CN111222516B - Method for extracting point cloud key outline features of printed circuit board - Google Patents

Method for extracting point cloud key outline features of printed circuit board Download PDF

Info

Publication number
CN111222516B
CN111222516B CN202010010772.9A CN202010010772A CN111222516B CN 111222516 B CN111222516 B CN 111222516B CN 202010010772 A CN202010010772 A CN 202010010772A CN 111222516 B CN111222516 B CN 111222516B
Authority
CN
China
Prior art keywords
point
points
point cloud
printed circuit
circuit board
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
CN202010010772.9A
Other languages
Chinese (zh)
Other versions
CN111222516A (en
Inventor
钟文彬
李旭瑞
孙思
刘光帅
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.)
Southwest Electronic Technology Institute No 10 Institute of Cetc
Original Assignee
Southwest Electronic Technology Institute No 10 Institute of Cetc
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 Southwest Electronic Technology Institute No 10 Institute of Cetc filed Critical Southwest Electronic Technology Institute No 10 Institute of Cetc
Priority to CN202010010772.9A priority Critical patent/CN111222516B/en
Publication of CN111222516A publication Critical patent/CN111222516A/en
Application granted granted Critical
Publication of CN111222516B publication Critical patent/CN111222516B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/06—Recognition of objects for industrial automation
    • 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)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Multimedia (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a method for extracting key outline features of a point cloud of a printed circuit board, and aims to provide a method which consumes less time and has high extraction efficiency and can extract key outline feature points of the point cloud. The invention is realized by the following scheme: defining boundary points by using an included angle of a connecting line of a query point and a neighboring point on a projection plane, preprocessing PCB point cloud data by adopting a straight-through filtering method, and establishing a kd-tree topological structure and a point cloud index vector for storing a clustering result; then, detecting the maximum plane area of the printed circuit board by using a random sampling consistency algorithm, calculating the Euclidean distance between each adjacent point and the seed point, comparing the normal vector included angle between the two points, converting the edge folding point into a boundary point, and extracting a point cloud data characteristic boundary 1; separating key features on the space, and extracting a point cloud data feature boundary 2; and combining the boundary points 1 and 2 of the two parts together, and extracting the key outline characteristics of the point cloud to obtain the result of the PCB key outline.

