A method, device and equipment for point cloud segmentation of complex casing cavity

Abstract

The present invention discloses a method, device and equipment for point cloud segmentation of a complex casing cavity, which belongs to the field of point cloud data segmentation. The present invention first obtains an initial cavity segmentation point cloud set C1 based on a conventional CAD model area search segmentation method; then, the duplicate coordinate points in the set C1 are deleted, and a clustering denoising method is adopted considering the characteristics of the casing cavity to obtain a sparse cavity point cloud segmentation set C2; the edge contour of the set C2 is further estimated and expanded to obtain a boundary extension point cloud set B2; finally, the set C2 is divided into two parts, internal points and boundary points, according to the set B2, and the internal points are searched and expanded in a radius neighborhood, which effectively increases the data volume of the complex casing cavity segmentation. In this way, the present invention introduces the estimation and boundary expansion of the segmentation cavity point cloud and the expansion of the casing cavity point cloud segmentation data volume on the basis of the conventional CAD model area search segmentation method, so that the final cavity segmentation data is more and the segmentation effect is better.

Description

Technical Field

[0001] The present invention belongs to the field of point cloud data segmentation, and more specifically, relates to a method, device and equipment for point cloud segmentation of a complex casing cavity. Background Art

[0002] Point cloud segmentation involves separating a specific region from a complete point cloud. This region can be the topographic point cloud of a specific object in a scene point cloud or a specific feature surface in a specific object point cloud. By applying point cloud segmentation algorithms, the required portion can be easily extracted for analysis, resulting in accurate results. This is particularly true in the aerospace field, where curved parts with multiple deep cavities, such as complex casings, require separate segmentation and analysis of each cavity to determine the correct machining allowance.

[0003] Mainstream methods for point cloud data segmentation include the Random Sampling Consensus (RANSAC) segmentation algorithm, the Euclidean clustering segmentation algorithm, and the CAD model-based region search segmentation algorithm. The RANSAC segmentation algorithm requires the mathematical equation for the features to be segmented, and achieves segmentation by continuously iterating to find the optimal coefficient that satisfies this mathematical equation. Therefore, this method is only suitable for segmenting regular features, such as planes and cylinders. The Euclidean clustering algorithm clusters points based on the Euclidean distance between adjacent points, resulting in multiple cluster sets, thus achieving segmentation. This method requires a clear separation between the segmented area and the entire object and is not suitable for segmenting connected areas. The CAD model-based region search segmentation algorithm first aligns the CAD model with the object point cloud, discretizes the feature faces to be segmented into points, and then performs a nearest neighbor search for corresponding points in the complete point cloud of the object for each discrete point. The searched points are then added to the segmentation set to complete the segmentation. This method can segment any type of feature surface because it is supported by a CAD model. However, since a large number of duplicate points will be found and added to the segmentation set during the corresponding point search process, this method has the disadvantages of sparse segmentation data and poor segmentation effect. Summary of the Invention

[0004] In response to the defects of the existing technology and the need for improvement, the present invention provides a method, device and equipment for point cloud segmentation of a complex casing cavity. The method mainly includes CAD discrete point cloud and corresponding point neighbor search, duplicate point deletion and Euclidean clustering denoising of the segmented cavity point cloud, boundary estimation and boundary expansion of the segmented cavity point cloud, and expansion of the data volume of the casing cavity point cloud segmentation. It aims to solve the problems of insufficient segmentation data and poor segmentation effect existing in the regional search segmentation algorithm based on the CAD model.

[0005] To achieve the above objectives, in a first aspect, the present invention provides a method for point cloud segmentation of a complex casing cavity, comprising:

[0006] S1, the CAD surface corresponding to the cavity to be segmented is discretely sampled into N points, the overall point cloud set W of the casing is used as the search space, and the discrete N points are used as seed points to perform a corresponding point nearest neighbor search to obtain the initial cavity segmentation point cloud set C1;

[0007] S2, delete the duplicate coordinate points in set C1, and cluster the remaining points to obtain multiple cluster sets; select the top M cluster sets with the largest number of cluster points as the sparse cavity point cloud segmentation set C2;

[0008] S3, estimate the boundary of set C2 to obtain the boundary point set B1 of the cavity to be segmented; use set C2 as the search space, and perform a radius neighborhood search with each point in set B1 as a seed point, and merge the searched points and the seed points into the boundary extended point cloud set B2;

[0009] S4, using set W as the search space, iterate each point in set C2. If the point is in set B2, it is directly added to the final cavity segmentation point cloud set F; otherwise, a radius neighborhood search is performed on the point, and the point and all the searched neighboring points are added to set F.

