CN112417618B - Four-dimensional space-time perception large-scene free bending pipeline detection and point cloud completion method - Google Patents

Four-dimensional space-time perception large-scene free bending pipeline detection and point cloud completion method Download PDF

Info

Publication number
CN112417618B
CN112417618B CN202011312309.6A CN202011312309A CN112417618B CN 112417618 B CN112417618 B CN 112417618B CN 202011312309 A CN202011312309 A CN 202011312309A CN 112417618 B CN112417618 B CN 112417618B
Authority
CN
China
Prior art keywords
pipeline
dimensional
point
point cloud
time
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
CN202011312309.6A
Other languages
Chinese (zh)
Other versions
CN112417618A (en
Inventor
张洁
孙军华
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.)
Beihang University
Beijing Technology and Business University
Original Assignee
Beihang University
Beijing Technology and Business 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 Beihang University, Beijing Technology and Business University filed Critical Beihang University
Priority to CN202011312309.6A priority Critical patent/CN112417618B/en
Publication of CN112417618A publication Critical patent/CN112417618A/en
Application granted granted Critical
Publication of CN112417618B publication Critical patent/CN112417618B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/10—Geometric CAD
    • G06F30/17—Mechanical parametric or variational design
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • 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
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T17/00—Three-dimensional [3D] modelling for computer graphics
    • G06T17/20—Finite element generation, e.g. wire-frame surface description, tesselation
    • 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
    • G06F2111/10—Numerical modelling
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2113/00—Details relating to the application field
    • G06F2113/14—Pipes

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Geometry (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Computer Hardware Design (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computer Graphics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Medical Informatics (AREA)
  • Image Analysis (AREA)
  • Length Measuring Devices By Optical Means (AREA)

Abstract

本发明公开一种四维时空感知的大场景自由弯曲管路检测及点云补全方法，在待检场景的三维点云中均匀获取三维采样点并建立球形支撑域，在三维点云块中构建点对式圆柱特征集，根据特征聚类进行二类决策，实现三维管道表面特征点的自动探测；在四维时空域建立管道的参数化几何模型和时序参数估计模型，根据几何模型求解管道特征点所在的局部管道段轴线作为时域起始状态，结合时序参数估计模型分别向起始时刻的轴线两端迭代进行连续的管道段参数估计，获取融合三维几何信息和时域相关信息的三维自由弯曲管道表面点云及其几何参数；根据几何模型重建管道表面点云并迭代优化，实现对原始场景中管道点云的补全。本发明具有计算高效、抗点云噪声的特点。

Figure 202011312309

The invention discloses a four-dimensional spatio-temporal perception of a large-scene free bending pipeline detection and point cloud completion method. The three-dimensional sampling points are uniformly obtained in the three-dimensional point cloud of the scene to be inspected, a spherical support domain is established, and the three-dimensional point cloud block is constructed in the three-dimensional point cloud. Point-to-point cylindrical feature set, make two-class decision based on feature clustering, realize automatic detection of three-dimensional pipeline surface feature points; establish a parametric geometric model and time-series parameter estimation model of pipeline in four-dimensional space-time domain, and solve pipeline feature points according to the geometric model The axis of the local pipeline section where it is located is used as the initial state of the time domain, and the time-series parameter estimation model is used to iterate the two ends of the axis at the initial time to perform continuous pipeline parameter estimation, and obtain the three-dimensional free bending that integrates three-dimensional geometric information and time-domain related information. Pipeline surface point cloud and its geometric parameters; reconstruct the pipeline surface point cloud according to the geometric model and iteratively optimize it to complete the pipeline point cloud in the original scene. The invention has the characteristics of high computational efficiency and resistance to point cloud noise.