Description

Method for extracting key outline characteristics of point cloud of printed circuit board
Technical Field
The invention belongs to the photoelectric industry, and particularly relates to a method for extracting a point cloud key contour feature of a printed circuit board.
Background
Key contour extraction is one of the contents of advanced subject research such as computer graphics and computer vision. The description and extraction of the 3D point cloud features are the most basic and key parts in the point cloud information processing. The point cloud data has one more dimension compared with the two-dimensional digital image, so the three-dimensional laser scanning technology is a measuring mode with space three-dimensional information. With the improvement of the acquisition precision of the 3D point cloud data and the reduction of the cost, the three-dimensional scanning technology provides a new idea for solving the practical engineering application for engineers in multiple fields. By introducing measurement into a three-dimensional space, target structure parameters which cannot be obtained or are not suitable for being obtained by a traditional measuring instrument and digital image analysis can be obtained, so that the phenotype information of a Printed Circuit Board (PCB) is more comprehensive, the workload of respectively collecting images from multiple angles in the traditional method is reduced to a certain extent, meanwhile, tasks such as collecting size information of typical characteristics of a target and reconstructing the surface can be completed only by a complete point cloud model, and the processing speed of subsequent work is improved. The contour detection technology is particularly important in the appearance and quality detection of some precision workpieces, such as the detection of vehicle bodies, the detection of precision part machining and the like. In order to obtain plane contour feature points from massive point cloud data and realize model reconstruction based on contour line features, in the prior art, point clouds are layered according to the extraction rate of the points, an angular deviation method is provided to obtain curvature mutation points of each layer of data, and a contour line model of the point cloud is generated through a homotopic algorithm. Since curvature break point acquisition is susceptible to noise, this method has poor noise immunity. For this reason, an iterative algorithm based on RE and RP is proposed, and shape factors are defined to calculate shape errors of each layer of data, so as to adjust the thickness of point cloud layering, thereby realizing the adaptation of layering division, but the generation efficiency of contour feature lines is low. To solve the problem of low generation efficiency, Liu et al propose a curve model method (IPCM) based on intermediate points to integrate RE and RP., which is high in efficiency, but does not solve the multi-ring phenomenon in the point cloud slice. Through integrating RE and RP, the process of surface fitting and STL file generation is avoided, the application field is mostly mechanical manufacturing, and the purpose is to use a curve model as the input of an RP system. Most algorithms for processing of point cloud identification, segmentation, resampling, registration, curved surface reconstruction and the like in the prior art seriously depend on the result of feature description and extraction. For a Printed Circuit Board (PCB), the overall size of the PCB is small, the number of miniature components is large, the drawing form of a lead wire is not fixed, so that a point cloud model is relatively small, the number of detailed features is large, the feature lines are dense, when a two-dimensional image of the PCB is taken as a processing object, multiple and multi-angle image acquisition is often required, and the efficiency is low; in addition, some existing 3D point cloud processing methods are too cumbersome in process and large in calculated amount and storage amount because three-dimensional images need to be converted into two-dimensional images. Therefore, how to accurately and efficiently extract the key outline features of the point cloud of the Printed Circuit Board (PCB) by using the three-dimensional image is a challenging subject. The extraction of the key outline characteristic line of the printed circuit board is the basis and the key of the quality detection of the printed circuit board, and plays a very important role in the detection of the photoelectric industry. Some existing contour feature extraction methods applied to the field of point cloud still need to project acquired three-dimensional point cloud data into a two-dimensional image, so that key contour features are indirectly extracted by using a digital image processing technology, intermediate conversion steps are added, the advantage that the three-dimensional point cloud data can quickly acquire target structure parameters cannot be exerted, and the efficiency is low.
At present, methods directly used for extracting key outlines of Printed Circuit Boards (PCBs) exist, but the methods do not relate to three-dimensional point clouds, do not map three-dimensional graphics into two-dimensional images for processing, and only simply perform traditional digital image processing on the two-dimensional images directly acquired. Most of the existing printed circuit board contour detection methods are based on traditional digital image processing, namely, the original data processed by a feature extraction algorithm are two-dimensional images. For example, the following chinese patent publications: the Chinese patent publication No. CN109472271A [ P ]. 2019-03-15, entitled "printed circuit board image contour extraction method and device", proposes a method and device for extracting a printed circuit board contour, the method firstly carries out binarization processing on an original gray level image, carries out convolution processing on the binary image by using a Gaussian difference (LOG) operator to obtain a corresponding convolution value, and finally obtains the contour of the original gray level image according to edge points, sub-pixel edge points, the gradient and amplitude of the edge points and the gradient and amplitude of the sub-pixel edge points.
Chinese patent publication No. CN109934837A [ P ] 2019-06-15 discloses a method for extracting the outline of a 3D plant leaf. The method comprises the steps of firstly placing a 3D point cloud model in a space coordinate system, then projecting the point cloud model onto a two-dimensional plane, adjusting a projected two-dimensional image to be parallel to the 3D point cloud model, finally finding out 3D points corresponding to the outline of the two-dimensional leaf through a shortest distance matching algorithm, and connecting the outlines formed by all the 3D points to obtain the outline of the plant leaf.
Chinese patent publication No. CN106780524A [ P ] 2017-05-31 discloses a method for automatically extracting a three-dimensional point cloud road boundary. The method comprises the steps of firstly carrying out hyper-voxel division on the whole three-dimensional point cloud data, and then extracting road boundary points by respectively using an alpha-shape algorithm and an energy minimization algorithm based on graph cut.
Photoelectron, laser, 2013, 24(4):740 and 745 printed circuit board photoelectric image edge detection research literature containing noise and being fuzzy proposes an image fusion edge information extraction method combining wavelet transformation and Canny edge detection. The method completely detects the edge of the image through the steps of median filtering, enhancing deblurring, wavelet decomposition, image fusion, image drying removal and the like.
