A method for calculating paint coating area based on 3D scanning technology

Abstract

The invention discloses a method for calculating the area of ​​paint based on three-dimensional scanning technology, which belongs to the technical field of paint area calculation. The method comprises: based on object surface point cloud data obtained by a three-dimensional laser scanner, a point cloud processing algorithm is used to construct a point cloud model of the object surface, surface shape characteristics and reflection intensity characteristics are analyzed, the point cloud boundary of the paint area is extracted, a point cloud model of the paint surface is established, a normal vector of a triangular network model is calculated by a differential geometry method, and a region is segmented to divide the surface into a plane region and a curved surface region. The area of ​​the plane region of the paint is calculated by accumulation, and the area of ​​the curved surface region is obtained by a numerical integration method. Small feature regions are extracted based on a convolutional network, and a local point cloud surface is reconstructed by an Alpha shape algorithm to calculate the small feature area. The total paint area of ​​the painted object is generated by comprehensively calculating the areas of the plane region, the curved surface region and the small feature region of the paint.

Description

Technical Field

[0001] The invention belongs to the technical field of paint coating area calculation, and in particular relates to a paint coating area calculation method based on three-dimensional scanning technology. Background Art

[0002] Accurate calculation of paint usage is crucial to controlling cost and quality. Traditional methods of calculating paint area rely on manual measurement and empirical estimation, which is inefficient and has large errors. Using 3D scanning technology to obtain point cloud information on the surface of an object can achieve fast and accurate calculation of the paint area. However, the surface of large objects is complex, with a large number of detailed features such as curved surfaces, grooves, bolt holes, etc., which makes the point cloud data huge and unevenly distributed.

[0003] Calculating the surface area directly based on the original point cloud is not only inefficient, but also difficult to ensure the accuracy of the calculation. How to extract key feature information from massive point cloud data and design an efficient and stable surface area calculation algorithm is a key technical problem that needs to be solved urgently. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a method for calculating the paint coating area based on three-dimensional scanning technology to solve the problems existing in the above-mentioned prior art.

[0005] To achieve the above object, the present invention provides a method for calculating the paint coating area based on three-dimensional scanning technology, comprising:

[0006] Acquire surface point cloud data of the object after painting using a three-dimensional laser scanner, calculate the surface point cloud data based on a point cloud preprocessing algorithm, and obtain a surface point cloud model of the object;

[0007] Extracting the paint area point cloud boundary of the object surface point cloud model according to the object surface shape characteristics and reflection intensity characteristics to obtain the paint surface point cloud model;

[0008] Constructing a triangulated network model of the painted object according to the paint surface point cloud model;

[0009] Calculate the normal vector of each triangular face of the triangular network model according to the differential geometry method, perform region segmentation on the triangular face based on the normal vector, and generate a paint coating plane region and a paint coating curved surface region;

[0010] Accumulating the areas of the triangular facets of the paint coating plane region to obtain the total area of ​​the paint coating plane region, and calculating the total area of ​​the paint coating curved surface region according to a numerical integration method;

[0011] Extracting feature areas of the surface point cloud data based on a convolutional network model to obtain fine feature areas of paint and coatings, reconstructing local point cloud surfaces of the fine feature areas of paint and coatings based on an Alpha shape algorithm, and calculating fine feature areas of paint and coatings based on the local point cloud surfaces;

[0012] The paint area is generated according to the total area of ​​the paint plane area, the total area of ​​the paint curved surface area and the small feature area of ​​the paint.

[0013] Preferably, the process of calculating the surface point cloud data based on a point cloud preprocessing algorithm to obtain a point cloud model of the object surface includes:

[0014] Use a 3D laser scanner to scan the painted object and obtain 3D point cloud data of the surface;

[0015] Statistically analyzing the features of the three-dimensional point cloud data of the surface, determining the judgment threshold of noise points and outliers, and if the feature values ​​of some points in the point cloud data exceed the threshold range, they are judged as noise points or outliers;

[0016] The noise points or outliers are removed to obtain a point cloud model of the object surface.

[0017] Preferably, the process of removing the noise points or outliers includes:

[0018] The noise points or outliers are processed based on a filtering algorithm. During the filtering process, the neighborhood points of each point are analyzed. If the point is determined to be a noise point, the noise point is removed from the point cloud data.

[0019] The point cloud data is divided into multiple clusters through clustering. If the distance between a point and all clusters exceeds the threshold, it is judged as an outlier and is removed from the point cloud data.

[0020] Preferably, the process of obtaining the paint surface point cloud model includes: using the surface shape characteristics and reflection intensity characteristics of the object, targeting the material and color differences between the paint and other medium surfaces, using a point cloud algorithm based on machine learning to identify and filter, extracting the point cloud boundaries of the paint area, and obtaining the paint surface point cloud model.

