An edge enhancement based semi-supervised k-means point cloud segmentation method

Abstract

The application provides a point cloud segmentation method based on edge enhancement and semi-supervised k-means, and belongs to the technical field of point cloud data processing. The application realizes adaptive high-quality segmentation of point cloud data by carrying out denoising processing on original point cloud data, analyzing the change of normal vectors in the point cloud data set, extracting geometric edges in the point cloud data, obtaining a closed region formed by an edge point cloud set, selecting a centroid by morphological closing operation, and segmenting the point cloud data set by using an optimized K-means clustering algorithm. The morphological closing operation effectively improves the edge continuity, the centroid initialization strategy based on the closed region reduces the iteration number and avoids manual preset clustering number, and the introduction of the edge constraint term improves the boundary matching accuracy. Compared with the traditional method, the method provided by the application has higher stability and accuracy in the point cloud segmentation method of splicing detectors.

Description

Technical Field

[0001] This invention belongs to the field of point cloud data processing technology, specifically relating to a semi-supervised k-means point cloud segmentation method based on edge enhancement. Background Technology

[0002] With the rapid development of astronomy, large-aperture, wide-field-of-view, and wide-band astronomical optical telescopes have gradually become key technological means to solve cutting-edge scientific problems. To detect more distant and fainter celestial objects, the apertures of ground-based astronomical telescopes continue to increase; simultaneously, to enhance the ability to observe wider areas of the sky, their fields of view are constantly expanding. This trend has led to a significant increase in focal plane array size. However, due to the current limitations in the manufacturing size of individual detectors, mosaicking detector technology has become the mainstream solution for constructing large-area focal plane arrays.

[0003] In this context, the flatness of the stitched detector is extremely important. For a given optical system, the tilt error of the CCD target surface must be strictly controlled within the depth of focus; otherwise, some areas will deviate from the optimal imaging plane, resulting in uneven image sharpness and failing to meet the requirements of high-precision imaging. Generally, the flatness of the stitched focal plane needs to be controlled within the range of 20–30 μm. In actual integration, multiple measurements and fine-tuning are usually required to achieve the required stitching specifications.

[0004] High-precision point cloud segmentation is a key step in the point cloud data processing method for stitched detectors, and it is the core link to achieve the above requirements. In recent years, research on point cloud segmentation has proposed a variety of improvement methods. Traditional point cloud segmentation methods mainly rely on geometric constraints and statistical rules to manually design the features of objects, dividing the original point cloud data into several non-overlapping groups of regions to correspond to various objects in the scene. For example, segmentation methods based on edge information, model fitting, and region growing are traditional methods, but their accuracy is low in complex point cloud processing. With the development of computer technology, the technology of using deep learning to process point cloud data has become very mature and has achieved good results, such as projection-based, voxel-based, and point-based segmentation. These methods require transforming the point cloud into a regular structure suitable for convolutional neural network processing, and then using the network model for prediction. Although deep learning-based methods have shown obvious advantages in general scene segmentation, they have problems in the special application scenario of large-aperture telescope stitched detectors. Point cloud data needs to be labeled point by point, which is time-consuming and manual. This characteristic determines that traditional methods are still irreplaceable in the field of stitched detectors.

[0005] Existing traditional segmentation methods still have shortcomings in terms of engineering applicability to stitched detectors: Adaptive region growing algorithms based on neighborhood density reduce dependence on initial seed points by dynamically adjusting growth criteria. However, this method requires pre-extraction of ground point clouds and has poor adaptability to unprocessed stitched point clouds. A city point cloud segmentation process based on the PCL library is proposed, combining voxel filtering and Euclidean clustering to extract building and terrain features. This method has high computational efficiency but is sensitive to point cloud density distribution and prone to oversegmentation. A point cloud segmentation algorithm based on octree voxelization and multi-stage region growing is also proposed. This method achieves efficient spatial indexing through the octree structure, but it is sensitive to the selection of initial seed points, which may affect stability. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and propose a point cloud segmentation method based on edge enhancement semi-supervised k-means.

