[go: up one dir, main page]

CN116703989B - Point cloud processing method for keeping non-rigid registration based on deep learning and self-adaptive topology - Google Patents

Point cloud processing method for keeping non-rigid registration based on deep learning and self-adaptive topology

Info

Publication number
CN116703989B
CN116703989B CN202310577670.9A CN202310577670A CN116703989B CN 116703989 B CN116703989 B CN 116703989B CN 202310577670 A CN202310577670 A CN 202310577670A CN 116703989 B CN116703989 B CN 116703989B
Authority
CN
China
Prior art keywords
point cloud
contour
measurement point
points
contour measurement
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202310577670.9A
Other languages
Chinese (zh)
Other versions
CN116703989A (en
Inventor
罗明
裴昊男
周文静
王泽宇
张璞玉
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Northwestern Polytechnical University
Original Assignee
Northwestern Polytechnical University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Northwestern Polytechnical University filed Critical Northwestern Polytechnical University
Priority to CN202310577670.9A priority Critical patent/CN116703989B/en
Publication of CN116703989A publication Critical patent/CN116703989A/en
Application granted granted Critical
Publication of CN116703989B publication Critical patent/CN116703989B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
    • G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
    • G06T3/4038—Image mosaicing, e.g. composing plane images from plane sub-images
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/13—Edge detection
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/04—Constraint-based CAD
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/08—Probabilistic or stochastic CAD
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2200/00—Indexing scheme for image data processing or generation, in general
    • G06T2200/32—Indexing scheme for image data processing or generation, in general involving image mosaicing
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10028—Range image; Depth image; 3D point clouds
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Physics (AREA)
  • General Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Length Measuring Devices With Unspecified Measuring Means (AREA)
  • Length Measuring Devices By Optical Means (AREA)

Abstract

为克服不规则截面环形构件的截面轮廓测量点云中存在的严重异常值、缺失与噪声叠加而导致的截面测量精度难以保证，而传统点云处理方法难以对测量点云有效处理的问题，本发明提出了一种基于深度学习与自适应拓扑保持非刚配准的点云处理方法，利用PointNet++网络深度学习的强函数拟合能力和特征学习能力，构建异常值分割点网，通过分层网络架构与多尺度分组提取局部精细特征，实现轮廓测量点云的连续异常值剔除；并在非刚配准期间求解基于概率密度估计函数建立的目标函数中引入全局拓扑约束与局部拓扑约束，解决了非刚性配准涉及的复杂形变问题。最终通过非刚性配准后的CAD模型点云替代测量点云，得到能满足几何质量评估精度的点云。

To overcome the difficulties in ensuring cross-sectional measurement accuracy due to severe outliers, missing values, and noise superposition in the cross-sectional profile measurement point cloud of irregular-section annular components, and the difficulty of effectively processing the measured point cloud using traditional point cloud processing methods, this paper proposes a point cloud processing method based on deep learning and adaptive topology-preserving non-rigid registration. This method utilizes the strong function fitting and feature learning capabilities of the PointNet++ network deep learning to construct an outlier segmentation point network. This method extracts local fine features through a layered network architecture and multi-scale grouping to achieve continuous outlier removal from the profile measurement point cloud. Furthermore, during the non-rigid registration process, global and local topological constraints are introduced into the objective function established based on the probability density estimation function, resolving the complex deformation issues associated with non-rigid registration. Ultimately, the measured point cloud is replaced by the non-rigidly registered CAD model point cloud, resulting in a point cloud that meets the accuracy requirements for geometric quality assessment.

