Method and device for detecting surface of plate

Abstract

The invention discloses a plate surface detection method and device, relates to the technical field of computer science and engineering, and aims to solve the problem of poor accuracy of acquired plate surface data in the prior art. The method comprises the steps of obtaining three-dimensional coordinate information data of the surface of a plate, and performing data preprocessing on the three-dimensional coordinate information data to obtain first three-dimensional point cloud data; rasterization processing is further carried out on the first three-dimensional point cloud data, target feature information in the first three-dimensional point cloud data is reserved, and second three-dimensional point cloud data is obtained; performing defect filtering processing on the second three-dimensional point cloud data according to a preset defect filtering strategy to obtain plate type three-dimensional point cloud data of the surface of the plate; the defect filtering processing comprises one or more of down-sampling, point cloud curved surface reconstruction and up-sampling processing; on the basis, the three-dimensional point cloud data of the surface of the plate can be obtained more accurately, and the accuracy of obtaining the surface data of the plate is improved.

Description

Technical Field

[0001] The invention relates to the field of computer science and engineering technology, and in particular to a method and device for detecting a plate surface. Background Art

[0002] In the field of industrial manufacturing, metal sheets are widely used in important fields such as automobile manufacturing, aerospace, and building structures. The surface quality of metal sheets directly affects the performance and quality of products. Therefore, sheet surface inspection has become a crucial quality control task in the manufacturing industry, which involves the early detection and timely treatment of various types of defects on the surface of metal sheets. In the sheet manufacturing process, various defects such as cracks, bubbles, inclusions, scratches, etc. are prone to appear on the surface of the sheet due to factors such as processing technology, processing technology, processing equipment, and external environment. These defects may have a serious impact on the performance and appearance of metal products, and even affect the service life and safety of the products.

[0003] The traditional method of detecting and measuring the surface of a plate mainly relies on two-dimensional images or one-dimensional data to obtain detection data. When using two-dimensional image processing technology, it is limited by factors such as ambient lighting, angle and reflection of the material surface, resulting in poor accuracy and reliability of the detection results. In addition, since defects on the surface of the plate are easily interfered by factors such as iron oxide, it is easy to miss or misdetect by relying solely on two-dimensional image detection. In addition, the traditional method has certain limitations in dealing with transmission vibration interference. For example, due to image blur or distortion caused by vibration, the traditional image processing algorithm cannot accurately extract the characteristic information of the plate surface. Especially for the plate surface with complex geometric shapes, the traditional measurement method cannot provide enough information for accurate measurement and analysis. Therefore, a more advanced detection method is needed to solve the problem of poor accuracy of the plate surface detection data obtained due to missed detection, misdetection and interference when detecting the plate surface in the existing technology. Summary of the invention

[0004] The object of the present invention is to provide a method and device for detecting the surface of a plate, so as to solve the problem of poor accuracy of obtaining plate surface data in the prior art.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] In a first aspect, the present invention provides a method for detecting a surface of a plate material, which may include:

[0007] Obtaining three-dimensional coordinate information data of the plate surface;

[0008] Performing data preprocessing on the three-dimensional coordinate information data to obtain first three-dimensional point cloud data; the data preprocessing includes filtering processing and / or point cloud coordinate conversion;

[0009] rasterizing the first three-dimensional point cloud data, retaining target feature information in the first three-dimensional point cloud data, and obtaining second three-dimensional point cloud data;

[0010] According to a preset defect filtering strategy, the second three-dimensional point cloud data is subjected to defect filtering processing to obtain the plate type three-dimensional point cloud data for the surface of the plate; the defect filtering processing includes one or more of downsampling, point cloud surface reconstruction and upsampling processing.

[0011] In a second aspect, the present invention provides a detection device for a plate surface, which may include:

[0012] An acquisition module, the acquisition module is used to acquire three-dimensional coordinate information data of the surface of the plate;

[0013] A preprocessing module, the preprocessing module is used to perform data preprocessing on the three-dimensional coordinate information data to obtain first three-dimensional point cloud data; the data preprocessing includes filtering processing and / or point cloud coordinate conversion;

[0014] A rasterization module, the rasterization module is used to perform rasterization processing on the first three-dimensional point cloud data, retain the target feature information in the first three-dimensional point cloud data, and obtain second three-dimensional point cloud data;

[0015] A defect filtering module, wherein the defect filtering module is used to perform defect filtering processing on the second three-dimensional point cloud data according to a preset defect filtering strategy to obtain the plate type three-dimensional point cloud data for the surface of the plate; the defect filtering processing includes one or more of downsampling, point cloud surface reconstruction and upsampling processing.

[0016] Compared with the prior art, the present invention provides a detection method for the surface of a plate, which obtains three-dimensional coordinate information data of the surface of the plate; performs data preprocessing on the three-dimensional coordinate information data to obtain first three-dimensional point cloud data; the data preprocessing includes filtering processing and / or point cloud coordinate conversion; further performs rasterization processing on the first three-dimensional point cloud data, retains the target feature information in the first three-dimensional point cloud data, and obtains second three-dimensional point cloud data; according to a preset defect filtering strategy, performs defect filtering processing on the second three-dimensional point cloud data to obtain the plate type three-dimensional point cloud data of the plate surface; the defect filtering processing includes downsampling, point cloud surface reconstruction and upsampling processing One or more of the above; based on this, firstly, by acquiring the three-dimensional coordinate information data of the plate surface, compared with the traditional plate surface detection and measurement method that relies on two-dimensional images or one-dimensional data, rich three-dimensional morphological data of the plate surface can be obtained, such as plate type problems, defect problems, etc. can be efficiently detected, and the detection results are stable and reliable; further filtering processing, point cloud coordinate conversion, rasterization processing and defect filtering processing are performed on the three-dimensional point cloud data, which can eliminate interference and other factors in the plate transmission process, and can obtain more accurate three-dimensional point cloud data of the plate surface, ensure the accuracy and reliability of plate surface detection, and improve the accuracy of obtaining plate surface data. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0018] Figure 1 A first flow chart of a method for detecting a surface of a plate provided by the present invention;