[0010] Furthermore, in the above S1, the CAD model of the single cavity of the casing is a triangular mesh model. For any triangular face, a point p is randomly sampled in the triangular face. r-n Expressed as:

[0011]

[0012] Among them, p 1-nd ,p 2-nd ,p 3-nd are the coordinates of the three vertices of the triangle, and λ1 and λ2 are random values ​​that satisfy the 0-1 normal distribution.

[0013] Furthermore, in S2, duplicate coordinate points in the set C1 are deleted, specifically:

[0014] Take set C1 as the search space and perform a neighbor search on each point in set C1. If the distance between the point and its nearest neighbor is less than the duplicate point distance judgment threshold τ repeat , then delete the point.

[0015] Furthermore, in S2, the remaining points are clustered to obtain multiple cluster sets, specifically:

[0016] S21, randomly select a point p from the set C'1 after deleting duplicate coordinate points c1-s, search to get its k1 nearest neighbor points, if the nearest neighbor point to point p c1-s The distance is less than the clustering threshold τ cluster , then add the point to cluster e s 1≤s≤S, S is the total number of points in the set C'1;

[0017] S22, select e s The k1 nearest neighbor search is repeated for the other points in the cluster and the judgment is made until no points that meet the requirements are added to the cluster e. s Then one clustering is completed and cluster e s The points in are deleted from the set C'1;

[0018] S23, repeat S21 and S22 until there are no remaining points in the set C'1, and finally obtain multiple cluster sets E = {e1, e2, ..., e s ,…}.

[0019] Furthermore, in S3, the boundary of the set C2 is estimated to obtain the boundary point set B1 of the cavity to be segmented, specifically:

[0020] For each coordinate point p in the set C2 c2-i , search and get its k2 nearest neighbor points p c2-j (1≤j≤k2), and use point p c2-i Perform least squares plane fitting with the coordinates of its neighboring points;

[0021] Point p c2-i and its k2 nearest neighbor points p c2-j (1≤j≤k2) is projected onto the fitting plane to obtain the plane projection coordinate p' c2-i and p' c2-j ;

[0022] After completing the plane point projection, use p' c2-i is the origin, point p' c2-i and the first neighbor projection point p' c2-1 Constructed vector is the x-axis, the normal of the fitting plane is the y-axis, and the vector The vector product with the normal of the fitting plane is the z-axis, and the local coordinate system is constructed; for the remaining neighboring projection points p c2-j (2≤j≤k2) and point p' c2-i Constructing vectors Solve for vectors separately The angle α between the x-axis and the y-axis t ,β t (t=1,2…k2-1);

[0023] According to β t The value of αt Make corrections:

[0024] After the correction α t Sort and solve adjacent α after sorting t The angle difference γ t , and obtain the maximum angle difference γ max ; If the maximum angle difference γ max Greater than the boundary point judgment threshold τ θ , then point p c2-i Determine it as a boundary point and keep it in the boundary point set B1.

[0025] Furthermore, in said S4, after obtaining the final cavity segmentation point cloud set F, the method further includes: deleting duplicate coordinate points in the set F.

[0026] To achieve the above objectives, in a second aspect, the present invention provides a device for segmenting a point cloud of a complex casing cavity, comprising:

[0027] The module for determining the initial cavity segmentation point cloud set C1 is used to discretely sample the CAD surface corresponding to the cavity to be segmented into N points, use the overall point cloud set W of the casing as the search space, and use the discrete N points as seed points to perform a corresponding point nearest neighbor search to obtain the initial cavity segmentation point cloud set C1;

[0028] The sparse cavity point cloud segmentation set C2 determination module is used to delete duplicate coordinate points in set C1 and cluster the remaining points to obtain multiple cluster sets; the top M cluster sets with the largest number of cluster points are selected as the sparse cavity point cloud segmentation set C2;

[0029] The boundary extension point cloud set B2 determination module is used to estimate the boundary of the set C2 to obtain the boundary point set B1 of the cavity to be segmented; using the set C2 as the search space and each point in the set B1 as the seed point to perform a radius neighborhood search, the searched points and the seed points are merged together to form the boundary extension point cloud set B2;

[0030] The final cavity segmentation point cloud set F determination module is used to iteratively traverse each point in the set C2 using the set W as the search space. If the point is in the set B2, it is directly added to the final cavity segmentation point cloud set F; otherwise, a radius neighborhood search is performed on the point, and the point and all the searched neighboring points are added to the set F.