[0021] Preferably, the process of constructing a triangulated network model of the painted object according to the object surface point cloud model comprises:

[0022] Processing the object surface point cloud model according to a Poisson reconstruction algorithm to construct an initial high-precision triangular mesh model;

[0023] Calculating an error metric value for each vertex in the initial high-precision triangular mesh model, and determining a vertex that can be simplified based on the error metric value;

[0024] According to the simplifiable vertices, an edge collapse algorithm is used to merge adjacent triangular facets to obtain a triangular network model of the painted object.

[0025] Preferably, the process of merging adjacent triangular facets using the edge collapse algorithm includes: setting a simplification threshold during the mesh simplification process, controlling the degree of simplification of each edge collapse operation, performing error evaluation on the simplified triangular mesh model, calculating the geometric deviation between the two models before and after simplification, and if the deviation exceeds a preset threshold, adjusting the simplification parameters and re-simplifying the mesh.

[0026] Preferably, the process of performing region segmentation on the triangular facet based on the normal vector to generate a paint coating plane region and a paint coating curved surface region comprises:

[0027] According to the topological structure of the triangular mesh model, a normal vector of each triangular face is calculated using a differential geometry method;

[0028] According to the predetermined normal vector angle threshold and the normal vector division of each triangular facet, a paint coating plane area and a paint coating curved surface area are generated.

[0029] Preferably, the process of generating the paint coating plane area and the paint coating curved surface area according to the predetermined normal vector angle threshold and the normal vector division of each triangular facet comprises:

[0030] According to a predetermined normal vector angle threshold, determine whether the normal vector angle of adjacent triangular facets is less than the threshold. If the normal vector angle of adjacent triangular facets is less than the threshold, these facets are classified into the same area and marked as a plane sub-area. If the normal vector angle of adjacent triangular facets is greater than or equal to the threshold, these facets are classified into different areas and marked as a surface sub-area.

[0031] The paint coating plane area and the paint coating surface area are subdivided by clustering algorithm to obtain fine sub-areas;

[0032] According to the normal vector distribution characteristics of each sub-region, the principal direction of each sub-region is extracted by principal component analysis method;

[0033] According to the main direction and normal vector distribution characteristics, each sub-region is semantically annotated to obtain the paint plane region and paint curved surface region on the object surface.

[0034] Preferably, the process of extracting the feature area of ​​the surface point cloud data based on the convolutional network model to obtain the fine feature area of ​​the paint coating comprises:

[0035] Converting the surface point cloud data into voxelized point cloud data;

[0036] Extracting the voxelized point cloud data based on the sliding window mechanism of the convolutional network model to obtain features of the local area;

[0037] The characteristic boundaries of the local area are processed based on an image segmentation algorithm to obtain the fine characteristic area of ​​the paint coating.

[0038] Preferably, the process of reconstructing the local point cloud surface of the fine feature area of ​​the paint coating based on the Alpha shape algorithm and calculating the fine feature area of ​​the paint coating based on the local point cloud surface comprises:

[0039] Determine a feature area according to fine features on the surface of the object, and extract a local point cloud subset of the feature area from the original point cloud data;

[0040] Analyzing the point cloud density and feature size of the local point cloud subset of the feature area to determine the Alpha value in the Alpha shape algorithm;

[0041] Performing surface reconstruction on the local point cloud subset based on the Alpha value to obtain a local surface model representing small features;

[0042] The surface area of ​​the local surface model is calculated to generate a fine feature area of ​​the paint coating.

[0043] Compared with the prior art, the present invention has the following advantages and technical effects:

[0044] The present invention can effectively remove noise and external points in the point cloud through a point cloud preprocessing algorithm, simplify the point cloud data, and unify the point clouds scanned at different angles into the same coordinate system, providing a robust data basis for subsequent surface construction and three-dimensional solid model generation. The paint coating area is extracted by the surface shape characteristics and reflection intensity characteristics of the object, and the paint boundary can be identified more accurately. The region segmentation based on the normal vector can accurately distinguish the plane area and the surface area of ​​the paint coating. By calculating the area of ​​the plane area and the surface area respectively, the total area of ​​the paint coating can be obtained more accurately. The point cloud is essentially an unordered set of points. The convolutional network model of the present invention learns the spatial structure of the point cloud by constructing the relationship (edge) between the points, which is particularly suitable for processing complex point cloud data. The convolutional network model can effectively solve the arrangement invariance and density non-uniformity of the point cloud, and can learn the local geometric features between the points.

