Paint layer grinding and polishing method and system based on point cloud

Abstract

The invention belongs to the related technical field of grinding robot machining, and discloses a paint layer grinding and polishing method and system based on point cloud, and the method comprises the steps: S1, obtaining point cloud data of a to-be-machined surface to form a point cloud model, and carrying out the initial track planning according to the point cloud model; s2, primary processing is carried out; s3, whether an unprocessed area exists or not is judged, and if yes, image data of the to-be-processed surface are obtained; s4, fusing the image data and the point cloud data to generate a color point cloud; s5, extracting an unprocessed area according to the color point cloud, and performing secondary trajectory planning according to the point cloud data corresponding to the unprocessed area; s6, carrying out secondary processing; and S3 to S6 are repeated until all the areas of the to-be-machined surface are machined. According to the method, the image information and the point cloud information are subjected to multi-modal fusion, and secondary processing track planning and execution are performed based on the extracted under-processing area information, so that the problem of incomplete single processing is solved, and the processing effect is ensured.

Description

Technical Field

[0001] The present invention belongs to the technical field related to grinding robot processing, and more specifically, relates to a paint layer grinding and polishing method and system based on point cloud. Background Art

[0002] With the widespread application of industrial robots in the field of mechanical processing and manufacturing, industries such as automobiles, aviation, aerospace and shipbuilding have put forward higher requirements for robot processing technology. In these fields, large and complex curved components (such as high-speed rail bodies, aircraft skins, wind turbine blades and ship hulls, etc.) generally need to be treated with paint. For example, aircraft skins need to be removed and repainted during manufacturing and maintenance, and the ship hull needs to regularly remove the old paint for anti-corrosion treatment. These application scenarios place strict requirements on the accuracy, efficiency and quality of paint polishing.

[0003] For complex surfaces with low stiffness and differences between the actual model and the designed model, the commonly used processing method is to measure first and then process. Specifically, for example, first use a three-dimensional coordinate measuring machine or laser tracker to obtain the three-dimensional point cloud data of the workpiece, then plan the processing path based on the data, and finally complete the processing through industrial robots or large CNC machine tools. However, these systems and methods often only use the three-dimensional information of the workpiece, ignoring the surface condition of the workpiece after processing. As a result, if the paint layer is removed unevenly or incompletely after processing, it is impossible to carry out targeted secondary processing, and thus the processing effect cannot be guaranteed. Summary of the invention

[0004] In response to the above defects or improvement needs of the prior art, the present invention provides a point cloud-based paint layer grinding and polishing method and system, which is used to solve the problem that if the paint layer is removed unevenly or incompletely after processing by the existing paint layer grinding and polishing method, targeted secondary processing cannot be carried out, thereby failing to guarantee the processing effect.

[0005] To achieve the above object, according to one aspect of the present invention, a paint layer grinding and polishing method based on point cloud is provided, comprising:

[0006] S1, acquiring point cloud data of the surface to be processed by a point cloud scanner to form a point cloud model, and performing initial trajectory planning for polishing of the paint layer according to the point cloud model;

[0007] S2, initial processing according to the path planned by the initial trajectory;

[0008] S3, determining whether there is an unprocessed area on the surface to be processed, and if so, obtaining image data of the surface to be processed;

[0009] S4, fusing the image data with the point cloud data in the point cloud model to generate a color point cloud;

[0010] S5, extracting an unprocessed area of ​​the surface to be processed according to the colored point cloud, and performing secondary trajectory planning according to the point cloud data corresponding to the unprocessed area;

[0011] S6, performing secondary processing according to the path planned by the secondary trajectory;

[0012] Repeat S3-S6 until all areas of the surface to be processed are processed.

[0013] According to the paint layer grinding and polishing method based on point cloud provided by the present invention, S4 specifically includes:

[0014] Selecting a plurality of positioning points in the image data and acquiring point cloud data corresponding to the positioning points;

[0015] According to the plurality of positioning points and the corresponding point cloud data, the coordinate transformation matrix in the point cloud pixel projection process is optimized by using the least square method to obtain an optimized coordinate transformation matrix;

[0016] The optimized coordinate transformation matrix is ​​used to perform projection transformation between point cloud data and image data to achieve data fusion and generate color point cloud.

