Online path planning method and system for automatic polishing robot

Abstract

The application provides an automatic polishing robot online path planning method and system, a complex curved surface is locally divided into a simple curved surface through one-to-one mapping relationship of a plane and a curved surface established by orthographic projection, algorithm time complexity of searching model surface path point coordinates is reduced, operation speed is improved to meet real-time path planning requirements, path planning can be directly performed on point cloud data, accuracy of a reconstruction algorithm and model quality are low in requirements, three-dimensional path planning can be converted into two-dimensional plane path planning, a two-dimensional path in an arbitrary shape is generated, and automatic and efficient online path planning of a robot is ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of surface processing, in particular to an online path planning method and system of an automatic polishing robot. BACKGROUND

[0002] In the manufacturing industry, the automation technology of workpiece surface processing is a problem to be solved. In automobile manufacturing, only the coating detection and polishing process still need to be completed manually. The difficulty of realizing the automatic process of the above two processes lies in how to obtain the accurate three-dimensional coordinates of the workpiece surface and generate coordinate sequences according to the processing requirements, and guide the robot to move according to the coordinate sequences to realize the path planning of the workpiece surface. In the automatic polishing process, the robot needs to traverse the polishing area along a non-repeating path, that is, to complete the coverage path planning (CPP) of the surface.

[0003] In the past, the off-line planning method was mainly used to construct a digital model of the workpiece by using a three-dimensional modeling software, to calculate the polishing path on the idealized model surface by importing a surface analysis function, but there is a gap between the model and the actual workpiece, and the generated path needs to be manually adjusted, which cannot meet the precision and efficiency requirements of the polishing process. The line planning method uses a visual camera or a three-dimensional scanning device to reconstruct a three-dimensional model of the workpiece, and automatically generates a coordinate sequence of the processing path on the surface digital model represented by a point cloud or a grid. Due to the unordered nature of the three-dimensional point cloud, the point cloud index is irrelevant to the coordinates, and the positioning of each path point coordinate needs to traverse all the vertices. Moreover, the surface processing path needs to be close to the surface, not only to judge the distance between the path and the surface and the concave-convex nature of the surface, but also to consider the relative attitude change of the model, so it is difficult and challenging to automatically plan the path on the point cloud model surface. The existing online path planning method mainly uses surface fitting, optimization search and two-dimensional coordinate UV expansion, which takes a long time and is difficult to meet the requirements of real-time and accurate planning. SUMMARY

[0004] The purpose of the present application is to provide an automatic polishing robot online path planning method and system, obtain workpiece point cloud data, form a workpiece point cloud model, segment and extract a plurality of rectangular regions from the workpiece point cloud model, divide all rectangular regions into a grid map according to robot polishing path parameters, create rules to generate a two-dimensional path that traverses the grid map, collect and analyze workpiece surface images to obtain the boundary of the workpiece surface area to be polished, adjust path points of the two-dimensional path according to the boundary of the workpiece surface area to be polished, project the two-dimensional path on the point cloud surface to obtain a three-dimensional path, optimize path points of the three-dimensional path to obtain the actual polishing path of the robot on the workpiece surface, establish a one-to-one mapping relationship between the plane and the curved surface through orthogonal projection to locally segment the complex curved surface into a simple curved surface, reduce the algorithm time complexity of searching for model surface path point coordinates, improve the running speed to meet the real-time path planning requirements, can directly plan the path on the point cloud data, has low requirements for the reconstruction algorithm accuracy and model quality, can convert three-dimensional path planning into two-dimensional plane path planning, generate a two-dimensional path of any shape, and ensure automatic and efficient online path planning of the robot.

[0005] The present application is implemented by the following technical solutions:

[0006] The automatic polishing robot online path planning method comprises:

[0007] Obtain workpiece point cloud data to form a workpiece point cloud model, segment and extract a plurality of rectangular regions from the workpiece point cloud model,

[0008] Divide all rectangular regions into a grid map according to robot polishing path parameters, create rules to generate a two-dimensional path that traverses the grid map,

[0009] Collect and analyze workpiece surface images to obtain the boundary of the workpiece surface area to be polished, adjust path points of the two-dimensional path according to the boundary of the workpiece surface area to be polished,

[0010] Project the two-dimensional path on the point cloud surface to obtain a three-dimensional path, optimize path points of the three-dimensional path to obtain the actual polishing path of the robot on the workpiece surface.

[0011] Optionally, obtaining workpiece point cloud data to form a workpiece point cloud model, segmenting and extracting a plurality of rectangular regions from the workpiece point cloud model comprises:

[0012] Perform depth scanning and shooting on the workpiece to obtain point cloud data of the workpiece, perform abnormal data screening and elimination preprocessing on the point cloud data, and model the workpiece using the point cloud data to form a workpiece point cloud model,

[0013] The workpiece point cloud model is subjected to normal clustering segmentation to obtain a plurality of curved surface regions; and principal component analysis is used to fit each curved surface region as a rectangular region on a projection plane.

[0014] Optionally, all the rectangular regions are divided into a grid map according to the robot polishing path parameters; and a rule is created to generate a two-dimensional path traversing the grid map, including:

[0015] All the rectangular regions are divided into a grid map according to the step size of the robot polishing path, and each grid in the grid map adopts two-dimensional coordinates (u, v) indexing;

[0016] A zigzag or H-shaped path rule is created, and a two-dimensional path traversing the grid map is generated according to the path rule; the two-dimensional path is saved in the form of two arrays that can query each other, one of which is a one-dimensional array of path point coordinates saved in path order, and the other is a two-dimensional array of serial numbers in the path saved in coordinates.

[0017] Optionally, a workpiece surface image is collected and analyzed to obtain a boundary of a region to be polished on the workpiece surface; and path point adjustment is performed on the two-dimensional path according to the boundary of the region to be polished, including:

[0018] A workpiece surface image is collected with the normal direction of the rectangular region as the shooting optical axis, and image segmentation, contour extraction and convex polygon fitting are performed on the workpiece surface image to obtain the boundary of the region to be polished on the workpiece surface;

[0019] The contour line coordinates of the boundary of the region to be polished are normalized and transformed to the coordinate system of the grid map, and the path points in the two-dimensional path located outside the contour line are identified and filtered out, and all the path points of the two-dimensional path are reordered.