Description

Point cloud processing method for keeping non-rigid registration based on deep learning and self-adaptive topology
Technical Field
The invention belongs to the field of complex component measurement, and particularly relates to a point cloud processing method for keeping non-rigid registration based on deep learning and self-adaptive topology, which is particularly suitable for the field of geometric quality detection of irregular section annular components in high-end equipment such as aviation, aerospace, nuclear power and the like.
Background
The irregular cross section annular member is one of key members widely applied to high-end equipment such as aviation, aerospace, nuclear power and the like, and is difficult to detect due to the characteristics of irregular cross section, complex cross section shape and the like, so that forming quality cannot be guaranteed. Taking a high-temperature alloy W-shaped sealing ring as an example, the whole high-temperature alloy W-shaped sealing ring is of an annular closed structure, has small cross section outline dimension and typical circular characteristic and steep inclined wall characteristic, so that the geometric quality detection of the high-temperature alloy W-shaped sealing ring is seriously dependent on destructive sampling final detection, and the quality detection means in the forming process is lost. In recent years, optical measurement has attracted great interest by virtue of portability, flexibility, and high measurement accuracy. However, when the line laser profile sensor is used for nondestructively measuring the cross-sectional profile of the high-temperature alloy W-shaped sealing ring, the following problems exist:
1. Under the influence of internal and external environmental factors, the measurement point cloud generally has serious outliers (especially continuous outliers), missing and noise superposition problems, and the traditional point cloud processing method (such as denoising, repairing and the like) is difficult to cooperatively process, so that subsequent geometric quality evaluation cannot be performed. For example, the optical measurement method is easily affected by internal and external environmental factors, when the internal and external environments are not timely, the irregular section profile measurement point cloud generates a continuous outlier, the outlier is different from a common outlier (sparse isolated point deviated from the main body of the measurement point cloud), the appearance form of the outlier is a large number of continuous pseudo points deviated from the main body of the measurement point cloud, the conventional outlier removing method (such as an outlier removing method based on statistics or radius thought) cannot effectively remove the points, the true profile points are isolated due to local severe loss, and are mistakenly removed by a denoising algorithm, and the conventional interpolation-based deletion repairing method is interfered by the continuous outlier, so that the error repair is carried out between the continuous outlier and the profile points. The multi-view registration measurement is an effective method for solving the defect, denoising is carried out on the basis, but the multi-view registration often needs to arrange a plurality of sensors or design a multi-degree-of-freedom motion measurement platform, which is expensive and complex, and when a high-temperature alloy W-shaped seal ring with smaller cross section outline dimension is measured, the layout of the sensors and the measurement motion planning are limited, and the multi-view registration cannot be carried out.
2. The density dissimilarity and the disorder of the measurement point cloud are the key problems to be solved in the field of optical measurement, the measurement point cloud with the density dissimilarity and the disorder is unfavorable for subsequent processing and geometric quality assessment due to the fact that the irregular cross-section profile is complex and difficult to process due to the fact that the resampling and the sorting algorithm are seriously dependent on after fitting.
Therefore, processing and converting the measurement point cloud with serious abnormal values, missing and noise superposition problems into ordered and consistent density point cloud data becomes a key problem to be solved in order to realize accurate measurement of the cross section of the irregular cross section annular member.
Disclosure of Invention
In order to overcome the technical problems that the section measurement accuracy is difficult to guarantee due to the fact that serious abnormal values, defects and noise are superposed in the section profile measurement point cloud of the irregular section annular member, and the traditional point cloud processing method is difficult to effectively process the measurement point cloud, the invention provides a point cloud processing method based on deep learning and self-adaptive topology keeping non-rigid registration.
In order to achieve the above purpose, the invention adopts the following technical scheme:
The point cloud processing method based on deep learning and self-adaptive topology keeping non-rigid registration is characterized by being used for processing the original contour measurement point cloud of the irregular section annular member and comprising the following steps of:
dividing the original contour measurement point cloud into contour points and outlier points by adopting PointNet ++ network architecture, and removing the outlier points to obtain inner contour point cloud and outer contour point cloud after removing the outlier points;
step 2, splicing the inner contour point cloud and the outer contour point cloud with abnormal value points removed to obtain a spliced contour measurement point cloud;
And 3, under the condition of adding global and local constraints, keeping non-rigid registration of the CAD model point cloud self-adaptive topology of the irregular section annular member to the spliced profile measurement point cloud obtained in the step 2 to obtain a non-rigid registered CAD model point cloud, and replacing the profile measurement point cloud with the non-rigid registered CAD model point cloud to serve as a point cloud for subsequent geometric quality evaluation of the irregular section annular member.
Further, in step 1, the method for dividing the original contour measurement point cloud into contour points and outlier points by adopting PointNet ++ network architecture is as follows:
Firstly, gradually extracting local fine features of the original contour measurement point cloud along a hierarchical structure in multiple scales;
And then, gradually transmitting the local fine features of the sub-sampling points back to the original contour measurement point cloud to obtain the point scores of the original contour measurement point cloud, and dividing the original contour measurement point cloud into contour points and outlier points according to the point scores.
Further, the method for splicing the point clouds in the step 2 is as follows:
And carrying out coordinate transformation based on a pre-calibration result of a point cloud acquisition system for acquiring the original contour measurement point cloud, transforming the outer contour point cloud with the abnormal values removed into an inner contour point cloud coordinate system to realize point cloud splicing, or transforming the inner contour point cloud with the abnormal values removed into the outer contour point cloud coordinate system to realize point cloud splicing.
Further, the method for maintaining rigid registration of the adaptive topology in the step 3 is as follows:
firstly, transforming the CAD model point cloud of the irregular section annular member to the registration problem of the spliced profile measurement point cloud, and converting the registration problem into a probability density estimation function:
wherein:
x N×D represents the post-stitching contour measurement point cloud;
n represents the number of points in the spliced contour measurement point cloud;
d represents the dimension of the contour measurement point cloud after splicing;
x n represents the points within the post-stitching contour measurement point cloud;
m represents the number of points in the CAD model point cloud;
Omega is a weight coefficient representing the prior level of noise;
πM+1=ω;
m=M+1;
σ 2 is the anisotropic variance;
is a non-rigid transformation;
θ is a non-rigid transformation parameter;