[0045] The present invention comprehensively considers the complexity of the object surface and the limitations of data acquisition. Through the coordinated application of multiple algorithms, it realizes high-precision calculation of the surface area of ​​various complex objects, provides reliable data support for subsequent spraying operations, and has important engineering application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0047] Figure 1 This is a flow chart of a method for calculating the paint coating area based on three-dimensional scanning technology according to an embodiment of the present invention. DETAILED DESCRIPTION

[0048] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0049] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0050] Embodiment 1

[0051] like Figure 1 As shown, this embodiment provides a method for calculating the paint coating area based on three-dimensional scanning technology, including:

[0052] S101, obtaining surface point cloud data of the painted object using a three-dimensional laser scanner, and calculating the surface point cloud data based on a point cloud preprocessing algorithm to obtain a surface point cloud model of the object;

[0053] S102, extracting the paint area point cloud boundary of the object surface point cloud model according to the object surface shape characteristics and reflection intensity characteristics, to obtain the paint surface point cloud model;

[0054] S103, constructing a triangulated network model of the painted object according to the paint surface point cloud model;

[0055] S104, calculating the normal vector of each triangular face of the triangular network model according to a differential geometry method, and performing region segmentation on the triangular face based on the normal vector to generate a paint coating plane region and a paint coating curved surface region;

[0056] S105, accumulating the areas of the triangular facets of the paint coating plane region to obtain the total area of ​​the paint coating plane region, and calculating the total area of ​​the paint coating curved surface region according to a numerical integration method;

[0057] S106, extracting the characteristic region of the surface point cloud data based on the convolutional network model to obtain the fine characteristic region of the paint coating, reconstructing the local point cloud surface of the fine characteristic region of the paint coating based on the Alpha shape algorithm, and calculating the fine characteristic area of ​​the paint coating based on the local point cloud surface;

[0058] S107, generating a paint area according to the total area of ​​the paint plane area, the total area of ​​the paint curved surface area and the fine feature area of ​​the paint.

[0059] Specifically:

[0060] Obtain 3D point cloud data of the object to be sprayed: Use a 3D laser scanner to scan the object to be sprayed to obtain 3D point cloud data of its surface. During the scanning process, the scanner emits a laser beam, measures the time it takes for the laser beam to reflect back to the scanner, and calculates the 3D coordinates of each sampling point to form a point cloud data set.

[0061] The 3D laser scanner emits a laser beam to the surface of the object to be sprayed, determines the distance by measuring the flight time of the laser beam, and then obtains the 3D coordinates of the surface points of the object. For example, when scanning a painted door, the scanner will emit millions of laser beams. The time for each laser beam to return will be recorded and converted into distance, and finally form the point cloud data of the door surface. These point data contain the 3D coordinates (x, y, z) of each sampling point on the door surface. The original point cloud data obtained often contains noise points and outliers.

[0062] Based on the acquired point cloud data of the object surface, the Poisson reconstruction algorithm is used to construct an initial high-precision triangular mesh model to obtain detailed geometric shape information of the object surface. For the constructed initial triangular mesh model, the error metric value of each vertex is calculated to determine the vertices that have less impact on mesh simplification, thereby determining the triangular facets that can be reduced. Based on the determined simplifiable vertices, the edge collapse algorithm is used to merge adjacent triangular facets to reduce the number of triangular facets and obtain a simplified triangular mesh model.

[0063] Extract geometric features of the object surface, such as curvature, normal, surface smoothness, etc. These features help distinguish painted areas from other material surfaces (such as metal, plastic, wood, etc.). Since painted surfaces usually have different gloss and reflection properties, the reflection intensity can be extracted as an important feature for identification. Paint surfaces usually show high gloss or uniform reflection, while other materials (such as metal, wood) have different reflection properties.

[0064] The point cloud data is classified by material by analyzing shape features and reflection intensity features through support vector machine (SVM), and the paint area is distinguished from other material areas. Based on the classification results, the point cloud data is segmented using a clustering algorithm to distinguish the paint point cloud area from the non-paint area. Points close to each other are identified through density clustering to form paint areas. In the segmented paint area, the Canny edge detection is used to extract the boundaries of the paint. The boundaries can be detected by the change of surface normals in the point cloud or the change of distances between adjacent points. The boundaries are smoothed and optimized by an iterative optimization method to ensure the accuracy and smoothness of the boundary line.

[0065] The Poisson reconstruction algorithm can generate a high-precision triangular mesh model based on point cloud data and effectively restore the surface shape of the object. The principle is to convert the point cloud data into a scalar field, and then obtain a smooth surface by solving the Poisson equation, and then convert the surface into a triangular mesh model. For example, the Poisson reconstruction algorithm can be used to convert the scanned point cloud data of the painted object into a high-precision triangular mesh model of the painted object. This model can accurately reflect the detailed features of the painted object, such as the body shape and window contour. The initial triangular mesh model usually contains a large number of vertices and facets, and the data volume is large, which is not convenient for subsequent processing. Therefore, the mesh model needs to be simplified. The goal of simplification is to reduce the number of vertices and facets while maintaining the main geometric features of the model.

