Automatic identification method for heterogeneous features of complex three-dimensional CAD model

Abstract

The application relates to a kind of complex three-dimensional CAD model heterogeneous feature automatic identification method, comprising the following steps: S1, the point cloud of three-dimensional CAD model of the object to be detected and multiple types of heterogeneous features is dispersed, respectively obtain the overall point cloud of the object to be detected and the multi-specification template point cloud library;S2, establish adaptive sliding window, the overall point cloud of the object to be detected is slid window, generate first candidate window set;S3, according to the FPFH feature descriptor, the first candidate window set and the multi-specification template point cloud library are roughly matched and screened, obtain second candidate window set;S4, the point cloud in the candidate window of second candidate window set is matched with the multi-specification template point cloud library based on ICP calculation, obtain third candidate window set;S5, according to the maximum neighborhood radius and two-point distance of the candidate window in third candidate window set, carry out deduplication processing.The application can realize the fast, accurate and batch extraction of the heterogeneous features of complex three-dimensional CAD model.

Description

Technical Field

[0001] This invention relates to the field of 3D CAD model feature recognition technology, and in particular to an automated method for recognizing heterogeneous features of complex 3D CAD models. Background Technology

[0002] In the 3D design process of complex equipment or industrial products (such as aerospace vehicles, ship structures, and rail transportation equipment), there are usually a large number of heterogeneous structural features, such as connectors, supports, hole systems, pipe and valve assemblies, and individual equipment. These features play a crucial role in assembly process planning, tooling design, path planning, interference checking, digital assembly verification, and subsequent manufacturing process management. The accuracy of their identification and positioning directly affects the product design quality and R&D efficiency.

[0003] Currently, the identification of the aforementioned structural features in complex 3D CAD models still mainly relies on designers manually browsing the model and annotating each feature one by one. As the complexity of modern equipment design continues to increase, with massive model sizes, numerous components, and densely distributed features, manual methods are not only inefficient but also prone to omissions, misjudgments, or positioning errors. This leads to engineering risks such as design rework, assembly deviations, and tooling mismatches, increasing R&D cycles and manufacturing costs. Although some research has adopted geometric rule-based methods for feature detection, these methods still suffer from insufficient versatility, low automation, and limited recognition accuracy, resulting in limited practical application effectiveness.

[0004] Therefore, there is an urgent need for an automated method for identifying heterogeneous features of complex 3D CAD models, in order to achieve fast, accurate, and batch extraction of heterogeneous features. Summary of the Invention

[0005] To address the technical problems existing in the prior art, the present invention aims to provide an automated method for identifying heterogeneous features of complex 3D CAD models, enabling rapid, accurate, and batch extraction of heterogeneous features, and significantly improving the efficiency and reliability of the design process for complex equipment or industrial products.

[0006] To achieve the above-mentioned objectives, this invention provides an automated method for identifying heterogeneous features of complex 3D CAD models, comprising the following steps:

[0007] Step S1: Discretize the point cloud of the 3D CAD model of the object to be detected and the multi-type heterogeneous features to obtain the overall point cloud of the object to be detected and the point cloud library of multi-specification templates.

[0008] Step S2: Establish an adaptive sliding window based on the template point cloud in the multi-specification template point cloud library, slide the window over the overall point cloud of the object to be detected, and generate a first candidate window set that covers the key areas densely and the non-key areas sparsely.

[0009] Step S3: Based on the FPFH feature descriptors of each candidate window in the first candidate window set, perform coarse matching and filtering between the first candidate window set and the multi-specification template point cloud library to obtain the second candidate window set;

[0010] Step S4: Perform fine matching based on ICP calculation on the point clouds in the candidate windows of the second candidate window set and the multi-specification template point cloud library to obtain the third candidate window set;

[0011] Step S5: Perform deduplication processing on the third candidate window set based on the maximum neighborhood radius and the distance between two points of the candidate windows in the third candidate window set.

[0012] According to a technical solution of the present invention, in step S1, the point cloud is discretized as follows:

[0013]

[0014] in, Represents the original 3D CAD model. This represents the density function of the discretized point cloud. As the baseline density, These are the characteristic weighting coefficients. is the curvature weighting coefficient.