y m represents a point within the CAD model point cloud;
then, iteratively solving the anisotropic variance σ 2 and the non-rigid transformation parameter θ in the probability density estimation function by a modified EM algorithm until convergence, thereby obtaining an optimally aligned non-rigid transformation matrix At this time, the probability density estimation function is solved;
Finally, a non-rigid transformation matrix aligned by the optimization Non-rigidly registering the CAD model point cloud to the spliced contour measurement point cloud;
the improved EM algorithm is specifically as follows:
The probability of matching p old(m|xn of the point x n in the contour measurement point cloud with the point y m in the CAD model point cloud is calculated in step E),
Introducing global topological constraint and local topological constraint into an objective function established based on the probability density estimation function in the step M, and then minimizing the objective function after introducing constraint to find a new non-rigid transformation parameter theta and an anisotropic variance sigma 2.
Further, a step of updating a weight coefficient omega representing the prior level of the noise is added in the step M, so as to realize the self-adaption of the judgment noise in the registration process.
The invention also provides a point cloud processing system based on deep learning and self-adaptive topology keeping non-rigid registration, which is characterized by comprising the following steps:
The outlier dividing point network module is realized by adopting PointNet ++ network architecture and is used for dividing the contour points and outlier points in the original contour measurement point cloud, and eliminating the outlier points to obtain inner and outer contour point clouds with outlier points removed;
the point cloud splicing module is used for splicing the inner contour point cloud and the outer contour point cloud after abnormal value points are removed;
the self-adaptive topological keeping non-rigid registration module is used for non-rigidly registering the CAD model point cloud of the irregular section annular member to the contour measurement point cloud obtained after splicing to obtain the non-rigidly registered CAD model point cloud, and the non-rigidly registered CAD model point cloud is used for subsequent geometric quality evaluation of the irregular section annular member.
Further, the abnormal value division point network module comprises a first layer combination and a second layer combination which are sequentially arranged;
The first hierarchical combination comprises a plurality of groups of sampling layers, grouping layers and feature extraction layers which are sequentially arranged, and local fine features of the original contour measurement point cloud are gradually extracted along the hierarchical structure in a multi-scale mode through the sampling layers, the grouping layers and the feature extraction layers;
The second hierarchical combination comprises interpolation and cross-level jump links which are arranged corresponding to the first hierarchical combination, local fine features of the sub-sampling points are gradually transmitted back to the original contour measurement point cloud along the hierarchical structure through the interpolation and the cross-level jump links, each point score of the original contour measurement point cloud is obtained, and each point in the original contour measurement point cloud is divided into a contour point and an outlier point according to each point score.
Further, the adaptive topology preserving non-rigid registration module includes a non-rigid transformation matrixThe non-rigid transformation matrixNon-rigid registration of CAD model point cloud for irregular section annular member to post-spliced contour measurement point cloud, non-rigid transformation matrixThe method comprises the following steps of:
first, CAD model point cloud Assuming a Gaussian mixture model centroid, measuring a point cloud of the contour after splicingAssuming corresponding data, the CAD model point cloud is obtainedTransforming to the post-splicing contour measurement point cloudThe registration problem of (2) is converted into a probability density estimation function:
Wherein X N×D represents the contour measurement point cloud, N represents the points in the contour measurement point cloud, D represents the dimensions of the post-splice contour measurement point cloud, X n represents the points in the post-splice contour measurement point cloud (n=1,..once., N), M represents the points in the CAD model point cloud; ω is a weight coefficient representing the noise a priori level, pi M+1 = ω; m=M+1; σ 2 is the anisotropic variance; being a non-rigid transformation matrix, θ being a non-rigid transformation parameter; y m represents a point within the CAD model point cloud;
Then, the anisotropic variance sigma 2 and the non-rigid transformation parameter theta in the probability density estimation function are iteratively solved through the improved EM algorithm until convergence, and the probability density estimation function is solved completely at the moment, so that the optimization aligned non-rigid transformation matrix is obtained
The improved EM algorithm is specifically as follows:
Calculating the matching probability p old(m|xn of a point x n in the spliced profile measurement point cloud and a point y m in the CAD model point cloud in the step E of the EM algorithm;
The objective function Q (θ, σ 2) established based on the probability density estimation function in the M steps of the EM algorithm, and adding global and local topology constraints thereto, and then minimizing the constrained objective function Q 1(W,σ2) finds new parameters:
Wherein, the For the estimated current number of matching points p old(m|xn) is the prior matching probability,Alpha and lambda represent two trade-off parameters between two topology constraint terms.
The invention also provides a storage medium, on which a computer program is stored, characterized in that the computer program when run by a processor performs the above method.
The invention also provides electronic equipment which comprises a processor and a storage medium, wherein the storage medium is stored with a computer program, and the electronic equipment is characterized in that the computer program is executed by the processor to perform the method.
The beneficial effects of the invention are as follows:
1. The invention transforms (rotates, translates and deforms) the CAD model point cloud to the measurement point cloud based on the self-adaptive topology maintaining non-rigid registration. Based on the robustness of non-rigid registration, abnormal values, missing and noise can be automatically judged in the process of transforming the CAD model point cloud into the measurement point cloud. When the optimization alignment of the two is realized, namely, after non-rigid registration, the transformed CAD model point cloud is used for replacing the measured point cloud, the problems of abnormal value, missing and noise superposition in the point cloud, which are difficult to solve in the traditional point cloud processing method, are overcome, meanwhile, the processing result is ordered and consistent-density point cloud data, a good data basis is provided for accurately measuring the cross-section profile of the irregular cross-section annular member, and the subsequent geometric quality evaluation is facilitated. Six high-temperature alloy sealing rings are used as measuring objects, the average deviation between the processing point cloud and the traditional destructive measuring data is between 8 and 15 mu m, and the processing point cloud has very high precision.
2. The traditional non-rigid registration algorithm generally deals with abnormal values in the measurement point cloud in a pre-modeling mode, and the pre-modeling mode can effectively process common abnormal values, so that registration robustness is ensured. However, in irregular section contour measurement, internal and external environment discomfort often causes continuous outliers in a measurement point cloud, the continuous outliers are quite similar to contour point features, at the moment, a mode of modeling in advance cannot reject the continuous outliers and further cause subsequent non-rigid registration failure, and the problem solved by the traditional non-rigid registration does not relate to the continuous outliers, so that the method establishes a foundation for subsequent non-rigid registration by utilizing strong function fitting capacity and feature learning capacity of PointNet ++ network deep learning to improve the robustness of the non-rigid registration algorithm to the continuous outliers, constructs an outlier division point network, and extracts local fine features through a hierarchical network architecture and multi-scale grouping.