[0017] According to the paint layer grinding and polishing method based on point cloud provided by the present invention, obtaining the optimized coordinate transformation matrix also includes:

[0018] Selecting a plurality of feature points in the image data and determining true pixel values ​​of the feature points;

[0019] Based on the optimized coordinate transformation matrix of the positioning points, the point cloud model is projected onto the image plane to obtain projection pixel values ​​and corresponding projection depth values;

[0020] Based on the real pixel value of the feature point and the projected pixel value, determine the projected pixel value that best matches the feature point, and obtain the corresponding projected depth value;

[0021] Based on the real pixel value of the feature point, the corresponding projection depth value and the coordinate transformation matrix, obtaining the generated coordinate value of the feature point;

[0022] Acquire the distance between the generated coordinate value and the adjacent point cloud in the point cloud model as a projection point cloud error;

[0023] Reprojecting the generated coordinate value onto an image plane to obtain a reprojected pixel value, and obtaining a projection pixel error between the reprojected pixel value and a true pixel value;

[0024] Based on the projection point cloud error and the projection pixel error, the coordinate transformation matrix is ​​optimized twice to obtain the coordinate transformation matrix after the second optimization.

[0025] According to the paint layer polishing method based on point cloud provided by the present invention, multiple feature points are selected from the image data, specifically including:

[0026] Perform SIFT feature detection on the image data to obtain a plurality of feature points corresponding to the image data.

[0027] According to the paint layer grinding and polishing method based on point cloud provided by the present invention, obtaining image data of the surface to be processed in S3 specifically includes:

[0028] Using a camera to perform surround shooting on the surface to be processed to obtain a plurality of image data; wherein the plurality of image data cover the surface to be processed and there is an overlapping area between two adjacent image data;

[0029] Accordingly, S4 specifically includes:

[0030] The plurality of image data are respectively fused with the point cloud data in the point cloud model to generate a color point cloud.

[0031] According to the paint layer polishing method based on point cloud provided by the present invention, for a plurality of the image data, respectively, the image data are fused with the point cloud data in the point cloud model to generate a color point cloud, and further includes:

[0032] Performing SIFT feature detection and feature matching on the plurality of image data to obtain feature points of each image data and matching relationships between different image data;

[0033] Determining the first image data to be fused with the point cloud data in the point cloud model according to the number of feature points;

[0034] Then, the image data having a matching relationship are selected in sequence for data fusion.

[0035] According to the paint layer grinding and polishing method based on point cloud provided by the present invention, in S5, the unprocessed area of ​​the surface to be processed is extracted according to the color point cloud, which specifically includes:

[0036] According to the image color information in the color point cloud, segmentation is performed by setting a threshold; or a decision tree is established by a random forest method to perform segmentation; and an unprocessed point cloud is obtained by segmentation to form an unprocessed area.

[0037] According to the paint layer grinding and polishing method based on point cloud provided by the present invention, the secondary trajectory planning in S5 according to the point cloud data corresponding to the unprocessed area specifically includes:

[0038] Divide the unprocessed area into grids, calculate the surface area of ​​each point cloud in the grid and determine whether the grid area needs secondary processing; if necessary, extract the center point of the point cloud in the grid area and store it in the point set to be processed; if not, skip the grid until all grids are traversed;

[0039] The traversal order of all points in the set of points to be processed is optimized to obtain the optimal traversal path as the path for secondary trajectory planning.

[0040] According to the paint layer grinding and polishing method based on point cloud provided by the present invention, the objective function X is optimized by traversing all points in the set of points to be processed in order. * for:

[0041]

[0042] Among them, d k is the path cost in optimization; l is the distance between two adjacent points; L is the set threshold, ω k is the set penalty coefficient.

[0043] According to another aspect of the present invention, a point cloud-based paint layer grinding and polishing system is provided, the system comprising a memory and a processor, the memory storing a computer program, and the processor executing any one of the above-mentioned point cloud-based paint layer grinding and polishing methods when executing the computer program.