[0007] To achieve the above objectives, the present invention adopts the following specific technical solution:

[0008] A semi-supervised k-means point cloud segmentation method based on edge enhancement includes the following steps:

[0009] S1. Obtain raw point cloud data, perform noise reduction processing on the raw point cloud data, and obtain a three-dimensional point cloud dataset;

[0010] S2. Analyze the changes in the normal vectors in the point cloud dataset, extract the geometric edges in the point cloud data, and obtain the edge point cloud set;

[0011] S3. In the closed region formed by the aggregation of edge points, the centroid is selected by morphological closing operation as the initial center point of the K-means clustering optimization function;

[0012] S4. Optimize the K-means clustering algorithm based on the edge point cloud;

[0013] S5. The optimized K-means clustering algorithm is used to segment the point cloud dataset, and the segmentation results of the point cloud are output.

[0014] Furthermore, the step of obtaining the edge point cloud set in step S2 is as follows:

[0015] Select any point p in the point cloud dataset i The local surface covariance matrix formed by its k nearest neighbors is shown in formula (1):

[0016]

[0017] in, Let represent the centroid of the neighborhood point set, and k represent the number of neighborhood points;

[0018] Performing eigenvalue decomposition on the covariance matrix, the eigenvector corresponding to the smallest eigenvalue is the normal vector n. i Then the normal vector of any point in the cloud dataset is as shown in formula (2):

[0019]

[0020] Where λ1, λ2, and λ3 represent the eigenvalues ​​of the covariance matrix, and λ1 ≤ λ2 ≤ λ3;

[0021] By comparing p i Its neighboring point p j The boundary is identified by the difference in normal vectors when p i Its neighboring point p j When the normal vector is greater than a preset angle threshold, the geometric edges in the point cloud data are extracted, and the edge point cloud set is shown in formula (3):

[0022]

[0023] E represents the edge point cloud, n i Point p i The normal vector, n j Point p j The normal vector, τ represents the preset angle threshold.

[0024] Furthermore, the step of obtaining the initial center point in step S3 is as follows:

[0025] The edge point cloud is projected onto a two-dimensional plane to obtain a two-dimensional edge projection point set, and the closed region is identified using a contour extraction method.

[0026] Discretize the two-dimensional edge projection point set into a binary image;

[0027] The binary image is processed using morphological closing operations to fill edge gaps and form a closed region with continuous edges;

[0028] Match each continuously closed region with the original point cloud to extract the point set inside the region;

[0029] The geometric centroid of each internal point set is selected as the initial centroid of the K-means clustering optimization function.

[0030] Furthermore, step S4 specifically includes:

[0031] Optimize the objective function of the K-means clustering algorithm:

[0032] An edge penalty term is added to the objective function of the K-means clustering algorithm, as shown in formula (4):

[0033]

[0034] Among them, w ij ∈{0,1} represents a binary membership variable function, when point p i The value is 1 if it belongs to cluster j, and 0 otherwise; p i c represents the i-th 3D point cloud data point within the closed region; j D represents the center point of the j-th cluster; E (p i ) = min q∈E ||p i -q|| indicates p i The nearest Euclidean distance to the edge point set E. λ represents the penalty coefficient, controlling the strength of the edge constraint, and is greater than 0; if point p i If the cluster is located near the edge, the center c of its cluster is penalized. j This forces the cluster center away from the edge;

[0035] Optimize the cluster centers of the K-means clustering algorithm:

[0036] By taking the partial derivative of the optimized objective function and setting it to zero, the optimized cluster centers are obtained.

[0037] Optimize the iterative function of the K-means clustering algorithm:

[0038] The optimized cluster center is shifted in the opposite direction along the edge. If the shift of the optimized cluster center is less than a set threshold, the iteration is terminated, and the optimization of the iteration function is completed.

[0039] Furthermore, the specific processing method of the morphological closing operation is: using the dilation-erosion operation of the circular structuring element.

[0040] Furthermore, the method for obtaining raw point cloud data is as follows: scan using a dual-probe structure with a differential structure, and obtain raw point cloud data through a real-time differential algorithm.

[0041] Furthermore, the method for denoising the original point cloud data is: a radius filtering algorithm based on the neighborhood.