3. When the measured profile differs greatly from the corresponding CAD model, the non-rigid registration involves complex deformations, resulting in registration failure and failure to complete the cross-sectional profile measurement. The invention solves the problem of complex deformation by adopting an improved EM algorithm to solve the probability density estimation function during non-rigid registration, specifically introduces global and local topological constraint terms into an objective function established based on the probability density estimation function, maintains global and local topology of the point cloud to be registered, and effectively solves the complex deformation.
4. According to the invention, the step of updating the weight coefficient is added in the M step in the EM algorithm, so that the self-adaption of noise in the non-rigid registration process can be realized, and the processing precision of the point cloud is further improved.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
Fig. 2 is a schematic diagram of a dual-line laser profile sensor alignment measurement system.
Fig. 3 is a schematic diagram of an irregular section inner contour point cloud a and an outer contour point cloud B.
Fig. 4 is a schematic diagram of an outlier split point network architecture constructed according to the present invention.
Fig. 5 is a schematic diagram of an inner and outer contour point cloud after outlier rejection.
Fig. 6 is a schematic diagram of a spliced contour measurement point cloud according to the present invention.
FIG. 7 is a schematic diagram of the final processing results of the present invention.
Reference numerals illustrate:
A 1-double-line laser contour sensor alignment measurement system and a 2-W-shaped superalloy sealing ring.
Detailed Description
The point cloud processing method and the point cloud processing system of the invention are further described below by taking the outline measurement point cloud processing of the irregular section of the W-shaped superalloy sealing ring as an example with reference to the accompanying drawings.
Referring to fig. 2, a point cloud acquisition system, such as a dual-line laser profile sensor, is used to acquire an original profile measurement point cloud of an irregular section of a W-shaped superalloy seal ring 2, the profile measurement point cloud including an inner profile point cloud a and an outer profile point cloud B (see fig. 3).
Respectively representing the inner contour point cloud and the outer contour point cloud asAndWherein: And Respectively representing an nth measuring point in the inner contour point cloud and the outer contour point cloud; Representing a D-dimensional real vector space, N 1、N2 representing points in the point cloud, respectively, and A, B representing measurement coordinate systems of the inner contour point cloud and the outer contour point cloud, respectively.
The method for processing the outline measurement point cloud of the irregular section of the W-shaped superalloy sealing ring by adopting the method provided by the invention comprises the following steps:
step 1, outlier segmentation
Dividing an original contour measurement point cloud into contour points and outlier points by adopting PointNet ++ network architecture, removing outlier points in the original contour measurement point cloud, and obtaining an inner contour point cloud after outlier point removalAnd an outline point cloudN' 1、N′2 represents the points in the inner contour point cloud and the outer contour point cloud after abnormal value points are removed.
Step 2, point cloud stitching
Inner contour point cloud after eliminating spliced outlier pointsAnd an outline point cloudWherein, the Respectively as dotsThe coordinate values in the x-axis are,Respectively as dotsCoordinate value of z axis to obtain contour measurement point cloud after splicingX n represents points in the post-stitching contour measurement point cloud.
In this embodiment, point cloud stitching is performed based on a pre-calibration result (the pre-calibration method is a known method) of a point cloud acquisition system, such as a line laser contour sensor, on a position measurement system, and an outer contour point cloud with outliers removed is transformed into an inner contour point cloud coordinate system to implement point cloud stitching, and contour measurement point cloud is obtained after stitching
Step 3, self-adaptive topology keeping non-rigid registration
Under the condition of adding global and local constraints, the CAD model point cloud of the W-shaped superalloy sealing ring is obtainedNon-rigid registration is carried out until contour measurement point cloud obtained after splicing in step 2Obtaining a non-rigid registered CAD model point cloudAnd replacing the spliced contour measurement point cloud with the non-rigid registered CAD model point cloud to serve as the point cloud for the subsequent geometric quality evaluation of the W-shaped superalloy sealing ring. Wherein y m represents the mth point in the CAD model point cloud; representing the D-dimensional real vector space, and M represents the points of the CAD model point cloud.
In this embodiment, the CAD model point cloud of the W-shaped superalloy seal ring is generated from a corresponding CAD model.
Besides the point cloud processing method, the invention also provides a point cloud processing system based on deep learning and self-adaptive topology keeping non-rigid registration, and the system can process the contour measurement point cloud of the irregular section of the W-shaped superalloy sealing ring. The point cloud processing system comprises an outlier segmentation point network module, a point cloud splicing module and a self-adaptive topology keeping non-rigid registration module.
The outlier dividing point network module is realized by adopting PointNet ++ network architecture and is used for dividing the contour points and outlier points in the original contour measurement point cloud, and eliminating the outlier points to obtain inner and outer contour point clouds with outlier points removed;
As shown in fig. 4, the outlier partitioning point network module adopts a PointNet ++ network architecture, and includes a first hierarchy combination and a second hierarchy combination which are sequentially set.
The first hierarchical combination comprises a plurality of groups of sampling layers, grouping layers and feature extraction layers which are sequentially arranged, and local fine features of an input point cloud (originally acquired inner contour point cloud and outer contour point cloud) are gradually extracted along the hierarchical structure in a multi-scale mode through the sampling layers, the grouping layers and the feature extraction layers. Wherein:
the sampling layer selects a local area centroid from an original input point cloud (i.e., a point cloud acquired by a point cloud acquisition system) through iterative furthest point sampling. Before the original input point cloud inputs the first hierarchical combination, the number of segmentation categories needs to be set to be equal to 2 in advance.
The grouping layer searches for multi-scale surrounding adjacent points of the centroid of the local area through a Ball query function, and establishes a local area set with multiple scales.
The feature extraction layer is PointNet network units and extracts the local fine features of the point cloud from the local area set of multiple scales. The PointNet network unit comprises a first T-net layer, a second T-net layer, a plurality of multi-layer perceptron MLPs and a feature fusion layer which are sequentially arranged.
The second hierarchical combination comprises interpolation and cross-level jump links which are correspondingly arranged with the first hierarchical combination, local fine features of sub-sampling points are gradually transmitted back to the original input point cloud along the hierarchical structure through the interpolation and the cross-level jump links, each point has two scores which are respectively an outlier point score and a contour point score, if the outlier point score of a certain point is high, the point is an outlier point, and if the contour point score of the certain point is high, the point is a contour pointAnd an outline point cloudAs shown in fig. 5.