[0044] In general, compared with the prior art, the above technical solutions conceived by the present invention provide a paint layer grinding and polishing method and system based on point cloud:

[0045] 1. It is proposed to use perception fusion technology to fuse image information and point cloud information in a multi-modal manner as the basis for processing trajectory planning. Through the data fusion of image and point cloud, and the introduction of under-processed area detection and extraction steps, secondary processing trajectory planning and execution are carried out based on the extracted under-processed area information, which effectively solves the problem of incomplete paint removal in a single processing, realizes closed-loop control of the processing process, and ensures the processing effect;

[0046] 2. The present invention significantly improves the automation level of robot processing, greatly improves processing efficiency by reducing manual intervention and avoiding repeated measurements and adjustments;

[0047] 3. The present invention can effectively cope with the challenges of insufficient rigidity of complex curved surface workpieces and deviations between actual models and design models, and has strong adaptability and robustness; in addition, through accurate detection and secondary processing of under-processed areas, the processing accuracy and surface quality of the workpiece are further guaranteed, and high-standard process requirements are met. The closed-loop processing of weak rigid components such as aircraft skins can be well realized, and the processing efficiency can be significantly improved and the processing quality can be ensured. It provides an efficient and accurate solution for the automated processing of complex curved surface workpieces, and has broad application prospects;

[0048] 4. The processing method proposed in the present invention is not only applicable to the field of robot grinding and polishing, but can also be extended to processing scenarios such as painting, and has strong versatility and promotion value. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 is a flow chart of a paint layer grinding and polishing method based on point cloud constructed according to an embodiment of the present invention;

[0050] Figure 2 is a schematic structural diagram of a paint layer grinding and polishing system based on point cloud constructed according to an embodiment of the present invention;

[0051] Figure 3 is a schematic diagram of trajectory planning using point cloud constructed according to an embodiment of the present invention;

[0052] Figure 4 is a schematic diagram of the effect of multimodal fusion of images and point clouds constructed according to an embodiment of the present invention;

[0053] Figure 5 It is a schematic diagram of performing point cloud segmentation on a color point cloud to extract an under-processed area constructed according to an embodiment of the present invention;

[0054] Figure 6 It is a trajectory schematic diagram of secondary trajectory planning based on the point cloud of the under-processed area constructed according to an embodiment of the present invention;

[0055] Throughout the drawings, the same reference numerals are used to denote the same elements or structures, wherein:

[0056] 1-Point cloud scanner; 2-Camera; 3-Force-controlled grinding head; 4-Industrial robot. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0058] See also Figure 1 and Figure 2 This embodiment provides a paint layer grinding and polishing method based on point cloud, and the paint layer grinding and polishing method based on point cloud includes:

[0059] S1, acquiring point cloud data of the surface to be processed by a point cloud scanner 1 to form a point cloud model, and performing initial trajectory planning for polishing of the paint layer according to the point cloud model;

[0060] S2, initial processing according to the path planned by the initial trajectory;

[0061] S3, determining whether there is an unprocessed area on the surface to be processed, and if so, obtaining image data of the surface to be processed;

[0062] S4, fusing the image data with the point cloud data in the point cloud model to generate a color point cloud;

[0063] S5, extracting an unprocessed area of ​​the surface to be processed according to the colored point cloud, and performing secondary trajectory planning according to the point cloud data corresponding to the unprocessed area;

[0064] S6, performing secondary processing according to the path planned by the secondary trajectory;

[0065] S3-S6 are repeated until all areas of the surface to be processed are processed, thereby completing the closed-loop control of the processing process.

[0066] The paint layer polishing method provided in this embodiment specifically acquires point cloud data through a point cloud scanner 1 and plans a processing trajectory. After a round of processing is completed, if there are some areas that are not completely removed, the surface image data of the workpiece is collected through a camera 2, and the image data and the point cloud data are multimodally fused to generate a color point cloud. Based on the color point cloud, the information of the under-processed area is extracted, and then the secondary processing trajectory is planned and the secondary processing is performed. If the processing effect still does not meet expectations, the above steps are repeated until the paint surface is completely removed, thereby ensuring the processing effect.

[0067] Furthermore, the multimodal data fusion of image data and point cloud data is achieved by using the structure from motion framework, using the point cloud as a reconstruction constraint, relying on feature point correspondence to align the image to the provided point cloud, and then back-projecting the pixels to the corresponding point cloud to obtain point cloud data with color texture.

[0068] S4 specifically includes:

[0069] Selecting a plurality of positioning points in the image data and acquiring point cloud data corresponding to the positioning points;

[0070] According to the plurality of positioning points and the corresponding point cloud data, the coordinate transformation matrix in the point cloud pixel projection process is optimized by using the least square method to obtain an optimized coordinate transformation matrix;

[0071] The optimized coordinate transformation matrix is ​​used to perform projection transformation between point cloud data and image data to achieve data fusion and generate color point cloud.