[0042] The present invention can achieve the following technical effects:

[0043] 1. The point cloud segmentation method based on edge enhancement semi-supervised k-means provided by this invention can achieve adaptive high-quality segmentation of point cloud data by fusing morphological processing of edge features and improved clustering algorithms.

[0044] 2. The point cloud segmentation method based on edge enhancement semi-supervised k-means provided by this invention can effectively improve edge continuity through morphological closing operation, reduce the number of iterations based on the centroid initialization strategy of closed regions, and avoid manually pre-setting the number of clusters. The introduction of edge constraint terms improves the accuracy of boundary matching.

[0045] 3. The point cloud segmentation method based on edge enhancement semi-supervised k-means provided by this invention can select the centroid as the initial center point in the closed region formed by the edge points, thereby reducing the uncertainty of the initial position selection, avoiding the centroid from falling on the edge or empty region, controlling the centroid to not be selected repeatedly, and enhancing stability.

[0046] 4. The point cloud segmentation method based on edge-enhanced semi-supervised k-means provided by this invention can effectively extract spatial structure regions from edge information and determine a unique initial centroid for each region, thereby avoiding clustering instability caused by random initialization in the K-means algorithm and improving consistency with the real structure.

[0047] 5. Compared with traditional methods, the point cloud segmentation method based on edge enhancement semi-supervised k-means provided by this invention has higher stability and accuracy in point cloud segmentation using stitched detectors. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of the method flow of the point cloud segmentation method based on edge enhancement semi-supervised k-means according to an embodiment of the present invention;

[0049] Figure 2 This is a schematic diagram of a point cloud dataset of a large-aperture stitched detector according to an embodiment of the present invention;

[0050] Figure 3 This is a schematic diagram of edge point aggregation of a large-aperture splicing detector according to an embodiment of the present invention;

[0051] Figure 4 This is a schematic diagram of the result of morphological operation on the edge point cloud of a large-aperture spliced ​​detector according to an embodiment of the present invention;

[0052] Figure 5 This is a schematic diagram of the segmentation result of a large-aperture splicing detector according to an embodiment of the present invention;

[0053] Figure 6 This is a schematic diagram of the height cloud map of the segmentation result of the large-aperture splicing detector according to an embodiment of the present invention;

[0054] Figure 7 This is a schematic diagram of the algorithm flow of the point cloud segmentation method based on edge enhancement semi-supervised k-means according to an embodiment of the present invention. Detailed Implementation

[0055] In the following description, embodiments of the invention will be described with reference to the accompanying drawings. In the description below, the same modules are denoted by the same reference numerals. Where the same reference numerals are used, their names and functions are also the same. Therefore, their detailed description will not be repeated.

[0056] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not constitute a limitation thereof.

[0057] The following is combined with Figures 1-7 The specific working method of the point cloud segmentation method based on edge enhancement semi-supervised k-means in the embodiments of the present invention will be described in detail below:

[0058] This invention, using point cloud data collected on a large-aperture telescope mosaic detector as an example, details how a semi-supervised k-means point cloud segmentation method based on edge enhancement can effectively segment point clouds in a large-aperture telescope mosaic detector dataset. In this embodiment, the large-aperture telescope mosaic detector data measurement device consists of an air-supported platform system, a motion system, and a dual-probe optical measurement system, achieving sub-micron level accuracy. A host computer control software based on the QT framework was developed to achieve synchronous and accurate acquisition of the motion system's real-time position and the dual-probe height data. The base platform system uses an active air-bearing vibration-isolated optical platform to provide stability support for the measurement. The motion system employs a MISUMI E-RSZ12-8S-X360-Y600-DE30-XC5-YC5 high-precision two-dimensional platform, achieving ±5μm repeatability and ±10μm straightness within a 600mm × 360mm travel range. The measurement system is configured with a dual-probe differential structure: a Keyence CL-3000 controller drives two probes, CLP015N / CLP015, for measurement. The second reference probe measures the point cloud data of a plane mirror with a surface shape better than λ / 20 in real time with a high-precision linearity of ±0.41μm; this data is recorded as platform error noise. The first working probe performs a large-area scan of the stitching detector, obtaining the final 3D raw point cloud data of the large-aperture telescope stitching detector through a real-time differential algorithm. Other methods in the prior art can also be used to acquire the 3D raw point cloud data; this invention does not limit this method.