The above-mentioned hierarchical structure means that the input point cloud is divided into N areas, each area has a central point, each area extracts an area characteristic through a network to represent the characteristic of the central point, then the N central points are continuously divided into M sub-areas, the sub-areas also have M central points, the characteristics of the M sub-areas are continuously extracted through the network to represent the characteristic of the M central points, and so on, the dividing times depend on the set layer number of the hierarchical combination.
Local fine features of sub-sampling points refer to features of center points in the partitioned areas.
The point cloud splicing module is used for splicing the inner contour point cloud and the outer contour point cloud after abnormal value points are removed, and is realized based on a pre-calibration result (the pre-calibration method is the existing known method) of a line laser contour sensor alignment measurement system and comprises a rotation matrix and a translation vector:
Wherein, the
In the formula,Representation ofTransforming to the coordinates of a coordinate system A; The rotation matrix from the outer contour point cloud coordinate system B to the inner contour point cloud coordinate system A is represented; The method comprises the steps of representing translation vectors from an outer contour point cloud coordinate system B to an inner contour point cloud coordinate system A, enabling theta T to represent included angles between the outer contour point cloud coordinate system B and the inner contour point cloud coordinate system A, and enabling a and B to represent translation amounts of the outer contour point cloud coordinate system B to the inner contour point cloud coordinate system A in the directions of x axis and z axis; Representation of Coordinate values of the x-axis in the inner contour point cloud coordinate system a,Representation ofCoordinate values of the z-axis in the inner contour point cloud coordinate system a.
By rotating the matrixTranslation vectorThe external contour point cloud is subjected to coordinate transformation and unified to an internal contour point cloud coordinate system A, so that splicing is realized, and a contour measurement point cloud is obtained after splicingIn other embodiments, the same method may be used to transform the coordinates of the inner contour point cloud, and unify the coordinates to the outer contour point cloud coordinate system B, so as to achieve stitching.
The self-adaptive topology-keeping non-rigid registration module is used for non-rigidly registering the CAD model point cloud of the W-shaped superalloy sealing ring to the spliced contour measurement point cloud to obtain the non-rigidly registered CAD model point cloud, and the non-rigidly registered CAD model point cloud is used for subsequent geometric quality evaluation of the W-shaped superalloy sealing ring.
The adaptive topology-preserving non-rigid registration module comprises a non-rigid transformation matrixThe non-rigid transformation matrixNon-rigid registration of CAD model point cloud for W-shaped superalloy sealing ring to post-splicing contour measurement point cloud, non-rigid transformation matrixThe method comprises the following steps of:
first, CAD model point cloud Assuming a Gaussian mixture model centroid, measuring a point cloud of the contour after splicingAssuming corresponding data, the CAD model point cloud is obtainedTransforming to the post-splicing contour measurement point cloudThe registration problem of (2) is converted into a probability density estimation function:
Wherein X N×D represents the contour measurement point cloud, N represents the points in the contour measurement point cloud, D represents the dimensions of the post-splice contour measurement point cloud, X n represents the points in the post-splice contour measurement point cloud (n=1,..once., N), M represents the points in the CAD model point cloud; ω is a weight coefficient representing the noise a priori level, pi M+1 = ω; m=M+1; σ 2 is the anisotropic variance; For a non-rigid transformation matrix, θ is a non-rigid transformation parameter, and y m represents a point within the CAD model point cloud (m=1.., M).
Then, the anisotropic variance sigma 2 and the non-rigid transformation parameter theta in the probability density estimation function are iteratively solved by a modified EM algorithm (expectation maximization algorithm) until convergence, at which time the probability density estimation function is solved to obtain an optimally aligned non-rigid transformation matrix
The EM algorithm is specifically:
Calculating the matching probability p old(m|xn of a point x n in the spliced profile measurement point cloud and a point y m in the CAD model point cloud in the step E of the EM algorithm;
Minimizing the objective function Q (θ, σ 2) built based on the probability density estimation function in the M steps of the EM algorithm finds new parameters:
In the formula, For the estimated current number of matching points p old(m|xn) is the prior matching probability,
The improved EM algorithm is characterized in that global topological constraint and local topological constraint are introduced into an objective function Q (theta, sigma 2) established based on a probability density estimation function in the M steps of the traditional EM algorithm, global local topology is coherent point drift, local topological constraint is local linear embedding, and the global topological constraint and the local topological constraint are introduced in a regularization term mode, namely a global topological constraint regularization term E GL and a local topological constraint regularization term E LO are introduced into the objective function to keep a global local topological structure of a matching point cloud so as to cope with complex deformation generated in non-rigid registration.
After introducing global topological constraint and local topological constraint, the objective function Q (theta, sigma 2) established based on the probability density estimation function is converted into:
Wherein, alpha and lambda represent two trade-off parameters between two topological constraint terms, according to the empirical value, the alpha value is suggested to be 100, and the lambda value is 5000000.
After the iteration of the EM algorithm is stopped, the optimally aligned non-rigid transformation matrix is obtained after the iteration is stoppedNon-rigidly registering the CAD model point cloud to the spliced contour measurement point cloud to obtain a non-rigidly registered CAD model point cloud Y' M×D:
Y′M×D＝YM×D+GW
Wherein G represents a kernel matrix, W represents a coefficient matrix of the kernel, and Y' M×D is a non-rigidly registered CAD model point cloud
Non-rigid registered CAD model point cloudAs shown in fig. 7, as can be seen from the comparison of the original point clouds in fig. 7 and fig. 3, the point clouds shown in fig. 7 have eliminated abnormal values in the original point clouds, and have completed the splicing, missing repair and noise correction of the point cloud data, so that the CAD model point clouds after non-rigid registration are utilizedSubstitute post-splice contour measurement point cloudThe irregular cross-section profile of the W-shaped superalloy sealing ring 2 can be accurately measured.
As a further optimization, the invention can also increase the update weight coefficient in the M steps of the EM algorithmThe self-adaption of judgment noise in the registration process can be realized, so that the processing precision of the point cloud is improved.
On the other hand, the invention also provides a storage medium, on which a computer program is stored, which, when being executed by a processor, is capable of executing the point cloud processing method provided by the invention.
In still another aspect, the present invention further provides an electronic device, including a processor and a storage medium, where the storage medium stores a computer program, and when the computer program is executed by the processor, the computer program is capable of executing the point cloud processing method provided by the present invention.