Figure 202011312309

Description

Four-dimensional space-time perception large-scene free bending pipeline detection and point cloud completion method
Technical Field
The invention relates to a free bending pipeline characteristic detection and point cloud completion method for large-scene three-dimensional point cloud, belongs to the field of industrial vision, and particularly relates to the field of three-dimensional vision.
Background
In the field of industrial manufacturing and assembly, large-sized operating objects, such as aircraft engines, are often faced. The surface structure of the machine body is complex, the types and the quantity of parts are various and are densely and alternately distributed, and each part plays an important role in the running quality and the safety of the machine body. The pipeline is an important part on the aircraft engine, is densely and alternately distributed on an aircraft engine body, and is mainly used for conveying media such as oil bodies, gases and the like to the engine and the aircraft to ensure the stable operation of the engine.
The problem of micro deformation of an aircraft engine pipeline in the manufacturing and operating processes can occur due to the influence of the temperature and the pressure of the operating environment; due to the influence of the assembly precision, each pipeline may have a problem that the clearance between the pipeline and the main machine body or the pipeline group is not qualified. The above problem is the goal of constant monitoring throughout the manufacturing, assembly, and operational maintenance phases of an aircraft engine pipeline. The manufacturing precision, the installation position precision and the operation reliability of the pipeline are important to the performance and the service life of the aircraft engine. At present, in the field of engine manufacturing and assembling, a specific instrument is mostly manually held to detect the geometric appearance and the mounting clearance of a pipeline, and due to the fact that the size of a machine body is large, the detection mode has the problems of inconvenience in operation and low efficiency; meanwhile, manual detection can only be realized in a pipeline fixed-point detection mode, the whole pipeline is difficult to detect quickly at one time, and the applicability is limited. Therefore, the rapid and intelligent monitoring of the shape and the geometric parameters of the conduit on the body has important value for each link of manufacturing and assembling of the aircraft engine.
The large-scale body has a large number of pipelines on the surface, the pipelines are long and freely bent, fastening or connecting pieces with different sizes and appearances are distributed on the pipelines, and the pipelines are partially shielded. Automatically detecting three-dimensional pipelines in such a complex and large scene is a very challenging task. Most of the existing automatic pipeline detection technologies are developed based on two-dimensional image data. For example, two-dimensional pipeline detection or segmentation in images is achieved by manually building features or automatically learning pipeline features through a deep neural network model. Because the surfaces of industrial manufacturing bodies such as engines lack textures and have high light reflection problems, high exposure or shadow areas usually exist on a two-dimensional image obtained from the surface of the body, the quality of the two-dimensional pipeline image is seriously influenced by the optical imaging factors, and more rigorous requirements are provided for the two-dimensional pipeline detection and reconstruction technology. Regarding the three-dimensional pipeline technology, patent 102410811B obtains a multi-view pipeline image through a multi-view stereo vision technology, extracts the pipeline edge in the image and reconstructs the three-dimensional pipeline axis; patent CN108801175B has constructed pipeline space axis perspective projection model, acquires the pipeline image pair through binocular vision sensor, rebuilds out the three-dimensional axis of pipeline through stereo matching, and then realizes pipeline clearance monitoring. The two-dimensional image is still used as original data, the geometric parameters of the three-dimensional pipeline can be obtained through a double/multi-view dimension reconstruction process, and because the original two-dimensional data does not have the real scale and three-dimensional geometric shape information of the pipeline, the invention can not optimize or complement the reconstructed and calculated pipeline parameters by using the real three-dimensional data as reference. At present, the detection and reconstruction of a large-scale manufacturing body surface pipeline by taking three-dimensional point cloud as original data still have a less explored problem. The three-dimensional data contains original geometric information of the target, has higher information dimensionality and richer modal information, and has important significance for acquiring a complete three-dimensional pipeline model and analyzing. Compared with a two-dimensional image, the three-dimensional target avoids the problems of target scale change, posture influence, deformation and the like, and has the essential advantage of a data layer. The invention designs a free bending pipeline detection and point cloud completion technology oriented to a three-dimensional point cloud scene, and has important value for automatically, efficiently, accurately and completely monitoring the three-dimensional appearance state of a pipeline with complex distribution on the surface of a large-scale machine.
Disclosure of Invention
The invention solves the problems: aiming at the detection problem of the three-dimensional freely bent pipeline on the surface of the large industrial manufactured part, the method for detecting the large-scene freely bent pipeline and completing the point cloud by sensing in four dimensions in a time-space domain is provided.
The technical scheme of the invention is as follows: a method for detecting a large-scene free bent pipeline and complementing point clouds in a time-space domain four-dimensional sensing mode comprises the steps of firstly detecting surface characteristic points of a three-dimensional pipeline by adopting a point-to-point pipeline characteristic clustering method, respectively establishing a geometric parameter model and a time sequence parameter estimation model of the free bent pipeline in a four-dimensional time-space domain, detecting the point clouds of the three-dimensional free bent pipeline on a large manufacturing body surface, and calculating the axis and the mean radius of the point clouds. And generating a complete three-dimensional pipeline model by detecting parameters, and performing iterative optimization on the three-dimensional pipeline model by minimizing registration errors of the generated model and the original point cloud model. The invention provides a three-dimensional pipeline modeling and detecting method for sensing four-dimensional time-space domain information, which can accurately detect a three-dimensional freely-bent pipeline in a large-scale industrial scene and has high efficiency and wide applicability.
The technical scheme of the invention is that a time-space domain four-dimensional perception large-scene free bending pipeline detection and point cloud completion method comprises the following steps:
a, uniformly sampling three-dimensional points in three-dimensional point cloud of a scene to be detected, establishing a spherical support domain with a certain radius by taking each sampling point as a center, and extracting three-dimensional point cloud blocks covered by each support domain; calculating three-dimensional normal vectors of all points in the three-dimensional point cloud block and aggregating the three-dimensional normal vectors with three-dimensional position information of all three-dimensional points to form a six-dimensional point cloud block;
b, randomly sampling six-dimensional point pairs for multiple times in the six-dimensional point cloud block in the step a, and enabling the included angles of normal vectors of the six-dimensional point pairs to meet set threshold value constraint; according to the geometric relation of the cylinder, the axis and the radius of the cylinder can be solved by a pair of six-dimensional points, so that the axis and the radius of the cylinder corresponding to the six-dimensional points sampled for multiple times at random are respectively solved, and the radii of the cylinder calculated for multiple times are aggregated into a point-to-point type cylinder characteristic set;