[0059] In this embodiment of the invention, a point cloud segmentation method based on edge-enhanced semi-supervised k-means is used, such as... Figure 1 As shown, it includes the following steps:

[0060] S1. Obtain the raw point cloud data, perform noise reduction processing on the raw point cloud data, and obtain a 3D point cloud dataset.

[0061] Point cloud data acquisition is susceptible to environmental interference, abnormal laser echoes, and sampling errors, resulting in isolated points, spikes, and other noise. To remove outliers introduced by noise, occlusion, or missampling, the technical solution provided in this invention uses dual-probe differential noise reduction technology to eliminate platform noise, then acquires the original point cloud data. A neighborhood-based radius filtering algorithm is then used to preprocess the original point cloud data to remove outliers, ultimately obtaining a high-quality 3D point cloud dataset for subsequent processing. Figure 2 As shown in the figure, the colored bars on the right represent the height information of the stitched detector point cloud dataset, and X and Y represent the X-axis and Y-axis coordinates of the point cloud dataset, respectively.

[0062] The basic idea of ​​the neighborhood-based radius filtering algorithm is: if the number of neighboring points of a certain point within a given radius r is less than a preset threshold N. min If the point is an outlier, it is considered an outlier and is removed.

[0063] Let p be a point in the point cloud. i Its coordinates in three-dimensional space are p i =(x i ,y i ,z i If ), then count the number of its neighboring points N within a given radius r. i The judgment criteria for this method are shown in the following formula (1):

[0064] N i =|{p j ∈P|||p i -p j ||≤r}| (1)

[0065] If the following conditions are met, then p is considered to be... i Outliers:

[0066] N i <N min

[0067] Where P represents the point cloud dataset, ||·|| represents the Euclidean distance, r represents the neighborhood radius, and N min p represents the threshold for the number of neighbors. j Represents any point p i The neighboring points.

[0068] S2. Analyze the changes in the normal vectors in the point cloud dataset, extract the geometric edges in the point cloud data, and obtain the edge point cloud set.

[0069] This invention employs edge detection using normal vectors, identifying object boundaries by analyzing changes in local surface normal vectors in a point cloud. In continuous surface regions, the direction of the normal vectors changes gradually between adjacent points; however, at object edges, the direction of the normal vectors changes drastically. By detecting changes in the angle between normal vectors, geometric edges in the point cloud dataset can be effectively detected.

[0070] Select any point p in the point cloud dataset i The local surface covariance matrix formed by its k nearest neighbors is shown in Equation (2):

[0071]

[0072] in, p represents the centroid of the neighborhood point set; k represents the number of neighborhood points; p j Represents any point p i The neighboring points.

[0073] Performing eigenvalue decomposition on the covariance matrix, the eigenvector corresponding to the smallest eigenvalue is the normal vector n. i Then the normal vector of any point in the cloud dataset is as shown in formula (3):

[0074]

[0075] Where λ1, λ2, and λ3 represent the eigenvalues ​​of the covariance matrix, and λ1 ≤ λ2 ≤ λ3;

[0076] By comparing p i Its neighboring point p j The boundary is identified by the difference in normal vectors when p i Its neighboring point p j When the normal vector is greater than a preset angle threshold, the geometric edges in the point cloud data are extracted, and the edge point cloud set is shown in formula (3):

[0077]

[0078] E represents the edge point cloud, n i Point p i The normal vector, n j Point p j The normal vector, τ represents the preset angle threshold.

[0079] In this embodiment of the invention, an edge point cloud is obtained, such as... Figure 3 As shown, the units for both the X-axis and Y-axis coordinates are millimeters.

[0080] S3. In the closed region formed by the aggregation of edge points, the centroid is selected through morphological closing operation as the initial center point of the K-means clustering optimization function.