[0019] Figure 2 A second flow chart of a method for detecting a surface of a plate provided by the present invention;

[0020] Figure 3 A two-dimensional image of the actual surface of a plate is involved in a plate surface detection method provided by the present invention;

[0021] Figure 4 A first processing effect diagram of a plate surface detection method provided by the present invention;

[0022] Figure 5 A second processing effect diagram of a plate surface detection method provided by the present invention;

[0023] Figure 6 A third processing effect diagram of a plate surface detection method provided by the present invention;

[0024] Figure 7 A schematic structural diagram of a detection device for a plate surface provided by the present invention. DETAILED DESCRIPTION

[0025] In order to clearly describe the technical solutions of the embodiments of the present invention, in the embodiments of the present invention, words such as "first" and "second" are used to distinguish the same items or similar items with basically the same functions and effects. For example, the first threshold and the second threshold are only used to distinguish different thresholds, and their order is not limited. Those skilled in the art can understand that words such as "first" and "second" do not limit the quantity and execution order, and words such as "first" and "second" do not necessarily limit them to be different.

[0026] It should be noted that, in the present invention, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0027] In the present invention, "at least one" means one or more, and "plurality" means two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. The character " / " generally indicates that the associated objects before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, the combination of a and b, the combination of a and c, the combination of b and c, or the combination of a, b and c, where a, b, c can be single or multiple.

[0028] With the continuous development of computer vision and deep learning technology, people have begun to explore the use of three-dimensional data for surface quality inspection and analysis to effectively make up for the shortcomings of traditional methods. By using three-dimensional point cloud data, the surface geometry, tiny details and defects can be more comprehensively captured, thereby improving the accuracy and reliability of inspection. Converting three-dimensional point cloud data into depth maps can further meet the needs of industrial manufacturing and visual inspection fields for data formats, making the data easier to process and analyze.

[0029] However, there are some challenges and limitations in existing technologies, such as:

[0030] 1. The complexity of the plate surface leads to insufficient detection information: For some plate surface shape problems, traditional detection methods may not provide enough information for accurate measurement and analysis. Especially when there is a shape problem at the defect location, the lack of depth information makes detection more difficult.

[0031] 2. Limitations of image processing technology: Image processing technology can often only detect two-dimensional defects on the surface. Surface defects are easily interfered by factors such as iron oxide. Traditional image processing algorithms may not be able to accurately extract the characteristic information of the plate surface. Relying solely on two-dimensional image detection is prone to missed detection and false detection.

[0032] 3. Vibration interference during sheet transmission: In practical applications, sheet detection during transmission is often affected by vibration and deformation. The three-dimensional point cloud data obtained by line laser scanning often contains noise and distortion such as vibration information. Traditional point cloud to depth map conversion methods based on plane fitting are greatly affected by this and cannot meet the accuracy and quality requirements of depth map generation.

[0033] In view of this, the present invention provides a method and device for detecting the surface of a plate, which can solve the problems in the prior art of extracting accurate three-dimensional information of the plate surface from the raw data of the line laser sensor and the vibration interference existing in the plate during the transmission and detection process, and can obtain more accurate three-dimensional point cloud data, ensure the accuracy and reliability of plate surface detection, and improve the accuracy of obtaining plate surface data.

[0034] Next, the technical solution of the present invention is described in detail with reference to the accompanying drawings:

[0035] See also Figure 1 , Figure 1 This is a first flow chart of a method for detecting the surface of a plate provided by the present invention. The execution subject is a server or terminal device equipped with the technical solution disclosed in the embodiment of the present invention, such as a detection service platform or a handheld detection device.

[0036] exist Figure 1 In, the method may include:

[0037] Step 110: Obtain three-dimensional coordinate information data of the surface of the plate.

[0038] In step 110, a line laser sensor may be used to scan the surface of the plate to obtain three-dimensional information data of the plate surface under actual working conditions; wherein the data format of the three-dimensional information acquisition data of the metal plate surface detected by the line laser sensor is a CSV format file.

[0039] The present invention uses a high-precision line laser sensor to scan and obtain three-dimensional data of the plate surface, thereby achieving high-precision detection of the plate surface. For complex surface conditions of metal plates, rich three-dimensional morphological data of the plate surface can be obtained. For example, plate shape problems, defect problems, etc. can be efficiently detected, and the detection results are stable and reliable.

[0040] Step 120: performing data preprocessing on the three-dimensional coordinate information data to obtain first three-dimensional point cloud data; the data preprocessing includes filtering processing and / or point cloud coordinate conversion.

[0041] Step 130: performing rasterization processing on the first three-dimensional point cloud data, retaining target feature information in the first three-dimensional point cloud data, and obtaining second three-dimensional point cloud data.

[0042] Step 140: According to a preset defect filtering strategy, the second three-dimensional point cloud data is subjected to defect filtering processing to obtain the three-dimensional point cloud data of the plate type on the surface of the plate; the defect filtering processing includes one or more of downsampling, point cloud surface reconstruction and upsampling processing.