[0020] Optionally, the two-dimensional path is projected on the point cloud surface to obtain a three-dimensional path; and path point optimization is performed on the three-dimensional path to obtain an actual polishing path of the robot on the workpiece surface, including:

[0021] The two-dimensional path is projected on the point cloud surface to transform the path point coordinates of the two-dimensional path into three-dimensional coordinates, thereby obtaining a three-dimensional path;

[0022] The path point coordinates of the three-dimensional path are subjected to interpolation and / or point supplementing processing to obtain an actual polishing path of the robot on the workpiece surface.

[0023] An online path planning system of an automatic polishing robot, comprising:

[0024] A point cloud model construction module is configured to acquire workpiece point cloud data and form a workpiece point cloud model;

[0025] a rectangular region segmentation module configured to segment and extract a plurality of rectangular regions from the workpiece point cloud model;

[0026] a grid map generation module configured to divide all the rectangular regions into a grid map according to robot polishing path parameters;

[0027] a two-dimensional path generation module configured to create rules to generate a two-dimensional path traversing the grid map;

[0028] a boundary determination module configured to collect and analyze workpiece surface images to obtain a boundary of a region to be polished on the workpiece surface;

[0029] a two-dimensional path adjustment module configured to adjust path points of the two-dimensional path according to the boundary of the region to be polished;

[0030] a three-dimensional path generation and optimization module configured to project the two-dimensional path on a point cloud surface to obtain a three-dimensional path, and to optimize path points of the three-dimensional path to obtain an actual polishing path of the robot on the workpiece surface.

[0031] Optionally, the point cloud model construction module is configured to obtain workpiece point cloud data to form a workpiece point cloud model, including:

[0032] performing depth scanning and shooting on a workpiece to obtain point cloud data of the workpiece, performing abnormal data screening and elimination preprocessing on the point cloud data, and modeling the workpiece by using the point cloud data to form a workpiece point cloud model;

[0033] the rectangular region segmentation module is configured to segment and extract a plurality of rectangular regions from the workpiece point cloud model, including:

[0034] performing normal clustering segmentation on the workpiece point cloud model to obtain a plurality of curved surface regions, and fitting each curved surface region as a rectangular region on a projection plane by using principal component analysis.

[0035] Optionally, the grid map generation module is configured to divide all the rectangular regions into a grid map according to robot polishing path parameters, including:

[0036] dividing the grid map according to a step size of the robot polishing path, and using two-dimensional coordinates (u, v) to index each grid in the grid map;

[0037] the two-dimensional path generation module is configured to create rules to generate a two-dimensional path traversing the grid map, including:

[0038] Create a zigzag or zigzag path rule, and generate a two-dimensional path that traverses the grid map according to the path rule. The two-dimensional path is stored in two arrays that can query each other. One array is a one-dimensional array of path point coordinates stored with the path order as the index, and the other array is a two-dimensional array of the sequence number in the path stored with the coordinate as the index.

[0039] Optionally, the boundary determination module is used to acquire and analyze workpiece surface images to obtain the boundary of the area to be polished on the workpiece surface, including:

[0040] The workpiece surface image is acquired with the normal direction of the rectangular region as the shooting optical axis. The workpiece surface image is then segmented, contour extracted, and convex polygon fitted to obtain the boundary of the area to be polished on the workpiece surface.

[0041] The two-dimensional path adjustment module is used to adjust the path points of the two-dimensional path according to the boundary of the area to be polished, including:

[0042] The coordinates of the outline of the area to be polished are normalized and transformed to the coordinate system of the grid map. Path points located outside the outline in the two-dimensional path are identified and filtered out. Then, all path points in the two-dimensional path are reordered.

[0043] Optionally, the 3D path generation and optimization module is used to project the 2D path onto a point cloud surface to obtain a 3D path; and to optimize the 3D path by optimizing the path points to obtain the actual grinding path of the robot on the workpiece surface, including:

[0044] The two-dimensional path is projected onto the point cloud surface, thereby transforming the path point coordinates of the two-dimensional path into three-dimensional coordinates, thus obtaining the three-dimensional path.

[0045] Interpolation and / or point supplementation are performed on the coordinates of the path points of the three-dimensional path to obtain the actual grinding path of the robot on the surface of the workpiece.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] The online path planning method and system of the automatic polishing robot provided by the application have low algorithm time complexity for searching path point coordinates on a model surface, high running speed, and can meet the requirement of real-time path planning; the online path planning is simple to implement, can be directly performed on point cloud data, and discrete point clouds facilitate segmentation of complex models, thereby avoiding the problem of re-triangulation after segmentation of a network model; the online path planning is suitable for projecting any non-repeated two-dimensional path, can be combined with other two-dimensional plane map path planning algorithms, and can set path planning boundaries through visual identification of a polishing area and automatically implement path planning within the boundaries; the online path planning is not only suitable for automatic polishing, but also can be used for other surface processing technologies, and has great scene applicability. BRIEF DESCRIPTION OF DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description only constitute some embodiments of the application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor. Among them:

[0049] Figure 1 FIG. 1 is a flowchart of the online path planning method of the automatic polishing robot provided by the application.

[0050] Figure 2 FIG. 2 is a two-dimensional path of a grid map.

[0051] Figure 3 FIG. 3 is a real shape contour of a workpiece.

[0052] Figure 4 FIG. 4 is a contour line adjustment two-dimensional path of a polishing area boundary of a workpiece surface.

[0053] Figure 5 FIG. 5 is a start state diagram of projecting a two-dimensional path on a point cloud surface.

[0054] Figure 6 FIG. 6 is a final state diagram of projecting a two-dimensional path on a point cloud surface.