Claims (8)

1.基于深度学习与自适应拓扑保持非刚配准的点云处理方法，其特征在于，用于对不规则截面环形构件的原始轮廓测量点云进行处理，包括以下步骤：1. A point cloud processing method based on deep learning and adaptive topology-preserving non-rigid registration, characterized in that it is used to process the original contour measurement point cloud of an irregular cross-section annular component, comprising the following steps: 步骤1：采用PointNet++网络架构将所述原始轮廓测量点云划分为轮廓点和异常值点，将所述异常值点剔除，得到剔除异常值点后的内、外轮廓点云；Step 1: Use the PointNet++ network architecture to divide the original contour measurement point cloud into contour points and outlier points, remove the outlier points, and obtain the inner and outer contour point clouds after removing the outlier points; 步骤2：将剔除异常值点后的内、外轮廓点云拼接，得到拼接后轮廓测量点云；Step 2: Splice the inner and outer contour point clouds after removing outlier points to obtain the spliced contour measurement point cloud; 步骤3：在添加全局和局部约束的情况下，将所述不规则截面环形构件的CAD模型点云自适应拓扑保持非刚性配准至步骤2得到的拼接后轮廓测量点云，得到非刚性配准后的CAD模型点云，用所述非刚性配准后的CAD模型点云替代轮廓测量点云作为后续对所述不规则截面环形构件的几何质量评估的点云；Step 3: Adaptively topologically maintain non-rigid registration of the CAD model point cloud of the irregular-section annular component to the spliced contour measurement point cloud obtained in step 2 while adding global and local constraints, to obtain a non-rigidly registered CAD model point cloud, and use the non-rigidly registered CAD model point cloud to replace the contour measurement point cloud as the point cloud for subsequent geometric quality assessment of the irregular-section annular component; 所述自适应拓扑保持非刚性配准的方法为：The method of adaptive topology preserving non-rigid registration is: 首先，将所述不规则截面环形构件的CAD模型点云变换至所述拼接后轮廓测量点云的配准问题，转换为概率密度估计函数：First, the registration problem of transforming the CAD model point cloud of the irregular cross-section annular component to the contour measurement point cloud after splicing is converted into a probability density estimation function: 其中：in: XN×D表示拼接后轮廓测量点云；X N × D represents the contour measurement point cloud after stitching; N表示拼接后轮廓测量点云内的点数；N represents the number of points in the contour measurement point cloud after stitching; D表示拼接后轮廓测量点云维度；D represents the dimension of the contour measurement point cloud after stitching; xn表示拼接后轮廓测量点云内的点；x n represents the point in the contour measurement point cloud after stitching; M表示CAD模型点云内的点数；M represents the number of points in the CAD model point cloud; ω为代表噪声先验水平的权系数； ω is the weight coefficient representing the prior level of noise; πM+1＝ω；π M+1 =ω; m＝M+1；m＝M+1； σ2为各向异性方差；σ 2 is the anisotropy variance; 为非刚性变换； is a non-rigid transformation; θ为非刚性变换参数；θ is the non-rigid transformation parameter; ym表示CAD模型点云内的点；y m represents a point in the point cloud of the CAD model; 然后，通过改进的EM算法迭代求解所述概率密度估计函数中的各向异性方差σ2和非刚性变换参数θ，直至收敛，从而获得最优化对齐的非刚性变换矩阵此时概率密度估计函数求解完成；Then, the anisotropic variance σ2 and the non-rigid transformation parameter θ in the probability density estimation function are iteratively solved by the improved EM algorithm until convergence, thereby obtaining the optimally aligned non-rigid transformation matrix At this point, the probability density estimation function is solved; 最后，通过所述最优化对齐的非刚性变换矩阵将CAD模型点云非刚性配准至拼接后轮廓测量点云；Finally, the non-rigid transformation matrix of the optimized alignment is Non-rigid registration of the CAD model point cloud to the spliced contour measurement point cloud; 所述改进的EM算法具体为：The improved EM algorithm is specifically as follows: 在E步中计算轮廓测量点云中的点xn与CAD模型点云中的点ym的匹配概率pold(m|xn)，In step E, the matching probability p old (m|x n ) between the point x n in the contour measurement point cloud and the point y m in the CAD model point cloud is calculated. 在M步中基于概率密度估计函数建立的目标函数中引入全局拓扑约束与局部拓扑约束，然后最小化引入约束后的目标函数，寻找新的非刚性变换参数θ和各向异性方差σ2。In the M-step, global and local topological constraints are introduced into the objective function established based on the probability density estimation function. Then, the objective function after the constraints are introduced is minimized to find new non-rigid transformation parameters θ and anisotropic variance σ 2 . 2.根据权利要求1所述的基于深度学习与自适应拓扑保持非刚配准的点云处理方法，其特征在于：步骤1中采用PointNet++网络架构将所述原始轮廓测量点云划分为轮廓点和异常值点的方法为：2. The point cloud processing method based on deep learning and adaptive topology-preserving non-rigid registration according to claim 1, characterized in that: in step 1, the method of using the PointNet++ network architecture to divide the original contour measurement point cloud into contour points and outlier points is: 首先，沿层次结构多尺度逐步提取所述原始轮廓测量点云的局部精细特征；Firstly, local fine features of the original contour measurement point cloud are gradually extracted along the hierarchical structure at multiple scales; 然后，将子采样点的局部精细特征逐步传播回所述原始轮廓测量点云，得到原始轮廓测量点云的各点得分，根据各点得分，将原始轮廓测量点云划分为轮廓点和异常值点。Then, the local fine features of the sub-sampling points are gradually propagated back to the original contour measurement point cloud to obtain the score of each point of the original contour measurement point cloud. According to the score of each point, the original contour measurement point cloud is divided into contour points and outlier points. 3.根据权利要求2所述的基于深度学习与自适应拓扑保持非刚配准的点云处理方法，其特征在于：所述步骤2点云拼接的方法为：3. The point cloud processing method based on deep learning and adaptive topology-preserving non-rigid registration according to claim 2, characterized in that the point cloud stitching method in step 2 is: 基于用于采集原始轮廓测量点云的点云采集系统的预标定结果进行坐标变换，将剔除异常值后的外轮廓点云变换至内轮廓点云坐标系中实现点云拼接，或者将剔除异常值后的内轮廓点云变换至外轮廓点云坐标系中实现点云拼接。Based on the pre-calibration results of the point cloud acquisition system used to acquire the original contour measurement point cloud, coordinate transformation is performed, and the outer contour point cloud after removing outliers is transformed into the inner contour point cloud coordinate system to achieve point cloud splicing, or the inner contour point cloud after removing outliers is transformed into the outer contour point cloud coordinate system to achieve point cloud splicing. 4.根据权利要求3所述的基于深度学习与自适应拓扑保持非刚配准的点云处理方法，其特征在于：在所述M步中增加更新代表噪声先验水平的权系数ω的步骤，以实现配准过程中判断噪声的自适应。4. The point cloud processing method based on deep learning and adaptive topology-preserving non-rigid registration according to claim 3 is characterized in that: a step of updating the weight coefficient ω representing the prior level of noise is added in the M step to achieve adaptive noise judgment during the registration process. 5.