c, performing feature clustering on the point-to-point cylindrical feature set corresponding to each point cloud block, judging whether the point cloud block to which the point-to-point cylindrical feature set belongs is a three-dimensional pipeline surface point according to the ratio of the inner points of the clustered maximum clusters, and realizing automatic detection of the three-dimensional pipeline surface feature points in the scene point cloud;
d, establishing a parameterized geometric model of the pipeline in a four-dimensional space-time domain, expressing the freely bent pipeline as a discrete parameterized cylindrical section in the three-dimensional space domain and a time sequence polymerization in a one-dimensional time domain, wherein the time sequence polymerization is a time sequence extension process of the cylindrical section taking local point cloud where the feature points are located as an initial moment, and the cylindrical section at each moment is parameterized into an axial vector and a cylinder radius in the three-dimensional space domain;
e, establishing a time sequence parameter estimation model of the pipeline in a four-dimensional space-time domain, and respectively performing extension and parameter estimation on the cylindrical sections from the initial time to the two ends of the current axis in an iteration mode, wherein the parameter estimation of the cylindrical sections at all times is a fusion filtering result of time domain prediction parameters and resolving parameters of the geometric model in the three-dimensional space domain according to the step d, and finally detecting and obtaining free bending pipeline surface point clouds for sensing four-dimensional space-time information and parameterized model representations thereof, namely the axis and the radius of the three-dimensional pipeline;
f, sampling a point on the three-dimensional pipeline axis obtained by solving in the step e as an initial time point according to the pipeline parameterized geometric model established in the step d, and reconstructing a complete three-dimensional pipeline surface in a time sequence extending mode at certain time intervals; and (3) taking the minimum non-rigid registration error of the reconstructed pipeline model and the original three-dimensional point cloud scene as a target function, iteratively solving to obtain the optimal three-dimensional pipeline parameters, and finally realizing the detection and point cloud completion of the freely bent pipeline in the three-dimensional point cloud scene.
Further, in the step a, three-dimensional points { p) are uniformly sampled in the three-dimensional point cloud of the scene to be detectedi∈R3}(R3Representing a three-dimensional spatial domain) with each sample point piAs a center, establishing a spherical supporting domain with a certain radius r and extracting a three-dimensional point cloud block S covered by the supporting domaini＝{pij∈R3And calculating each three-dimensional point p in the point cloud blockijOf three-dimensional normal vector nij∈R3And aggregating with three-dimensional position information to form six-dimensional point cloud block S'i＝{p′ij∈R6}，R6Denotes a six-dimensional spatial domain of p'ij＝{pij,nij}。
Further, the step b is specifically realized as follows:
(1) for each six-dimensional point cloud block S'iRandomly sampling six-dimensional point pair p'im,p′inAnd enabling the included angle of the normal vector of the six-dimensional point pair to meet the set threshold value constraint
Figure BDA0002790194220000031
Wherein n isimAnd ninRespectively are normal vectors of the six-dimensional point pairs, and epsilon is an included angle threshold value so as to ensure the accuracy of subsequent six-dimensional point pair type cylindrical feature calculation;
(2) and (3) calculating the cylinder characteristic parameters of each six-dimensional point pair: the cylinder axis and radius are calculated by first calculating the normal to both directions nim,ninThe vector axis of ═ nim×ninLet a plane normal to axis be P1(ii) a Then, p 'is calculated over one sampling point'imNormal direction n ofimAnd is perpendicular to the plane P1Plane P of2Calculating another sampling point p'inNormal direction n ofinIn a straight line to the plane P2Is a point on the axis of the cylindrical section, and is p 'to the sampling point'inThe distance between the point cloud blocks is the radius of the cylindrical section, the solved radius of the cylinder is used as a characteristic, and the radius of the cylinder obtained by resolving the six-dimensional point pairs sampled for many times is aggregated into a point-to-point type cylindrical characteristic set of the point cloud block.
Further, the step c specifically includes:
if one point cloud block is a three-dimensional pipeline surface point cloud, the following points exist: the characteristics in the point-to-point type cylindrical characteristic set constructed on the point cloud block have obvious density center and aggregation, the characteristic components in the point-to-point type cylindrical characteristic set corresponding to each point cloud block are clustered, whether the point cloud block to which the point-to-point type cylindrical characteristic set belongs is a three-dimensional pipeline surface point or not is judged according to the internal point proportion of the maximum cluster after clustering, and the automatic detection of the three-dimensional pipeline surface characteristic points in the scene point cloud is realized.
Further, the step d is in four dimensionsEstablishing a parameterized geometric model of the pipeline in the airspace, and enabling the ith freely-bent pipeline pipe to be pipeiExpressed as parameterized cylindrical segments seg discrete over one-dimensional time domain titTime-sequential polymerization of (i.e.
Figure BDA0002790194220000041
The aggregation is a time sequence extension process which takes the local pipeline section where the characteristic point detected in the step c is as the initial, and the pipeline section seg at each moment titParameterized in three-dimensional spatial domain as segit＝f(cenit,dirit,rit) In which cenitIs an axial center point of the pipe section, diritIs axial to the pipe section, ritPipe segment radius.
Further, the step e is specifically implemented as follows:
(1) establishing a time sequence parameter estimation model of the pipeline in a four-dimensional time-space domain, performing fusion estimation on two parameters of the axis and the axis of the pipeline section in the three-dimensional space domain at each moment, wherein the time sequence parameter estimation model fuses two parts of parameter estimation results, and the time sequence parameter estimation model is respectively a predicted value of the pipeline section parameter at the current moment and a geometric solution value of the pipeline parameter at the current moment for the pipeline section parameter at the previous moment, and the model is as follows:
Figure BDA0002790194220000042
wherein
Figure BDA0002790194220000043
And
Figure BDA0002790194220000044
respectively obtaining the optimal axis and axis estimation value of the pipeline section at the last moment;
Figure BDA0002790194220000045
and
Figure BDA0002790194220000046
respectively the most of the pipeline segments at the current momentPreferred axis and axis estimates; a. the6×6A pipeline parameter prediction matrix is used for predicting pipeline parameters at the next moment at a certain time interval along the axial direction of the pipeline; k6×6Is a gain matrix; h6×6Is an observation matrix of the pipeline parameters; n 'of'tAnd dir'tD, respectively resolving the axis and the axis value of the pipeline section obtained at the current moment according to the parameterized geometric model in the step d;