[0055] Figure 7 FIG. 7 is a structure diagram of the online path planning system of the automatic polishing robot provided by the application. DETAILED DESCRIPTION

[0056] In order to make the above objectives, characteristics and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application are described in detail below with reference to the drawings. It can be understood that the specific embodiments described herein are only used to explain the present application, but not to limit the present application. In addition, it should be noted that, for the purpose of description, only the parts related to the present application are shown in the drawings, but not all the structures. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the present application.

[0057] The terms "comprising" and "having" and any variations thereof in the present application are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units is not limited to the listed steps or units, but optionally further includes steps or units not listed, or optionally further includes other steps or units inherent to these processes, methods, products or devices.

[0058] In this paper, the "embodiment" means that the specific features, structures or characteristics described in conjunction with the embodiment can be included in at least one embodiment of the present application. The phrase appears at various places in the specification does not necessarily refer to the same embodiment, nor is it independent or alternative to other embodiments. Those skilled in the art explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0059] Please refer to Figure 1 An embodiment of the present application provides an automatic polishing robot online path planning method. The automatic polishing robot online path planning method comprises:

[0060] Obtain workpiece point cloud data to form a workpiece point cloud model; segment and extract a plurality of rectangular regions from the workpiece point cloud model;

[0061] According to the robot polishing path parameters, all rectangular regions are divided into a grid map; a rule is created to generate a two-dimensional path traversing the grid map;

[0062] Collect and analyze the workpiece surface image to obtain the boundary of the workpiece surface to be polished; adjust the path points of the two-dimensional path according to the boundary of the workpiece surface to be polished;

[0063] Project the two-dimensional path on the point cloud surface to obtain a three-dimensional path; optimize the path points of the three-dimensional path to obtain the actual polishing path of the robot on the workpiece surface.

[0064] The automatic polishing robot online path planning method of the above embodiment reduces the algorithm time complexity of searching the model surface path point coordinates, improves the running speed to meet the real-time path planning requirements, can directly plan the path on the point cloud data, has low requirements on the reconstruction algorithm accuracy and model quality, can convert three-dimensional path planning into two-dimensional plane path planning, generate a two-dimensional path of any shape, and ensure automatic and efficient online path planning of the robot.

[0065] In another embodiment, workpiece point cloud data is acquired to form a workpiece point cloud model; the workpiece point cloud model is segmented to extract a plurality of rectangular regions, including:

[0066] The workpiece is subjected to depth scanning and shooting to obtain point cloud data of the workpiece; the point cloud data is subjected to abnormal data screening and elimination preprocessing; the workpiece is modeled using the point cloud data to form a workpiece point cloud model;

[0067] The workpiece point cloud model is subjected to normal clustering segmentation to obtain a plurality of curved surface regions; principal component analysis is used to fit each curved surface region into a rectangular region on a projection plane.

[0068] In actual operation, the workpiece (especially the workpiece surface) can be subjected to depth scanning and shooting using a depth camera to obtain point cloud data of the workpiece in the global range. Considering that the point cloud data collected during depth scanning and shooting is inevitably affected by internal factors of the depth camera itself and external environmental factors, there are abnormal data such as error data in the point cloud data. In order to ensure the correctness of the point cloud data, the point cloud data is subjected to abnormal data screening and elimination preprocessing to eliminate error data such as excessively large or small data in the point cloud data. Then, the workpiece is modeled using the point cloud data to obtain a workpiece point cloud model, thereby realizing three-dimensional shape representation of the workpiece. The workpiece point cloud model is further segmented into a plurality of curved surface regions according to direction clustering, and the surface curvature of each curved surface region changes little, with a maximum normal angle less than 90 degrees. Then, the principal component analysis (PCA) method is used to fit each curved surface region into a rectangular region with the maximum area on a projection plane S, and a two-dimensional plane coordinate system is established with the projection plane S as the XY plane to transform the point cloud coordinates corresponding to the point cloud data into the above coordinate system. In the specific transformation process, the coordinate value x and the coordinate value y of all point cloud coordinates (x, y, z) remain unchanged, and the coordinate value z of all point cloud coordinates (x, y, z) is zero, thereby realizing transformation of the point cloud coordinates from three-dimensional to two-dimensional.

[0069] In another embodiment, according to the robot polishing path parameters, all rectangular regions are divided into a grid map; a rule is created to generate a two-dimensional path traversing the grid map, including:

[0070] Divide all rectangular areas into a grid map according to the step size of the robot grinding path, and each grid in the grid map is indexed by two-dimensional coordinates (u, v);

[0071] Create a "zigzag" or "I-shaped" path rule. According to the path rule, generate a two-dimensional path that traverses the grid map; the storage format of the two-dimensional path is two arrays that can query each other. One array is a one-dimensional array of path point coordinates saved with the path sequence as the index, and the other array is a two-dimensional array of the sequence numbers saved in the path with the coordinates as the index.

[0072] In actual operation, please refer to Figure 2 , divide all rectangular areas into a grid map according to the step size grid_step of each movement during the movement of the robot along the grinding path (for example, corresponding to Figure 2 the side length of the square in it is 0.2), and each grid in the grid map is indexed and calibrated by two-dimensional coordinates (u, v). Then create a "zigzag" or "I-shaped" path rule. Specifically, Figure 2 for the "I-shaped" path rule, according to the above path planning, generate a two-dimensional path that traverses the entire grid map (such as Figure 2 the blue two-dimensional path in). Also, set the storage format of the above two-dimensional path to two arrays that can query each other. One array is a one-dimensional array of path point coordinates (u, v) saved with the path sequence as the index Point(index)=(u, v), and the other array is a two-dimensional array of the sequence numbers saved in the path with the coordinates (u, v) as the index Index(u, v)=index. If a certain coordinate is not in the path, then index=-1. The above storage format of the two-dimensional path can be used to describe two-dimensional paths of any shape, realizing the applicability and compatibility to different workpiece surface shape scenarios.