[0072] Furthermore, obtaining the optimized coordinate transformation matrix also includes:

[0073] Selecting a plurality of feature points in the image data and determining true pixel values ​​of the feature points;

[0074] Based on the optimized coordinate transformation matrix of the positioning points, the point cloud model is projected onto the image plane to obtain projection pixel values ​​and corresponding projection depth values;

[0075] Based on the real pixel value of the feature point and the projected pixel value, determine the projected pixel value that best matches the feature point, and obtain the corresponding projected depth value;

[0076] Based on the real pixel value of the feature point, the corresponding projection depth value and the coordinate transformation matrix, obtaining the generated coordinate value of the feature point;

[0077] Acquire the distance between the generated coordinate value and the adjacent point cloud in the point cloud model as a projection point cloud error;

[0078] Reprojecting the generated coordinate value onto an image plane to obtain a reprojected pixel value, and obtaining a projection pixel error between the reprojected pixel value and a true pixel value;

[0079] Based on the projected point cloud error and the projected pixel error, the coordinate transformation matrix is ​​optimized twice to obtain the coordinate transformation matrix after the second optimization. The point cloud is projected onto the image plane using the coordinate transformation matrix after the second optimization, thereby achieving data fusion.

[0080] In some specific embodiments, selecting a plurality of feature points in the image data specifically includes:

[0081] Perform SIFT feature detection on the image data to obtain a plurality of feature points corresponding to the image data.

[0082] In some specific embodiments, obtaining image data of the surface to be processed in S3 specifically includes:

[0083] Using the camera 2 to perform surround shooting on the surface to be processed, a plurality of image data are obtained; wherein the plurality of image data cover the surface to be processed and there is an overlapping area between two adjacent image data;

[0084] Accordingly, S4 specifically includes:

[0085] The plurality of image data are respectively fused with the point cloud data in the point cloud model to generate a color point cloud.

[0086] In some specific embodiments, for the plurality of image data, respectively fusing them with the point cloud data in the point cloud model to generate a color point cloud, further comprising:

[0087] Performing SIFT feature detection and feature matching on the plurality of image data to obtain feature points of each image data and matching relationships between different image data;

[0088] Determining the first image data to be fused with the point cloud data in the point cloud model according to the number of feature points;

[0089] Then, the image data with matching relationship are selected in turn for data fusion. After the first image data and the point cloud data are fused, an image with matching relationship with the first image data, i.e., an image with overlapping area, can be selected as the next image to be fused, and the known point cloud data of the first image can be used in the overlapping area to select the positioning point of the next image to be fused, and then the coordinate transformation matrix of the next image can be optimized; and so on.

[0090] Specifically, in S5, the unprocessed area of ​​the surface to be processed is extracted according to the color point cloud, that is, the point cloud segmentation method specifically includes:

[0091] According to the image color information in the color point cloud, segmentation can be performed simply by setting a threshold; or by using a random forest method to establish a decision tree for segmentation; and an unprocessed point cloud is obtained through segmentation to form an unprocessed area.

[0092] Furthermore, the secondary trajectory planning method is to voxelize the point cloud of the under-processed area and extract it as points to be processed, and then obtain a better processing trajectory through optimization algorithms such as genetic algorithms.

[0093] In S5, the secondary trajectory planning based on the point cloud data corresponding to the unprocessed area specifically includes:

[0094] Divide the unprocessed area into grids, calculate the surface area of ​​each point cloud in the grid and determine whether the grid area needs secondary processing; if necessary, extract the center point of the point cloud in the grid area and store it in the point set to be processed; if not, skip the grid until all grids are traversed;

[0095] The traversal order of all points in the set of points to be processed is optimized to obtain the optimal traversal path as the path for secondary trajectory planning.

