Path planning method and system for customizing key point location based on 3D vision

Through customizing the path planning method of key point positions, combining 2D and 3D vision technology, point cloud data is obtained and robot motion trajectory is formed, which solves the problem that traditional technology cannot obtain depth information and manual correction time-consuming and labor-intensive problems, and improves the flexibility of path planning and practicality.

CN119088029APending Publication Date: 2024-12-06SHANGHAI ZHIYIN AUTOMATION TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411243908.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-05
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

Traditional 2D vision technology cannot obtain the depth information of objects and cannot meet the needs of complex working environments. In addition, traditional robot path correction methods require manual teaching, which is time-consuming and labor-intensive.

Method used

Using a custom key point path planning method, the point cloud data of the target object is obtained for preprocessing, and converted into 2D images. The key point is selected using the human-computer interactive interface, determine the ROI area, fit the plane, and obtain the robot moving position Data forms the robot's motion trajectory.

Benefits of technology

The combination of 2D and 3D vision technology is achieved, which improves the flexibility and practicality of path planning, reduces the time and labor of manual correction, and is suitable for complex operating environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119088029A_ABST
    Figure CN119088029A_ABST
Patent Text Reader

Abstract

The invention discloses a path planning method and system for customizing a key point location. The method comprises the following steps: step 1, acquiring original point cloud data of a target object; 2, performing data preprocessing operation on the original point cloud data to obtain preprocessed point cloud data; step 3, converting the preprocessed point cloud data into a 2D image so as to obtain a two-dimensional coordinate corresponding to the point cloud data; 4, displaying the 2D image through a human-computer interaction interface; step 5, selecting a target key point on the 2D image to obtain a corresponding ROI (Region of Interest); step 6, determining a point cloud fitting plane according to the ROI; 7, according to the plane normal vector determined by the point cloud fitting plane and the original point normal vector of the robot tool coordinate system, robot moving pose data are obtained; and step 8, obtaining a robot movement track according to the robot movement pose data. The method is high in flexibility, high in practicability and wide in application prospect.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of machine vision technology, and in particular to a method and system for self-defined key point path planning. Background Art

[0002] With the continuous development of modern industry and robotics technology, intelligent robots have been widely used in industrial production. Combining industrial cameras with robots is an important trend in the future development of robots. At present, the common 2D vision processing object is a flat image or video, which only contains spatial information on the X-axis and Y-axis, and cannot obtain the depth information of the object, and cannot meet the complex working environment. 3D vision not only contains spatial information on the X-axis and Y-axis, but also includes depth information on the Z-axis, which can obtain more comprehensive information.

[0003] In the robot mobile path planning and control related technologies, when the detected object is offset, or path points need to be added or subtracted, the traditional method requires manual teaching and correction of the robot path, which is time-consuming and labor-intensive.

[0004] Therefore, it is necessary to develop a visual technology that combines 2D and 3D to solve the problems raised above. Summary of the invention

[0005] The present invention provides a path planning method for customizing key points to overcome the above-mentioned technical problems.

[0006] The technical solution of the present invention is:

[0007] A path planning method for custom key points, characterized by comprising:

[0008] Step 1: Obtain the original point cloud data of the target object;

[0009] Step 2: performing data preprocessing on the original point cloud data to obtain preprocessed point cloud data;

[0010] Step 3: Convert the pre-processed point cloud data into a 2D image, thereby obtaining the two-dimensional coordinates corresponding to the point cloud data;

[0011] Step 4: displaying the aforementioned 2D image through a human-computer interaction interface;

[0012] Step 5: Select the target key points on the aforementioned 2D image to obtain the corresponding ROI area;

[0013] Step 6: Determine the point cloud fitting plane according to the aforementioned ROI area;

[0014] Step 7: Obtain the robot movement posture data according to the plane normal vector determined by the point cloud fitting plane and the normal vector of the origin of the robot tool coordinate system;

[0015] Step 8: Obtain the robot motion trajectory based on the robot movement posture data.