[0043] In step 120 to step 140, firstly, the three-dimensional information acquisition data of the metal plate surface detected by the warp laser sensor is converted to generate three-dimensional point cloud data of the actual surface of the metal plate, that is, the CSV format file is converted into a PCD format file; the three-dimensional point cloud data of the actual surface of the metal plate is further filtered and the point cloud coordinates are transformed, so as to eliminate the noise influence generated when transmitting and detecting the surface of the metal plate, and obtain the actual three-dimensional acquisition point cloud of the metal plate surface; the three-dimensional acquisition point cloud data is further rasterized, so as to make the three-dimensional acquisition point cloud data easier to process and analyze, and at the same time retain the main feature information, It provides a better foundation for subsequent 3D point cloud processing tasks; finally, the 3D rasterized point cloud is subjected to point cloud transformation such as filtering out defect factors, such as downsampling, point cloud surface reconstruction and upsampling, so as to filter out defect information on the surface of the metal plate and the inevitable influence of factors such as vibration of the plate during the detection process, and obtain a theoretical 3D point cloud of the actual plate shape of the metal plate surface; based on this, the present invention can eliminate factors such as vibration interference during the transmission of the plate, adapt to various challenging detection strategies in the industrial environment, and can accurately obtain 3D point cloud data of the actual surface of the plate during the transmission process, thereby ensuring the accuracy and reliability of plate surface detection.

[0044] Further, in step 120, data preprocessing is performed on the three-dimensional coordinate information data to obtain first three-dimensional point cloud data, which may include performing data preprocessing on the three-dimensional point cloud data in the following manner.

[0045] Method 1: convert the three-dimensional coordinate information data into three-dimensional point cloud coordinates to obtain third three-dimensional point cloud data; that is, convert the point cloud data into PCD format by loading the CSV file.

[0046] As an example, since the line laser sensor used generates a CSV file containing only the three-dimensional coordinate information of the metal sheet surface, its data structure type example is Therefore, it is necessary to process the data structure in the CSV file and convert the three-dimensional coordinate information (x, y, z) into the three-dimensional coordinates of each point cloud, so as to generate the three-dimensional acquisition point cloud data of the metal sheet surface.

[0047] Method 2: Use a straight-through filtering method to perform data filtering processing on the third three-dimensional point cloud data, filter out outlier point cloud data in the third three-dimensional point cloud data, and obtain fourth three-dimensional point cloud data; based on the surface area of ​​the plate, scale the coordinates of the fourth three-dimensional point cloud data according to the proportional scaling method to obtain fifth three-dimensional point cloud data, and the surface area corresponding to the fifth three-dimensional point cloud data is the surface area of ​​the plate.

[0048] Specifically, filtering and transforming the three-dimensional point cloud model data on the surface of the metal plate can eliminate the noise generated when transmitting and testing the surface of the metal plate, and obtain the actual three-dimensional point cloud collected on the surface of the metal plate. First, the third three-dimensional point cloud data is filtered by the straight-through filtering method, that is, the three-dimensional point cloud data collected on the surface of the metal plate is filtered directly, so as to filter out the outliers in the third three-dimensional point cloud data on a large scale and obtain the fourth three-dimensional point cloud data; the point cloud coordinates of the fourth three-dimensional point cloud data are further transformed to regularly scale the three-dimensional point cloud coordinates to the actual metal plate surface size lx×ly, and obtain the fifth three-dimensional point cloud data; wherein the surface area corresponding to the fifth three-dimensional point cloud data is the surface area of ​​the plate. Since the metal plate is inevitably placed crookedly during the actual transmission and detection, the three-dimensional point cloud is aligned in this step, which is conducive to the subsequent visualization observation and analysis of the three-dimensional point cloud data.

[0049] Method three, using a straight-through filtering method to perform data filtering processing on the fifth three-dimensional point cloud data, filtering out outlier point cloud data in the fifth three-dimensional point cloud data, and obtaining sixth three-dimensional point cloud data; based on the surface area of ​​the plate, the sixth three-dimensional point cloud data is scaled according to a proportional scaling method to obtain seventh three-dimensional point cloud data; the seventh three-dimensional point cloud data is determined as the first three-dimensional point cloud data.

[0050] In method three, based on the technical means of method two, fifth three-dimensional point cloud data consistent with the actual size of the plate is obtained; the fifth three-dimensional point cloud data is processed by straight-through filtering to collect outliers on the edge of the point cloud through small-scale filtering to obtain sixth three-dimensional point cloud data; the filtered point cloud coordinates are further regularized and scaled to the actual surface size of the metal plate to obtain seventh three-dimensional point cloud data; what is obtained at this time is the actual three-dimensional point cloud model data of the metal plate surface that can be used for subsequent analysis.

[0051] Furthermore, it is necessary to rasterize the 3D point cloud data to make it easier to process and analyze, while retaining the main feature information, providing a better foundation for subsequent 3D point cloud processing tasks.

[0052] Specifically, in step 130, the first three-dimensional point cloud data is rasterized, and the target feature information in the first three-dimensional point cloud data is retained to obtain the second three-dimensional point cloud data, which may include: projecting the first three-dimensional point cloud data onto the XOY plane to obtain the first two-dimensional point cloud data for the first three-dimensional point cloud data; creating a first grid point cloud on the XOY plane, wherein each intersection of the grids is a grid point cloud, and the step length between the grids is a first preset step length; using the kd tree algorithm to rasterize the first grid point cloud, taking any grid point cloud as a circle point, searching the first two-dimensional point cloud data according to a first preset radius, to obtain the first two-dimensional point cloud closest to the grid point cloud; determining the height value of the first two-dimensional point cloud closest to the grid point cloud as the height value of the grid point cloud, to obtain the second three-dimensional point cloud data.

[0053] As an example, the second three-dimensional point cloud data retaining the target feature information in the first three-dimensional point cloud data may be obtained by the following method.

[0054] First, the height value information of each point in the 3D acquisition point cloud data is saved, and then the 3D acquisition point cloud is projected on the XOY plane, and the total number of projected acquisition point clouds m0 is obtained, so that the subsequent 3D grid point cloud 1 can perform neighboring search and processing on the acquisition point cloud.

[0055] Secondly, create a three-dimensional grid point cloud 1 on the XOY plane, where each grid intersection is a point cloud and the step length between grids is the first preset step length; the first preset step length can be It can also be calculated based on historical data; C1 represents a constant used to increase stability.