[0031] To achieve the above-mentioned objectives, in a third aspect, the present invention provides an electronic device, comprising: a processor; and a memory storing a computer executable program, wherein when the program is executed by the processor, the processor executes the complex casing cavity point cloud segmentation method as described in the first aspect.

[0032] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects:

[0033] The present invention first obtains an initial cavity segmentation point cloud set C1 based on a conventional CAD model area search segmentation method; then, the duplicate coordinate points in set C1 are deleted, and a clustering denoising method is adopted considering the characteristics of the casing cavity, effectively removing outliers and isolated points in the original segmentation result, and obtaining a sparse cavity point cloud segmentation set C2; further edge contour estimation and expansion are performed on set C2 to obtain a boundary extension point cloud set B2; finally, according to set B2, set C2 is divided into two parts: internal points and boundary points, and a radius neighborhood search and expansion is performed on the internal points, effectively increasing the data volume of the complex casing cavity segmentation. In this way, based on the conventional CAD model area search segmentation method, the present invention introduces the estimation and boundary extension of the segmentation cavity point cloud and the expansion of the casing cavity point cloud segmentation data volume, so that the final cavity segmentation data obtained is more and the segmentation effect is better. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A flow chart of a complex casing cavity point cloud segmentation method provided by an embodiment of the present invention;

[0035] Figure 2(a) and Figure 2(b) are the complete casing point cloud model and the single cavity CAD model, respectively;

[0036] Figure 3 This is the sparse cavity point cloud segmentation result obtained using the original segmentation method;

[0037] Figure 4 This is the result of edge contour estimation for sparse cavity point cloud;

[0038] Figure 5 This is the result of expanding the edge contour of the cavity point cloud;

[0039] Figure 6 This is the point cloud segmentation result of the casing cavity obtained by using the improved segmentation method;

[0040] Figure 7 A comparison of the amount of segmentation data obtained using the original and improved segmentation methods;

[0041] Figure 8(a) and Figure 8(b) show the segmentation results obtained using the original and improved segmentation methods, respectively. DETAILED DESCRIPTION

[0042] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely for the purpose of explaining the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

[0043] In the present invention, the terms "first", "second", etc. (if any) in the present invention and the drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0044] Example 1

[0045] See Figure 1 The present invention provides a method for point cloud segmentation of a complex casing cavity, which includes operations S1 to S4.

[0046] In operation S1, the CAD surface corresponding to the cavity to be segmented is discretely sampled into N points, the overall point cloud set W of the casing is used as the search space, and the discrete N points are used as seed points to perform a corresponding point nearest neighbor search to obtain the initial cavity segmentation point cloud set C1.

[0047] Before operation S1, the CAD model of the casing part and the point cloud model are also registered.

[0048] As shown in FIG2(a) and FIG2(b), this embodiment takes the single cavity CAD model of the casing as a triangular mesh model as an example. In fact, it can also be a polygon, a curved surface, etc. Assume that the single cavity triangular facet set is V = {v1, v2…v m Set the number of target points for discrete sampling to N, generate one point randomly each time, and repeat the iteration N times. In the nth (1≤n≤N) discrete process, randomly extract an index value id (1≤id≤m) according to the subscript range of the triangle face set to obtain the target face v to be discretized id According to the three vertex coordinates p of the target patch 1-nd ,p 2-nd ,p 3-nd , randomly sample a point p in the triangle r-n :

[0049]

[0050] In the above formula, λ1 and λ2 are random values ​​that satisfy the 0-1 normal distribution.

[0051] After the discretization of the single cavity CAD model of the casing is completed, the discrete point cloud set T is obtained. The overall point cloud set W of the casing measured by actual scanning is used as the search space to construct a kdtree. For each point in the set T, the nearest corresponding point in W is searched and retained to obtain the initial cavity segmentation point cloud set C1.

[0052] Operation S2: delete the duplicate coordinate points in set C1 and cluster the remaining points to obtain multiple cluster sets; select the top M cluster sets with the largest number of cluster points as the sparse cavity point cloud segmentation set C2.