[0096] In some specific embodiments, a method for polishing a paint layer on a complex curved surface based on point cloud specifically comprises the following steps:

[0097] 1) Obtain point cloud model

[0098] After the workpiece to be processed is installed, the point cloud scanner 1 is used to scan the entire workpiece, and the overall point cloud data of the surface to be processed of the workpiece is obtained to form a point cloud model and converted into the robot coordinate system for a unified coordinate system;

[0099] 2) Initial processing

[0100] The point cloud model is used to directly plan the robot processing trajectory, and the planned trajectory points are transmitted to the robot; the trajectory of a certain implementation plan is shown as follows Figure 3 As shown. The force-controlled grinding head 3 and the industrial robot 4 are used to complete the grinding and polishing of the workpiece;

[0101] 3) Determine whether processing is completed

[0102] It is judged whether the paint surface is completely removed by manual observation or image detection. If it is completely removed, the processing is terminated; otherwise, the following steps are performed; the image detection can be performed by shooting an image of the surface to be processed and judging whether the paint surface is completely removed by setting a threshold value according to the color information of the image;

[0103] 4) Multimodal data fusion

[0104] Use camera 2 to capture images around the workpiece, ensuring that each image has at least 30% overlap with the adjacent image. The number of images taken varies depending on the size of the workpiece being processed, and generally at least 15 images are required. Then all image data and the workpiece point cloud are multimodally fused to obtain point cloud data with color textures. The result of a multimodal data fusion implementation is shown below: Figure 4 shown.

[0105] The specific scheme is as follows: ① First, perform SIFT feature detection and feature matching on the image sequence of all image data to obtain the feature points of each image and the matching relationship between each image. ② Then select a better image among the images with more feature points, such as the image with the most feature points, as the first image to be fused, select at least 4 points in the image as positioning points and find corresponding points in the point cloud to form a positioning point pair. The positioning point can be a point at the corner or center of the image to facilitate finding corresponding points in the point cloud model. The specific number and position are not limited. The projection formula from the point cloud to the image plane is as follows:

[0106]

[0107] Among them, p i =[ui ,v i ,1] T (i=1,2,...,n) is the homogeneous coordinate of the i-th positioning point in the pixel coordinate system, is the coordinate of the i-th positioning point in the point cloud coordinate system (world coordinate system), ω i is the depth of the point in the camera 2 coordinate system, K is the intrinsic parameter matrix of camera 2, and T is the coordinate transformation matrix from the world coordinate system to the camera 2 coordinate system. Due to the error, the above formula is not strictly true. It is necessary to construct the least squares optimization problem of the reprojection error, calculate the camera 2 pose of the image, that is, the coordinate transformation matrix, and use the least squares method to optimize the objective function T * as follows:

[0108]

[0109] By using the optimized camera 2 pose of the positioning point, the approximate position of the image in the point cloud can be determined, and used as the starting position for the next step of the algorithm.

[0110] ③ In the next step, the image and posture are used to obtain the initial 3D points of the feature points on the image, that is, to generate coordinate values. The details are as follows: Set the known point cloud model as P = {(p i ,n i )|i=1...N},p i is the point cloud coordinate, n i is the corresponding normal vector, and the real pixel value of the feature point on the image is Project the point cloud model onto the image plane using formula (1) to obtain the corresponding projected pixel value For any feature point, the true pixel value I m , find the best matching projection pixel I P (The shortest distance can be used as the best matching standard), and the projected depth information, that is, the corresponding projected depth value combined with the real pixel value, is used to generate the initial feature point 3D coordinate q through formula (1) i That is, the generated coordinate value corresponding to the feature point. Use KNN search to establish the matching relationship between the generated coordinate value and the neighboring point cloud, and calculate the distance E from the generated coordinate value to the plane composed of a preset number of neighboring point clouds. d , the distance can be calculated by the normal vector of the point cloud coordinates. Then the generated coordinate value is reprojected back to the image plane, and its reprojection error, that is, the projection pixel error E, is calculated. r The formula is as follows:

[0111]

[0112] Where (u i ,v i ) is the real pixel coordinate corresponding to the feature point; The reprojected pixel coordinates corresponding to the generated coordinate values ​​of the feature points; by optimizing the reprojection error, the objective function of the secondary optimization using the feature points is:

[0113] min∑(E r +E d )(4)

[0114] Further optimize T.

[0115] ④ In the next step, based on the matching relationship, select the image that overlaps the most with the current image and has not been accessed as the image data for the next calculation. Continue to optimize the pose T of the next image based on the matching relationship between the two images and the feature point-3D point correspondence of the previous image. The method and principle are as shown in formulas (1) and (2). Then, the new 3D point is calculated by triangulating the two images to generate the coordinate value. Use KNN search to establish the matching relationship between the reconstructed 3D point and the neighboring point cloud, and calculate the distance E from the point to the plane composed of the neighboring point cloud. d The next step is to reproject the reconstructed 3D points back to the current image and calculate the reprojection error E according to formula (3): r Similarly, the pose T of the current image is further optimized using a method similar to formula (4).