[0066] Calculating the error metric for each vertex is a key step in mesh simplification. The error metric reflects the degree of influence of the vertex on the shape of the model. For example, a vertex located in a flat area usually has a smaller error metric because it has little influence on the shape of the model; while a vertex located at a sharp corner or edge usually has a larger error metric because it has a greater influence on the shape of the model. By calculating the error metric, you can determine which vertices can be removed or merged to reduce the number of patches.

[0067] For example, if the deviation between the simplified model to be tested and the original model is too large, it is necessary to reduce the simplification threshold and simplify it again. Feature information of the object surface, such as curvature, normal vector, etc., can be extracted from the simplified triangular mesh model. This feature information can be used for subsequent object recognition and analysis tasks. For example, the type of recognition can be determined based on the curvature characteristics of the surface of the object to be tested itself, or the lighting effect of the object surface after painting can be calculated based on the normal vector information. Finally, the simplified triangular mesh model is stored in a standard 3D model file format, such as OBJ, STL, etc.

[0068] According to the topological structure of the triangular mesh model, the differential geometry method is used to calculate the normal vector of each triangular facet; according to the predetermined normal vector angle threshold, it is judged whether the normal vector angle of adjacent triangular facets is less than the threshold; if the normal vector angle of adjacent triangular facets is less than the threshold, these facets are classified into the same area and marked as paint plane area; if the normal vector angle of adjacent triangular facets is greater than or equal to the threshold, these facets are classified into different areas and marked as paint surface area; the paint plane area and paint surface area are further subdivided by clustering algorithm to obtain finer sub-areas; according to the normal vector distribution characteristics of each sub-area, the principal component analysis method is used to extract the main direction of each sub-area; according to the main direction and normal vector distribution characteristics, each sub-area is semantically annotated to obtain the paint plane area and paint surface area on the object surface.

[0069] The triangular mesh model is composed of a series of triangular patches, which are connected to form the surface of the object. To analyze the surface characteristics of the object, it is necessary to process these triangular patches. First, the normal vector of each triangular patch needs to be calculated. The normal vector is a vector perpendicular to the plane of the triangle, which can be used to describe the direction of the triangle.

[0070] Specifically, the coordinates of the three vertices of a triangle are (1, 1, 1), (2, 1, 1), and (1, 2, 1). The cross product operation can be used to obtain the normal vector of the triangle as (0, 0, 1). Next, it is necessary to determine the normal vector angle of adjacent triangular facets based on a predetermined normal vector angle threshold. This threshold can be adjusted according to the actual application scenario. Specifically, the threshold can be set to 30 degrees. If the normal vector angle of two adjacent triangular facets is less than 30 degrees, the two facets are considered to belong to the same area and are marked as a paint coating plane area. Specifically, the normal vectors of two adjacent triangular facets are (0, 0, 1) and (1, 1, 1), and their angle is less than 30 degrees, so the two facets can be marked as a paint coating plane area. Conversely, if the normal vector angle of two adjacent triangular facets is greater than or equal to 30 degrees, the two facets are considered to belong to different areas and are marked as paint coating surface areas. Specifically, the normal vectors of two adjacent triangular facets are (0, 0, 1) and (1, 0, 0), and the angle between them is 90 degrees, so these two facets can be marked as paint surface areas.

[0071] In order to obtain more refined sub-regions, clustering algorithms can be used to further subdivide the paint plane region and the paint surface region. Clustering algorithms can divide triangular facets into different sub-regions based on their normal vectors, curvature and other features. Specifically, a paint plane region may contain multiple sub-planes, which can be distinguished by clustering algorithms. Similarly, a paint surface region may also contain multiple sub-surfaces with different curvatures, which can also be distinguished by clustering algorithms. For each sub-region, the principal component analysis method can be used to extract its main direction. The main direction can be used to describe the overall orientation of the sub-region. Specifically, if the normal vector of a sub-region is mainly concentrated near (0, 0, 1), its main direction can be approximated as (0, 0, 1). Extracting the main direction can help better understand the shape characteristics of the sub-region. Finally, semantic annotations can be performed on each sub-region based on the main direction and normal vector distribution characteristics, specifically annotating a paint plane region as "desktop" and annotating a paint surface region as "sphere". This can facilitate subsequent object recognition and analysis tasks. Specifically, when identifying an object placed on a desktop, the desktop area can be identified first, and then the object recognition can be performed in the area, thereby improving the recognition efficiency. Through the above steps, the triangular mesh model can be segmented into different paint plane areas and paint surface areas, and each area can be semantically labeled.