[0073] In another embodiment, collect and analyze the workpiece surface image to obtain the boundary of the area to be polished on the workpiece surface; according to the boundary of the area to be polished, adjust the path points of the two-dimensional path, including:

[0074] Collect the workpiece surface image with the normal direction of the rectangular area as the shooting optical axis, perform image segmentation, contour extraction and convex polygon fitting on the workpiece surface image to obtain the boundary of the area to be polished on the workpiece surface;

[0075] Normalize and transform the contour line coordinates of the boundary of the area to be polished to the coordinate system where the grid map is located, identify and filter out the path points outside the contour line in the two-dimensional path, and then reorder all the path points of the two-dimensional path.

[0076] In actual operation, the high-definition camera can be used to capture the workpiece surface image in the normal direction of the rectangular region (i.e., the axis direction of the high-definition camera is parallel to the normal direction of the rectangular region), and the workpiece surface image is subjected to image segmentation, contour extraction and convex polygon fitting to obtain the boundary of the workpiece surface area to be polished. Please refer to Figure 3 and Figure 4 , which are the contour line adjustment two-dimensional path of the real shape contour of the workpiece and the boundary of the workpiece surface area to be polished, respectively, wherein Figure 4 The contour line adjustment two-dimensional path of the real shape contour of the workpiece is determined based on Figure 3 . Specifically, the contour line coordinates of the area to be polished are normalized and transformed to the coordinate system of the grid map, so as to identify and filter out the path points in the two-dimensional path located outside the contour line, and then reorder all the path points of the two-dimensional path, wherein if the area to be polished cannot be identified, the rectangular region is directly used as the boundary, thereby obtaining the two-dimensional path as shown in Figure 4 , which provides a reliable basis for subsequent optimization of the three-dimensional path.

[0077] In another embodiment, the two-dimensional path is projected on the point cloud surface to obtain a three-dimensional path; the path points of the three-dimensional path are optimized to obtain the actual polishing path of the robot on the workpiece surface, including:

[0078] The two-dimensional path is projected on the point cloud surface, so as to transform the path point coordinates of the two-dimensional path into three-dimensional coordinates, thereby obtaining the three-dimensional path;

[0079] The path point coordinates of the three-dimensional path are subjected to interpolation and / or point supplementing processing to obtain the actual polishing path of the robot on the workpiece surface.

[0080] In actual operation, in order to avoid searching for coordinates in three-dimensional space, in fact, the point cloud is projected on a two-dimensional plane to search for the nearest path point. The aforementioned point cloud data has been transformed to the projection plane coordinate system, and only the Z coordinate of the point needs to be set to zero to obtain the projection point, as shown in Figure 5 and Figure 6The start state and the final state of projecting the two-dimensional path on the point cloud surface are shown respectively; in the start state, the two-dimensional path is aligned with the point cloud surface; in the final state, the two-dimensional path is completely projected on the point cloud surface. All vertices of the point cloud model are traversed, the (x, y) coordinates of each vertex are calculated to obtain the grid (u, v) coordinates of the vertex, the calculation method is (u, v) = (x / grid_step, y / grid_step), the path serial number index corresponding to the coordinates (u, v) is queried, and the three-dimensional coordinates of the vertex are replaced by the corresponding path point coordinates Point(index), that is, Point(index(x / grid_step, y / grid_step)) = (x, y, z), and the normal of the vertex is output, so as to obtain the three-dimensional path. Because there may be multiple vertices falling in the same grid, the vertex closest to the center of the grid is taken, and an array Dis(index) is needed to save the minimum distance between each path point and the vertex. The present application only needs to traverse all vertices once to find the mapping from two-dimensional coordinates to three-dimensional coordinates, assuming that the number of point cloud vertices is n, then the time complexity of the algorithm is O(n), and all vertices are calculated independently, and parallel computing is supported to accelerate, so as to realize real-time path planning.

[0081] There is also a situation that the local point density of the reconstructed point cloud model is low, which may cause some path points to be missing or the distance between the path points and the vertices to be too large, and the path point coordinates are optimized by interpolation and / or point filling. The coordinate optimization of the path point adopts triangular barycentric coordinate interpolation, searches for the triangular grid closest to the path projection point P, obtains the coordinates of the three vertices ABC, calculates the areas S ABC 、S PBC 、S PCA 、S PAB , the z value of the path point coordinate The calculation formula is , wherein 、 、 , respectively, the z value of the coordinates of points A, B and C. If the distance between two adjacent points of the path is large and the normal changes greatly, the polishing path may deviate from the workpiece surface, and a path point can be filled between the two points, and the coordinates of the filled path point are determined by the intersection point of the tangent lines of the two points along the path direction.

[0082] In one embodiment, before performing the step of "obtaining workpiece point cloud data and forming a workpiece point cloud model", the method can further include the following steps:

[0083] Step A1, scanning the workpiece by using a laser scanning shooting method, collecting the scanned information, and obtaining the original acquisition signal of the workpiece.

[0084] Step A2, based on the improved Lambertian reflection model, separates and quantifies the diffuse reflection and specular reflection components of the workpiece material, and corrects the original collected signal by normalization processing to obtain the corrected reflection signal intensity, thereby eliminating the interference of scanning conditions.

[0085] In the formula (1), the corrected reflection signal intensity of each point on the workpiece being scanned is obtained.