[0081] exist Figure 3 As shown in the edge point cloud obtained based on normal vector differences, the vector estimation method can accurately capture geometric edges with significant curvature changes in the point cloud, such as the red point set. However, the original edge points exhibit local breaks, such as the break on the left side. Therefore, the technical solution of this embodiment of the invention needs to fill the edge gaps to form a complete closed structure.

[0082] Project the edge point cloud onto a two-dimensional plane to obtain the two-dimensional edge projection point set ε. XY , ε XY ={(x i ,y i )|P i ∈E i}, use contour extraction methods to identify closed regions.

[0083] The two-dimensional edge projection point set ε XY Discretized into a binary image B(x,y), where 1 represents the position of the edge point, it can be expressed as formula (4):

[0084]

[0085] The binary image is processed using morphological closing operations to fill edge gaps, making the edges more continuous and forming a closed structure, which can be expressed as formula (5):

[0086]

[0087] in, This indicates an expansion operation. This indicates the erosion operation; S is a structural element.

[0088] In this embodiment of the invention, the edge gaps were successfully filled and a complete closed structure was formed by the expansion-erosion operation of a 3×3 circular structural element. The morphological operation results are as follows: Figure 4 As shown.

[0089] Obtain the closed region P k The point set R within the region k Defined as formula (6):

[0090] R k ={p i ∈P|proj xy (p i )∈P k} (6)

[0091] Among them, projxy This represents the projection of a point onto the xy plane.

[0092] Match each enclosed region with the original point cloud to extract the point set within that region. For each region R... k Initial cluster centers are selected within the cluster to initialize the K-means algorithm. In this embodiment of the invention, the geometric centroid is used as the initial center point, as expressed in formula (7):

[0093]

[0094] Where c k Let |R| be the center of the k-th region. k | indicates the number of points in the region.

[0095] This yields a set of initial cluster centers, represented by formula (8), which serve as the initial center points for the K-means clustering optimization function.

[0096] C = {c1, c2, ..., c} k} (8)

[0097] After obtaining the edge point cloud, this invention proposes a K-means clustering initialization method based on an edge supervision mechanism. Unlike the traditional K-means initialization method that uses random centroids, this invention selects the centroid as the initial center point within the closed region formed by the edge points, thereby reducing the uncertainty of the initial position selection, avoiding centroids falling on edges or empty regions, controlling the centroids to not be repeatedly selected, and enhancing stability.

[0098] S4. Optimize the K-means clustering algorithm based on the edge point cloud.

[0099] Optimize the objective function of the K-means clustering algorithm:

[0100] An edge penalty term is added to the objective function of the K-means clustering algorithm, as shown in formula (9):

[0101]

[0102] Among them, w ij ∈{0,1} represents a binary membership variable function, when point p i The value is 1 if it belongs to cluster j, and 0 otherwise; p i c represents the i-th 3D point cloud data point within the closed region; j D represents the center point of the j-th cluster; E (p i ) = min q∈E ||p i -q|| indicates pi The nearest Euclidean distance to the edge point set E. λ represents the penalty coefficient, controlling the strength of the edge constraint, and is greater than 0; if point p i If the cluster is located near the edge, the center c of its cluster is penalized. j This forces the cluster center away from the edge.

[0103] The existing K-means clustering algorithm updates the center by minimizing the squared distance within the cluster, but the method provided in this embodiment of the invention needs to consider edge constraints at the same time, and can effectively extract spatial structure regions from edge information.

[0104] Optimize the cluster centers of the K-means clustering algorithm:

[0105] Taking the partial derivative of the optimized objective function J and setting it to zero, we can obtain the optimized cluster centers, as shown in formula (10):

[0106]

[0107] Among them, gradient term Characterization point p i The nearest direction to the edge can be approximated by formula (11):

[0108]

[0109] Where, q * This represents point p. i The corresponding nearest edge point, that is, the nearest point on the edge.

[0110] In this embodiment of the invention, the edge penalty term is a correction term that causes the cluster center to shift in the opposite direction of the edge, ensuring the accuracy of the segmentation boundary.