[0072] According to the type of the target area, determine whether it is a paint plane area or a paint surface area; if it is a paint plane area, divide it into multiple triangular facets, calculate the area of ​​each triangular facet by geometric methods, and accumulate the areas of all triangular facets to obtain the total area of ​​the paint plane area; if it is a paint surface area, discretize the surface by numerical integration method, and divide the surface into multiple small surface units; calculate the area of ​​each surface unit according to the numerical integration formula of Gaussian quadrature method or Simpson method, and accumulate the areas of all surface units to obtain the total area of ​​the paint surface area; in the integral calculation process of Gaussian quadrature method or Simpson method, improve the accuracy and efficiency of numerical integration by selecting appropriate integration points and weights; for complex paint surface areas, an adaptive integration method can be used to automatically adjust the integration step and order according to the local characteristics of the surface to balance the calculation accuracy and efficiency.

[0073] Specifically, suppose there is a rectangular plane, which can be divided into two triangles. The area of ​​each triangle can be calculated by multiplying the base by the height and then dividing by two. Then the areas of the two triangles are added together to get the area of ​​the entire rectangle. For the paint surface area, due to its complex geometric shape, a numerical integration method is required to calculate the area. The surface can be discretized into multiple small surface units, each of which is approximately a small plane or surface patch. The areas of these small surface units can be accurately calculated by Gauss quadrature method or Simpson method. Specifically, consider a part of a sphere, which can be divided into multiple small triangles, each of which is determined by three points on the sphere. Using Simpson's method, the area of ​​each small triangle can be estimated by selecting several integration points inside each small triangle and calculating the distance between the normal vector of these points and the center of the sphere. Then the areas of all small triangles are accumulated to get the area of ​​the entire spherical part. In the process of numerical integration, choosing appropriate integration points and weights is the key to improving integration accuracy and efficiency. Specifically, when calculating the area of ​​a complex surface, different numbers and positions of integration points can be selected according to the local curvature of the surface to adapt to the local geometric characteristics. For particularly complex paint surface areas, an adaptive integration method can also be used to automatically adjust the integration step size and order according to the local characteristics of the surface, which can ensure the accuracy of the calculation and improve the calculation efficiency.

[0074] According to the small features on the surface of the object, such as bolt holes and grooves, the feature area that needs to be focused on is determined, and the local point cloud subset of the area is extracted from the original point cloud data. For the extracted local point cloud subset, by analyzing its point cloud density and feature size, the Alpha value parameter in the Alpha shape algorithm is adaptively determined to balance the computational efficiency and reconstruction accuracy. Using the determined Alpha value, the Alpha shape algorithm is used to reconstruct the surface of the local point cloud subset to obtain a local surface model that can accurately represent the small features. Based on the reconstructed local surface model, its surface area is calculated as the area measurement of the small feature area of ​​the paint coating on the surface of the object for subsequent analysis and application.

[0075] Convert the original point cloud data of small features on the surface of the object into a voxel grid. Voxelization is done by dividing the space into a regular cubic grid and then mapping the point cloud data to these voxels according to the point cloud density of each grid cell. This process can be achieved using a voxelization algorithm.

[0076] When extracting small feature regions, convolutional neural networks are used to automatically learn and extract local features of point cloud data. The spatial features of the input data are extracted through the convolution layer. For voxelized point cloud data, the convolution layer learns information such as the geometry and surface texture in the local area. In order to capture local features, the voxelized point cloud data is scanned using a sliding window mechanism. The sliding window slides in three-dimensional space, traversing the entire point cloud voxel grid, and extracting features of the local area from each window. The features of each local area are obtained through convolution operations, which can be understood as information extracted from each small area in the voxel grid. Through the training of the convolutional network, the model learns how to extract useful local features from the voxelized point cloud data, especially small paint coating features (such as coating thickness changes, surface irregularities, etc.). The convolution operation captures the geometry, surface details and possible material differences of the point cloud data in each small window.

[0077] The convolutional neural network includes a one-dimensional dilated convolution module, an LSTM module and a fully connected neural network module. The training process uses cross entropy CE as the loss function and uses the back propagation algorithm to optimize the model.

[0078] The one-dimensional dilated convolution module performs convolution in the time dimension. First, the 66 channels are converted to 128 channels through a one-dimensional convolution, which plays an upsampling role. The following six layers of one-dimensional dilated convolution maintain 128 channels, and the receptive field is continuously expanded through convolution to obtain a preliminary feature map.

[0079] Input data: T×66 sequence, time series length is T, the number of feature vectors is 66;

[0080] Upsampling layer: 128 convolution kernels of length 5, dilation factor 1, input T×66, output T×128;

[0081] The first layer of dilated convolution: 128 convolution kernels of length 5, dilation factor 2, input T×128, output T×128;

[0082] The second layer of dilated convolution: 128 convolution kernels of length 5, dilation factor 4, input T×128, output T×128;

[0083] The third layer of dilated convolution: 128 convolution kernels of length 5, dilation factor 6, input T×128, output T×128;

[0084] The fourth layer of dilated convolution: 128 convolution kernels of length 5, dilation factor 2, input T×128, output T×128;

[0085] The fifth layer of dilated convolution: 128 convolution kernels of length 5, dilation factor 4, input T×128, output T×128;

[0086] The sixth layer of dilated convolution: 128 convolution kernels of length 5, dilation factor 6, input T×128, output T×128.