基于深度学习与自适应拓扑保持非刚配准的点云处理系统，其特征在于，包括：5. A point cloud processing system based on deep learning and adaptive topology-preserving non-rigid registration, characterized by including: 异常值分割点网模块，采用PointNet++网络架构实现，用于分割原始轮廓测量点云中的轮廓点和异常值点，并将所述异常值点剔除，得到剔除异常值点后的内、外轮廓点云；The outlier segmentation point network module is implemented using the PointNet++ network architecture and is used to segment the contour points and outlier points in the original contour measurement point cloud, and remove the outlier points to obtain the inner and outer contour point clouds after removing the outlier points; 点云拼接模块，用于拼接异常值点剔除后的内、外轮廓点云；Point cloud stitching module, used to stitch the inner and outer contour point clouds after outlier points are removed; 自适应拓扑保持非刚配准模块，用于将不规则截面环形构件的CAD模型点云非刚性配准至拼接后得到的轮廓测量点云，得到非刚性配准后的CAD模型点云，所述非刚性配准后的CAD模型点云用于后续对所述不规则截面环形构件的几何质量评估；An adaptive topology-preserving non-rigid registration module is used to non-rigidly register the CAD model point cloud of the irregular-section annular component to the contour measurement point cloud obtained after splicing, thereby obtaining a non-rigidly registered CAD model point cloud. The non-rigidly registered CAD model point cloud is used for subsequent geometric quality assessment of the irregular-section annular component. 自适应拓扑保持非刚配准模块包括非刚性变换矩阵该非刚性变换矩阵T用于将不规则截面环形构件的CAD模型点云非刚性配准至拼接后轮廓测量点云；非刚性变换矩阵通过下述方法获取：Adaptive topology preserving non-rigid registration module including non-rigid transformation matrix The non-rigid transformation matrix T is used to non-rigidly register the CAD model point cloud of the irregular cross-section annular component to the contour measurement point cloud after splicing; the non-rigid transformation matrix Obtained through the following methods: 首先，将CAD模型点云假设为高斯混合模型质心，将拼接后轮廓测量点云假设为相应的数据，将CAD模型点云变换至拼接后轮廓测量点云的配准问题转换为概率密度估计函数：First, the CAD model point cloud Assuming the centroid of the Gaussian mixture model, the contour measurement point cloud after splicing Assuming that it is the corresponding data, the CAD model point cloud Transform to the spliced contour measurement point cloud The registration problem is converted into a probability density estimation function: 其中，XN×D表示轮廓测量点云；N表示轮廓测量点云内的点数；D表示拼接后轮廓测量点云维度；xn表示拼接后轮廓测量点云内的点(n＝1,...,N)；M表示CAD模型点云内的点数；ω为代表噪声先验水平的权系数；πM+1＝ω；m＝M+1；σ2为各向异性方差；为非刚性变换矩阵，θ为非刚性变换参数；ym表示CAD模型点云内的点；Where XN ×D represents the contour measurement point cloud; N represents the number of points in the contour measurement point cloud; D represents the dimension of the contour measurement point cloud after stitching; xn represents the points in the contour measurement point cloud after stitching (n=1,...,N); M represents the number of points in the CAD model point cloud; ω is the weight coefficient representing the prior level of noise; π M+1 = ω; m＝M+1； σ 2 is the anisotropy variance; is the non-rigid transformation matrix, θ is the non-rigid transformation parameter; y m represents the point in the CAD model point cloud; 然后，通过改进的EM算法迭代求解概率密度估计函数中的各向异性方差σ2和非刚性变换参数θ，直至收敛，此时概率密度估计函数求解完成，获得最优化对齐的非刚性变换矩阵T；Then, the anisotropic variance σ2 and the non-rigid transformation parameter θ in the probability density estimation function are iteratively solved by the improved EM algorithm until convergence. At this time, the probability density estimation function is solved and the non-rigid transformation matrix T of the optimal alignment is obtained; 所述改进的EM算法具体为：The improved EM algorithm is specifically as follows: 在EM算法的E步中计算拼接后轮廓测量点云中的点xn与CAD模型点云中的点ym的匹配概率pold(m|xn)；In the E step of the EM algorithm, the matching probability p old (m|x n ) between the point x n in the spliced contour measurement point cloud and the point y m in the CAD model point cloud is calculated; 在EM算法的M步中基于概率密度估计函数建立的目标函数Q(θ,σ2)，并为其添加全局拓扑约束和局部拓扑约束，然后最小化添加约束后的目标函数Q1(W,σ2)寻找新参数：In the M step of the EM algorithm, the objective function Q(θ,σ 2 ) is established based on the probability density estimation function, and global and local topological constraints are added to it. Then, the objective function Q 1 (W,σ 2 ) with added constraints is minimized to find new parameters: 其中，为估计的当前匹配点数；pold(m|xn)为先验匹配概率，α与λ表示两个拓扑约束项之间的两个权衡参数。in, is the estimated number of current matching points; p old (m|x n ) is the prior matching probability, α and λ represent two trade-off parameters between the two topological constraints. 6.根据权利要求5所述的基于深度学习与自适应拓扑保持非刚配准的点云处理系统，其特征在于：所述异常值分割点网模块包括依次设置的第一层次组合和第二层次组合；6. The point cloud processing system based on deep learning and adaptive topology-preserving non-rigid registration according to claim 5, characterized in that: the outlier segmentation point network module includes a first-level combination and a second-level combination arranged in sequence; 第一层次组合包括多组依次设置的采样层、分组层和特征提取层，通过采样层、分组层与特征提取层沿层次结构多尺度逐步提取原始轮廓测量点云的局部精细特征；The first level combination includes multiple groups of sampling layers, grouping layers and feature extraction layers arranged in sequence, through which the local fine features of the original contour measurement point cloud are gradually extracted along the hierarchical structure at multiple scales; 第二层次组合包括与第一层次组合对应设置的插值和跨级跳跃链接，通过插值和跨级跳跃链接将子采样点的局部精细特征沿层次结构逐步传播回原始轮廓测量点云，获得原始轮廓测量点云的各点得分，根据各点得分，将原始轮廓测量点云中各点划分为轮廓点与异常值点。The second-level combination includes interpolation and cross-level skip links set corresponding to the first-level combination. Through interpolation and cross-level skip links, the local fine features of the sub-sampling points are gradually propagated back to the original contour measurement point cloud along the hierarchical structure to obtain the score of each point in the original contour measurement point cloud. According to the score of each point, the points in the original contour measurement point cloud are divided into contour points and outlier points. 7.存储介质，所述存储介质上存储有计算机程序；其特征在于：所述计算机程序被处理器运行时执行权利要求1-4任一所述的方法。7. A storage medium storing a computer program; wherein the computer program executes the method according to any one of claims 1 to 4 when executed by a processor. 8.电子设备，包括处理器和存储介质；所述存储介质上存储有计算机程序；其特征在于：所述计算机程序被所述处理器运行时执行权利要求1-4任一所述的方法。8. An electronic device comprising a processor and a storage medium; the storage medium stores a computer program; and the computer program executes the method according to any one of claims 1 to 4 when executed by the processor.
CN202310577670.9A 2023-05-22 2023-05-22 Point cloud processing method for keeping non-rigid registration based on deep learning and self-adaptive topology Active CN116703989B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310577670.9A CN116703989B (en) 2023-05-22 2023-05-22 Point cloud processing method for keeping non-rigid registration based on deep learning and self-adaptive topology