[0111] The technical solution provided in this invention combines edge information to adjust the optimization function. An edge penalty term is added to the traditional K-means objective function, forming a new optimization objective. This avoids errors in edge detection that might introduce incorrect constraints and affect clustering and segmentation results. The iterative function of the K-means clustering algorithm is optimized as follows:

[0112] The optimized cluster center is shifted in the opposite direction along the edge. If the shift of the optimized cluster center is less than the set threshold, the iteration is terminated and the optimization of the iteration function is completed. The calculation formula (12) is shown below.

[0113]

[0114] in, Let represent the coordinates of the cluster center at iteration t; ε represents the cluster center coordinates at iteration t+1; ε represents the set threshold.

[0115] S5. The optimized K-means clustering algorithm is used to segment the point cloud dataset, and the segmentation results of the point cloud are output.

[0116] The final point cloud segmentation result of the large-aperture telescope mosaic detector in this embodiment of the invention is as follows: Figure 5 and Figure 6 As shown in the figure, the colored bars on the right represent the height information of the stitched detector point cloud dataset. The point cloud segmentation result processed by the method provided in this embodiment of the invention exhibits excellent geometric feature preservation capabilities. Figure 5 It can be seen that the stitching detector module can identify 16 segmented results. From Figure 6 It can be seen that in the three-dimensional height map, the height information mapped by the color gradient reproduces the differences in planar deformation characteristics and tilt angles between different modules of the stitching detector.

[0117] The method flowchart in the embodiment of the present invention is as follows: Figure 7 As shown, the technical solution provided by the embodiments of the present invention can effectively extract spatial structure regions from edge information and determine a unique initial centroid for each region, thereby avoiding the clustering instability caused by random initialization in the K-means algorithm in the prior art and improving the consistency with the real structure.

[0118] This invention verifies the effectiveness of its technical solution through comparative experiments, comparing the traditional K-means algorithm and the DBSCAN algorithm from existing technologies. This invention uses silhouette coefficient and edge matching degree as core evaluation indicators, while also counting the number of iterations. The silhouette coefficient, a general indicator for segmentation quality assessment, has a value range of [-1, 1], reflecting the combined performance of intra-class compactness and inter-class separation; a larger value indicates a more reasonable clustering structure. Edge matching degree is specifically designed for stitching detector scenarios, quantifying the algorithm's ability to preserve geometric boundary features; a value closer to 1 indicates a higher degree of agreement between cluster boundaries and real physical edges. The comparison results are shown in Table 1.

[0119] Table 1

[0120] algorithm Initial centroid Number of iterations Profile coefficient Edge matching degree Traditional k-means yes 15 0.51 0.65 DBSCAN no -- 0.58 0.72 Technical solution of the present invention no 6 0.62 0.85

[0121] As shown in Table 1, the technical solution provided in this embodiment of the invention achieves a contour coefficient of 0.62 using a semi-supervised k-means algorithm based on edge enhancement, which is 21.6% higher than the traditional K-means algorithm. The edge matching degree reaches 0.85, outperforming the compared algorithms. In terms of efficiency, the technical solution provided in this embodiment of the invention converges in only 6 iterations, reducing the number of iterations by 56.8% compared to the traditional K-means algorithm, and completely eliminates the need for manually presetting the initial centroid. The traditional K-means algorithm is limited by the random selection of the initial centroid, and its contour coefficient is significantly higher than that of the technical solution provided in this embodiment of the invention. Although DBSCAN does not require presetting the number of clusters, it is sensitive to the neighborhood radius and the minimum number of points. As shown in Table 1, its edge matching degree drops to 0.72 in the light scattering region, and it cannot directly output the centroid position, making it difficult to meet the requirements of precise measurement.

[0122] Therefore, the technical solution proposed in this embodiment of the invention, a point cloud segmentation method based on edge-enhanced semi-supervised k-means, achieves adaptive high-quality segmentation of point cloud data by fusing morphological processing of edge features with an improved clustering algorithm. Experiments show that morphological closing operations effectively improve edge continuity, the centroid initialization strategy based on closed regions reduces the number of iterations by 56.8%, and avoids manually pre-setting the number of clusters; the introduction of edge constraint terms improves boundary matching accuracy by more than 57%. Compared with traditional methods, this method has higher stability and accuracy in point cloud segmentation using stitched detectors.