[0015] According to one technical solution of the present invention, step S2 specifically includes:

[0016] Step S21: Using the principal component direction and curvature distribution of each template point cloud in the multi-specification template point cloud library as input, establish an adaptive sliding window for each template point cloud;

[0017] Step S22: Using the adaptive sliding window, slide the window across the entire point cloud of the object to be detected to generate several sets of candidate windows, forming the first candidate window set.

[0018] According to one technical solution of the present invention, in step S21, the adaptive sliding window is an ellipsoidal window with a window size of... Represented as:

[0019]

[0020] in, Point cloud template In the principal component direction The length of the upper part, The sequence number of the principal component direction. ; Point cloud template The average point spacing of the point itself.

[0021] According to a technical solution of the present invention, in step S22, the adaptive step size of the adaptive sliding window along the principal component direction during window sliding is... Represented as:

[0022]

[0023] In the formula, This represents the average Gaussian curvature of the point cloud in the current window. and These are respectively template point clouds The high curvature threshold and low curvature threshold are obtained by calculating Gaussian curvature and performing statistical analysis.

[0024] According to a technical solution of the present invention, in step S3, the coarse matching screening specifically includes:

[0025] Step S31: Extract FPFH feature descriptors from the point cloud within each group of candidate windows in the first candidate window set, and calculate the feature distance between the extracted FPFH feature descriptors and the FPFH template vector of the corresponding template point cloud.

[0026] Step S32: Determine the minimum feature distance of the candidate window. If the minimum feature distance of the current candidate window is greater than the coarse matching screening threshold, discard the current candidate window.

[0027] Step S33: Obtain the second candidate window set based on the coarse matching and filtering results.

[0028] According to one technical solution of the present invention, the formula for calculating the feature distance is as follows:

[0029]

[0030] In the formula, The result is the feature distance calculation, where N is the dimension of the FPFH feature descriptor. The first point cloud in the current candidate window Each FPFH histogram component Point cloud template The One FPFH histogram component;

[0031] According to one technical solution of the present invention, step S4 specifically includes:

[0032] Step S41: Establish the point-to-surface ICP objective function, which is expressed as:

[0033]

[0034] In the formula, The candidate window points are the point clouds within the candidate window. The point in the middle; The template points are the template point cloud. Middle and candidate window points The nearest corresponding point; For template points The unit normal vector at that location, and These are the rotation and translation matrices of the template point cloud relative to the candidate window, respectively.

[0035] The minimum error The calculation method is as follows

[0036]

[0037] In the formula, Point cloud within the candidate window Total number of points;

[0038] Step S42: Perform ICP calculation on the candidate windows in the second candidate window set and all template point clouds in the multi-specification template point cloud library according to the ICP objective function, and obtain the ICP calculation results of the candidate windows and template point clouds. The ICP calculation results include the template point cloud number, transformation matrix, matching center and error value.

[0039] Step S43: Filter the second candidate window set according to the ICP calculation results to obtain the third candidate window set: If the minimum error value in the ICP calculation results of the current candidate window is less than the accuracy threshold, then retain the ICP calculation result corresponding to the minimum error value of the current candidate window and add the current candidate window to the third candidate set.

[0040] According to one technical solution of the present invention, step S5 specifically includes:

[0041] Step S51: For any candidate window k in the third candidate window set, calculate its maximum neighborhood radius. , is represented as:

[0042]

[0043] Step S52: Calculate the other candidate windows in the third candidate window set. Distance between two points and the current candidate window k Distance between two points The calculation formula is expressed as:

[0044]

[0045] in, and These are the three-dimensional center coordinates of the current candidate window and k other candidate windows l, respectively. and These are the minimum error values ​​between the current candidate window k and other candidate windows l, respectively.

[0046] Step S53: Determine whether other candidate windows l and the current candidate window k satisfy the following conditions. If so, then other candidate windows l are considered to be retained; otherwise, other candidate windows l are discarded.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] This invention provides an automated method for identifying heterogeneous features in complex 3D CAD models. The 3D CAD model is discretized into a point cloud model. By combining the designed sliding window, registration, and deduplication steps, the method identifies all entity cases of feature models within the overall model and can subsequently provide adaptable outputs such as 3D center coordinates, attitude angles, and matching errors. This invention solves the problem of insufficient versatility of traditional feature recognition methods by generating a candidate window set through an adaptive sliding window based on template point clouds. It can automatically complete the identification and positioning of various heterogeneous features such as connectors, supports, hole structures, pipe and valve assemblies, and single-unit equipment. It has advantages such as high accuracy and strong robustness, effectively improving the efficiency and reliability of the design process for complex equipment or industrial products, and enhancing the level of digital design and engineering automation for complex equipment or industrial products. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0050] Figure 1 A flowchart of an automated method for identifying heterogeneous features of complex 3D CAD models according to the present invention;

[0051] Figure 2 This is a schematic diagram of feature recognition of a 3D model of a sub-component of a mechanical equipment in Example 1, wherein... Figure 2 (a) is a schematic diagram of a 3D CAD model of a sub-component of a mechanical equipment. Figure 2(b) is the result of multi-type heterogeneous feature recognition obtained by the method of the present invention on this sub-component.