[0087] Furthermore, the LSTM module is used to further extract time series features;

[0088] The input is T×128, and the output is T×128;

[0089] The first fully connected layer: the input is 1×128 and the output dimension is 1×64;

[0090] The second fully connected layer: input is 1×64, output is 1×32;

[0091] The third fully connected layer: input is 1×32, output is 1×16;

[0092] The fourth fully connected layer: the input is 1×16, the output is 1×8, and 8 is the number of categories;

[0093] The fully connected neural network module resizes the Hidden Embedding into a long vector before calculation, and then sends it to the fully connected neural network.

[0094] Obtain the material and color information of the paint, and establish a point cloud reflection characteristic model based on the material and color attributes to describe the point cloud reflection characteristics under different materials and colors. Use the point cloud segmentation algorithm to divide the input raw point cloud data into multiple local point cloud areas, each of which corresponds to a local surface. For each local point cloud area, extract its shape features and reflection intensity features. The shape features include curvature, normal vector, etc., and the reflection intensity features include color, brightness, etc. The extracted local point cloud features are input into a pre-trained deep learning model. The model adopts an encoder-decoder structure. The encoder maps the local features to a low-dimensional feature space, and the decoder reconstructs the completed local point cloud based on the low-dimensional features. During the training process, the deep learning model uses a large amount of point cloud data of different materials and colors for training. The model parameters are optimized by minimizing the reconstruction error so that it can be intelligently completed according to local features. All the completed local point cloud areas are spliced ​​to obtain complete completed point cloud data, which are fused with the original point cloud to update the coordinates and reflection intensity information of each point in the point cloud. Surface reconstruction is performed on the completed point cloud, and the surface mesh model is obtained through a triangulation algorithm. Its surface area is calculated. Compared with the incomplete point cloud, the completed point cloud has higher surface area calculation accuracy and robustness.

[0095] Specifically, due to its high reflectivity, the metal surface will generate a strong reflection signal during laser scanning, while the matte surface has a weaker reflection. Suppose there is a set of point cloud data containing three materials: metal, plastic and wood. The reflection intensity of the metal surface may reach 8, plastic is 5, and wood is 3. With this information, a point cloud reflection characteristic model can be established to describe the point cloud reflection characteristics under different materials and colors. The point cloud segmentation algorithm is used to divide the original point cloud data into multiple local point cloud areas, each corresponding to a local surface. Common segmentation algorithms such as clustering segmentation based on Euclidean distance can divide the point cloud into several subsets according to spatial distance. Assuming that the original point cloud contains 100,000 points, after clustering segmentation, 100 local point cloud areas may be obtained, each containing about 1,000 points. For each local point cloud area, its shape features and reflection intensity features are extracted. Shape features include curvature and normal vectors. Curvature reflects the degree of curvature of the surface, and normal vectors describe the orientation of the surface. Specifically, the local point cloud area of ​​a bolt hole has a large curvature and a relatively concentrated distribution of normal vectors. Reflection intensity features include color and brightness. Color can be represented by RGB values, while brightness can be represented by reflection intensity values. Assuming that the RGB value of a certain area is (255, 0, 0) and the reflection intensity is 6, it means that the area is red and has strong reflection. The extracted local point cloud features are input into a pre-trained deep learning model, which adopts an encoder-decoder structure. The encoder maps local features to a low-dimensional feature space, and the decoder reconstructs the completed local point cloud based on the low-dimensional features. Assume that the encoder compresses the 1000-dimensional local features to 100 dimensions, and the decoder reconstructs 1000-dimensional point cloud data from the 100-dimensional features. During the training process, the model uses a large amount of point cloud data of different materials and colors for training, and optimizes the model parameters by minimizing the reconstruction error. Specifically, the training data contains 1000 samples, each of which contains 1000 points. The model is iteratively optimized to reduce the reconstruction error from the initial 1 to 01. All completed local point cloud areas are stitched together to obtain complete completed point cloud data, which are then merged with the original point cloud to update the coordinates and reflection intensity information of each point in the point cloud. During the stitching process, the relative position relationship between the local point clouds needs to be considered to ensure the spatial consistency of the stitched point cloud data. Assume that the stitched point cloud data contains 100,000 points, and the coordinates and reflection intensity information of each point are updated. Surface reconstruction is performed on the completed point cloud, and a surface mesh model is obtained through a triangulation algorithm. Triangulation algorithms such as Delaunay triangulation can convert point cloud data into continuous triangular meshes to construct a surface model of an object. Assuming that the reconstructed surface mesh model contains 50,000 triangles, the surface area of ​​the object can be obtained by calculating the sum of the areas of these triangles. Compared with the incomplete point cloud, the completed point cloud has higher surface area calculation accuracy and robustness.