[0086] (1-1)

[0087] In the formula (1), the corrected reflection signal intensity of each point on the workpiece being scanned is obtained. The corrected reflection signal intensity of the i-th point on the workpiece being scanned is represented by (dimension of intensity). The reference signal intensity (dimension of intensity) is a constant with clear physical meaning and intensity dimension, and its value can be an empirical value, such as the signal intensity value measured under standard conditions (specific distance, specific material standard plate, and specific light source power). The original reflection signal intensity of the i-th point is represented by (dimension of intensity). The maximum intensity value in the original emission signal intensity values corresponding to all points on the workpiece is represented by (dimension of intensity). The diffuse reflectivity of the material at the i-th point is represented by a coefficient between 0 and 1, and is dimensionless. The specular reflectivity of the material at the i-th point is represented by a coefficient between 0 and 1, and is dimensionless. The incident angle at the i-th point is represented by The observation angle at the i-th point is represented by The highlight index or glossiness coefficient of the material at the i-th point is represented by, and the larger the value, the smoother the material. The diffuse reflectivity of the material at the j-th point is represented by The specular reflectivity of the material at the j-th point is represented by The incident angle at the j-th point is represented by The observation angle at the j-th point is represented by The highlight index or glossiness coefficient of the material at the j-th point is represented by The set of all points on the workpiece surface is represented by

[0088] The principle of the above formula (1-1) is as follows:

[0089] The present application is an improved Lambertian and non-Lambertian mixed reflection model, and its core principle is to eliminate the interference of measurement conditions by multi-level normalization and reference scaling, and to realize adaptive signal correction of the material. The formula includes the following processing:

[0090] The first level normalization is realized by , which aims to eliminate the absolute intensity influence. Its principle is: the original signal intensity of the i-th point is divided by the maximum intensity of the whole field , which is a dimensionless number between [0, 1], which eliminates the overall intensity scaling effect caused by light source power, global distance, etc.

[0091] The second level normalization is realized by , which aims to eliminate the material reflection characteristics and local geometry influence. Its principle is:

[0092] The numerator in this term is the theoretical reflection model of the i-th point, which comprehensively considers diffuse reflection and specular reflection, and comprehensively represents the theoretical relative reflection ability of the point under the material and geometric posture.

[0093] The denominator in this term calculates the maximum value of the theoretical reflection model of all points in the whole field. The purpose of dividing the numerator by the maximum value is to normalize the local reflectivity difference caused by different materials (such as high-gloss metal and matte plastic) and different geometric postures (such as front and side), and finally obtain a material and geometric correction coefficient between (0, 1].

[0094] Finally, the reference scaling is realized by the product of , the calculation results of the above 1, 2, and 3, so that the final output value of the formula is the corrected reflection signal intensity of the correct i-th point.

[0095] The role of the above formula (1) is: 1, realize material adaptive correction: through the diffuse reflection coefficient, the specular reflection coefficient and the highlight index, adapt to complex materials such as reflection, transparency and light absorption; 2, realize signal normalization: eliminate scanning distance, light source intensity influence; 3, realize noise suppression precursor: provide weight for subsequent formulas (2-1), (2-2), (2-3), and improve the subsequent filtering accuracy.

[0096] Alternatively, the above formula (1) can also be expressed as the following form:

[0097] (1-2)

[0098] Wherein, Ni represents the unit normal vector at the i-th point (located at the i-th point on the surface of the object, and perpendicular to the unit vector of the object surface at the point); Li represents the unit incident light vector of the i-th point pointing to the light source; Vi represents the unit vector from the i-th point to the camera or sensor (defines the direction of the sensor observing the point); Ri represents the unit specular reflection vector at the i-th point (according to the law of reflection of light, the unit vector of the ideal reflection direction calculated from the incident light vector Li and the normal vector Ni, for example, its calculation formula can be​ Nj represents the unit normal vector at point j (a unit vector located at the j-th point on the object's surface and perpendicular to that point); Lj represents the unit incident light vector from the j-th point to the light source; Vj represents the unit vector from the j-th point to the camera or sensor (defining the direction in which the sensor observes that point); Rj represents the unit specular reflection vector at point j (a unit vector representing the ideal reflection direction calculated from the incident light vector Lj and the normal vector Nj according to the law of reflection).

[0099] Step A3: Perform weighted neighborhood adaptive filtering based on the corrected reflected signal intensity to suppress noise and obtain the filtered point cloud data of the workpiece.

[0100] Specifically, step A3 can be implemented as follows:

[0101] (2-1)

[0102] (2-1)

[0103] (2-3)

[0104] in, (2-4)

[0105] in, , , This represents the filtered coordinates of the i-th point. The x, y, and z coordinates of the i-th point; N is the number of neighboring points of the i-th point, and N is a preset positive integer, for example, N can be 20; Let j be the weight of the j-th neighboring point of the i-th point. , , The original coordinates of the j-th neighboring point of the i-th point. x, y, z coordinates Let be the mean of the corrected reflected signal intensity of all points in the neighborhood of the i-th point. Let be the standard deviation of the corrected reflected signal intensity of all points in the neighborhood of the i-th point.

[0106] The principle of the above formulas (2-1), (2-2), and (2-3) is: to calculate the weights based on the results output by formula (1-1) or (1-2) and to perform a dynamic weighted average on the neighborhood points;

[0107] The function has the following effects: 1. accurate noise suppression, and point cloud signal-to-noise ratio improvement; 2. feature preservation: through weight differentiation, edge blurring caused by traditional filtering is avoided, such as gear tooth profile, thin-walled structure, etc.; 3. data purification: output filtered point cloud, providing high-quality geometric data for subsequent formulas.

[0108] Step A4, for the missing area of the point cloud data of the workpiece, such as the blank area without valid coordinate data in the point cloud

[0109] , the missing point interpolation is performed according to the filtered point cloud data to fill the point cloud data of the missing area.