[0052] Figure 3 This is a schematic diagram of feature identification for an industrial pipeline system in Example 2, wherein... Figure 3 (a) is a schematic diagram of a three-dimensional CAD model of an industrial pipeline. Figure 3 (b) shows the results of identifying multiple types of heterogeneous features in the pipeline system obtained by the method of the present invention.

[0053] Figure 4 This is a schematic diagram of feature recognition for the overall structural model of a spacecraft system in Example 3, where... Figure 4 (a) is a schematic diagram of a three-dimensional CAD model of the overall structure of a spacecraft. Figure 4 (b) shows the results of multi-type heterogeneous feature recognition obtained by the method of the present invention on the spacecraft system. Detailed Implementation

[0054] The description of the embodiments in this specification should be taken in conjunction with the accompanying drawings, which should form part of the complete specification. In the drawings, the shape or thickness of the embodiments may be exaggerated and may be indicated in a simplified or convenient manner. Furthermore, parts of the various structures in the drawings will be described separately; it is worth noting that elements not shown in the figures or not described in words are in a form known to those skilled in the art.

[0055] The descriptions of the embodiments herein, including any references to directions and orientations, are for ease of description only and should not be construed as limiting the scope of the invention. The following description of preferred embodiments involves combinations of features, which may exist independently or in combination; the invention is not particularly limited to the preferred embodiments. The scope of the invention is defined by the claims.

[0056] like Figure 1 As shown, the present invention provides an automated method for identifying heterogeneous features of complex 3D CAD models, comprising the following steps:

[0057] Step S1: Discretize the point cloud of the 3D CAD model of the object to be detected and the multi-type heterogeneous features to obtain the overall point cloud of the object to be detected and the point cloud library of multi-specification templates, respectively; wherein, the overall point cloud of the object to be detected is represented as The multi-specification template point cloud library is represented as ,in The number of point clouds in the template;

[0058] In step S1, the point cloud is discretized as follows:

[0059]

[0060] in, The original 3D CAD model is referred to in this invention, which includes the 3D CAD model of the object to be detected and the 3D CAD model with multiple types of heterogeneous features. This represents the density function of the discretized point cloud. As the baseline density, These are the characteristic weighting coefficients. These are the curvature weighting coefficients. In the point cloud discretized using the above method, the density of points with obvious features or curvature is higher, thereby improving the recognition accuracy and efficiency of heterogeneous features.

[0061] Step S2: Establish an adaptive sliding window based on the template point cloud in the multi-specification template point cloud library, slide the window on the overall point cloud of the object to be detected, and generate a first candidate window set that covers the key areas densely and the non-key areas sparsely.

[0062] Step S2 specifically includes:

[0063] Step S21: Use a multi-specification template point cloud library The principal component orientation and curvature distribution of each template point cloud are used as inputs to establish an adaptive sliding window for each template point cloud.

[0064] The adaptive sliding window is an ellipsoidal window, and its ellipsoidal sliding window size is... , is represented as:

[0065]

[0066] in, Point cloud template In the principal component direction The length of the upper part, The sequence number of the principal component direction. ; Point cloud template The average point spacing of the point itself.

[0067] The sliding window generated in the above manner is an ellipsoidal window, which fits the feature shape better than the traditional cube, reducing redundant searches and thus improving the accuracy and efficiency of feature recognition.

[0068] Step S22: Utilize the established adaptive sliding window to detect the overall point cloud of the object to be detected. Perform a sliding window operation to generate several groups of candidate windows, which together form the first candidate window set.