[0053] In the process of deleting duplicate points, C1 is used as the search space to construct a kdtree, and a neighbor search is performed on each point in the set C1. If the distance between the point and its nearest neighbor is less than the duplicate point distance judgment threshold τ, repeat , then delete the point; where τ repeat Take a smaller value, usually τ repeat =1×10 -5 .

[0054] After deleting duplicate points, there are still some outliers and isolated points in the segmented set. If they are not removed, it will affect the calculation results of the margin of the inner and outer walls of the entire cavity and the support plate surface. Since the inner and outer walls of the cavity and the support plate surface constitute three large connected domains, the clustering denoising method can effectively remove isolated and outliers. The specific denoising process is as follows:

[0055] S21, randomly select a point p from the set C'1 after deleting duplicate coordinate points c1-s , search to get its k1 nearest neighbor points, if the nearest neighbor point to point p c1-s The distance is less than the clustering threshold τ cluster , then add the point to cluster e s 1≤s≤S, S is the total number of points in the set C'1;

[0056] S22, select e s The k1 nearest neighbor search is repeated for the other points in the cluster and the judgment is made until no points that meet the requirements are added to the cluster e. s Then one clustering is completed and cluster e s The points in are deleted from the set C'1;

[0057] S23, repeat S21 and S22 until there are no remaining points in the set C'1, and finally obtain multiple cluster sets E = {e1, e2, ..., e s ,…}.

[0058] Furthermore, the cluster set E is sorted in descending order according to the number of cluster points. According to the characteristics of the single cavity (three connected domains of the inner and outer walls and the left and right support plates), the first three clusters in the cluster set are selected as the sparse cavity point cloud segmentation set C2 = {e1, e2, e3}, as shown in Figure 3 shown.

[0059] In operation S3, the boundary of set C2 is estimated to obtain the boundary point set B1 of the cavity to be segmented; set C2 is used as the search space, and each point in set B1 is used as a seed point to perform a radius neighborhood search, and the searched points and seed points are merged together to form the boundary extended point cloud set B2.

[0060] For each coordinate point p in C2 c2-i When determining whether it is a boundary point, first search for the k2 nearest neighbor points p of the point c2-j (1≤j≤k2). Then, the coordinates of this point and its neighboring points are used to perform a least squares plane fit to obtain the fitted plane equation:

[0061] Ax+By+Cz+D=0

[0062] In the above formula, A, B, C, and D are the coefficients of the fitting plane equation, and the unit normal vector of the plane is

[0063] Point p c2-i =(x c2-i ,y c2-i ,z c2-i ) and its k2 nearest neighbor points p c2-j =(x c2-j ,y c2-j ,z c2-j ) is projected onto the fitting plane to obtain the plane projection coordinate p' c2-i =(x' c2-i ,y' c2-i ,z' c2-i ) and p' c2-j =(x' c2-j ,y' c2-j ,z' c2-j ):

[0064]

[0065] The above formula is to find the projection point p' c2-i Specific formula, projection point p' c2-j The solution is the same.

[0066] After completing the plane point projection, use p' c2-i is the origin, point p' c2-i and the first neighbor projection point p' c2-1 Constructed vector is the x-axis, the normal of the fitting plane is the y-axis, and the vector The vector product with the plane normal is the z-axis, and the local coordinate system is constructed. For the remaining neighboring projection points p c2-j (2≤j≤k2) and point p' c2-i Constructing vectors Solve for vectors separately The angle α between the x-axis and the y-axis t ,β t (t=1,2…k2-1).

[0067] To ensure The angle with the x-axis is clockwise, and further calculation is required based on β t The value of α t Make corrections:

[0068] Further correction of α t Sort in ascending order and solve adjacent α after sorting t The angle difference γ t :

[0069]

[0070] Set the boundary point judgment threshold τ θ , the angle difference γ t Sort in ascending order to obtain the maximum angle difference γ max If the maximum angle difference γ max Greater than the boundary point judgment threshold τ θ Then the point is determined to be a boundary point and retained in the boundary point set B1, such as Figure 4 shown.

[0071] The points in the boundary point set B1 are only the contours of the outermost circle of the cavity. The sparse cavity point cloud segmentation set C2 is further used as the search space, and the appropriate expansion radius r is selected. k For every point b in B1 i Perform radius neighborhood search, if the midpoint of C2 is to b i The distance is less than the expansion radius r k , then add it to the extended boundary band B2. After completing the iterative search of each point, a cavity boundary contour band with a certain width is obtained, such as Figure 5 shown.