[0123] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0124] Although embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

[0125] The specific embodiments of the present invention described above do not constitute a limitation on the scope of protection of the present invention. Any other corresponding changes and modifications made in accordance with the technical concept of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A semi-supervised k-means point cloud segmentation method based on edge enhancement, characterized in that, Includes the following steps: S1. Obtain raw point cloud data, perform noise reduction processing on the raw point cloud data, and obtain a three-dimensional point cloud dataset; S2. Analyze the changes in the normal vectors in the point cloud dataset, extract the geometric edges in the point cloud data, and obtain the edge point cloud set; S3. In the closed region formed by the aggregation of edge points, the centroid is selected by morphological closing operation as the initial center point of the K-means clustering optimization function; S4. Optimize the K-means clustering algorithm based on the edge point cloud; S5. The optimized K-means clustering algorithm is used to segment the point cloud dataset, and the segmentation results of the point cloud are output. Step S4 is as follows: Optimize the objective function of the K-means clustering algorithm: An edge penalty term is added to the objective function of the K-means clustering algorithm, as shown in formula (4): in, Represents a binary membership variable function, when the point The value is 1 if it belongs to cluster j, and 0 otherwise; This represents the i-th 3D point cloud data point within the closed area; This represents the center point of the j-th cluster; express The nearest Euclidean distance to the edge point set E; This represents the penalty coefficient, controlling the strength of the edge constraint; it is greater than 0. If the point... If the cluster is located near the edge, the center of its cluster is penalized. This forces the cluster center away from the edge; Optimize the cluster centers of the K-means clustering algorithm: By taking the partial derivative of the optimized objective function and setting it to zero, the optimized cluster centers are obtained. Optimize the iterative function of the K-means clustering algorithm: The optimized cluster center is shifted in the opposite direction along the edge. If the shift of the optimized cluster center is less than a set threshold, the iteration is terminated, and the optimization of the iteration function is completed.

2. The point cloud segmentation method according to claim 1, characterized in that, The step of obtaining the edge point cloud in step S2 is as follows: Select any point in the point cloud dataset The local surface covariance matrix formed by its k nearest neighbors is shown in formula (1): in, Let represent the centroid of the neighborhood point set, and k represent the number of neighborhood points; The eigenvalues ​​of the covariance matrix are eigenvalues, and the eigenvectors corresponding to the smallest eigenvalues ​​are the normal vectors. Then the normal vector of any point in the cloud dataset is as shown in formula (2): in, Let represent the eigenvalues ​​of the covariance matrix, and . ; By comparison Its neighboring points The boundary is identified by the difference in normal vectors. Its neighboring points When the normal vector is greater than a preset angle threshold, the geometric edges in the point cloud data are extracted, and the edge point cloud set is shown in formula (3): E represents the edge point cloud. Point The normal vector, Point The normal vector, This indicates the preset angle threshold.

3. The point cloud segmentation method according to claim 2, characterized in that, The step of obtaining the initial center point in step S3 is as follows: The edge point cloud is projected onto a two-dimensional plane to obtain a two-dimensional edge projection point set, and the closed region is identified using a contour extraction method. Discretize the two-dimensional edge projection point set into a binary image; The binary image is processed using morphological closing operations to fill edge gaps and form closed regions with continuous edges. Match each continuously closed region with the original point cloud to extract the point set inside the region; The geometric centroid of each internal point set is selected as the initial centroid of the K-means clustering optimization function.

4. The point cloud segmentation method according to claim 3, characterized in that, The specific processing method of the morphological closing operation is as follows: using the dilation-erosion operation of the circular structuring element.

5. The point cloud segmentation method according to claim 1, characterized in that, The method for obtaining raw point cloud data is as follows: scan using a dual-probe structure with a differential structure, and obtain raw point cloud data through a real-time differential algorithm.

6. The point cloud segmentation method according to claim 5, characterized in that, The method for denoising the original point cloud data is: a radius filtering algorithm based on the neighborhood.

Images