[0069] Among them, the adaptive step size along the principal component direction j during the adaptive sliding window sliding process. Represented as:

[0070]

[0071] In the formula, This represents the average Gaussian curvature of the point cloud in the current window. and These are respectively template point clouds The method calculates Gaussian curvature and performs statistical analysis to obtain high curvature and low curvature thresholds. It achieves adaptive step size calculation, with a small step size in feature-dense regions to improve detection accuracy and a large step size in feature-sparse regions to improve detection efficiency.

[0072] Step S3: Based on the FPFH feature descriptors of each candidate window in the first candidate window set, perform coarse matching and filtering between the first candidate window set and the multi-specification template point cloud library to obtain the second candidate window set;

[0073] For each candidate window, extract FPFH (Fast Point Feature Histogram) feature descriptors from the point cloud and calculate the feature distance with the FPFH template vector in the multi-specification template point cloud library. If the distance of the current window is less than the threshold, proceed to step S4; otherwise, discard the current window.

[0074] Specifically, step S3 includes:

[0075] Step S31: Generate the point cloud within each group of candidate windows in the first candidate window set. FPFH feature descriptors are extracted, and feature distances are calculated between the extracted FPFH feature descriptors and the corresponding FPFH template vectors of the template point cloud.

[0076] Among them, feature distance The calculation formula is:

[0077]

[0078] In the formula, N is the dimension of the FPFH feature descriptor. The first point cloud in the current candidate window Each FPFH histogram component Point cloud template The Each FPFH histogram component.

[0079] Step S32: Determine the minimum feature distance of the candidate window. If the minimum feature distance of the current candidate window is greater than the coarse matching screening threshold, i.e. Discard the current candidate window

[0080] Step S33: Obtain the second candidate window set based on the coarse matching and filtering results.

[0081] The above method can be used to quickly filter point cloud windows that are similar to the template features, thus achieving coarse matching filtering.

[0082] Step S4: Perform fine matching based on ICP calculation on the point clouds in the candidate windows of the second candidate window set and the multi-specification template point cloud library to obtain the third candidate window set;

[0083] Specifically, step S4 includes:

[0084] Step S41: Establish the ICP objective function from point to surface;

[0085] Step S42: Perform ICP calculation on the candidate windows in the second candidate window set and all template point clouds in the multi-specification template point cloud library according to the ICP objective function, and obtain the ICP calculation results of the candidate windows and template point clouds. The ICP calculation results include the template point cloud number, transformation matrix, matching center and error value.

[0086] Step S43: Filter the second candidate window set according to the ICP calculation results to obtain the third candidate window set: If the minimum error value in the ICP calculation results of the current candidate window is less than the accuracy threshold, then retain the ICP calculation result corresponding to the minimum error value of the current candidate window and add the current candidate window to the third candidate set.

[0087] For the point cloud of each candidate window filtered by coarse matching Establish a point-to-surface ICP objective function and perform ICP calculations against all templates in the template library. If the minimum error value of the current candidate window is... If so, retain the result and record the template number under the ICP calculation result. Transformation matrix Matching Center With minimum error .

[0088] The established ICP objective function is expressed as follows:

[0089]

[0090] In the formula, Candidate window points, i.e., the point cloud within the candidate window. The point in the middle; Template points, i.e., template point cloud Middle and candidate window points The nearest corresponding point For template points The unit normal vector at that location, and These are the rotation and translation matrices of the template relative to the candidate window, respectively.

[0091] The minimum error The calculation method is as follows

[0092]

[0093] In the formula, Point cloud within the current candidate window The total number of points in the cloud. High-precision detection of template point clouds can be achieved using the above method.

[0094] Step S5: Based on the maximum neighborhood radius and the distance between two points of the candidate windows in the third candidate window set, perform deduplication processing on the third candidate window set: Calculate the maximum neighborhood radius of the candidate windows based on the ICP calculation result parameters retained in step S4. Distance between two points If the maximum neighborhood radius of the candidate window Distance between two points satisfy If the current window is selected, the current window will be retained; otherwise, the ICP calculation results of the current window will be discarded to eliminate duplicate detection.

[0095] Step S5 specifically includes:

[0096] Step S51: For any candidate window k in the third candidate window set, calculate its maximum neighborhood radius. In order to determine the range of duplicate results, The calculation method is as follows:

[0097]

[0098] Step S52: Calculate the other candidate windows in the third candidate window set. Distance between two points and the current candidate window k To determine other candidate windows Is it within the scope of duplicate result determination for the current candidate window k? The calculation method is as follows:

[0099]

[0100] in, and These are the three-dimensional center coordinates of the current candidate window and k other candidate windows l, respectively. and These are the minimum root mean square errors obtained by the current candidate window and k other candidate windows l in the fine registration process.