[0016] Furthermore, in the aforementioned step 1, the original point cloud data of the target object is obtained through a 3D camera.

[0017] Furthermore, the data preprocessing operations in the aforementioned step 2 include: downsampling operation, pass-through filtering operation, radius filtering operation, statistical filtering operation, and conditional filtering operation; the user can flexibly select one of the preprocessing operations or a combination of operations according to the point cloud data processing requirements, that is, the specific application requirements.

[0018] Furthermore, the specific steps of converting the pre-processed point cloud data into a 2D image, i.e., a two-dimensional image, in the aforementioned step 3 are as follows:

[0019] Step 31, project the preprocessed point cloud data onto the XOY plane and obtain the extreme values ​​of the plane where the two-dimensional point cloud is located. Each extreme point is represented by x_max, x_min, y_max, y_min, and then the coordinate range of the two-dimensional image is determined, so as to realize the conversion of each three-dimensional point (x, y, z) into a two-dimensional point (x, y);

[0020] Step 32, calculating the Z-axis height difference of the preprocessed point cloud data in the Z-axis direction and normalizing it so that the Z coordinate value is mapped to the range of 0-1;

[0021] Step 33: Convert the normalized Z coordinate value into a pixel value to obtain a 2D image.

[0022] The human-computer interaction interface in the aforementioned step 4 is programmed in a programming language to form a corresponding interface program, and the aforementioned interface program can display the 2D image and obtain the key points selected by the user; the programming language includes but is not limited to any one of HTML, CSS, JavaScript, QT, and C#.

[0023] Furthermore, the specific steps of obtaining the ROI area corresponding to the 2D image in the above step 5 include:

[0024] The key points selected by the user are obtained through the interface program, and the ROI area corresponding to each key point is obtained; the interface coordinates of the key point are (x, y), and the ROI area is drawn with the key point as the center, w as the width, and h as the height.

[0025] Furthermore, the specific steps of determining the point cloud fitting plane according to the ROI area in the aforementioned step 6 include:

[0026] The specific steps of determining the point cloud fitting plane according to the ROI area in the aforementioned step 6 include:

[0027] Step 61: According to the plurality of ROI regions obtained above, each of the ROI regions is mapped to a corresponding 3D point cloud data, so as to fit the point cloud of each of the ROI regions into a plurality of planes;

[0028] Step 62: Filter out the point cloud set with the most plane point cloud data as the plane finally fitted in the ROI area.

[0029] Furthermore, the plane fitting method in step 61 adopts a random sampling consensus algorithm or a least squares method, which needs to be selected according to the quality of the point cloud data and the requirements of the application scenario.

[0030] Furthermore, in the aforementioned step 7, the specific steps of obtaining the robot movement posture data according to the plane normal vector determined by the point cloud fitting plane and the origin normal vector of the robot tool coordinate system include:

[0031] In the above step 7, the specific steps of obtaining the robot movement posture data according to the plane normal vector determined by the point cloud fitting plane and the origin normal vector of the robot tool coordinate system include:

[0032] Step 71, obtaining the plane normal vector and the robot tool coordinate system origin normal vector;

[0033] Step 72, obtain the rotation angle and the center vector according to the aforementioned plane normal vector and the normal vector of the robot tool coordinate system origin, wherein the normal vector m of the robot workpiece coordinate system origin is set, and the rotation angle θ is calculated according to the dot product of the aforementioned plane normal vector n and the normal vector m of the robot tool coordinate system origin, that is, by n×m=|n|×|m|×cosθ, the angle θ between vectors n and m is obtained, n×m represents the dot product of vectors n and m, |n|, |m| are the modulus lengths of vectors n and m respectively; calculate the center vector u according to the cross product between the aforementioned plane normal vector n and the normal vector m of the robot tool coordinate system origin, and the center vector u is perpendicular to the normal vector n and the normal vector m;

[0034] Step 73, taking the center vector u as the rotation center axis, rotating at an angle of θ, and obtaining the corresponding rotation vector w, w=θu; that is, the rotation vector w refers to a rotation of θ radians in the direction of the rotation center axis u;

[0035] Step 74: Obtain the required movement posture data of the robot according to the quaternion index mapping formula.