[0110] Among them, step A4 can be implemented as:

[0111] Step A41, determining the missing area of the point cloud data of the workpiece;

[0112] Specifically: 1. determine the distance threshold, which is R1 (R1 can be 3-5 times the global point cloud average distance, which is the average value of the distance between all points in the point cloud and their nearest neighbors); 2. Calculate the average distance of A (A can be 20-50) nearest neighbors around each point in the point cloud, and determine the size relationship between the average distance corresponding to each point and the distance threshold. The point with the corresponding average distance greater than the distance threshold is regarded as a potential missing boundary point; cluster the potential missing boundary points to form a closed contour, which is the contour of the missing area;

[0113] Step A42, taking the boundary of the missing area as a reference, expanding the ring-shaped area with a width of R2 times the global point cloud average distance (R2 can be 3-20) as the candidate neighborhood of the missing area;

[0114] Step A43, determine whether the point cloud density in the candidate neighborhood is greater than or equal to the global point cloud density (the point cloud density of other areas on the workpiece except the missing area); if yes, the candidate neighborhood is regarded as the neighborhood of the missing area, and step A44 is continued; if not, output a re-scanning prompt;

[0115] Step A44, uniformly select Q boundary points on the boundary of the missing area, Q is equal to or greater than 3;

[0116] Step A45, uniformly selecting M points in the neighborhood of the missing region as reference points; fitting a minimum three-dimensional bounding box of the missing region through the boundary points, generating three-dimensional grid points in the minimum three-dimensional bounding box at a preset interval, each grid point corresponding to a sampling point, determining the coordinates of the interpolation points in the missing region according to the distance from each sampling point to the Q boundary points, the distance from each reference point to the Q boundary points, the coordinates of each reference point, and the normal vector change rate corresponding to each reference point;

[0117] wherein the coordinates of the interpolation points in the missing region can be calculated by the following formula:

[0118] (3-1)

[0119] (3-2)

[0120] (3-3)

[0121] wherein, (3-4)

[0122] wherein, , , represents the x, y, z coordinate values (coordinate values in a three-dimensional coordinate system) of the interpolation point coordinates obtained by traversing the tth sampling point according to the above formula . , , represents the x, y, z coordinate values of the kth reference point coordinate . M represents the total number of reference points, which can be 20-50; represents the weight of the kth reference point, which represents the influence of the kth reference point on the current interpolation point , The greater the value, the greater the influence of the reference point on the final interpolation result. p represents the distance attenuation coefficient, which is greater than or equal to 1.

[0123] The functions of formula (3-1), (3-2), and (3-3) are: high-fidelity completion of point cloud missing region: intelligently generating three-dimensional coordinate points in the missing region according to the known point cloud around the missing region, effectively solving the data missing problem caused by material or occlusion, providing a complete and accurate three-dimensional model for path planning, and ensuring that the robot polishing path generated subsequently can closely fit the workpiece surface, thereby improving the polishing precision.

[0124] The steps A1-A4 solve the core problems of signal distortion, noise interference and data loss in the acquisition of point cloud data of the workpiece with complex material, provide an accurate and complete data basis for the online path planning of the automatic polishing robot, and improve the planning accuracy.

[0125] The value of p can be determined in advance according to the material of the workpiece, for example, as shown in Table 1:

[0126] Table 1

[0127]

[0128] Referring to Figure 7 The automatic polishing robot online path planning system provided by an embodiment of the application. The automatic polishing robot online path planning system comprises:

[0129] A point cloud model construction module configured to acquire workpiece point cloud data and form a workpiece point cloud model;

[0130] A rectangular region segmentation module configured to segment and extract a plurality of rectangular regions from the workpiece point cloud model;

[0131] A grid map generation module configured to divide all the rectangular regions into a grid map according to robot polishing path parameters;

[0132] A two-dimensional path generation module configured to create a rule to generate a two-dimensional path for traversing the grid map;

[0133] A boundary determination module configured to acquire and analyze a workpiece surface image to obtain a boundary of a workpiece surface region to be polished;

[0134] A two-dimensional path adjustment module configured to adjust path points of the two-dimensional path according to the boundary of the workpiece surface region to be polished;

[0135] A three-dimensional path generation and optimization module configured to project the two-dimensional path on a point cloud surface to obtain a three-dimensional path, and to optimize path points of the three-dimensional path to obtain an actual polishing path of the robot on the workpiece surface.

[0136] In another embodiment, the point cloud model construction module is configured to acquire workpiece point cloud data and form a workpiece point cloud model, and comprises:

[0137] The workpiece is subjected to depth scanning and shooting to obtain point cloud data of the workpiece, the point cloud data is subjected to abnormal data screening and elimination preprocessing, and the workpiece is modeled using the point cloud data to form a workpiece point cloud model;

[0138] The rectangular region segmentation module is configured to segment and extract a plurality of rectangular regions from the workpiece point cloud model, and comprises:

[0139] The normal clustering segmentation is performed on the workpiece point cloud model to obtain a plurality of surface regions; and principal component analysis is used to fit each surface region as a rectangular region on a projection plane.

[0140] In another embodiment, the grid map generation module is configured to divide all the rectangular regions into a grid map according to the robot polishing path parameters, including:

[0141] The grid map is divided according to the step size of the robot polishing path, and each grid in the grid map is indexed by two-dimensional coordinates (u, v);

[0142] The two-dimensional path generation module is configured to create a rule to generate a two-dimensional path traversing the grid map, including:

[0143] The zigzag or "work" type path rule is created, and the two-dimensional path traversing the grid map is generated according to the path rule; the two-dimensional path is saved in the form of two arrays that can query each other, one of which is a one-dimensional array of path point coordinates indexed by path order, and the other is a two-dimensional array of path sequence numbers indexed by coordinates.

[0144] In another embodiment, the boundary determination module is configured to collect and analyze the workpiece surface image to obtain the boundary of the workpiece surface to be polished, including:

[0145] The workpiece surface image is collected with the normal direction of the rectangular region as the shooting optical axis, and the workpiece surface image is segmented, the contour is extracted, and the convex polygon is fitted to obtain the boundary of the workpiece surface to be polished;

[0146] The two-dimensional path adjustment module is configured to adjust the path points of the two-dimensional path according to the boundary of the region to be polished, including:

[0147] The contour line coordinates of the region to be polished are normalized and transformed to the coordinate system of the grid map, the path points of the two-dimensional path located outside the contour line are identified and filtered out, and all the path points of the two-dimensional path are reordered.