[0101] Step S53: Determine whether other candidate windows l and the current candidate window k satisfy the following conditions. If satisfied Then it is considered a candidate window. With the current window Not belonging to the same cluster, the window to be selected Redundant features are retained, while those that are not are discarded. This judgment process precisely filters out redundant features, improving recognition accuracy.

[0102] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0103] Example 1:

[0104] This embodiment uses a 3D CAD model of a sub-component of mechanical equipment as the test object to verify the applicability of the present invention in general industrial component feature recognition scenarios. The 3D CAD model of the sub-component is as follows: Figure 2 As shown in (a). The steps are as follows:

[0105] (1) Discretize the three-dimensional CAD model of the mechanical sub-component into a point cloud with controllable density. Simultaneously, the multi-template 3D CAD models (including but not limited to connectors, bolts, etc.) are discretized into template point clouds using the same sampling strategy. The point cloud discretization method is as follows:

[0106]

[0107] Wherein, the baseline density is Feature weighting coefficients curvature weighting coefficient .

[0108] (2) Establish an adaptive sliding window for the entire point cloud. A sliding window was used to generate a set of 2847 candidate windows that densely cover key areas and sparsely cover non-key areas.

[0109] (3) For the point cloud within each candidate window Extract its FPFH feature descriptor and compare it with the corresponding template point cloud library. Feature distance is calculated using the FPFH template vector. A coarse screening threshold is set. If the current window If the window is selected, proceed to the next step; otherwise, discard the window. A total of 214 windows are retained through this process. The calculation method is as follows:

[0110]

[0111] In the formula, The first point cloud in the current candidate window Each FPFH histogram component Point cloud template The Each FPFH histogram component.

[0112] (4) For each of the coarse matching filters Establish a point-to-surface ICP objective function and perform ICP calculations against all templates in the template library. Set the ICP calculation threshold. If the current candidate window If so, retain the result and record the template number under that result. Transformation matrix Matching Center With minimum error This step yielded 16 candidate windows. The established ICP objective function is:

[0113]

[0114] In the formula, Point cloud within the candidate window The point in the middle, Point cloud template Zhongyu The nearest corresponding point For template points The unit normal vector at that location, and These are the rotation and translation matrices of the template relative to the window, respectively.

[0115] (5) For each result retained in step (4), calculate its maximum neighborhood radius. Distance between two points ,like If the result is positive, the current result is retained; otherwise, the result is discarded to eliminate duplicate detections. After this deduplication step, 13 detection results in 4 categories are obtained, and the center deviation is compared with the actual location. No missed tests and no over-tests.

[0116] The recognition result of this embodiment is as follows: Figure 2 As shown in (b), various heterogeneous features, including type A connector 201, type B connectors 202 and 203, type C connector 204, and bolt 205, were accurately detected and displayed. This embodiment demonstrates that the method has excellent detection and deduplication capabilities for various heterogeneous features in the 3D CAD model of mechanical equipment parts.

[0117] Example 2:

[0118] This embodiment modifies Embodiment 1 by changing the test object to a 3D CAD model of an industrial pipeline system, further verifying the applicability of the present invention in the scenario of feature recognition for pipeline equipment. The 3D CAD model of the pipeline system is as follows: Figure 3 As shown in (a).

[0119] (1) Discretize the CAD model of a pipeline system into a point cloud with controllable density. Simultaneously, three types of 3D CAD models (including but not limited to pipes, valves, and storage tanks) are discretized into multi-specification template point cloud libraries using the same sampling strategy. .

[0120] (2) Establish an adaptive sliding window for the entire point cloud. A sliding window is used to generate a set of 2322 candidate windows that cover key areas densely and non-key areas sparsely.

[0121] (3) For the point cloud within each candidate window Extract its FPFH feature descriptor and compare it with the corresponding template point cloud library. Feature distance is calculated using the FPFH template vector. A coarse screening threshold is set. This process retains a total of 242 windows.

[0122] (4) For each of the coarse matching filters Establish a point-to-surface ICP objective function and perform ICP calculations against all templates in the template library. Set the ICP calculation threshold. This step yields a total of 10 candidate windows.