[0072] Operation S4, using set W as the search space, iterates through each point in set C2. If the point is in set B2, it is directly added to the final cavity segmentation point cloud set F; otherwise, a radius neighborhood search is performed on the point, and the point and the searched neighboring points are all added to set F.

[0073] According to the cavity boundary expansion point cloud set B2, the sparse cavity point cloud segmentation set C2 can be divided into two parts: internal points and boundary points. For each point in C2, the complete casing point cloud W is used as the search space. If a point in C2 is located in the boundary zone, it is not expanded and is directly added to the single cavity complete segmentation set F. On the contrary, if it is located in the internal point, the appropriate expansion search radius r is taken. e Perform a spatial radius neighborhood search and add the point and the searched radius r e All the neighborhood points in are added to the segmentation set F.

[0074] After completing the expansion of all points in C2, the cavity complete point cloud data F will still contain a certain amount of duplicate coordinate points. The same operation of deleting duplicate points is adopted, and after filtering, the final casing cavity point cloud data can be obtained, such as Figure 6 shown.

[0075] To verify the effectiveness of the proposed improved casing segmentation method, a segmentation comparison was conducted using the original segmentation method and the improved method. The original segmentation method is a CAD model-based region search segmentation method. Its main segmentation process involves discretizing the cavity CAD model and then performing nearest neighbor search segmentation. The experimental data is a replica of a certain aircraft engine casing. The scanned and measured complete casing point cloud contains 3,799,887 points, and the number of discrete CAD point clouds used for segmentation is 200,000. Figure 7 Comparing the results of segmenting a single cavity of the casing using two different methods, the CAD model-based region search segmentation method (the original segmentation method) can only segment 111,870 valid data points, while the proposed improved segmentation method can segment 250,915 data points, an increase of approximately 2.2 times compared to the original segmentation method. Figures 8(a) and 8(b) show a visual comparison of the results of the two segmentation methods. It can be clearly seen that the cavity point cloud obtained by the original segmentation method contains a large number of sparse holes and incomplete surface morphology, while the cavity point cloud obtained by the improved segmentation method has dense data and complete surface morphology.

[0076] Example 2

[0077] A complex casing cavity point cloud segmentation device, comprising:

[0078] The module for determining the initial cavity segmentation point cloud set C1 is used to discretely sample the CAD surface corresponding to the cavity to be segmented into N points, use the overall point cloud set W of the casing as the search space, and use the discrete N points as seed points to perform a corresponding point nearest neighbor search to obtain the initial cavity segmentation point cloud set C1;

[0079] The sparse cavity point cloud segmentation set C2 determination module is used to delete duplicate coordinate points in set C1 and cluster the remaining points to obtain multiple cluster sets; the top M cluster sets with the largest number of cluster points are selected as the sparse cavity point cloud segmentation set C2;

[0080] The boundary extension point cloud set B2 determination module is used to estimate the boundary of the set C2 to obtain the boundary point set B1 of the cavity to be segmented; using the set C2 as the search space and each point in the set B1 as the seed point to perform a radius neighborhood search, the searched points and the seed points are merged together to form the boundary extension point cloud set B2;

[0081] The final cavity segmentation point cloud set F determination module is used to iteratively traverse each point in the set C2 using the set W as the search space. If the point is in the set B2, it is directly added to the final cavity segmentation point cloud set F; otherwise, a radius neighborhood search is performed on the point, and the point and all the searched neighboring points are added to the set F.

[0082] The relevant technical solutions are the same as those in Example 1 and will not be described again here.

[0083] Example 3

[0084] An electronic device includes: a processor; and a memory storing a computer-executable program. When the program is executed by the processor, the processor executes the complex casing cavity point cloud segmentation method as described in the first embodiment.

[0085] The relevant technical solutions are the same as those in Example 1 and will not be described again here.