[0116] ⑤ In the next step, continue to add related images, repeat the above steps, and calculate the poses of all images in the point cloud coordinate system. In this way, the point cloud can be projected onto the image to obtain its corresponding color information, and the single point cloud coordinates can be expanded into the form of (X, Y, Z, R, G, B) to obtain point cloud data with color texture.

[0117] 5) Secondary processing

[0118] According to the difference between the processed area and the under-processed area, the color information in the color texture point cloud is used to segment the point cloud and extract the point cloud of the under-processed area; Figure 4 The point cloud of the unprocessed area segmentation is shown as follows Figure 5 As shown in the figure, the extracted point cloud of the under-processed area is subjected to secondary trajectory planning, and the planned trajectory points are transmitted to the robot for secondary processing. Figure 5 The trajectory obtained by performing secondary trajectory planning on the under-processed point cloud shown in Figure 6 Then determine whether the processing is completed, otherwise repeat the 4-5 process.

[0119] The specific method of secondary trajectory planning is as follows: set a suitable grid size that can be determined according to the processing volume of the processing equipment, divide the extracted point cloud of the under-processed area into grids, calculate the surface area of ​​each point cloud in the grid and determine whether the area needs secondary processing. If necessary, extract the center point of the point cloud in the area and store it in the processing point set P. If not, skip the grid until all grids are traversed. Thus, the path planning problem of secondary processing is converted into the problem of how to traverse the processing point set P in the optimal order. For grinding and polishing, if the distance between two consecutive points is too large, the robot needs to be lifted up, walk empty, and then fall down for processing during operation. Therefore, there will be an invalid stage of empty tool operation. In order to minimize this phenomenon in secondary planning, it is necessary to arrange the order of the processing point set reasonably. Assuming that a certain order is X = [1,2,3,…k,…n], then according to the order, the distance between a point and its next point can be expressed as:

[0120] l=||p k -p k+1 || (5)

[0121] Then the objective function X is optimized by traversing all points in the set of points to be processed. * for:

[0122]

[0123] Among them, d k is the path cost in optimization; l is the distance between two adjacent points; L is the set threshold, ω k is the set penalty coefficient. The objective optimization function is solved by a heuristic algorithm, for example, by considering the sequence X = [1, 2, 3, ... k, ... n] as a chromosome, a relatively satisfactory result can be obtained by a genetic algorithm.

[0124] An embodiment of the present invention also provides a point cloud-based paint layer grinding and polishing system, the system comprising a memory and a processor, the memory storing a computer program, and the processor executing any one of the above-mentioned point cloud-based paint layer grinding and polishing methods when executing the computer program.

[0125] like Figure 2As shown, a complex curved surface paint layer grinding and polishing system based on point cloud, the system also includes a point cloud scanner 1, a camera 2, a force-controlled grinding head 3 and an industrial robot 4; wherein the point cloud scanner 1 is used to obtain the three-dimensional morphological data of the workpiece, and the camera 2 is an RGB camera 2, which is used to obtain the color texture image information of the workpiece, and there is no real-time requirement between the two, that is, no synchronous triggering is required; the data acquisition of the camera 2 and the point cloud scanner 1 is asynchronous, that is, there is no need to synchronously collect point cloud data and image data. The robot is a six-degree-of-freedom industrial robot 4, and the force-controlled grinding head 3 is set on the end flange of the robot. The multi-degree-of-freedom movement of the grinding head is realized through the robot joint movement, and the processing of the workpiece is completed together.

[0126] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A paint layer grinding and polishing method based on point cloud, characterized in that: include: S1, acquiring point cloud data of the surface to be processed by a point cloud scanner to form a point cloud model, and performing initial trajectory planning for polishing of the paint layer according to the point cloud model; S2, initial processing according to the path planned by the initial trajectory; S3, determining whether there is an unprocessed area on the surface to be processed, and if so, obtaining image data of the surface to be processed; S4, fusing the image data with the point cloud data in the point cloud model to generate a color point cloud; S5, extracting an unprocessed area of ​​the surface to be processed according to the colored point cloud, and performing secondary trajectory planning according to the point cloud data corresponding to the unprocessed area; S6, performing secondary processing according to the path planned by the secondary trajectory; Repeat S3-S6 until all areas of the surface to be processed are processed.