[0148] In another embodiment, the three-dimensional path generation and optimization module is configured to project the two-dimensional path on the point cloud surface to obtain a three-dimensional path; and the path points of the three-dimensional path are optimized to obtain the actual polishing path of the robot on the workpiece surface, including:

[0149] The two-dimensional path is projected on the point cloud surface to transform the path point coordinates of the two-dimensional path into three-dimensional coordinates, thereby obtaining a three-dimensional path;

[0150] The path point coordinates of the three-dimensional path are interpolated and / or supplemented to obtain the actual polishing path of the robot on the workpiece surface.

[0151] The automatic polishing robot online path planning system of the present application is consistent with the operation and effect of the automatic polishing robot online path planning method described above. The automatic polishing robot online path planning system will not be repeated here.

[0152] Overall, the automatic polishing robot online path planning method and system establish a one-to-one mapping relationship between planes and curved surfaces by forward projection to divide complex curved surfaces into simple curved surfaces, reduce the algorithm time complexity of searching model surface path point coordinates, improve the running speed to meet the real-time path planning requirements, can directly plan the path on the point cloud data, has low requirements for the accuracy of the reconstruction algorithm and the quality of the model, can convert three-dimensional path planning into two-dimensional plane path planning, generate two-dimensional paths of any shape, and ensure automatic and efficient online path planning of the robot.

[0153] The above is only one specific embodiment of the present application, and any improvement made on the basis of the concept of the present application is considered to be within the scope of protection of the present application.

Claims

1. A method for online path planning of an automatic polishing robot, characterized in that, include: Acquire workpiece point cloud data to form a workpiece point cloud model; segment and extract several rectangular regions from the workpiece point cloud model. Based on the robot's polishing path parameters, all rectangular areas are divided into a grid map; rules are created to generate a two-dimensional path that traverses the grid map. Acquire and analyze workpiece surface images to obtain the boundary of the area to be polished on the workpiece surface; adjust the path points of the two-dimensional path according to the boundary of the area to be polished. The two-dimensional path is projected onto the point cloud surface to obtain a three-dimensional path; The path points of the three-dimensional path are optimized to obtain the actual grinding path of the robot on the workpiece surface; Before acquiring workpiece point cloud data and forming a workpiece point cloud model, the method further includes the following steps: Step A1, the workpiece is scanned by using a laser scanning shooting mode, information scanned is collected, and original collection signals of the workpiece are obtained; Step A2, based on an improved Lambert reflection model, diffuse reflection and specular reflection components of the material of the workpiece are separated and quantified, the original collection signals are corrected in reflection signal intensity by normalization processing, and the corrected reflection signal intensity is obtained; Step A3, based on the corrected reflection signal intensity, weighted neighborhood adaptive filtering processing is performed to suppress noise, so that filtered point cloud data of the workpiece is obtained; Step A4, for a missing area of the point cloud data of the workpiece, missing point interpolation is performed according to the filtered point cloud data, and the point cloud data of the missing area is filled, including: Step A41, the missing area of the point cloud data of the workpiece is determined; Step A42, taking the boundary of the missing area as a reference, a ring-shaped area with a width of R2 times of the global point cloud average spacing is expanded outward as a candidate neighborhood of the missing area; Step A43, it is judged whether the point cloud density in the candidate neighborhood is greater than or equal to the global point cloud density, if yes, the candidate neighborhood is taken as the neighborhood of the missing area, and Step A44 is continuously executed, if not, a re-scanning prompt is output ​ Step A44: Uniformly select Q boundary points on the boundary of the missing region, where Q is equal to or greater than 3; Step A45: Uniformly select M points as reference points in the neighborhood of the missing region; Fit the minimum three-dimensional bounding box of the missing region using the boundary points, and generate three-dimensional grid points at preset intervals within the minimum three-dimensional bounding box, with each grid point corresponding to a sampling point. Determine the coordinates of the interpolation points within the missing region based on the distances from each sampling point to the Q boundary points, the distances from each reference point to the Q boundary points, the coordinates of each reference point, and the rate of change of the normal vector corresponding to each reference point.

2. The online path planning method for an automated grinding robot as described in claim 1, characterized in that: Acquire workpiece point cloud data to form a workpiece point cloud model; segment and extract several rectangular regions from the workpiece point cloud model, including: The workpiece is subjected to depth scanning and imaging to obtain point cloud data of the workpiece; the point cloud data is preprocessed by filtering and removing abnormal data; the workpiece is modeled using the point cloud data to form a workpiece point cloud model. The workpiece point cloud model is segmented by normal clustering to obtain several curved surface regions; each curved surface region is fitted into a rectangular region on the projection plane using principal component analysis.

3. The online path planning method for an automated grinding robot as described in claim 2, characterized in that: Based on the robot's polishing path parameters, all rectangular areas are divided into a grid map; rules are created to generate a two-dimensional path that traverses the grid map, including: All rectangular areas are divided into grid maps according to the step size of the robot's grinding path, and each grid in the grid map is indexed by two-dimensional coordinates (u, v); Create a zigzag or zigzag path rule, and generate a two-dimensional path that traverses the grid map according to the path rule; the two-dimensional path is stored in two arrays that can query each other, one array is a one-dimensional array of path point coordinates stored with the path order as the index, and the other array is a two-dimensional array of the sequence number in the path stored with the coordinate as the index.

4. The online path planning method for an automated grinding robot as described in claim 1, characterized in that: Acquire and analyze images of the workpiece surface to obtain the boundary of the area to be polished on the workpiece surface; Based on the boundary of the area to be polished, the path points of the two-dimensional path are adjusted, including: Collecting a workpiece surface image with a normal direction of the rectangular region as a shooting optical axis, performing image segmentation, contour extraction and convex polygon fitting on the workpiece surface image to obtain a boundary of a to-be-ground region of the workpiece surface; Normalizing contour line coordinates of the boundary of the to-be-ground region to a coordinate system in which the grid map is located, identifying and filtering out path points in the two-dimensional path located outside the contour line, and reordering all path points of the two-dimensional path.