[0086] It will be easily understood by those skilled in the art that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A complex casing cavity point cloud segmentation method, characterized in that: The following steps are involved: S1, discretely sample the CAD surface corresponding to the cavity to be segmented into points, based on the overall point cloud of the casing As the search space, the discrete Points are used as seed points to perform a neighbor search for corresponding points to obtain the initial cavity segmentation point cloud set. ; S2, delete the collection The repeated coordinate points in , and cluster the remaining points to obtain multiple cluster sets; Select the top M cluster sets with the largest number of cluster points as the sparse cavity point cloud segmentation set ; S3, for collections Perform boundary estimation to obtain the boundary point set of the cavity to be segmented ; to gather is the search space, with the set Each point in the radius neighborhood is a seed point for search, and the searched points and seed points are merged into a boundary extended point cloud set. ; S4, to gather Iterate through the collection to search the space For every point in the set In the final cavity segmentation point cloud set, it is directly added to the final cavity segmentation point cloud set Otherwise, perform a radius neighborhood search on the point and add the point and all the searched neighboring points to the set middle; In the S3, the set Perform boundary estimation to obtain the boundary point set of the cavity to be segmented , specifically: Pair Collection Each coordinate point , search for it Neighbors , and use the point Perform least squares plane fitting with the coordinates of its neighboring points; Point and its Neighbors Project onto the fitting plane to obtain the plane projection coordinates and ; After completing the plane point projection, is the origin, point Projected point with the first nearest neighbor Constructed vector is the x-axis, the normal of the fitting plane is the y-axis, and the vector The vector product with the normal of the fitting plane is the z-axis, and the local coordinate system is constructed; for the remaining neighboring projection points Respectively with point Constructing vectors ; Solve for vectors separately Angles with the x-axis and y-axis ; according to The value pair Make corrections: ; After the correction Sort and solve adjacent The angle difference between , and obtain the maximum angle difference ; If the maximum angle difference Greater than the boundary point judgment threshold , then point Determine it as a boundary point and keep it in the boundary point set middle.

2. The complex casing cavity point cloud segmentation method according to claim 1, characterized in that: In the above S1, the CAD model of the single cavity of the casing is a triangular mesh model. For any triangular face, a point is randomly sampled in the triangular face. Expressed as: in, are the coordinates of the three vertices of the triangle. is a random value that satisfies the 0-1 normal distribution.

3. The complex casing cavity point cloud segmentation method according to claim 1, characterized in that: In S2, the deletion set The repeated coordinate points in are: By collection is the search space, for the set Perform a neighbor search for each point in the , if the distance between the point and its nearest neighbor is less than the duplicate point distance judgment threshold , then delete the point.

4. The complex casing cavity point cloud segmentation method according to claim 1 or 3, characterized in that: In S2, the remaining points are clustered to obtain multiple cluster sets, specifically: S21, the set after deleting duplicate coordinate points Randomly select a point , search for it neighbor points, if the neighbor point to point The distance is less than the clustering threshold , then add the point to the cluster middle; , For collection Total number of midpoints; S22, select Re-do the other point Neighbor search and judgment are performed until no points that meet the requirements are added to the cluster , then complete one clustering and cluster Points in the set Delete in; S23, repeat S21 and S22 until the collection There are no remaining points in the final result, and multiple cluster sets are obtained. .

5. The complex casing cavity point cloud segmentation method according to claim 1, characterized in that: In S4, the final cavity segmentation point cloud set is obtained After that, it also includes: Deleting the collection Duplicate coordinate points in .

6. A point cloud segmentation device for a complex casing cavity, characterized in that: The method for executing the complex casing cavity point cloud segmentation method according to any one of claims 1 to 5 comprises: Initial cavity segmentation point cloud collection Determination module, used to discretely sample the CAD surface corresponding to the cavity to be segmented points, based on the overall point cloud of the casing As the search space, the discrete Points are used as seed points to perform a neighbor search for corresponding points to obtain the initial cavity segmentation point cloud set. ; Sparse cavity point cloud segmentation set Determine module, used to delete collections The repeated coordinate points in the cluster are clustered to obtain multiple cluster sets; the top M cluster sets with the largest number of cluster points are selected as the sparse cavity point cloud segmentation set ; Boundary Extended Point Cloud Collection Determine the module for the collection Perform boundary estimation to obtain the boundary point set of the cavity to be segmented ; to gather is the search space, with the set Each point in the radius neighborhood is a seed point for search, and the searched points and seed points are merged into a boundary extended point cloud set. ; Final cavity segmentation point cloud collection Determine the module for collecting Iterate through the collection to search the space For every point in the set In the final cavity segmentation point cloud set, it is directly added to the final cavity segmentation point cloud set Otherwise, perform a radius neighborhood search on the point and add the point and all the searched neighboring points to the set middle.

7. An electronic device, characterized in that: include: processor; A memory storing a computer executable program, wherein when the program is executed by the processor, the processor executes the complex casing cavity point cloud segmentation method according to any one of claims 1 to 5.

Images