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310577670.9A CN116703989B (en) 2023-05-22 2023-05-22 Point cloud processing method for keeping non-rigid registration based on deep learning and self-adaptive topology

Publications (2)

Publication Number Publication Date
CN116703989A CN116703989A (en) 2023-09-05
CN116703989B true CN116703989B (en) 2025-07-29

Family

ID=87834911

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310577670.9A Active CN116703989B (en) 2023-05-22 2023-05-22 Point cloud processing method for keeping non-rigid registration based on deep learning and self-adaptive topology

Country Status (1)

Country Link
CN (1) CN116703989B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118279363B (en) * 2024-05-29 2024-09-06 西北工业大学 Processing and quality assessment method of online measurement point cloud of thin-walled complex components
CN118505763B (en) * 2024-07-17 2024-11-05 浙江大学 3D surface rigid registration method, system and electronic device assisted by geometric information
CN119803341A (en) * 2024-12-19 2025-04-11 武汉理工大学 Intelligent measurement method and device for cross-sectional profile of large special-shaped ring parts using structured light

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111080680A (en) * 2019-12-29 2020-04-28 苏州体素信息科技有限公司 A patient-oriented three-dimensional chest organ reconstruction method and system
CN114882083A (en) * 2022-04-25 2022-08-09 西北工业大学 Method and device for preprocessing profile point cloud of complex special-shaped section thin-wall annular part

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP3451291B1 (en) * 2017-09-01 2020-08-26 Tata Consultancy Services Limited System and method for shape constrained 3d point cloud registration
CN115578408A (en) * 2022-07-28 2023-01-06 四川大学 Point cloud registration blade profile optical detection method, system, equipment and terminal

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111080680A (en) * 2019-12-29 2020-04-28 苏州体素信息科技有限公司 A patient-oriented three-dimensional chest organ reconstruction method and system
CN114882083A (en) * 2022-04-25 2022-08-09 西北工业大学 Method and device for preprocessing profile point cloud of complex special-shaped section thin-wall annular part

Also Published As

Publication number Publication date
CN116703989A (en) 2023-09-05

Similar Documents

Publication Publication Date Title
CN116703989B (en) Point cloud processing method for keeping non-rigid registration based on deep learning and self-adaptive topology
CN104484508B (en) Optimizing method for noncontact three-dimensional matching detection of complex curved-surface part
Huo et al. Research on product surface quality inspection technology based on 3D point cloud
CN116468767B (en) Airplane surface reconstruction method based on local geometric features and implicit distance field
CN114119922A (en) A complex contour defect verification method based on point cloud 3D reconstruction
CN112652003A (en) Three-dimensional point cloud registration method based on RANSAC measure optimization
CN120628058B (en) IMU auxiliary depth SLAM method and system integrating language-vision multi-mode perception
Li et al. A novel simplification method of point cloud with directed Hausdorff distance
CN106354935B (en) Complex curved surface parts matching detection method based on electron outside nucleus probability density distribution
CN120374991A (en) CAD drawing segmentation parameter adjustment method and system combined with region identification
CN118537506A (en) Composite 3D surface reconstruction method
CN117953232A (en) Three-dimensional point cloud simplifying method based on boundary point dimension reduction extraction
Cheng et al. Boosting global-local feature matching via anomaly synthesis for multi-class point cloud anomaly detection
Jiang et al. Efficient match pair retrieval for large-scale UAV images via graph indexed global descriptor
CN120805514B (en) Inorganic mineral casting stress analysis method and system based on virtual simulation system
CN113112471A (en) Target detection method based on RI-HOG characteristics and quick pyramid
Zhao et al. Probing a point cloud based expeditious approach with deep learning for constructing digital twin models in shopfloor
Hu et al. Visual localization of robotic end effector via fusion of 3D Gaussian Splatting and heuristic optimization algorithm
CN118314086B (en) Automatic matching method and system for wafer reference layout and SEM image
CN117853766B (en) Tunnel fracture coplanarity matching method and system based on tunnel face and borehole image
CN111724298B (en) Dictionary optimization and mapping method for digital rock core super-dimensional reconstruction
CN118628499A (en) A method for aircraft engine blade defect detection based on network architecture search
CN117952998A (en) Method and system for generating three-dimensional model contour line
CN116310204A (en) A Hierarchical Depth Peeling Method for 3D Point Cloud Models
Hu et al. Solving the point cloud registration based on deep learning

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant