coordinate origin for scanning, and obtain a rotation matrix and a translation vector.

4. The anchor hole positioning method according to claim 3, characterized in that: The method of using the pre-scanned standard round-end rectangular anchor point cloud to perform similarity matching processing on the pre-processed point cloud to obtain the point cloud matching degree includes: The ICP iterative closest point algorithm is used for registration to obtain the transformation matrix T, including the rotation matrix R and the translation vector t. The matching degree m is determined by calculating the registration error. The registration error is calculated using the root mean square error to obtain the matching degree. The root mean square error calculation method is: Where N is the number of point pairs, d i is the Euclidean distance between the i-th corresponding point pair.

5. The anchor hole positioning method according to claim 1, characterized in that: The normal analysis of the pre-processed point cloud information is performed to obtain the point cloud plane normal evaluation degree and plane parameters, including: The preprocessed point cloud information is subjected to PCA principal component analysis to obtain the plane normal vector n1. The PCA principal component analysis uses the SVD singular value decomposition method to obtain the point cloud normal vector n1. The preprocessed point cloud information is subjected to RANSAC random sampling consistency analysis to obtain the plane normal vector n2 and plane parameters. The plane parameter form is: Ax+By+Cz+D=0 A, B, C are the plane normal vectors, and D is the distance from the plane to the origin; The normal evaluation degree of the point cloud plane is the angle β between n1 and n2 n .

6. The anchor hole positioning method according to claim 4, characterized in that: The transformation matrix T can be obtained through ICP registration, and the transformation matrix T includes the rotation matrix R and the translation vector t: Among them, x, y, z are the coordinates of the center point of the pre-scanned standard circular end rectangular anchor point, and the coordinate value is (0, 0, 0). The point cloud coordinates are transformed into a coordinate system with the center point as the origin; x', y', z' are the coordinates of the anchor point center relative to the initial global coordinate system, which are used as the center point coordinates.

7. The anchor hole positioning method according to claim 1, characterized in that: The step of randomly scattering points in the plane according to the plane parameters and the expanded target frame, and training the Gaussian process model using the point cloud obtained by the random scattering points, includes: According to the given plane parameters and the expanded target box, randomly scatter points on the plane to generate a point cloud dataset. The generated point cloud dataset (X, Y, Z) is used to train the Gaussian process model, with X and Y as input and Z as output, to learn the distribution and characteristics of the points. The Gaussian process model selects a radial basis function kernel, whose kernel function parameters include a length scale; a large length scale indicates that the kernel function is insensitive to changes in input data, that is, the model is relatively smooth; a small length scale indicates that the kernel function is sensitive to changes in input data, that is, the model may have large fluctuations; the kernel function is: Among them, l is the length dimension parameter, ||x i -x j || is the Euclidean distance between the input data points.

8. The anchor hole positioning method according to claim 1, characterized in that: Whether the Gaussian process regression through the center point meets the standards includes: The Gaussian process regression result is the predicted value point_pred and the uncertainty standard deviation sigma. If the center point coordinates (x', y', z') are already obtained, the Gaussian process regression model is used for prediction, where the input value is (x', y'), the predicted value is z, i.e., the height, and the corresponding uncertainty standard deviation sigma; The 95% confidence interval (point_pred-1.96*sigma, point_pred+1.96) is determined by the output prediction value. If z' is within the 95% confidence interval, the center point coordinates (x', y', z') can be determined; if z' is not within the 95% confidence interval, the next target is processed.

9. An anchor hole positioning device, characterized in that: The device comprises: An acquisition module is used to acquire visual and point cloud information data that needs to be processed for target alignment; An expansion module is used to obtain the target frame coordinates through vision-based target detection and to expand the target frame; A preprocessing module is used to obtain a point cloud of round-end rectangular anchor holes within the target frame after expansion processing, and to preprocess the point cloud; A matching degree module is used to use the pre-scanned standard round-end rectangular anchor point cloud to perform similarity matching processing on the pre-processed point cloud to obtain the point cloud matching degree; The normal evaluation module is used to perform normal analysis on the pre-processed point cloud information to obtain the point cloud plane normal evaluation degree and plane parameters; The comprehensive weighted evaluation module is used to weight the point cloud matching degree and the point cloud normal evaluation degree to obtain the total cost, and process the circular end rectangular anchor point point cloud according to the total cost to obtain the circular end rectangular anchor point normal vector; The center coordinate calculation module is used to perform coordinate transformation on the normal vector of the circular end rectangle anchor point to calculate the center coordinate of the circular end rectangle; The training module is used to randomly scatter points in the plane according to the plane parameters and the expanded target box, and train the Gaussian process model using the point cloud obtained by the random scattering; The judgment module is used to determine whether the Gaussian process regression of the center point meets the standard and determine the center coordinates of the rectangular anchor hole point at the circular end.

10. An anchor hole positioning system, characterized in that: The system includes a sensor module, a computing platform and a wire-controlled chassis; wherein the sensor module includes a depth camera and a binocular vision positioning camera, and the computing platform uses ubunut and ROS system to run the fusion anchor hole positioning method described in any one of claims 1 to 8 to determine the center coordinates of the anchor point; the computing platform plans the path according to the center coordinates of the anchor point, uses the MPC controller to output control signals including speed and steering angular velocity, and realizes movement through CAN bus and wire-controlled chassis communication.

Images