[0056] Finally, the kd tree algorithm is used to efficiently implement rasterization processing. Each grid point cloud is used as a circle point, and a first preset radius, such as radius r0, is used to search for adjacent projected acquisition point clouds in the first two-dimensional point cloud data. Then, the height value of the collected nearest projected acquisition point cloud is assigned to the grid point cloud 1, so as to obtain a three-dimensional rasterized acquisition point cloud and its point cloud size m1, that is, the point cloud size of the second three-dimensional point cloud data is m1.

[0057] Furthermore, in step 140, the three-dimensional rasterized point cloud can be subjected to point cloud transformation such as filtering out defect factors, such as downsampling, point cloud surface reconstruction, and upsampling processing, so as to filter out defect information on the surface of the metal plate and the inevitable influence of factors such as vibration of the plate during the detection process, and obtain a theoretical three-dimensional point cloud of the actual plate shape of the metal plate surface.

[0058] Specifically, the defect filtering processing of the second three-dimensional point cloud data according to the preset defect filtering strategy to obtain the plate type three-dimensional point cloud data for the surface of the plate may include: downsampling the second three-dimensional point cloud data using a voxel grid filtering method to filter out defect data in the second three-dimensional point cloud data to obtain eighth three-dimensional point cloud data; based on the point cloud three-dimensional coordinates of the eighth three-dimensional point cloud data, reconstructing the point cloud surface of the eighth three-dimensional point cloud data using the least moving squares method to obtain ninth three-dimensional point cloud data; and resampling and remodeling the ninth three-dimensional point cloud data using a kd tree algorithm model to obtain the plate type point cloud data on the surface of the plate.

[0059] As an example, the plate point cloud data of the plate surface can be obtained by the following method.

[0060] First, the second three-dimensional point cloud data is downsampled using the voxel grid filtering method, so as to quickly process the rasterized point cloud data and filter out defective information in the rasterized point cloud on a large scale, and at the same time obtain the downsampled point cloud size m2, that is, the size of the eighth three-dimensional point cloud data is m2.

[0061] Secondly, the point cloud surface is reconstructed by using the least moving squares method based on the downsampled 8th 3D point cloud data. The adjacent point information is searched for by the search radius, and then the three-dimensional point cloud surface model with the defect information filtered out is reconstructed to obtain the model point cloud size m3, that is, the point cloud size of the ninth three-dimensional point cloud data is m3.

[0062] Finally, the kd tree algorithm is used to upsample the ninth three-dimensional point cloud data to reconstruct the three-dimensional point cloud surface model, and the height value z2 of each point cloud after the data sampling and reconstruction of the ninth three-dimensional point cloud data is calculated. Each rasterized acquisition point cloud in the ninth three-dimensional point cloud data is taken as a circle point, and the search radius is To search for the reconstructed model point cloud data in the neighborhood of the point, the formula can be used:

[0063]

[0064] Calculate the height values ​​of each point cloud after data sampling and remodeling of the ninth three-dimensional point cloud data; wherein z2 represents the height values ​​of each point cloud after data sampling and remodeling of the ninth three-dimensional point cloud data, z0 is the height value of the most recently reconstructed model point cloud searched, and z1 is the average value of all searched point cloud height values.

[0065] Furthermore, for each rasterized point cloud, the formula can be used:

[0066]

[0067] Correct the coordinate z value; where t0 is the defect depth threshold of the rasterized acquisition point cloud.

[0068] It should be noted that in step 110 to step 140, the various types of three-dimensional point cloud data generated in the above steps are also visualized and saved. Visualization allows us to intuitively observe the actual conditions of the metal sheet surface at each stage of the detection process, making it easier to understand and analyze the data.

[0069] By using the above sampling and reconstruction model method, the plate type point cloud data of the plate surface is obtained, and the size of the three-dimensional point cloud is consistent with the size of the original three-dimensional rasterized acquisition point cloud; thereby achieving more accurate three-dimensional point cloud data for the plate surface, ensuring the accuracy and reliability of plate surface detection, and improving the accuracy of obtaining plate surface data; solving the problem that the prior art relies on two-dimensional image detection and prone to missed detection and false detection, as well as the limitations in processing transmission vibration interference, resulting in poor accuracy of the acquired plate surface detection data; and the method and equipment proposed in the present invention only require the use of a line laser sensor to scan and obtain metal plate surface information, with a simple and efficient design, suitable for deployment in industrial environments, and rapid detection of metal plate surface defects in terms of shape, position, defect depth, etc., improving the flexibility and efficiency of detection, reducing process costs, and providing an effective technical solution for the problem of metal plate surface detection.

[0070] Furthermore, after obtaining the 3D point cloud data of the plate surface with high accuracy and reliability, a high-quality 2D depth image can be created. Figure 2 , Figure 2 A second flow chart of a method for detecting a surface of a plate material provided by the present invention.

[0071] exist Figure 2 In, the method may include:

[0072] Step 210: Create three-dimensional defect depth point cloud data based on the second three-dimensional point cloud data and the three-dimensional point cloud data of the plate surface.

[0073] In step 210, based on the second three-dimensional point cloud data generated in steps 110 to 140 and the three-dimensional point cloud data of the plate surface, the point cloud data height difference between the second three-dimensional point cloud data and the three-dimensional point cloud data of the plate surface is determined to generate a three-dimensional defect depth point cloud.

[0074] It should be noted that the three-dimensional defect depth point cloud may only include defect information on the surface of the metal sheet, such as shape and size information, position information, and depth information relative to the surface of the metal sheet.

[0075] Step 220: rasterizing the three-dimensional defect depth point cloud data to obtain a tenth three-dimensional point cloud data.

[0076] In step 220, the three-dimensional defect depth point cloud data can be rasterized to obtain the tenth three-dimensional point cloud data, making the three-dimensional acquisition point cloud data easier to process and analyze, and providing a better basis for the subsequent two-dimensional grayscale image conversion processing of the three-dimensional defect depth point cloud.

[0077] Step 230: Based on the tenth three-dimensional point cloud data, image preprocessing is performed on the tenth three-dimensional point cloud data according to a preset grayscale image processing strategy to obtain a two-dimensional grayscale image of the surface of the plate.

[0078] In step 230, the three-dimensional rasterized defect depth point cloud can be converted into a two-dimensional grayscale image, so as to convert the three-dimensional defect information on the surface of the metal plate into a two-dimensional grayscale image, which is further beneficial for intuitively studying and analyzing the defect information and for using the grayscale image in subsequent machine vision and deep learning technology fields.

[0079] Through steps 210 to 230, the present invention can directly generate a high-quality two-dimensional depth image that reflects the actual surface condition of the plate, which can effectively overcome many limitations in traditional two-dimensional image detection, provide high-quality two-dimensional depth images for multimodal detection of plate surface quality, and provide more reliable and effective data support for applications in fields such as industrial manufacturing and visual inspection.

[0080] Further, in step 210, since the point cloud sizes of the second three-dimensional point cloud data obtained in step 140 and the plate type three-dimensional point cloud data of the plate surface are consistent, and only the z values ​​of each point cloud are different, the z values ​​of the two point clouds are differentiated to obtain a three-dimensional defect depth point cloud containing only the defect information of the metal plate surface.

[0081] As an example, establishing three-dimensional defect depth point cloud data based on the second three-dimensional point cloud data and the plate type three-dimensional point cloud data of the plate surface may include: determining first position information of a first point cloud in the second three-dimensional point cloud data based on the second three-dimensional point cloud data; determining a second point cloud in the plate type three-dimensional point cloud data of the plate surface that is identical to the first position information based on the first position information and the plate type three-dimensional point cloud data of the plate surface; calculating the height difference between multiple first point clouds and multiple second point clouds to obtain a set of height differences between the second three-dimensional point cloud data and each point cloud in the plate type three-dimensional point cloud data of the plate surface; establishing the three-dimensional defect depth point cloud data based on the set of height differences between the second three-dimensional point cloud data and each point cloud in the plate type three-dimensional point cloud data of the plate surface.

[0082] Further, in step 220, the three-dimensional defect depth point cloud data is rasterized to obtain a tenth three-dimensional point cloud data, which may include: projecting the three-dimensional defect depth point cloud data onto the XOY plane to obtain second two-dimensional defect point cloud data for the three-dimensional defect depth point cloud data; creating a second grid point cloud on the XOY plane, wherein each intersection of the grids is a grid point cloud, and the step length between the grids is a second preset step length; using the kd tree algorithm to rasterize the second grid point cloud, including: taking any one of the grid point clouds as a circle point, searching for the second two-dimensional defect point cloud data projected on the XOY plane according to a second preset radius, to obtain the two-dimensional defect point cloud closest to the grid point cloud; using the height value of the two-dimensional defect point cloud closest to the grid point cloud as the depth value of the grid point cloud, to obtain the tenth three-dimensional point cloud data.

[0083] As an example, the height value information of each point cloud in the three-dimensional defect depth point cloud data can be first saved, and then the three-dimensional defect depth point cloud can be projected onto the XOY plane to obtain a second two-dimensional defect point cloud data for the three-dimensional defect depth point cloud data, and the total number of collected point clouds m4 of the projection can be obtained.

[0084] Secondly, a grid point cloud 2 is created on the XOY plane, where each grid intersection is a point cloud, and the grid step is the second preset step step1; each grid point represents a pixel point.

[0085] Finally, the kd tree algorithm is used to efficiently implement rasterization processing, with each grid point cloud as a circle and a radius of Search for the adjacent projected defect depth point cloud for the search radius, and then use the height value of the nearest projected defect depth point cloud collected as the depth value h of the grid point to obtain the rasterized defect depth point cloud data, that is, the tenth three-dimensional point cloud data; wherein the rasterized defect depth point cloud size is m5, the size is lx'×ly'; the maximum value of the depth value in the grid point cloud is z_max, the minimum value is z_min, and the difference is lz'. It should be noted that in order to facilitate the subsequent grayscale image conversion, the depth value h can be converted into a grayscale value.

[0086] Further, in step 230, it is necessary to convert the rasterized defect depth point cloud data into a two-dimensional grayscale image; that is, the tenth three-dimensional point cloud data is image preprocessed according to a preset grayscale image processing strategy to obtain a two-dimensional grayscale image of the surface of the plate, which may include: based on the tenth three-dimensional point cloud data, using an open source computer vision library to perform image preprocessing on the tenth three-dimensional point cloud data to generate a two-dimensional grayscale image of the surface of the plate; the image preprocessing includes adjusting any one of the image plane resolution pixel size, the image resolution size, and the grayscale value of each pixel in the image.

[0087] As an example, OpenCV can be used to generate and visualize a two-dimensional grayscale image of the plate surface, that is, the image preprocessing of the tenth three-dimensional point cloud data is performed according to a preset grayscale image processing strategy to obtain a two-dimensional grayscale image of the plate surface. The formula can be used:

[0088]

[0089] Determine the plane resolution pixel size p0 of the image; wherein lx' represents the horizontal axis size of the tenth dimensional point cloud data, ly' represents the vertical axis size of the tenth dimensional point cloud data, and m5 represents the size of the tenth dimensional point cloud data.

[0090] The formula can be used:

[0091]

[0092]

[0093] Determine the resolution size W and H of the image; wherein W represents the width of the image resolution, H represents the length of the image resolution, lx' represents the horizontal axis size of the tenth three-dimensional point cloud data, ly' represents the vertical axis size of the tenth three-dimensional point cloud data, and p0 represents the plane resolution pixel size of the image.

[0094] A two-dimensional grayscale image is generated by using formula (3) to formula (5). For each pixel in the two-dimensional grayscale image, the formula can be used:

[0095]

[0096] Determine the grayscale value of each pixel; where g represents the grayscale value of the pixel, h represents the depth value of the grid point, z_min represents the minimum depth value of all grid points in the tenth three-dimensional point cloud data, and lz' represents the difference between the maximum and minimum depth values ​​of all grid points in the tenth three-dimensional point cloud data.

[0097] Based on this, the three-dimensional rasterized defect depth point cloud is converted into a two-dimensional grayscale image to generate a high-quality two-dimensional depth image. The present invention directly generates a high-quality two-dimensional depth image that reflects the actual surface of the plate, which can effectively overcome many limitations in traditional two-dimensional image detection, provide high-quality two-dimensional depth images for multimodal detection of plate surface quality, and provide more reliable and effective data support for applications in fields such as industrial manufacturing and visual inspection.

[0098] It should be noted that, in step 210 to step 230, it also includes visually displaying and saving various three-dimensional point cloud data models generated in the above steps and the final two-dimensional grayscale image. This can not only detect the plate shape problems on the surface of the metal plate in real time, but also make the surface quality problems of the metal plate more specific, comprehensive, intuitive and clear, and clear at a glance. At the same time, it also greatly helps to analyze and understand the surface defect characteristics of the depth map, help to improve the contrast of the image, improve the efficiency of metal plate surface defect detection, and reduce detection costs.

[0099] In steps 110 to 140, and steps 210 to 230, the present invention makes full use of the rich functions of three-dimensional point cloud data processing, including data processing, filtering reconstruction, fitting transformation and visualization, etc., which provides strong support for the efficient realization of the generation of three-dimensional point cloud to depth map of the plate surface considering the transmission vibration interference in the detection method, and enables the operator to intuitively observe each stage of the detection process and find problems in time, thereby improving the accuracy and efficiency of detection. In addition, the present invention provides a new technical solution for the surface multimodal detection of the plate, which can obtain the three-dimensional point cloud model of the metal plate surface, the three-dimensional defect point cloud model that filters out the influence of the plate shape and the detection environment factors, and the two-dimensional grayscale map of the defect, thereby reducing the detection cost, improving the accuracy, reliability, applicability and efficiency of the detection, and providing new ideas and solutions for upgrading the metal plate surface detection method.

[0100] Furthermore, in order to fully describe the plate surface detection data that can be efficiently and accurately obtained by using the method of the present invention, the present invention also provides the following evaluation indicators, which are used to quantitatively analyze the grayscale image generated by the method model disclosed in this specification.

[0101] Evaluation indicator 1: Mean Squared Error (MSE): The average value of the square of the pixel difference between two images. The calculation formula is as follows:

[0102]

[0103] Where M and N represent the height and width of the image, and I1(i, j) and I2(i, j) represent the pixel values ​​of image 1 and image 2 at position (i, j).

[0104] Evaluation indicator 2: Peak Signal-to-Noise Ratio (PSNR): A common indicator for measuring image quality. The higher the value, the better the image quality. It is usually measured in decibels (dB). Generally speaking, a PSNR value between 20dB and 30dB is considered a good result. The mathematical formula is as follows:

[0105]

[0106] Among them, MAX represents the maximum possible value of the image pixel (usually 255 in an 8-bit grayscale image), and MSE represents the mean square error of the image.

[0107] Evaluation indicator 3: Structural Similarity Index (SSIM): An indicator used to measure the structural similarity of two images. It takes into account three aspects: brightness, contrast, and structure. Its value is between -1 and 1. The closer it is to 1, the more similar the two images are. Generally speaking, an SSIM value greater than 0.9 is considered a good result. Its mathematical expression is:

[0108]

[0109] Among them, x and y represent the local windows of the two images, μ x and μ y represents the mean of x and y, σ x and σ y represents the variance of x and y, σ xy represents the covariance of x and y, C1 and C2 represent constants used to increase stability.

[0110] In order to more clearly show that the technical solution provided by the present invention can improve the accuracy of obtaining the surface data of the plate, thereby generating a higher quality two-dimensional depth map; please refer to Figures 3 to 6 , Figure 3 A two-dimensional image of the actual surface of a plate is involved in a plate surface detection method provided by the present invention; Figure 4 A first processing effect diagram of a plate surface detection method provided by the present invention; Figure 5 A second processing effect diagram of a plate surface detection method provided by the present invention; Figure 6 This is a diagram showing the third processing effect of a method for detecting the surface of a plate material provided by the present invention.

[0111] exist Figure 3 middle, Figure 3 This is an image of the plate surface taken by an industrial camera. The white marks in the image are defective areas on the plate surface. Figure 4 The picture shown is a schematic diagram of a two-dimensional image obtained by processing the three-dimensional information acquisition data of the metal plate surface detected by the warp laser sensor using the technical means 1 described in Table 1; the technical means 1 is the technical means of converting the loaded CSV file into point cloud data in step 120 of this specification, and the technical means of converting the three-dimensional rasterized defect depth point cloud into a two-dimensional grayscale image in step 230; Figure 5 The picture shown is a schematic diagram of a two-dimensional image obtained by processing the three-dimensional information collection data of the metal plate surface detected by the warp laser sensor using the technical means 2 described in Table 1; the technical means 2 is the technical means disclosed in step 120 and step 230 disclosed in this specification; Figure 6The picture shown is a schematic diagram of a two-dimensional image obtained by processing the three-dimensional information collection data of the metal plate surface detected by the warp laser sensor using the technical means 3 described in Table 1; the technical means 3 is the technical means of using steps 110 to 140 and steps 210 to 230 in this specification.

[0112] right Figures 4 to 5 Evaluation indicators 1-3 were used to evaluate and the evaluation results shown in Table 1 were obtained.

[0113] Table 1 shows the quantitative analysis of different processing methods for the three-dimensional point cloud data collected on the plate surface.

[0114]

[0115]

[0116] Further, according to Figures 3 to 6 , combined with Table 1, we can get: Figure 4 The technical means 1 used in the above is to directly generate the defect depth map from the collected three-dimensional point cloud data of the plate surface. Obviously, it is unable to filter out the problems of detection vibration, plate shape, and uneven placement of plates during detection; Figure 5 The technical means 2 used in the present application is the effect of generating the sink depth map by using the conventional processing methods such as simple filtering, which obviously still cannot filter out the problems such as vibration and plate shape during detection; Figure 6 The technical means 3 used in the text is the effect of the sink depth map generated by the method of the present application. It obviously and effectively filters out the problems such as vibration and plate shape during detection, and can more accurately obtain the three-dimensional point cloud data of the plate surface, thereby improving the accuracy of obtaining the plate surface data.

[0117] In the second aspect, based on the same technical concept as the first aspect, the present invention also provides a detection device for the surface of a plate. Figure 7 , Figure 7 A schematic structural diagram of a detection device for a plate surface provided by the present invention.

[0118] exist Figure 7 In the embodiment, the device may include:

[0119] An acquisition module 710, wherein the acquisition module 710 is used to acquire three-dimensional coordinate information data of the surface of the plate;

[0120] A preprocessing module 720, the preprocessing module 720 is used to perform data preprocessing on the three-dimensional coordinate information data to obtain first three-dimensional point cloud data; the data preprocessing includes filtering processing and / or point cloud coordinate conversion;

[0121] A rasterization module 730, wherein the rasterization module 730 is used to perform rasterization processing on the first three-dimensional point cloud data, retain target feature information in the first three-dimensional point cloud data, and obtain second three-dimensional point cloud data;

[0122] The defect filtering module 740 is used to perform defect filtering processing on the second three-dimensional point cloud data according to a preset defect filtering strategy to obtain the plate type three-dimensional point cloud data for the surface of the plate; the defect filtering processing includes one or more of downsampling, point cloud surface reconstruction and upsampling processing.

[0123] The present invention proposes a detection device for the surface of a plate material, which obtains the three-dimensional coordinate information data of the surface of the plate material through an acquisition module, firstly uses a preprocessing module to perform data preprocessing on the three-dimensional coordinate information data to obtain first three-dimensional point cloud data; the data preprocessing includes filtering processing and / or point cloud coordinate conversion; further uses a rasterization module to perform rasterization processing on the first three-dimensional point cloud data, retains the target feature information in the first three-dimensional point cloud data, and obtains second three-dimensional point cloud data; finally, uses a defect filtering module to perform defect filtering processing on the second three-dimensional point cloud data according to a preset defect filtering strategy to obtain the plate type three-dimensional point cloud data of the plate surface; the defect filtering processing includes one or more of downsampling, point cloud surface reconstruction and upsampling processing; it is realized that rich three-dimensional morphological data of the plate surface can be stably and reliably obtained, and by performing filtering processing, point cloud coordinate conversion, rasterization processing and defect filtering processing on the three-dimensional point cloud data, interference and other factors in the plate transmission process can be eliminated; the three-dimensional point cloud data of the plate surface can be obtained more accurately, the accuracy and reliability of plate surface detection are ensured, and the accuracy of obtaining plate surface data is improved.

[0124] Although the present invention is described herein in conjunction with various embodiments, in the process of implementing the claimed invention, those skilled in the art may understand and implement other variations of the disclosed embodiments by viewing the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other components or steps, and "one" or "an" does not exclude multiple situations. A single processor or other unit may implement several functions listed in the claims. Certain measures are recorded in different dependent claims, but this does not mean that these measures cannot be combined to produce good results.

[0125] Although the present invention has been described in conjunction with specific features and embodiments thereof, it is apparent that various modifications and combinations may be made thereto without departing from the spirit and scope of the present invention. Accordingly, this specification and the accompanying drawings are merely exemplary illustrations of the present invention as defined by the appended claims and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present invention. Obviously, those skilled in the art may make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, the present invention is intended to include such modifications and variations if they fall within the scope of the claims of the present invention and their equivalents.

Claims

1. A method for detecting the surface of a plate, characterized in that: include: Obtaining three-dimensional coordinate information data of the plate surface; Performing data preprocessing on the three-dimensional coordinate information data to obtain first three-dimensional point cloud data; The data preprocessing includes filtering processing and / or point cloud coordinate conversion; rasterizing the first three-dimensional point cloud data, retaining target feature information in the first three-dimensional point cloud data, and obtaining second three-dimensional point cloud data; According to a preset defect filtering strategy, the second three-dimensional point cloud data is subjected to defect filtering processing to obtain the plate type three-dimensional point cloud data for the surface of the plate; the defect filtering processing includes one or more of downsampling, point cloud surface reconstruction and upsampling processing.

2. The method according to claim 1, characterized in that The step of preprocessing the three-dimensional coordinate information data to obtain first three-dimensional point cloud data includes: Performing three-dimensional point cloud coordinate conversion on the three-dimensional coordinate information data to obtain third three-dimensional point cloud data; Performing data filtering processing on the third three-dimensional point cloud data by adopting a straight-through filtering method to filter out outlier point cloud data in the third three-dimensional point cloud data to obtain fourth three-dimensional point cloud data; Based on the surface area of ​​the plate, scaling the fourth three-dimensional point cloud data according to a proportional scaling method to obtain fifth three-dimensional point cloud data, wherein the surface area corresponding to the fifth three-dimensional point cloud data is the surface area of ​​the plate; Performing data filtering processing on the fifth three-dimensional point cloud data by adopting a straight-through filtering method to filter out outlier point cloud data in the fifth three-dimensional point cloud data, thereby obtaining sixth three-dimensional point cloud data; Based on the surface area of ​​the plate, the sixth three-dimensional point cloud data is scaled in accordance with a proportional scaling method to obtain seventh three-dimensional point cloud data; and the seventh three-dimensional point cloud data is determined as the first three-dimensional point cloud data.