(2) c, performing cylinder fitting on the local point cloud block where the pipeline characteristic points are located and obtained through detection in the step c to obtain a parameterized representation seg of the pipeline sectioni0＝{ceni0,diri0,ri0Where ceni0Is the axial center point of the pipeline section, diri0Is axial to the pipe section, ri0Pipe segment radius) as starting time t0The state of the pipeline parameters; according to the time sequence parameter estimation model, continuously extending the pipeline sections and estimating parameters from the initial moment to the two ends of the current axis in an iteration mode; the estimation of the pipeline section parameter at each moment t is performed by a fusion filtering result of a time domain prediction parameter and a three-dimensional space domain geometric resolving parameter, and finally, the free bending pipeline surface point cloud for sensing four-dimensional space-time information and the parameterized model representation thereof, namely the axis and the radius of a three-dimensional pipeline are detected, wherein the radius of the whole pipeline adopts the radius { r } of the pipeline section at each momentitThe largest cluster mean obtained via mean clustering.
Further, the step f is specifically realized as follows:
(1) sampling a point on the three-dimensional pipeline axis calculated in the step e as an initial time point, and reconstructing a complete three-dimensional pipeline surface in a time sequence extending mode at certain time intervals according to the pipeline parameterized geometric model in the step d to realize the completion of pipeline point cloud in the original scene;
(2) and (3) performing iterative optimization on the axis and radius parameters of the reconstructed three-dimensional pipeline by taking the minimum non-rigid registration error of the reconstructed model and the original three-dimensional point cloud scene as a target function to obtain a complete pipeline point cloud model with higher quality and geometric accuracy, and finally realizing detection and point cloud completion of the freely bent pipeline in the three-dimensional point cloud scene.
Compared with the prior art, the method for detecting the characteristics of the freely bent pipeline and completing the point cloud in the large-scale industrial scene has the advantages that the detection and point cloud completion method is carried out based on three-dimensional point cloud data, the data modality is not influenced by environmental illumination and shadow change, the three-dimensional point cloud data presents the geometric shape information of the real scale of the target, and the problems of target scale change, imaging projection deformation and the like in a two-dimensional image are solved; the time-sequence free bending pipeline detection effectively utilizes the prediction capability of a time domain on the pipeline characteristics of a three-dimensional space domain, and is a time-space four-dimensional pipeline detection method; the pipeline feature construction form based on the point-to-point method is simple and efficient, the calculation complexity is low, and the method is suitable for industrial large-scene application occasions.
Compared with the prior art, the invention has the advantages that:
(1) compared with a pipeline detection technology based on a two-dimensional image, the pipeline detection method based on the two-dimensional image can detect and obtain a pipeline model with real geometric dimension and complete three-dimensional shape information, avoids the influence of environmental illumination and shadow change in the detection process, and has the characteristic of scale and attitude invariance;
(2) the geometric parameter calculation process of the three-dimensional pipeline integrates the calculation value of the pipeline section in the three-dimensional space domain and the predicted value of the one-dimensional time domain at each moment, and compared with the pipeline parameter calculation process based on a single domain, the pipeline axis obtained by calculation is smoother; due to the prediction of the time domain dimension, the pipeline detection process can overcome the obstruction and shielding of the pipeline connecting piece, so that more complete and continuous whole pipeline detection is realized;
(3) the pipeline geometric parameterized model of four-dimensional space-time perception provided by the invention expresses the pipeline as time-sequence space aggregation of discrete pipeline segments, so that the pipeline geometric parameterized model is particularly suitable for freely-bent pipeline detection; the point-to-point pipeline characteristic detection provided by the invention only has the characteristic of small calculated amount, and is particularly suitable for large industrial scenes.
Drawings
FIG. 1 is a flow chart of the overall implementation of the present invention;
FIG. 2 is a four-dimensional time-space domain modeling diagram of a freely bent pipeline;
FIG. 3 is a diagram of the results of automatic detection of pipeline surface points based on point-to-point feature clustering;
FIG. 4 is a three-dimensional free-form curvature pipeline point cloud and its axis detected from a three-dimensional point cloud scene;
FIG. 5 is a point cloud completion result diagram of a freely curved pipeline in a three-dimensional point cloud scene.
Detailed Description
The technical solutions in the embodiments of the present invention will be described clearly and completely with reference to the accompanying drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, rather than all embodiments, and all other embodiments obtained by a person skilled in the art based on the embodiments of the present invention belong to the protection scope of the present invention without creative efforts.
The general implementation flowchart of a specific embodiment of the present invention, as shown in fig. 1, specifically includes the following steps:
step 11: uniformly sampling three-dimensional points in the three-dimensional point cloud of a scene to be detected, establishing a spherical support domain with a certain radius by taking each sampling point as a center, and extracting three-dimensional point cloud blocks covered by each support domain; and calculating the three-dimensional normal vector of each point in the point cloud block and aggregating the three-dimensional normal vector with the three-dimensional position information to form the six-dimensional point cloud block.
Step 12: for each six-dimensional point cloud block, randomly sampling six-dimensional point pairs pair for multiple timesi＝{p′im,p′inAnd enabling the included angle of the normal vector of the point pair to meet certain threshold value constraint
Figure BDA0002790194220000061
So as to ensure the accuracy of the subsequent cylindrical feature solution. In this embodiment, the normal angle threshold of the point pair is set to 30 °, and 50 times of sampling of the random six-dimensional point cloud blocks are performed in each point cloud block;
and (3) calculating the cylinder characteristic parameters of each six-dimensional point pair: the cylinder axis and radius are calculated by first calculating the normal to both directions nim,ninThe vector axis of ═ nim×ninLet a plane normal to axis be P1(ii) a Then, p 'is calculated over one sampling point'imNormal direction n ofimAnd is perpendicular to the plane P1Plane P of2Calculating another sampling point p'inNormal direction n ofinIn a straight line to the plane P2Is a point on the axis of the cylindrical section, and is p 'to the sampling point'inThe distance between the point cloud blocks is the radius of the cylindrical section, the solved radius of the cylinder is used as a characteristic, and the radius of the cylinder obtained by resolving the six-dimensional point pairs sampled for many times is aggregated into a point-to-point type cylindrical characteristic set of the point cloud block. In this embodiment, 50 random samples are performed in each cloud block, so that a 50-dimensional point-to-point cylindrical feature set is finally formed.
Step 13: and performing feature clustering on the point-to-point cylindrical feature set corresponding to each point cloud block, and judging whether the point cloud block to which the point-to-point cylindrical feature set belongs is a three-dimensional pipeline surface point cloud block or not according to the internal point proportion of the clustered maximum cluster, wherein a mean value clustering algorithm is adopted in the embodiment, the clustering bandwidth is set to be 0.0015, and the maximum cluster point proportion threshold of the pipeline feature points is judged to be 0.7. Fig. 2 shows a central point of a cloud block of a three-dimensional pipeline surface point detected from a three-dimensional point cloud on the local surface of an engine, wherein a scene to be detected is the local surface of an engine body, 3 pipelines are distributed on the surface, as indicated by the numbers 1 to 3 in the figure, wherein the number 1 pipeline and the number 2 pipeline are isolated due to the fixation of the same clamp member, a joint is arranged at the end of the number 3 pipeline, and the number 1 pipeline and the number 3 pipeline are staggered; the points with larger particles in fig. 2 are the central points of the three-dimensional pipeline surface point cloud blocks automatically detected in the local scene of the engine body, and the points are distributed on the pipeline surface without error detection points.