[0036] The present invention also provides a user-defined key point path planning system, which is characterized by comprising:

[0037] An acquisition unit capable of acquiring original point cloud data of a target object;

[0038] A data preprocessing unit, which can perform a data preprocessing operation on the aforementioned original point cloud data to obtain preprocessed point cloud data;

[0039] An image conversion unit, which can convert the pre-processed point cloud data into a 2D image, thereby obtaining the two-dimensional coordinates corresponding to the point cloud data;

[0040] An image display unit, which can provide a display interface to display the aforementioned 2D image and obtain customized key target points;

[0041] A ROI region definition unit, which can obtain a number of ROI regions corresponding to the aforementioned 2D image, and the ROI regions are determined by custom key points;

[0042] A point cloud fitting plane processing unit, which can determine a point cloud fitting plane based on the aforementioned ROI area;

[0043] The robot motion trajectory acquisition unit can obtain the robot movement posture data according to the plane normal vector determined by the point cloud fitting plane and the robot tool coordinate system origin normal vector; and then obtain the robot motion trajectory according to the robot movement posture data.

[0044] Beneficial effects of the present invention:

[0045] The present invention selects 2D image information through a human-machine interactive interface to determine the key points of robot movement, thereby forming a robot movement path. The method has greater flexibility, strong practicality, and a relatively broad application prospect. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0047] Figure 1 is a flowchart of the steps of the method corresponding to the present invention;

[0048] Figure 2 The original point cloud data image corresponding to the specific example of the present invention;

[0049] Figure 3 A target point cloud data image corresponding to a specific example of the present invention;

[0050] Figure 4 A schematic diagram of converting point cloud data corresponding to a specific example of the present invention into a 2D image;

[0051] Figure numerals: 1-original point cloud data image; 2-preprocessed target point cloud data image; 3-target point cloud data converted into 2D image; 4-ROI area; 5-robot movement trajectory. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] In order to solve the above problems, Figure 1-Figure 4 The present invention proposes a new type of path planning method for custom key points, which is characterized by comprising the following steps:

[0054] Step 1: Obtain the original point cloud data of the target object;

[0055] Step 2: performing data preprocessing on the original point cloud data to obtain preprocessed point cloud data;

[0056] Step 3: Convert the pre-processed point cloud data into a 2D image, thereby obtaining the two-dimensional coordinates corresponding to the point cloud data;

[0057] Step 4: displaying the aforementioned 2D image through a human-computer interaction interface;

[0058] Step 5: Select the target key points on the aforementioned 2D image to obtain the corresponding ROI area;

[0059] Step 6: Determine the point cloud fitting plane according to the aforementioned ROI area;

[0060] Step 7: Obtain the robot movement posture data according to the plane normal vector determined by the point cloud fitting plane and the normal vector of the origin of the robot tool coordinate system;

[0061] Step 8: Obtain the robot motion trajectory based on the robot movement posture data.

[0062] Preferably, in the aforementioned step 1, original point cloud data of the target object is acquired through a 3D camera.

[0063] Preferably, in the aforementioned step 2, a data preprocessing operation is performed on the aforementioned original point cloud data to obtain preprocessed point cloud data, and the point cloud preprocessing operation is used to optimize the point cloud data, reduce the amount of calculation and improve the processing efficiency, so as to facilitate subsequent processing and analysis, and the specific contents include:

[0064] The data preprocessing operations mentioned above include: downsampling operation, straight-through filtering operation, radius filtering operation, statistical filtering operation, and conditional filtering operation; users can flexibly select one of the preprocessing operations or a combination of operations according to the point cloud data processing requirements, that is, the specific application requirements, to achieve the best data processing effect;

[0065] 1) Downsampling: Downsampling is to reduce the number of points in the point cloud to reduce the computational complexity and storage space and speed up the subsequent processing;

[0066] 2) Pass-Through Filtering: Pass-Through filtering is performed in the X-axis, Y-axis and Z-axis directions, that is, the point cloud is filtered according to the coordinate value of the point (usually the height in the Z-axis direction), and the points below or above the specified range are removed, thereby eliminating irrelevant interference background;

[0067] 3) Radius Outlier Removal: Within a given radius, if the number of points around a point is less than the set threshold, the point is considered an outlier and removed. This method can effectively remove outliers and interference points and is often used for point cloud cleaning;

[0068] 4) Statistical Outlier Removal: Statistical filtering calculates the mean and standard deviation of the number of points in the neighborhood of each point, and then removes those points that deviate from the mean by a certain standard deviation. It can identify and remove outliers based on the statistical characteristics of the local neighborhood.

[0069] 5) Conditional Filtering: It is a more flexible filtering method that allows users to set conditions to filter points based on various attributes of the point cloud (such as color, intensity, curvature, etc.). For example, you can retain or remove specific types of points by setting a color range or intensity threshold. It is suitable for scenarios that require data cleaning or feature extraction based on complex rules.

[0070] Preferably, in the aforementioned step 3, the pre-processed point cloud data is converted into a 2D image, i.e., a two-dimensional image; the specific contents of the aforementioned step 3 are:

[0071] Step 31, project the preprocessed point cloud data onto the XOY plane and obtain the extreme value of the plane where the two-dimensional point cloud is located; specifically, the preprocessed cloud data obtained in step 2 is projected onto the xoy plane, and the extreme value of the plane where the two-dimensional point cloud is located is obtained, that is, the maximum and minimum values ​​of these points in the X and Y directions are found, and each extreme point is represented by x_max, x_min, y_max, y_min, and the coordinate range of the two-dimensional image is determined, so as to realize the conversion of each three-dimensional point (x, y, z) into a two-dimensional point (x, y);

[0072] Step 32, calculate the Z-axis height difference of the pre-processed point cloud data in the Z-axis direction and normalize it, so that the Z coordinate value is mapped to the range of 0-1; specifically, the maximum and minimum values ​​of the point cloud data in the Z-axis direction are obtained to determine the height difference on the Z axis, and the Z coordinates of all points are normalized by using this height difference so that the Z coordinate value is mapped to the range of 0-1;

[0073] Step 33, converting the normalized Z coordinate value into a pixel value to obtain a 2D image; specifically, linearly mapping the normalized Z coordinate value into a pixel value range of 0-255 and assigning a value to adapt to the grayscale value representation of the image pixel.

[0074] The human-computer interaction interface in the aforementioned step 4 is programmed in a programming language to form a corresponding interface program, and the aforementioned interface program can display the 2D image and obtain user-defined points based on an interaction mechanism given in the interface program; the user obtains user-defined key point information by clicking the mouse or touching the display screen; the programming language includes but is not limited to any one of HTML, CSS, JavaScript, QT, and C# languages.

[0075] Preferably, in the aforementioned step 5, the ROI area corresponding to the aforementioned 2D image is obtained, and the ROI area is determined by customizing key points. The shape of the ROI area is not specifically limited, and a circle, a rectangle, a polygon, etc. can all be used as the ROI area in the present invention; and the number of ROI areas can be increased or deleted according to actual point requirements; the specific steps of obtaining the ROI area corresponding to the aforementioned 2D image include: through the interface program, by clicking the mouse or touching the display screen, click on the 2D image displayed on the human-computer interaction interface to obtain the ROI (region of interest) area of ​​the clicked point, i.e., the key point; that is, by clicking the 2D image displayed in step 4 with the mouse, the (x i ,y i ) coordinates, and draw the ROI area with the point as the center, w as the width, and h as the height. The number of ROI areas can be increased or deleted according to the actual point requirements.

[0076] The design principles and steps of the interaction mechanism given in the above interface program are:

[0077] Step 51, creating a user interface Web application for 2D image interaction and display, such as building a Web application through HTML or JavaScript, and adding necessary buttons and controls in the Web application to complete the tasks of managing the addition, editing and deletion of ROI areas;

[0078] Step 52: Set an event listener that matches the 2D image interface. The event listener is used to capture the selected event when the user selects a certain position of the 2D image and obtain the coordinates of the selected image position, and save its coordinate information in the rois array;

[0079] Step 53: Draw the ROI region on the 2D image according to the coordinates of the aforementioned image position and the preset drawing rules, and store the aforementioned ROI region data in an ROI data structure, such as an ROI list or an ROI array, such as drawing a rectangular frame as the ROI region according to a given width w and height h, or drawing a circular frame as the ROI region according to a given radius; and displaying multiple ROI regions in different colors; the aforementioned ROI region data includes the center coordinates (x i ,y i ) and information such as width w and height h;

[0080] Preferably, the specific steps of determining the point cloud fitting plane according to the ROI area in the aforementioned step 6 include:

[0081] Step 61: According to the plurality of ROI regions obtained above, each of the ROI regions is mapped to a corresponding 3D point cloud data, so as to fit the point cloud of each of the ROI regions into a plurality of planes;

[0082] Step 62, filter out the point cloud set with the most plane point cloud data as the plane finally fitted in the ROI area, the plane equation corresponding to the plane is expressed as Ax+By+Cz+D=0, and the corresponding plane normal vector is expressed as vector n(A, B, C).

[0083] In the preferred step 61, the plane fitting method can adopt the "random sampling consensus algorithm" or the "least squares method", and the appropriate method needs to be selected according to the quality of the point cloud data and the requirements of the application scenario. If there are many noises or outliers in the data, the "random sampling consensus algorithm" may be a better choice; if the data is relatively pure and fast and accurate fitting is sought, the least squares method is more appropriate.

[0084] Among them, the working steps of the random sampling consensus algorithm in the scene of point cloud plane fitting generally include: random sampling: randomly select a small number of points from the point cloud (usually 3 points, because the minimum number of points to define a plane); model fitting: use these points to fit a plane model; inlier judgment: calculate the distance from other points to the fitting plane, and the points within a certain threshold are considered to be inliers; counting and evaluation: if the number of inliers exceeds a certain threshold, it is considered that a better model has been found. Otherwise, return to step 1 and continue to try. At the same time, after multiple rounds of iterations, the model with the most inliers is selected as the final plane model. Among them, the least squares method aims to find the best fitting model by minimizing the sum of squared errors. The working steps in the scene of point cloud plane fitting generally include: selecting an initial plane hypothesis or a set of initial values; calculating the distance from all points to the plane, and constructing an error function based on these distances. By solving the partial derivatives of this error function with respect to the plane parameters and setting it equal to zero, find the parameter value that minimizes the error.

[0085] Preferably, the specific steps of obtaining the robot movement posture data according to the plane normal vector determined by the point cloud fitting plane and the origin normal vector of the robot tool coordinate system in the aforementioned step 7 include:

[0086] Step 71, obtaining the plane normal vector and the robot tool coordinate system origin normal vector;

[0087] Step 72, obtain the rotation angle and the center vector according to the aforementioned plane normal vector and the normal vector of the robot tool coordinate system origin, wherein the normal vector m of the robot workpiece coordinate system origin is set, and the rotation angle θ is calculated according to the dot product (inner product) of the aforementioned plane normal vector n and the normal vector m of the robot tool coordinate system origin, that is, by n×m=|n|×|m|×cosθ, the angle θ between vectors n and m can be obtained, n×m represents the dot product of vectors n and m, |n|, |m| are the modulus lengths of vectors n and m respectively; calculate the center vector u according to the cross product (outer product) between the aforementioned plane normal vector n and the normal vector m of the robot tool coordinate system origin, and the center vector u is perpendicular to the normal vector n and the normal vector m;

[0088] Step 73, taking the center vector u as the rotation center axis, rotating at an angle of θ, and obtaining the corresponding rotation vector w, w=θu; that is, the rotation vector w refers to a rotation of θ radians in the direction of the rotation center axis u;

[0089] Step 74, the quaternion index mapping formula is obtained by expanding the Euler formula, and then the robot movement posture data is obtained and the control instructions of the corresponding robot motion trajectory are generated, so that the robot can reach the target posture. The quaternion index mapping here is a conversion method from a rotation vector (also called an axis-angle representation) to a quaternion representation, which avoids the singularity problem caused by using Euler angles in robot control and can handle continuous rotation changes more naturally.

[0090] The specific process is: First, obtain the real part of the quaternion according to the determined rotation vector w, that is, obtain the real part of the quaternion according to the cosine part of the Euler formula Secondly, obtain the imaginary part of the quaternion, that is, obtain the imaginary part of the quaternion according to the sin part of Euler's formula The expression of the quaternion obtained by combining the imaginary part and the real part is

[0091]

[0092] Step 8: Obtain the robot motion trajectory according to the robot movement posture data. Connect all robot movement posture data in series to form a continuous motion trajectory.

[0093] At the same time, the present invention also proposes a user-defined key point path planning system, which is characterized by comprising:

[0094] An acquisition unit capable of acquiring original point cloud data of a target object;

[0095] A data preprocessing unit, which can perform a data preprocessing operation on the aforementioned original point cloud data to obtain preprocessed point cloud data;

[0096] An image conversion unit, which can convert the pre-processed point cloud data into a 2D image, thereby obtaining the two-dimensional coordinates corresponding to the point cloud data;

[0097] An image display unit, which can provide a display interface to display the aforementioned 2D image and obtain customized key target points;

[0098] A ROI region definition unit, which can obtain a number of ROI regions corresponding to the aforementioned 2D image, and the ROI regions are determined by custom key points;

[0099] A point cloud fitting plane processing unit, which can determine a point cloud fitting plane based on the aforementioned ROI area;

[0100] The robot motion trajectory acquisition unit can obtain the robot movement posture data according to the plane normal vector determined by the point cloud fitting plane and the robot tool coordinate system origin normal vector; and then obtain the robot motion trajectory according to the robot movement posture data.

[0101] According to the above technical solutions, the present invention has greater flexibility, strong practicality and broad application prospects.

[0102] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A path planning method for custom key points, characterized in that: include: Step 1: Obtain the original point cloud data of the target object; Step 2: performing data preprocessing on the original point cloud data to obtain preprocessed point cloud data; Step 3: Convert the pre-processed point cloud data into a 2D image, thereby obtaining the two-dimensional coordinates corresponding to the point cloud data; Step 4: displaying the aforementioned 2D image through a human-computer interaction interface; Step 5: Select the target key points on the aforementioned 2D image to obtain the corresponding ROI area; Step 6: Determine the point cloud fitting plane according to the aforementioned ROI area; Step 7: Obtain the robot movement posture data according to the plane normal vector determined by the point cloud fitting plane and the normal vector of the origin of the robot tool coordinate system; Step 8: Obtain the robot motion trajectory based on the robot movement posture data.

2. The path planning method for custom key points according to claim 1, characterized in that: The data preprocessing operations in the aforementioned step 2 include: downsampling operation, pass-through filtering operation, radius filtering operation, statistical filtering operation, and conditional filtering operation; the user can flexibly select one of the preprocessing operations or a combination of operations according to the point cloud data processing requirements, that is, the specific application requirements.

3. The path planning method for custom key points according to claim 1, characterized in that: The specific steps of converting the pre-processed point cloud data into a 2D image, i.e., a two-dimensional image, in the aforementioned step 3 are as follows: Step 31, project the preprocessed point cloud data onto the XOY plane and obtain the extreme values ​​of the plane where the two-dimensional point cloud is located. Each extreme point is represented by x_max, x_min, y_max, y_min, and then the coordinate range of the two-dimensional image is determined, so as to realize the conversion of each three-dimensional point (x, y, z) into a two-dimensional point (x, y); Step 32, calculating the Z-axis height difference of the preprocessed point cloud data in the Z-axis direction and normalizing it so that the Z coordinate value is mapped to the range of 0-1; Step 33: Convert the normalized Z coordinate value into a pixel value to obtain a 2D image.

4. The path planning method for custom key points according to claim 1, characterized in that: The human-computer interaction interface in the aforementioned step 4 is programmed in a programming language to form a corresponding interface program, and the aforementioned interface program can display the 2D image and obtain key point data selected by the user; the programming language includes but is not limited to any one of HTML, CSS, JavaScript, QT, and C#.

5. The path planning method for custom key points according to claim 4, characterized in that: The specific steps of obtaining the ROI area corresponding to the 2D image in the above step 5 include: The key points selected by the user are obtained through the interface program, and the ROI area corresponding to each key point is obtained; the interface coordinates of the key point are (x, y), and the ROI area is drawn with the key point as the center, w as the width, and h as the height.

6. The path planning method for custom key points according to claim 1, characterized in that: The specific steps of determining the point cloud fitting plane according to the ROI area in the aforementioned step 6 include: Step 61: According to the plurality of ROI regions obtained above, each of the ROI regions is mapped to a corresponding 3D point cloud data, so as to fit the point cloud of each of the ROI regions into a plurality of planes; Step 62: Filter out the point cloud set with the most plane point cloud data as the plane finally fitted in the ROI area.

7. A path planning method for custom key points according to claim 6, characterized in that: The plane fitting method in step 61 adopts a random sampling consensus algorithm or a least squares method, which needs to be selected according to the quality of the point cloud data and the requirements of the application scenario.

8. The path planning method for custom key points according to claim 6, characterized in that: In the above step 7, the specific steps of obtaining the robot movement posture data according to the plane normal vector determined by the point cloud fitting plane and the origin normal vector of the robot tool coordinate system include: Step 71, obtaining the plane normal vector and the robot tool coordinate system origin normal vector; Step 72, obtain the rotation angle and the center vector according to the aforementioned plane normal vector and the normal vector of the robot tool coordinate system origin, wherein the normal vector m of the robot workpiece coordinate system origin is set, and the rotation angle θ is calculated according to the dot product of the aforementioned plane normal vector n and the normal vector m of the robot tool coordinate system origin, that is, by n×m=|n|×|m|×cosθ, the angle θ between vectors n and m is obtained, n×m represents the dot product of vectors n and m, |n|, |m| are the modulus lengths of vectors n and m respectively; calculate the center vector u according to the cross product between the aforementioned plane normal vector n and the normal vector m of the robot tool coordinate system origin, and the center vector u is perpendicular to the normal vector n and the normal vector m; Step 73, taking the center vector u as the rotation center axis, rotating at an angle of θ, and obtaining the corresponding rotation vector w, w=θu; that is, the rotation vector w refers to a rotation of θ radians in the direction of the rotation center axis u; Step 74: Obtain the required movement posture data of the robot according to the quaternion index mapping formula.

9. A path planning system for custom key points constructed according to the path planning method for custom key points according to any one of claims 1 to 8, characterized in that: include: An acquisition unit capable of acquiring original point cloud data of a target object; A data preprocessing unit, which can perform a data preprocessing operation on the aforementioned original point cloud data to obtain preprocessed point cloud data; An image conversion unit, which can convert the pre-processed point cloud data into a 2D image, thereby obtaining the two-dimensional coordinates corresponding to the point cloud data; An image display unit, which can provide a display interface to display the aforementioned 2D image and obtain customized key target points; A ROI region definition unit, which can obtain a number of ROI regions corresponding to the aforementioned 2D image, and the ROI regions are determined by custom key points; A point cloud fitting plane processing unit, which can determine a point cloud fitting plane based on the aforementioned ROI area; The robot motion trajectory acquisition unit can obtain the robot movement posture data according to the plane normal vector determined by the point cloud fitting plane and the robot tool coordinate system origin normal vector; and then obtain the robot motion trajectory according to the robot movement posture data.