3. The method according to claim 1, characterized in that The rasterizing the first three-dimensional point cloud data to retain target feature information in the first three-dimensional point cloud data to obtain second three-dimensional point cloud data includes: Projecting the first three-dimensional point cloud data onto an XOY plane to obtain first two-dimensional point cloud data for the first three-dimensional point cloud data; Creating a first grid point cloud on the XOY plane, wherein each intersection of the grids is a grid point cloud, and the step length between the grids is a first preset step length; The first grid point cloud is rasterized using a kd tree algorithm, including: taking any grid point cloud as a circle, searching the first two-dimensional point cloud data according to a first preset radius to obtain the first two-dimensional point cloud closest to the grid point cloud; determining the height value of the first two-dimensional point cloud closest to the grid point cloud as the height value of the grid point cloud to obtain the second three-dimensional point cloud data.

4. The method according to claim 1, characterized in that The step of performing defect filtering processing on the second three-dimensional point cloud data according to a preset defect filtering strategy to obtain the three-dimensional point cloud data of the plate type on the surface of the plate includes: downsampling the second three-dimensional point cloud data by using a voxel grid filtering method, filtering out defective data in the second three-dimensional point cloud data, and obtaining eighth three-dimensional point cloud data; Based on the three-dimensional coordinates of the point cloud of the eighth three-dimensional point cloud data, a least moving squares method is used to reconstruct a point cloud surface of the eighth three-dimensional point cloud data to obtain ninth three-dimensional point cloud data; The kd tree algorithm model is used to perform data sampling and remodeling on the ninth three-dimensional point cloud data to obtain the plate type point cloud data of the plate surface.

5. The method according to claim 4, characterized in that The method of performing data sampling and remodeling on the ninth three-dimensional point cloud data by using the kd tree algorithm model includes: Using the formula: Calculate the height values ​​of each point cloud after data sampling and remodeling of the ninth three-dimensional point cloud data; wherein z2 represents the height values ​​of each point cloud after data sampling and remodeling of the ninth three-dimensional point cloud data, z0 is the height value of the most recently reconstructed model point cloud searched, and z1 is the average value of all searched point cloud height values.

6. The method according to claim 1, characterized in that Also includes: Based on the second three-dimensional point cloud data and the three-dimensional point cloud data of the plate shape on the surface of the plate, three-dimensional defect depth point cloud data is established; rasterizing the three-dimensional defect depth point cloud data to obtain a tenth three-dimensional point cloud data; Based on the tenth three-dimensional point cloud data, image preprocessing is performed on the tenth three-dimensional point cloud data according to a preset grayscale image processing strategy to obtain a two-dimensional grayscale image of the surface of the plate.

7. The method according to claim 6, characterized in that The step of establishing three-dimensional defect depth point cloud data based on the second three-dimensional point cloud data and the three-dimensional point cloud data of the plate surface comprises: Based on the second three-dimensional point cloud data, determining first position information of a first point cloud in the second three-dimensional point cloud data; Based on the first position information and the three-dimensional point cloud data of the plate surface, determining a second point cloud in the three-dimensional point cloud data of the plate surface that is identical to the first position information; Calculating the height differences between the first point clouds and the second point clouds to obtain a set of height differences between the second three-dimensional point cloud data and each point cloud in the three-dimensional point cloud data of the plate surface; The three-dimensional defect depth point cloud data is established based on a set of height differences between the second three-dimensional point cloud data and each point cloud in the three-dimensional point cloud data of the plate surface.

8. The method according to claim 6, characterized in that The rasterizing process of the three-dimensional defect depth point cloud data to obtain the tenth three-dimensional point cloud data includes: Projecting the three-dimensional defect depth point cloud data onto an XOY plane to obtain second two-dimensional defect point cloud data for the three-dimensional defect depth point cloud data; Creating a second grid point cloud on the XOY plane, wherein each grid intersection is a grid point cloud, and the step length between grids is a second preset step length; The second grid point cloud is rasterized using the kd tree algorithm, including: taking any one grid point cloud as a circle, searching for the second two-dimensional defect point cloud data projected on the XOY plane according to a second preset radius, and obtaining the two-dimensional defect point cloud closest to the grid point cloud; using the height value of the two-dimensional defect point cloud closest to the grid point cloud as the depth value of the grid point cloud, and obtaining the tenth three-dimensional point cloud data.

9. The method according to claim 6, characterized in that The image preprocessing of the tenth three-dimensional point cloud data according to a preset grayscale image processing strategy to obtain a two-dimensional grayscale image of the surface of the plate includes: Based on the tenth 3D point cloud data, an open source computer vision library is used to perform image preprocessing on the tenth 3D point cloud data to generate a two-dimensional grayscale image of the surface of the plate; the image preprocessing includes adjusting any one of the image plane resolution pixel size, the image resolution size, and the grayscale value of each pixel in the image.

10. A detection device for the surface of a plate, characterized in that: include: An acquisition module, the acquisition module is used to acquire three-dimensional coordinate information data of the surface of the plate; A preprocessing module, the preprocessing module is used to perform data preprocessing on the three-dimensional coordinate information data to obtain first three-dimensional point cloud data; The data preprocessing includes filtering processing and / or point cloud coordinate conversion; A rasterization module, the rasterization module is used to perform rasterization processing on the first three-dimensional point cloud data, retain the target feature information in the first three-dimensional point cloud data, and obtain second three-dimensional point cloud data; A defect filtering module, wherein the defect filtering module is used to perform defect filtering processing on the second three-dimensional point cloud data according to a preset defect filtering strategy to obtain the plate type three-dimensional point cloud data for the surface of the plate; the defect filtering processing includes one or more of downsampling, point cloud surface reconstruction and upsampling processing.

Images