Step 14: and (3) establishing a pipeline parameterized geometric model in a four-dimensional time-space domain, wherein an upper right sub-graph is a rendering graph of the three-dimensional bent pipeline, and a main graph is a modeling graph, as shown in FIG. 3. The axis of the pipeline is taken as a time domain coordinate axis, and the freely bent pipeline is expressed as a discrete parameterized cylindrical section on a three-dimensional space domain and a time sequence aggregation of the parameterized cylindrical section on a one-dimensional time domain. The aggregation is a time sequence extension process which takes a pipeline section where the characteristic point is located as an initial point, and the pipeline section at each moment is parameterized into an axial vector and a cylinder radius in a three-dimensional space domain.
Step 15: according to the parameterized geometric model of the pipeline in the step 14, the time sequence parameter estimation of the pipeline is carried out in a four-dimensional time space domain based on a Kalman time domain filtering model, and the parameterized representation seg of the pipeline segment is carried out by using the local point cloud block where the pipeline characteristic point detected in the step 13 is locatedi0＝{ceni0,diri0,ri0And f, taking the state of the parameter as the starting time T of the pipeline detection being 0. In this schematic diagram, the starting time T is 0, and is placed at the end of the pipeline, and only the process of extending the axis of the pipeline to one end is explained in detail, (if the starting time is located in the middle section of the pipeline, the extending processes from the starting time to the two ends of the axis are consistent with this explanation, and are not described in detail). And (3) performing iterative extension on the cylindrical section from the initial moment along the T axis of the time domain coordinate system, wherein the cylindrical section parameter estimation at each moment is a fusion filtering result of time domain prediction parameters and three-dimensional space domain geometric resolving parameters. In this example, the time domain parameter estimation model is established based on the kalman filter model as follows:
Figure BDA0002790194220000071
wherein
Figure BDA0002790194220000072
And
Figure BDA0002790194220000073
respectively is the optimal axle center and the optimal axis of the pipeline segment at the last moment;
Figure BDA0002790194220000074
and
Figure BDA0002790194220000075
respectively obtaining the optimal axis and axis estimation results of the pipeline section at the current moment; a. the6×6Estimating a matrix for the pipeline axis, wherein the matrix is used for predicting the pipeline parameter at the next moment along the axial direction at a certain time interval; k6×6Is a Kalman gain matrix, which is at eachThe time is updated; h6×6Is an observation matrix of the pipeline parameters, which is set as a unit matrix in the example; n 'of'tAnd dir'tRespectively obtaining the axis and the axis of the pipeline section at the current moment according to the fitting of the three-dimensional point cloud block in the local support domain;
in fig. 3, point a is the axis point of the pipeline segment calculated at time t-1, and the predicted value of the pipeline segment parameter at time t can be obtained from the pipeline segment parameter at point a; according to the parameterized geometric model of the pipeline, the measured value of the parameter of the pipeline section at the time t can be calculated by local point cloud of the pipeline section at the time t, and the optimal parameter value of the pipeline section at the time t, namely the optimal parameter of the pipeline section where the point B is located, can be obtained by fusing the measured value and the predicted value. And performing iterative fusion calculation at each moment to finally detect and obtain the point cloud of the surface of the free bent pipeline sensing the four-dimensional space-time information and the parameterized model representation of the pipeline, namely the axis and the radius of the three-dimensional pipeline, wherein the radius of the whole pipeline is the radius { r } of the pipeline section at each momentitThe largest cluster mean obtained via mean clustering. The point cloud of the surfaces of the three freely bent pipelines detected in the engine scene of the example is shown in fig. 4, the axes of the three detected bent pipelines are displayed in the pipelines in a three-dimensional point mode, wherein the No. 1 and the No. 2 pipelines both span the partition and the shelter of the hoop connecting piece, and the complete extension detection of the whole pipeline is realized.
Step 16: sampling a point on each three-dimensional pipeline axis as an initial time point according to the pipeline parameterized geometric model established in the step 14, and reconstructing a complete three-dimensional pipeline surface in a time sequence extending mode at certain time intervals; with the minimum non-rigid registration error of the reconstructed pipeline model and the original three-dimensional point cloud scene as a target function, iteratively calculating optimal three-dimensional pipeline parameters, and finally realizing detection and point cloud completion of a freely bent pipeline in the three-dimensional point cloud scene, for example, fig. 5 is an example of the reconstructed and optimized three-dimensional freely bent pipeline model, wherein three complete pipeline models with the numbers of 1,2 and 3 in the figure respectively correspond to the reconstruction results of No. 1 to No. 3 pipelines in the original point cloud, and the reconstructed pipeline model has a complete pipeline surface compared with the pipeline point cloud in the original scene, so that the defect of the partial surface of the pipeline caused by self-shielding of the original point cloud under an acquisition view angle is compensated, and the function of completing the original pipeline point cloud is achieved; this example also illustrates that the present technique is applicable to pipes of various degrees of curvature.
The invention provides a four-dimensional space-time perception large-scene free bending pipeline detection and point cloud completion method which mainly comprises 6 steps of six-dimensional characteristic point sampling, point pair type cylindrical characteristic set construction, three-dimensional pipeline surface characteristic point detection, three-dimensional pipeline parameter initialization, four-dimensional space-time domain detection of a three-dimensional pipeline, and three-dimensional pipeline model completion and optimization. The method of the invention is tested on a three-dimensional point cloud model of the aeroengine, can detect freely bent pipelines distributed on the surface and carry out point cloud completion in a model generation mode, and meanwhile, the pipeline detection algorithm of the invention can cross over the pipeline connecting piece and other interferents to realize the detection of the whole pipeline. Therefore, the invention has theoretical feasibility and practical effectiveness. The invention depends on three-dimensional point cloud data to carry out pipeline detection in a three-dimensional mode, avoids the interference of ambient illumination, shadow and highlight body surface to the detection process, has stronger environmental adaptability, and is particularly suitable for the detection and model completion of multi-free bent pipelines in large-scale complex scenes.
Although illustrative embodiments of the present invention have been described above to facilitate the understanding of the present invention by those skilled in the art, it should be understood that the present invention is not limited to the scope of the embodiments, but various changes may be apparent to those skilled in the art, and it is intended that all inventive concepts utilizing the inventive concepts set forth herein be protected without departing from the spirit and scope of the present invention as defined and limited by the appended claims.

Claims (3)

1.一种四维时空感知的大场景自由弯曲管路检测及点云补全方法，其特征在于，包括以下步骤：1. a large scene free bending pipeline detection and point cloud completion method of four-dimensional space-time perception, is characterized in that, comprises the following steps: 步骤a、在待检场景的三维点云中均匀采样三维点，以各采样点为中心，建立具有一定半径的球形支撑域并提取各支撑域覆盖的三维点云块；计算三维点云块中各点的三维法向量并与各三维点的三维位置信息聚合，形成六维点云块；Step a, uniformly sample 3D points in the 3D point cloud of the scene to be inspected, establish a spherical support domain with a certain radius with each sampling point as the center, and extract the 3D point cloud blocks covered by each support domain; The three-dimensional normal vector of each point is aggregated with the three-dimensional position information of each three-dimensional point to form a six-dimensional point cloud block; 步骤b、在步骤a中的六维点云块中多次随机采样六维点对，并使得六维点对的法向量夹角满足设定的阈值约束；根据圆柱体几何关系，由一对六维点可解算出圆柱轴心和柱体半径，因此分别解算出多次随机采样的六维点对所对应的圆柱轴心和柱体半径，将多次采样解算的柱体半径聚合为一个点对式圆柱特征集；Step b, randomly sample six-dimensional point pairs in the six-dimensional point cloud block in step a, and make the normal vector angle of the six-dimensional point pair satisfy the set threshold constraint; according to the geometric relationship of the cylinder, a pair of The six-dimensional point can solve the cylinder axis and cylinder radius, so the cylinder axis and cylinder radius corresponding to the six-dimensional point pairs of multiple random sampling are calculated respectively, and the cylinder radius calculated by multiple sampling is aggregated as A point-to-point cylindrical feature set; 步骤c、对各点云块对应的点对式圆柱特征集进行特征聚类，根据聚类后最大簇的内点比例判断该点对式圆柱特征集所归属的点云块是否为三维管道表面点，实现在场景点云中自动探测三维管道表面特征点；Step c. Perform feature clustering on the point-to-point cylindrical feature set corresponding to each point cloud block, and determine whether the point cloud block to which the point-to-point cylindrical feature set belongs is a three-dimensional pipeline surface according to the inner point ratio of the largest cluster after clustering point, realize automatic detection of 3D pipeline surface feature points in the scene point cloud; 步骤d、在四维时空域中建立管道的参数化几何模型，将自由弯曲管道表达为三维空间域上离散的参数化圆柱段及在一维时间域上的时序性聚合，该时序性聚合是以特征点所在的局部点云为起始时刻圆柱段的时序性延伸过程，各时刻的圆柱段在三维空间域中参数化为轴向量和柱体半径；Step d, establish a parametric geometric model of the pipeline in the four-dimensional space-time domain, and express the free-bending pipeline as discrete parametric cylindrical segments in the three-dimensional space domain and temporal aggregation in the one-dimensional time domain. The local point cloud where the feature points are located is the sequential extension process of the cylinder segment at the starting time, and the cylinder segment at each time is parameterized as the axial vector and the cylinder radius in the three-dimensional space domain; 步骤e、在四维时空域中建立管道的时序参数估计模型，由起始时刻分别向当前轴线两端迭代进行圆柱段的延伸和参数估计，各时刻圆柱段的参数估计是由时域预测参数和三维空间域中根据步骤d几何模型的解算参数的融合滤波结果，最终检测得到感知四维时空信息的自由弯曲管道表面点云及其参数化的模型表示，即三维管路轴线和半径；Step e, establish the time series parameter estimation model of the pipeline in the four-dimensional space-time domain, and iteratively carry out the extension and parameter estimation of the cylindrical segment from the starting time to both ends of the current axis. In the three-dimensional space domain, according to the fusion filtering results of the calculation parameters of the geometric model in step d, the surface point cloud of the free-bending pipeline and its parameterized model representation that perceives four-dimensional space-time information are finally detected, that is, the three-dimensional pipeline axis and radius; 步骤f、根据步骤d所建立的管道参数化几何模型，在步骤e求解所得的三维管路轴线上采样一个点作为起始时刻点，按一定的时刻间隔以时序延伸的模式重建出完整的三维管道表面；以重建的管道模型和原始三维点云场景的非刚性配准误差最小为目标函数，迭代解算出最优的三维管道参数，最终实现三维点云场景中自由弯曲管道的检测和点云补全；Step f, according to the pipeline parameterized geometric model established in step d, sample a point on the three-dimensional pipeline axis obtained by the solution in step e as the starting time point, and reconstruct a complete three-dimensional pipeline in a time-series extension mode at certain time intervals. Pipe surface; take the minimum non-rigid registration error between the reconstructed pipe model and the original 3D point cloud scene as the objective function, iteratively solve the optimal 3D pipe parameters, and finally realize the detection and point cloud of free bending pipes in the 3D point cloud scene Completion; 所述步骤a，在待检场景的三维点云中均匀采样三维点{pi∈R3}，R3表示三维空间域，以各采样点pi为中心，建立具有一定半径r的球形支撑域并提取支撑域覆盖的三维点云块Si＝{pij∈R3},计算点云块中各个三维点pij的三维法向量nij∈R3并与其三维度位置信息聚合，形成六维度点云块S′i＝{p′ij∈R6},R6表示六维空间域，其中p′ij＝{pij,nij}；In the step a, uniformly sample three-dimensional points {pi ∈ R 3 } in the three-dimensional point cloud of the scene to be inspected, where R 3 represents the three-dimensional space domain, and a spherical support with a certain radius r is established with each sampling point p i as the center domain and extract the three-dimensional point cloud block S i ={p ij ∈ R 3 } covered by the support domain, calculate the three-dimensional normal vector n ij ∈ R 3 of each three-dimensional point p ij in the point cloud block and aggregate its three-dimensional position information to form Six-dimensional point cloud block S′ i ={p′ ij ∈R 6 }, R 6 represents a six-dimensional space domain, where p′ ij ={p ij ,n ij }; 所述步骤b具体实现过程如下：The specific implementation process of step b is as follows: (1)针对各六维点云块S′i，多次随机采样六维点对pair＝{p′im,p′in}，并使得六维点对的法向量夹角满足设定的阈值约束
Figure FDA0003476230860000021
其中nim和nin分别为六维点对的法向量，ε为夹角阈值，以保证后续六维点对式圆柱特征解算的精度；(1) For each six-dimensional point cloud S′ i , randomly sample the six-dimensional point pair pair={p′ im , p′ in } for many times, and make the normal vector angle of the six-dimensional point pair meet the set threshold constraint
Figure FDA0003476230860000021
where n im and n in are the normal vectors of the six-dimensional point pair, respectively, and ε is the included angle threshold to ensure the accuracy of the subsequent six-dimensional point-pair cylindrical feature solution; (2)计算各六维点对的圆柱特征参数：柱体轴心和半径，首先，计算同时垂直于两个法向{nim,nin}的向量axis＝nim×nin，令以axis为法向的平面为P1；然后，计算经过一个采样点p′im的法向nim且垂直于平面P1的平面P2，计算另一采样点p′in的法向nin所在直线到平面P2的交点，该交点即为圆柱段轴线上的一点，该交点与采样点p′in之间的距离即为圆柱段半径，将求解的柱体半径作为特征，将多次采样的六维点对解算所得的柱体半径聚合为该点云块的点对式圆柱特征集；(2) Calculate the cylindrical characteristic parameters of each six-dimensional point pair: the axis and radius of the cylinder. First, calculate the vector axis=n im ×n in which is perpendicular to the two normal directions {n im ,n in } at the same time, let The plane whose axis is the normal direction is P 1 ; then, the normal direction n im passing through one sampling point p′ im and the plane P 2 perpendicular to the plane P 1 is calculated, and the normal direction n in of another sampling point p′ in is calculated. The intersection of the straight line and the plane P 2 , the intersection is a point on the axis of the cylinder segment, and the distance between the intersection and the sampling point p'in is the radius of the cylinder segment, and the calculated cylinder radius is used as a feature. The cylinder radius obtained from the six-dimensional point pair solution is aggregated into the point-to-point cylinder feature set of the point cloud block; 所述步骤d，在四维时空域建立管道的参数化几何模型，将第i根自由弯曲管道pipei表达为一维时间域上离散的参数化圆柱段segit的时序性聚合，即
Figure FDA0003476230860000022
该聚合是以步骤c探测的特征点所在局部管道段为起始的时序性延伸过程，各时刻t的管道段segit在三维空间域中参数化表示为segit＝f(cenit,dirit,rit)，其中cenit为管道段的一个轴心点，dirit为管道段轴向，rit管道段半径；In the step d, a parameterized geometric model of the pipeline is established in the four-dimensional space-time domain, and the i-th free-bending pipeline pipe i is expressed as the time-series aggregation of discrete parameterized cylindrical segments seg it in the one-dimensional time domain, that is,
Figure FDA0003476230860000022
The aggregation is a sequential extension process starting from the local pipeline segment where the feature point detected in step c is located. The pipeline segment seg it at each time t is parameterized in the three-dimensional space domain as seg it =f(cen it ,dir it , r it ), where cen it is an axis point of the pipeline segment, dir it is the axial direction of the pipeline segment, and r it is the radius of the pipeline segment; 所述步骤e具体包含：The step e specifically includes: (1)在四维时空域建立管道的时序参数估计模型，对各时刻下三维空间域中的管道段的轴心和轴线两方面参数进行融合估计，该时序参数估计模型融合了两部分参数估计结果，分别为前一时刻的管道段参数对当前时刻管道段参数的预测值和当前时刻管道参数的几何解算值，模型如下：(1) Establish the time series parameter estimation model of the pipeline in the four-dimensional space-time domain, and perform fusion estimation on the parameters of the axis and axis of the pipeline section in the three-dimensional space domain at each moment. The time-series parameter estimation model integrates the two parts of the parameter estimation results. , which are the predicted value of the pipeline segment parameters at the previous moment to the pipeline segment parameters at the current moment and the geometric solution value of the pipeline parameters at the current moment, respectively. The model is as follows:
Figure FDA0003476230860000023
Figure FDA0003476230860000023
 其中
Figure FDA0003476230860000024
和
Figure FDA0003476230860000025
分别为上一时刻管道段最优轴心和轴线估计值；
Figure FDA0003476230860000026
和
Figure FDA0003476230860000027
分别为当前时刻管道段的最优轴心和轴线估计值；A是一个6×6维度的为管道参数预测矩阵，用于延管道轴向以一定的时间间隔对下一时刻管道参数的预测；K是一个6×6维度的增益矩阵；H是一个6×6维度的管道参数的观测矩阵；cen′t和dir′t分别为当前时刻根据步骤d参数化几何模型解算所得的管道段轴心和轴线值；in
Figure FDA0003476230860000024
and
Figure FDA0003476230860000025
are the estimated values of the optimal axis and axis of the pipeline segment at the last moment, respectively;
Figure FDA0003476230860000026
and
Figure FDA0003476230860000027
are the optimal axis and axis estimates of the pipeline section at the current moment, respectively; A is a 6×6 dimension prediction matrix for the pipeline parameters, which is used to predict the pipeline parameters at a certain time interval along the axis of the pipeline at the next moment; K is a gain matrix of 6 × 6 dimensions; H is an observation matrix of pipeline parameters of 6 × 6 dimensions; cen' t and dir' t are the pipeline segment axes obtained by solving the parameterized geometric model in step d at the current moment, respectively center and axis values; (2)对步骤c检测所得的管道特征点所在的局部点云块进行圆柱体拟合，获得管道段的参数化表示segi0＝{ceni0,diri0,ri0}，其中ceni0为管道段轴心点，diri0为管道段轴向，ri0管道段半径，作为起始时刻t0的管道参数状态；根据时序参数估计模型，由起始时刻分别向当前轴线两端迭代进行管道段的连续延伸和参数估计；各时刻t的管道段参数化过程是由时域预测参数和三维空间域几何解算参数的融合滤波结果，最终检测出感知四维时空信息的自由弯曲管道表面点云及其参数化的模型表示，即三维管路轴线和半径，其中整根管道的半径采用各时刻管道段半径{rit}经由均值聚类所得的最大簇均值。(2) Perform cylinder fitting on the local point cloud block where the pipeline feature points detected in step c are located, and obtain the parametric representation of the pipeline segment seg i0 ={cen i0 ,dir i0 ,r i0 }, where cen i0 is the pipeline The axis point of the segment, dir i0 is the axial direction of the pipeline segment, and r i0 is the radius of the pipeline segment, which is the pipeline parameter state at the starting time t 0 ; according to the time series parameter estimation model, the pipeline segment is iteratively performed from the starting time to both ends of the current axis. continuous extension and parameter estimation; the parameterization process of the pipeline section at each time t is the fusion filtering result of the time domain prediction parameters and the three-dimensional space domain geometric solution parameters, and finally detects the surface point cloud of the freely curved pipeline that perceives four-dimensional space-time information and Its parameterized model representation is three-dimensional pipeline axis and radius, wherein the radius of the entire pipeline adopts the maximum cluster mean obtained by mean clustering of the pipeline segment radius {r it } at each moment. 2.根据权利要求1所述的四维时空感知的大场景自由弯曲管路检测及点云补全方法，其特征在于：所述步骤c具体包括：2. The large-scene free-bending pipeline detection and point cloud completion method of four-dimensional space-time perception according to claim 1, wherein the step c specifically comprises: 若一个点云块为三维管道表面点云，则存在如下：该点云块上构建的点对式圆柱特征集中的特征应具有明显的密度中心和聚合性，对各点云块对应的点对式圆柱特征集中的特征分量进行聚类，根据聚类后最大簇的内点比例判断该点对式圆柱特征集所归属的点云块是否为三维管道表面点，实现在场景点云中自动探测三维管道表面特征点。If a point cloud block is a 3D pipeline surface point cloud, it exists as follows: The features in the point-to-point cylindrical feature set constructed on the point cloud block should have obvious density centers and aggregations. The feature components in the feature set of the cylindrical feature set are clustered, and according to the inner point ratio of the largest cluster after clustering, it is judged whether the point cloud block to which the point-to-point cylindrical feature set belongs is the surface point of the 3D pipeline, so as to realize the automatic detection of 3D pipeline in the scene point cloud. Pipe surface feature points. 3.根据权利要求1所述的一种四维时空感知的大场景自由弯曲管路检测及点云补全方法，其特征在于：所述步骤f具体包括：3. A four-dimensional space-time perception large-scene free-bending pipeline detection and point cloud completion method according to claim 1, characterized in that: the step f specifically comprises: (1)在步骤e解算出的三维管路轴线上采样一个点作为起始时刻点，根据步骤d的管道参数化几何模型，按一定的时刻间隔以时序延伸的模式重建出完整的三维管道表面，实现对原始场景中管道点云的补全；(1) Sampling a point on the three-dimensional pipeline axis calculated in step e as the starting time point, according to the pipeline parameterized geometric model in step d, reconstruct a complete three-dimensional pipeline surface in a time-series extension mode at certain time intervals , to complete the pipeline point cloud in the original scene; (2)以重建模型和原始三维点云场景的非刚性配准误差最小作为目标函数，对重建的三维管道轴线和半径参数进行迭代优化，获取更高质量和几何准确度的完整管道点云模型，最终实现三维点云场景中自由弯曲管道的检测和点云补全。(2) Taking the minimum non-rigid registration error between the reconstructed model and the original 3D point cloud scene as the objective function, iteratively optimizes the reconstructed 3D pipeline axis and radius parameters to obtain a complete pipeline point cloud model with higher quality and geometric accuracy , and finally realize the detection and point cloud completion of free bending pipes in 3D point cloud scenes.
CN202011312309.6A 2020-11-20 2020-11-20 Four-dimensional space-time perception large-scene free bending pipeline detection and point cloud completion method Active CN112417618B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011312309.6A CN112417618B (en) 2020-11-20 2020-11-20 Four-dimensional space-time perception large-scene free bending pipeline detection and point cloud completion method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011312309.6A CN112417618B (en) 2020-11-20 2020-11-20 Four-dimensional space-time perception large-scene free bending pipeline detection and point cloud completion method