2. The paint layer grinding and polishing method based on point cloud according to claim 1, characterized in that: S4 specifically includes: Selecting a plurality of positioning points in the image data and acquiring point cloud data corresponding to the positioning points; According to the plurality of positioning points and the corresponding point cloud data, the coordinate transformation matrix in the point cloud pixel projection process is optimized by using the least square method to obtain an optimized coordinate transformation matrix; The optimized coordinate transformation matrix is ​​used to perform projection transformation between point cloud data and image data to achieve data fusion and generate color point cloud.

3. The paint layer grinding and polishing method based on point cloud according to claim 2, characterized in that: Get the optimized coordinate transformation matrix, including: Selecting a plurality of feature points in the image data and determining true pixel values ​​of the feature points; Based on the optimized coordinate transformation matrix of the positioning points, the point cloud model is projected onto the image plane to obtain projection pixel values ​​and corresponding projection depth values; Based on the real pixel value of the feature point and the projected pixel value, determine the projected pixel value that best matches the feature point, and obtain the corresponding projected depth value; Based on the real pixel value of the feature point, the corresponding projection depth value and the coordinate transformation matrix, obtaining the generated coordinate value of the feature point; Acquire the distance between the generated coordinate value and the adjacent point cloud in the point cloud model as a projection point cloud error; Reprojecting the generated coordinate value onto an image plane to obtain a reprojected pixel value, and obtaining a projection pixel error between the reprojected pixel value and a true pixel value; Based on the projection point cloud error and the projection pixel error, the coordinate transformation matrix is ​​optimized twice to obtain the coordinate transformation matrix after the second optimization.

4. The paint layer grinding and polishing method based on point cloud according to claim 3, characterized in that: Selecting a plurality of feature points in the image data specifically includes: Perform SIFT feature detection on the image data to obtain a plurality of feature points corresponding to the image data.

5. The paint layer grinding and polishing method based on point cloud according to any one of claims 1 to 4, characterized in that: S3 obtains image data of the surface to be processed, specifically including: Using a camera to perform surround shooting on the surface to be processed to obtain a plurality of image data; wherein the plurality of image data cover the surface to be processed and there is an overlapping area between two adjacent image data; Accordingly, S4 specifically includes: The plurality of image data are respectively fused with the point cloud data in the point cloud model to generate a color point cloud.

6. The paint layer grinding and polishing method based on point cloud according to claim 5, characterized in that: For the plurality of image data, respectively, they are fused with the point cloud data in the point cloud model to generate a color point cloud, further comprising: Performing SIFT feature detection and feature matching on the plurality of image data to obtain feature points of each image data and matching relationships between different image data; Determining the first image data to be fused with the point cloud data in the point cloud model according to the number of feature points; Then, the image data having a matching relationship are selected in sequence for data fusion.

7. The paint layer grinding and polishing method based on point cloud according to any one of claims 1 to 4, characterized in that: In S5, the unprocessed area of ​​the surface to be processed is extracted according to the color point cloud, including: According to the image color information in the color point cloud, segmentation is performed by setting a threshold; or a decision tree is established by a random forest method to perform segmentation; and an unprocessed point cloud is obtained by segmentation to form an unprocessed area.

8. The paint layer grinding and polishing method based on point cloud according to any one of claims 1 to 4, characterized in that: In S5, the secondary trajectory planning based on the point cloud data corresponding to the unprocessed area specifically includes: Divide the unprocessed area into grids, calculate the surface area of ​​each point cloud in the grid and determine whether the grid area needs secondary processing; if necessary, extract the center point of the point cloud in the grid area and store it in the point set to be processed; if not, skip the grid until all grids are traversed; The traversal order of all points in the set of points to be processed is optimized to obtain the optimal traversal path as the path for secondary trajectory planning.

9. The paint layer grinding and polishing method based on point cloud according to claim 8, characterized in that: The objective function X is optimized by traversing all points in the set of points to be processed. * for: X * =argmin∑d k Among them, d k is the path cost in optimization; l is the distance between two adjacent points; L is the set threshold, ω k is the set penalty coefficient.

10. A paint layer grinding and polishing system based on point cloud, characterized in that: The system includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the point cloud-based paint layer grinding and polishing method described in any one of claims 1 to 9 is executed.

Images