[0096] Obtain the three-dimensional point cloud data of the target object, divide the surface of the object into areas of paint plane, paint surface and paint fine feature areas according to the point cloud data; for the paint plane area, calculate the area of ​​the paint plane area by plane fitting algorithm; for the paint surface area, calculate the area of ​​the paint surface area by surface integral algorithm; for the paint fine feature area, extract fine features by feature extraction algorithm and calculate the area of ​​fine features; determine whether there is a missing part on the surface of the object, if there is a missing part, fill in the missing part by three-dimensional reconstruction algorithm according to the local features of the object surface; add up the area of ​​the paint plane area, the area of ​​the paint surface area, the area of ​​the paint fine feature and the area of ​​the missing part after filling in and repairing to obtain the total surface area of ​​the object; use the random sampling consistency algorithm to evaluate the accuracy of the total surface area, if the accuracy does not reach the preset threshold, return to step one to re-divide the area and recalculate the area; output the total surface area result of the object.

[0097] For the flat area of ​​paint, specifically the roof of the car, the plane fitting algorithm can be used to fit the point cloud data to a plane, and then the area is calculated according to the parameters of the fitting plane, specifically the area of ​​the roof is 0.5 square meters. For the curved area of ​​paint, since the sealing seam is mainly located between the windshield and the car body, the surface integral algorithm is used to calculate its area. The point cloud data at the sealing seam of the windshield is triangulated, and then the area of ​​each triangle is calculated. Finally, the area of ​​all triangles is accumulated to obtain the area of ​​the sealing seam of the windshield, specifically the area of ​​the sealing seam of the windshield is 0.003 square meters. For the small feature area of ​​paint, specifically the shell and seam of the rearview mirror of the car, it is difficult to directly calculate the area due to its complex shape. First, the edge point cloud of the rearview mirror shell and seam is extracted, and then the edge point cloud is projected onto a plane by two-dimensional projection to calculate the projection area, specifically the area of ​​the rearview mirror shell and seam is 0.005 square meters. In the process of point cloud data acquisition, point cloud data may be missing due to occlusion or other factors. Specifically, the bottom of the car is missing because the scanner cannot fully cover it. In this case, it is necessary to use a 3D reconstruction algorithm to complete and repair the missing part based on the features of the surrounding point clouds. Specifically, the Poisson reconstruction algorithm is used to infer the shape of the missing part based on the point cloud data around the missing area, and new point cloud data is generated for filling. Then the area of ​​the bottom of the car after repair is calculated, and the specific area of ​​the bottom of the car is 0.3 square meters. The previously calculated paint and coating plane area, paint and coating curved surface area, paint and coating fine feature area, and the area of ​​the missing part after completion and repair are added up, specifically 0.5+0.003+0.005+0.3=0.808 square meters, to obtain the surface area of ​​the car.

[0098] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for calculating the paint coating area based on three-dimensional scanning technology, characterized in that: The following steps are involved: Acquire surface point cloud data of the object after painting using a three-dimensional laser scanner, calculate the surface point cloud data based on a point cloud preprocessing algorithm, and obtain a surface point cloud model of the object; Extracting the paint area point cloud boundary of the object surface point cloud model according to the object surface shape characteristics and reflection intensity characteristics to obtain the paint surface point cloud model; Constructing a triangulated network model of the painted object according to the paint surface point cloud model; Calculate the normal vector of each triangular face of the triangular network model according to the differential geometry method, perform region segmentation on the triangular face based on the normal vector, and generate a paint coating plane region and a paint coating curved surface region; Accumulating the areas of the triangular facets of the paint coating plane region to obtain the total area of ​​the paint coating plane region, and calculating the total area of ​​the paint coating curved surface region according to a numerical integration method; Extracting feature areas of the surface point cloud data based on a convolutional network model to obtain fine feature areas of paint and coatings, reconstructing local point cloud surfaces of the fine feature areas of paint and coatings based on an Alpha shape algorithm, and calculating fine feature areas of paint and coatings based on the local point cloud surfaces; The paint area is generated according to the total area of ​​the paint plane area, the total area of ​​the paint curved surface area and the small feature area of ​​the paint.

2. The paint coating area calculation method based on three-dimensional scanning technology according to claim 1 is characterized in that: The process of calculating the surface point cloud data based on the point cloud preprocessing algorithm to obtain the object surface point cloud model includes: Use a 3D laser scanner to scan the painted object and obtain 3D point cloud data of the surface; Statistically analyzing the features of the three-dimensional point cloud data of the surface, determining the judgment threshold of noise points and outliers, and if the feature values ​​of some points in the point cloud data exceed the threshold range, they are judged as noise points or outliers; The noise points or outliers are removed to obtain a point cloud model of the object surface.