Publications (2)

Publication Number Publication Date
CN112417618A CN112417618A (en) 2021-02-26
CN112417618B true CN112417618B (en) 2022-03-11

Family

ID=74778285

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011312309.6A Active CN112417618B (en) 2020-11-20 2020-11-20 Four-dimensional space-time perception large-scene free bending pipeline detection and point cloud completion method

Country Status (1)

Country Link
CN (1) CN112417618B (en)

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113205086B (en) * 2021-07-05 2021-09-28 武汉瀚迈科技有限公司 Method for identifying characteristic parameters of circular-section bent pipe type part based on ellipse fitting
CN114066859B (en) * 2021-11-18 2025-07-11 新拓三维技术(深圳)有限公司 A pipeline measurement method and device
CN114419012B (en) * 2022-01-25 2024-06-28 北京工商大学 Full range clearance measurement method and application of aircraft engine pipeline based on 3D point cloud
CN115546302B (en) * 2022-10-20 2024-06-14 上海人工智能创新中心 Point cloud data resolving method for local geometric modeling
CN116091559B (en) * 2022-12-01 2025-12-05 西北工业大学 An Assembly State Recognition Method Based on Optimization Perspective
CN116011107A (en) * 2023-01-10 2023-04-25 南京航空航天大学 Hole Feature Extraction Method, Device and System for Large Composite Components
CN116740169A (en) * 2023-06-27 2023-09-12 中国石油大学(北京) Methods, devices, electronic equipment and storage media for extracting central axis of pipelines
CN117058027A (en) * 2023-08-09 2023-11-14 苏州埃米迈德医疗科技有限公司 A method of repairing blood vessel geometric models based on point cloud data

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109147038A (en) * 2018-08-21 2019-01-04 北京工业大学 Pipeline three-dimensional modeling method based on three-dimensional point cloud processing
CN110889243A (en) * 2019-12-20 2020-03-17 南京航空航天大学 A three-dimensional reconstruction method and detection method of aircraft fuel tank based on depth camera

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9070216B2 (en) * 2011-12-14 2015-06-30 The Board Of Trustees Of The University Of Illinois Four-dimensional augmented reality models for interactive visualization and automated construction progress monitoring
US9852238B2 (en) * 2014-04-24 2017-12-26 The Board Of Trustees Of The University Of Illinois 4D vizualization of building design and construction modeling with photographs

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109147038A (en) * 2018-08-21 2019-01-04 北京工业大学 Pipeline three-dimensional modeling method based on three-dimensional point cloud processing
CN110889243A (en) * 2019-12-20 2020-03-17 南京航空航天大学 A three-dimensional reconstruction method and detection method of aircraft fuel tank based on depth camera

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
《基于轴线投影精确模型的弯管立体视觉测量方法》;孙军华等;《航空制造技术》;20190301;第62卷(第05期);全文 *
《面向RGB-D数据的4D-ICP点云配准方法》;苏本跃等;《南京大学学报(自然科学)》;20180730;第54卷(第04期);全文 *

Also Published As

Publication number Publication date
CN112417618A (en) 2021-02-26

Similar Documents

Publication Publication Date Title
CN112417618B (en) Four-dimensional space-time perception large-scene free bending pipeline detection and point cloud completion method
CN113205466B (en) Incomplete point cloud completion method based on hidden space topological structure constraint
CN109147038B (en) 3D Pipeline Modeling Method Based on 3D Point Cloud Processing
Song et al. View path planning via online multiview stereo for 3-d modeling of large-scale structures
CN118397000B (en) Dynamic data processing method and system based on industrial equipment
CN110992427B (en) A three-dimensional pose estimation method and positioning and grasping system for deformed objects
JP2024542591A (en) Method and system for reconstructing building information modeling (BIM) data for piping systems - Patents.com
CN114091628B (en) Three-dimensional point cloud up-sampling method and system based on double branch network
CN118781178B (en) A volume measurement method based on surface reconstruction and triple integral
CN119273868A (en) A real-time 3D fusion reconstruction method for ship comprehensive test scenes
CN114119690A (en) Point cloud registration method based on neural network reconstruction Gaussian mixture model
CN116557787B (en) Intelligent evaluation system and method for pipe network state
CN112883976A (en) Point cloud based semantic segmentation method, device and system and storage medium
CN118365969A (en) Intelligent robot-based all-condition urban underground pipe culvert detection method and system
CN120411364A (en) Three-dimensional reconstruction method and system for underwater targets based on three-dimensional imaging sonar images
Hu et al. Active implicit reconstruction using one-shot view planning
Xie et al. Evaluating loss landscapes from a topology perspective
CN120932221B (en) Platform license plate recognition method and system
CN119445291B (en) 3D point cloud task analysis method integrating local features and global context information
CN121147441A (en) A Smart Detection and Mapping Method for Underground Pipelines Based on Image Recognition
CN120725965A (en) Cable defect detection method using binocular intelligent inspection robot
CN119417865A (en) Optical flow calculation method based on global matching
Liao et al. A coupled 3D morphological reconstruction approach for surface microcrack in Si3N4 ceramic bearing roller based on adaptive nano feature extraction & multiscale depth fusion
CN121259219B (en) Multi-view visual characteristic reconstruction method for spring assembly facing complex working conditions
CN121392195B (en) Methods and Systems for Constructing and Optimizing Real-Scene 3D Models Using UAVs in Urban Planning

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