[0123] (5) For each result retained in step (4), calculate its maximum neighborhood radius. Distance between two points ,like If the result is positive, the current result is retained; otherwise, the result is discarded to eliminate duplicate detections. After this deduplication step, a total of 6 detection results in 4 categories are obtained, and the center deviation is compared with the actual location. No missed tests and no over-tests.

[0124] The difference between this embodiment and Embodiment 1 is that the test object is changed to a 3D CAD model of an industrial pipeline system. This embodiment identifies various heterogeneous feature targets such as pipe valves and storage tanks in the pipeline design, and the method is modified according to the different targets. , Parameter settings, etc. The recognition result of this embodiment is as follows: Figure 3 As shown in (b), multiple heterogeneous features, including type A pipe valves 301 and 302, type B pipe valves 303, type C pipe valves 304, and storage tank 305, were accurately detected and displayed. This embodiment further demonstrates that the method has excellent detection and deduplication capabilities for multiple heterogeneous features in the 3D CAD model of pipeline equipment.

[0125] Example 3:

[0126] This embodiment modifies Embodiments 1 and 2 by changing the test object to a comprehensive 3D CAD model of a complex piece of equipment, further verifying the applicability of the present invention in complex equipment feature recognition scenarios. The 3D CAD model of the complex equipment is as follows: Figure 4 As shown in (a).

[0127] (1) Discretize the overall three-dimensional CAD model of complex equipment into point clouds with controllable density. Simultaneously, multiple types of heterogeneous feature models (including but not limited to solar panels and antennas) are discretized into multi-specification template point cloud libraries using the same sampling strategy. .

[0128] (2) Establish an adaptive sliding window for the entire point cloud. A sliding window is used to generate a set of 1868 candidate windows that cover key areas densely and non-key areas sparsely.

[0129] (3) For each candidate window point cloud Extract its FPFH feature descriptor and compare it with the corresponding template point cloud library. Feature distance is calculated using the FPFH template vector. A coarse screening threshold is set. This process retains a total of 228 windows.

[0130] (4) For each of the coarse matching filters Establish a point-to-surface ICP objective function and perform ICP calculations against all templates in the template library. Set the ICP calculation threshold. This step yields a total of 20 candidate windows.

[0131] (5) For each result retained in step (4), calculate its maximum neighborhood radius. Distance between two points ,like If the result is positive, retain it; otherwise, discard it to eliminate duplicate detections. This step removes duplicates, resulting in 9 detection results across 4 categories. The center deviation is then compared with the actual location. No missed tests and no over-tests.

[0132] The difference between this embodiment and embodiments 1 and 2 is that the test object is changed to a complex overall 3D CAD model of equipment, and the method is changed according to the different targets. , Parameter settings, etc. The recognition result of this embodiment is as follows: Figure 4As shown in (b), various heterogeneous features, including solar panels 401 and 402, thruster protective covers 403, antennas 404 and 405, and clamping devices 406, were accurately detected and displayed. This embodiment further demonstrates that the method has excellent detection and deduplication capabilities for various heterogeneous features in complex equipment 3D CAD models.

[0133] Finally, it should be noted that the above description represents a preferred embodiment of the present invention. It should be pointed out that although preferred embodiments have been described, those skilled in the art, once they understand the basic inventive concept of the present invention, can make various improvements and modifications without departing from the principles described herein. These improvements and modifications should also be considered within the scope of protection of the present invention. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present invention.

Claims

1. A method for automatically identifying heterogeneous features of complex 3D CAD models, characterized in that, Includes the following steps: Step S1: Discretize the point cloud of the 3D CAD model of the object to be detected and the multi-type heterogeneous features to obtain the overall point cloud of the object to be detected and the point cloud library of multi-specification templates. Step S2: Establish an adaptive sliding window based on the template point cloud in the multi-specification template point cloud library, slide the window over the overall point cloud of the object to be detected, and generate a first candidate window set that covers the key areas densely and the non-key areas sparsely. Step S3: Based on the FPFH feature descriptors of each candidate window in the first candidate window set, perform coarse matching and filtering between the first candidate window set and the multi-specification template point cloud library to obtain the second candidate window set; Step S4: Perform fine matching based on ICP calculation on the point clouds in the candidate windows of the second candidate window set and the multi-specification template point cloud library to obtain the third candidate window set; Step S5: Perform deduplication processing on the third candidate window set based on the maximum neighborhood radius and the distance between two points of the candidate windows in the third candidate window set.