5. The automatic grinding robot online path planning method of claim 1, wherein: projecting the two-dimensional path on the point cloud surface to obtain a three-dimensional path; performing path point optimization on the three-dimensional path to obtain an actual grinding path of the robot on the workpiece surface, including: projecting the two-dimensional path on the point cloud surface to transform path point coordinates of the two-dimensional path into three-dimensional coordinates, thereby obtaining a three-dimensional path; performing interpolation and / or point supplementing processing on path point coordinates of the three-dimensional path to obtain an actual grinding path of the robot on the workpiece surface.

6. An online path planning system for an automatic polishing robot, characterized by including: a point cloud model construction module configured to obtain workpiece point cloud data and form a workpiece point cloud model; a rectangular region segmentation module configured to segment and extract a plurality of rectangular regions from the workpiece point cloud model; a grid map generation module configured to divide all the rectangular regions into a grid map according to robot grinding path parameters; a two-dimensional path generation module configured to create rules to generate a two-dimensional path that traverses the grid map; a boundary determination module configured to collect and analyze a workpiece surface image to obtain a boundary of a to-be-ground region of the workpiece surface; a two-dimensional path adjustment module configured to adjust path points of the two-dimensional path according to the boundary of the to-be-ground region; a three-dimensional path generation and optimization module configured to project the two-dimensional path on a point cloud surface to obtain a three-dimensional path; performing path point optimization on the three-dimensional path to obtain an actual grinding path of the robot on the workpiece surface; wherein, before obtaining workpiece point cloud data and forming a workpiece point cloud model, the following steps are further included: Step A1, the workpiece is scanned by using a laser scanning shooting mode, information scanned is collected, and original collection signals of the workpiece are obtained; Step A2, based on an improved Lambert reflection model, diffuse reflection and specular reflection components of the material of the workpiece are separated and quantified, the original collection signals are corrected in reflection signal intensity by normalization processing, and the corrected reflection signal intensity is obtained; Step A3, based on the corrected reflection signal intensity, weighted neighborhood adaptive filtering processing is performed to suppress noise, so that filtered point cloud data of the workpiece is obtained; Step A4, for a missing area of the point cloud data of the workpiece, missing point interpolation is performed according to the filtered point cloud data, and the point cloud data of the missing area is filled, including: Step A41, determining the missing area of the point cloud data of the workpiece; Step A42, taking the boundary of the missing area as a reference, expanding an annular area with a global point cloud average spacing R2 times as a width of the missing area as a candidate neighborhood of the missing area; Step A43, judging whether the point cloud density in the candidate neighborhood is greater than or equal to the global point cloud density, if yes, taking the candidate neighborhood as the neighborhood of the missing area, and continuing to perform Step A44, if not, outputting a re-scanning prompt; Step A44, performing missing point interpolation on the missing area based on the neighborhood of the missing area. step A44, uniformly selecting Q boundary points on the missing region boundary, wherein Q is equal to or greater than 3; step A45, uniformly selecting M points in the neighborhood of the missing region as reference points; fitting a minimum three-dimensional bounding box of the missing region through the boundary points, generating three-dimensional grid points in the minimum three-dimensional bounding box at a preset interval, each grid point corresponding to a sampling point, and determining the coordinates of the interpolation points in the missing region according to the distance from each sampling point to the Q boundary points, the distance from each reference point to the Q boundary points, the coordinates of each reference point, and the normal vector change rate corresponding to each reference point.

7. The automatic grinding robot online path planning system of claim 6, wherein: the point cloud model construction module is configured to obtain workpiece point cloud data and form a workpiece point cloud model, including: performing depth scanning and shooting on the workpiece to obtain point cloud data of the workpiece; performing abnormal data screening and rejection preprocessing on the point cloud data; and modeling the workpiece using the point cloud data to form a workpiece point cloud model. The rectangular region segmentation module is used to segment and extract several rectangular regions from the workpiece point cloud model, including: The workpiece point cloud model is segmented by normal clustering to obtain several curved surface regions; each curved surface region is fitted into a rectangular region on the projection plane using principal component analysis.

8. The online path planning system for an automated grinding robot as described in claim 6, characterized in that: The grid map generation module is used to divide all rectangular areas into grid maps based on the robot's grinding path parameters, including: All rectangular areas are divided into grid maps according to the step size of the robot's grinding path, and each grid in the grid map is indexed by two-dimensional coordinates (u, v); The two-dimensional path generation module is used to create rules to generate two-dimensional paths that traverse the grid map, including: Create a zigzag or zigzag path rule, and generate a two-dimensional path that traverses the grid map according to the path rule; the two-dimensional path is stored in two arrays that can query each other, one array is a one-dimensional array of path point coordinates stored with the path order as the index, and the other array is a two-dimensional array of the sequence number in the path stored with the coordinate as the index.

9. The online path planning system for an automated grinding robot as described in claim 6, characterized in that: The boundary determination module is used to acquire and analyze workpiece surface images to obtain the boundary of the area to be polished on the workpiece surface, including: The workpiece surface image is acquired with the normal direction of the rectangular region as the shooting optical axis. The workpiece surface image is then segmented, contour extracted, and convex polygon fitted to obtain the boundary of the area to be polished on the workpiece surface. The two-dimensional path adjustment module is used to adjust the path points of the two-dimensional path according to the boundary of the area to be polished, including: The coordinates of the outline of the area to be polished are normalized and transformed to the coordinate system of the grid map. Path points located outside the outline in the two-dimensional path are identified and filtered out. Then, all path points in the two-dimensional path are reordered.

10. The online path planning system for an automated grinding robot as described in claim 6, characterized in that: The 3D path generation and optimization module is used to project the 2D path onto the point cloud surface to obtain the 3D path. The path point optimization of the three-dimensional path yields the actual grinding path of the robot on the workpiece surface, including: The two-dimensional path is projected onto the point cloud surface, thereby transforming the path point coordinates of the two-dimensional path into three-dimensional coordinates, thus obtaining the three-dimensional path. Interpolation and / or point supplementation are performed on the coordinates of the path points of the three-dimensional path to obtain the actual grinding path of the robot on the surface of the workpiece.

Images