3. The paint coating area calculation method based on three-dimensional scanning technology according to claim 2 is characterized in that: The process of removing the noise points or outliers includes: The noise points or outliers are processed based on a filtering algorithm. During the filtering process, the neighborhood points of each point are analyzed. If the point is determined to be a noise point, the noise point is removed from the point cloud data. The point cloud data is divided into multiple clusters through clustering. If the distance between a point and all clusters exceeds the threshold, it is judged as an outlier and is removed from the point cloud data.

4. The paint coating area calculation method based on three-dimensional scanning technology according to claim 1 is characterized in that: The process of obtaining the paint surface point cloud model includes: using the surface shape characteristics and reflection intensity characteristics of the object, targeting the material and color differences between the paint and other medium surfaces, using a point cloud algorithm based on machine learning to identify and filter, extracting the point cloud boundaries of the paint area, and obtaining the paint surface point cloud model.

5. The paint coating area calculation method based on three-dimensional scanning technology according to claim 1 is characterized in that: The process of constructing a triangulated network model of the painted object according to the paint surface point cloud model includes: Processing the paint surface point cloud model according to a Poisson reconstruction algorithm to construct an initial high-precision triangular mesh model; Calculating an error metric value for each vertex in the initial high-precision triangular mesh model, and determining a vertex that can be simplified based on the error metric value; According to the simplifiable vertices, an edge collapse algorithm is used to merge adjacent triangular facets to obtain a triangular network model of the painted object.

6. The paint coating area calculation method based on three-dimensional scanning technology according to claim 5 is characterized in that: The process of merging adjacent triangular facets using the edge collapse algorithm includes: setting a simplification threshold during the mesh simplification process, controlling the simplification degree of each edge collapse operation, performing error evaluation on the simplified triangular mesh model, calculating the geometric deviation between the two models before and after simplification, and if the deviation exceeds the preset threshold, adjusting the simplification parameters and re-simplifying the mesh.

7. The method for calculating the paint coating area based on three-dimensional scanning technology according to claim 1 is characterized in that: The process of performing region segmentation on the triangular face based on the normal vector to generate a paint coating plane region and a paint coating curved surface region comprises: According to the topological structure of the triangular mesh model, a normal vector of each triangular facet is calculated using a differential geometry method; According to the predetermined normal vector angle threshold and the normal vector division of each triangular facet, a paint coating plane area and a paint coating curved surface area are generated.

8. The paint coating area calculation method based on three-dimensional scanning technology according to claim 7 is characterized in that: The process of generating the paint coating plane area and the paint coating curved surface area according to the predetermined normal vector angle threshold and the normal vector division of each triangular facet comprises: According to a predetermined normal vector angle threshold, determine whether the normal vector angle of adjacent triangular facets is less than the threshold. If the normal vector angle of adjacent triangular facets is less than the threshold, these facets are classified into the same area and marked as a plane sub-area. If the normal vector angle of adjacent triangular facets is greater than or equal to the threshold, these facets are classified into different areas and marked as a surface sub-area. The paint coating plane area and the paint coating surface area are subdivided by clustering algorithm to obtain fine sub-areas; According to the normal vector distribution characteristics of each sub-region, the principal direction of each sub-region is extracted by principal component analysis method; According to the main direction and normal vector distribution characteristics, each sub-region is semantically annotated to obtain the paint plane region and paint curved surface region on the object surface.

9. The paint coating area calculation method based on three-dimensional scanning technology according to claim 1 is characterized in that: The process of extracting the feature area of ​​the surface point cloud data based on the convolutional network model to obtain the fine feature area of ​​the paint coating comprises: Converting the surface point cloud data into voxelized point cloud data; Extracting the voxelized point cloud data based on the sliding window mechanism of the convolutional network model to obtain features of the local area; The characteristic boundaries of the local area are processed based on an image segmentation algorithm to obtain the fine characteristic area of ​​the paint coating.

10. The paint coating area calculation method based on three-dimensional scanning technology according to claim 1 is characterized in that: The process of reconstructing the local point cloud surface of the fine feature area of ​​the paint coating based on the Alpha shape algorithm and calculating the fine feature area of ​​the paint coating based on the local point cloud surface includes: Determine a feature area according to fine features on the surface of the object, and extract a local point cloud subset of the feature area from the original point cloud data; Analyzing the point cloud density and feature size of the local point cloud subset of the feature area to determine the Alpha value in the Alpha shape algorithm; Performing surface reconstruction on the local point cloud subset based on the Alpha value to obtain a local surface model representing small features; The surface area of ​​the local surface model is calculated to generate a fine feature area of ​​the paint coating.

Images