2. The method for automated identification of heterogeneous features of complex 3D CAD models according to claim 1, characterized in that, In step S1, the point cloud is discretized as follows: in, Represents the original 3D CAD model. This represents the density function of the discretized point cloud. As the baseline density, These are the characteristic weighting coefficients. is the curvature weighting coefficient.

3. The method for automated identification of heterogeneous features of complex 3D CAD models according to claim 2, characterized in that, Step S2 specifically includes: Step S21: Using the principal component direction and curvature distribution of each template point cloud in the multi-specification template point cloud library as input, establish an adaptive sliding window for each template point cloud; Step S22: Using the adaptive sliding window, slide the window across the entire point cloud of the object to be detected to generate several sets of candidate windows, forming the first candidate window set.

4. The method for automated identification of heterogeneous features of complex 3D CAD models according to claim 3, characterized in that, In step S21, the adaptive sliding window is an ellipsoidal window with a window size of... Represented as: in, Point cloud template In the principal component direction The length of the upper part, The sequence number of the principal component direction. ; Point cloud template The average point spacing of the point itself.

5. The method for automated identification of heterogeneous features of complex 3D CAD models according to claim 4, characterized in that, In step S22, the adaptive step size of the adaptive sliding window along the principal component direction during window sliding is... Represented as: In the formula, This represents the average Gaussian curvature of the point cloud in the current window. and These are respectively template point clouds The high curvature threshold and low curvature threshold are obtained by calculating Gaussian curvature and performing statistical analysis.

6. The method for automatically identifying heterogeneous features of complex 3D CAD models according to claim 1, characterized in that, In step S3, the coarse matching screening specifically includes: Step S31: Extract FPFH feature descriptors from the point cloud within each group of candidate windows in the first candidate window set, and calculate the feature distance between the extracted FPFH feature descriptors and the FPFH template vector of the corresponding template point cloud. Step S32: Determine the minimum feature distance of the candidate window. If the minimum feature distance of the current candidate window is greater than the coarse matching screening threshold, discard the current candidate window. Step S33: Obtain the second candidate window set based on the coarse matching and filtering results.

7. The method for automated identification of heterogeneous features of complex 3D CAD models according to claim 6, characterized in that, The formula for calculating feature distance is: In the formula, The result is the feature distance calculation, where N is the dimension of the FPFH feature descriptor. For the point cloud within the candidate window Each FPFH histogram component Point cloud template The Each FPFH histogram component.

8. The method for automated identification of heterogeneous features of complex 3D CAD models according to claim 1, characterized in that, Step S4 specifically includes: Step S41: Establish the point-to-surface ICP objective function, which is expressed as: In the formula, The candidate window points are the point clouds within the candidate window. The point in the middle; The template points are the template point cloud. Middle and candidate window points The nearest corresponding point; For template points The unit normal vector at that location, and These are the rotation and translation matrices of the template point cloud relative to the candidate window, respectively. The minimum error The calculation method is as follows In the formula, Point cloud within the candidate window Total number of points; Step S42: Perform ICP calculation on the candidate windows in the second candidate window set and all template point clouds in the multi-specification template point cloud library according to the ICP objective function, and obtain the ICP calculation results of the candidate windows and template point clouds. The ICP calculation results include the template point cloud number, transformation matrix, matching center and error value. Step S43: Filter the second candidate window set according to the ICP calculation results to obtain the third candidate window set: If the minimum error value in the ICP calculation results of the current candidate window is less than the accuracy threshold, then retain the ICP calculation result corresponding to the minimum error value of the current candidate window and add the current candidate window to the third candidate set.

9. The method for automatically identifying heterogeneous features of complex 3D CAD models according to claim 8, characterized in that, Step S5 specifically includes: Step S51: For any candidate window k in the third candidate window set, calculate its maximum neighborhood radius. , is represented as: Step S52: Calculate the other candidate windows in the third candidate window set. Distance between two points and the current candidate window k Distance between two points The calculation formula is expressed as: in, and These are the three-dimensional center coordinates of the current candidate window and k other candidate windows l, respectively. and These are the minimum error values ​​between the current candidate window k and other candidate windows l, respectively. Step S53: Determine whether other candidate windows l and the current candidate window k satisfy the following conditions. If so, then other candidate windows l are considered to be retained; otherwise, other candidate windows l are discarded.

Images