When point cloud data is acquired, due to influences caused by equipment precision, operator experience, environmental factors and the like, and influences of electromagnetic wave diffraction characteristics and surface properties of a measured object, some noise points inevitably appear in the point cloud data. By observing the actual collected printed circuit board data, the noise points are mainly outliers, i.e. noise points that are free outside the surface of the printed circuit board. The kd-Tree is a special data structure for dividing data points of a high-dimensional space, and is mainly applied to data searching of the high-dimensional space, such as: range search and k-nearest neighbor (KNN) search, wherein the range search is to give a query point and a distance threshold value and acquire all data points within the threshold value range; the KNN search is to give the number n of the query points and the number n of the search points and find out the number of the n nearest search points; if the two searches are implemented by the conventional method, all points in the data set may be exhausted in the worst case, and this method has the disadvantages that the structural information in the data set is not utilized at all, and when there are many data points, the search efficiency is not high.
The method applied to the extraction of the outline features in the point cloud field is directly applied to the extraction of the key outline feature points of the point cloud of the Printed Circuit Board (PCB), and the outline extraction effect is not ideal due to the structural particularity of the Printed Circuit Board (PCB). For example, although the point cloud slice is guided to the manually drawn contour line after CAD, the contour characteristics of the point cloud slice can be expressed, but it takes a lot of time to manually judge and manually connect.
Disclosure of Invention
Aiming at the defects of the prior art for extracting the key outline feature points of the point cloud of the printed circuit board, the invention provides the method which has less total time consumption and high outline feature point extraction efficiency and can efficiently extract the key outline feature points of the point cloud of the printed circuit board.
The above object of the present invention can be achieved by the following measures, a method for extracting a point cloud key contour feature of a printed circuit board, comprising the steps of:
scanning and acquiring point cloud data of the Printed Circuit Board (PCB) from multiple angles around the PCB, projecting the point cloud data onto a plane, and utilizing a query point and k neighbor points thereof on the projection plane
Figure BDA0002357071990000031
Angle theta of connecting lineiDefining boundary points, indexing variables i, preprocessing point cloud data of a Printed Circuit Board (PCB) by adopting a direct filtering method, establishing a kd-tree topological structure and a point cloud index vector C for storing a clustering result, and realizing an algorithm for quickly and accurately generating a point cloud contour curve model in a subsequent module; then, the maximum plane area of the printed circuit board is detected by using a random sample consensus (RANSAC) algorithm, and the distance p is searched by using a kd-treeqNearest k points, performing k neighbor search on each query point, and calculating each k neighbor point
Figure BDA0002357071990000032
And the seed point pqConverting the edge folding points into boundary points, extracting key outline features by using a boundary extraction algorithm, and extracting a key outline feature boundary 1 of the point cloud data of the printed circuit board; on the basis of finishing the maximum plane separation of the printed circuit board, utilizing an Euclidean clustering algorithm based on a normal vector included angle to divide the point cloud by adopting the point cloud clustering algorithm according to the Euclidean distance of discrete data points in the space, separating key features on the space, limiting clustering conditions by introducing the normal vector included angle, extracting a key contour feature boundary 2 of the point cloud data of the printed circuit board, and combining the boundary 1 and the boundary 2; according to the sequence Q [ j]Has seed point pqJudging whether the cluster meets the set distance threshold condition, if so, continuing to compare the k neighbor points
Figure BDA0002357071990000033
Normal to
Figure BDA0002357071990000034
And the seed point pqNormal to
Figure BDA0002357071990000035
The included angle between the data points still meets the condition, the data points at different positions are judged
Figure BDA0002357071990000036
And the seed point pqIs the same cluster class and new k neighbor points are arranged
Figure BDA0002357071990000037
Added to the current sequence Q [ j ]]In the method, the normal vector of the fitting plane is calculated by using the covariance matrix
Figure BDA0002357071990000038
Separately find the query point pqAnd its k neighbor points
Figure BDA0002357071990000039
Angle of line of (a) thetaiDetermining a sequence of angles theta from small to large as { theta ═ theta1,…,θiAnd merging the boundary points 1 and 2 of the two parts together, and extracting the point cloud key outline characteristics of the Printed Circuit Board (PCB) to obtain the result of the key outline of the PCB.
Compared with the prior art, the invention has the following beneficial technical effects:
the total time consumption is less. The invention utilizes RANSAC algorithm to detect the maximum plane area of the area in the printed circuit board, and utilizes kd-tree to search the distance pqNearest k points, performing neighbor search speed search on each query point, and calculating each k neighbor point
Figure BDA0002357071990000041
And the seed point pqExtracting a key outline characteristic boundary 1 of the point cloud data; on the basis of finishing the maximum plane separation of the printed circuit board, remaining point cloud data adopts an Euclidean clustering algorithm based on a normal vector included angle, the point cloud is segmented by adopting the point cloud clustering algorithm according to the Euclidean distance of discrete data points in space, key features are separated on the space, a key outline feature boundary 2 of the point cloud data is extracted, edge folding points are converted into boundary points, then the boundary extraction algorithm is used for extracting key outline features, finally, the extracted key outline information of the printed circuit board PCB is neat, and the total time consumption is small.
The extraction efficiency is high. The invention provides a method for extracting key contour features aiming at the current situation that the existing method has an unsatisfactory effect on extracting the three-dimensional point cloud contour of a Printed Circuit Board (PCB), which is different from the traditional two-dimensional digital image processing technology, does not relate to the mapping of a three-dimensional image to a two-dimensional plane in the intermediate process, and has originality in work. By using query point and its k adjacent points on projection plane
Figure BDA0002357071990000042
Angle theta of connecting lineiDefining boundary points, preprocessing the point cloud data of the printed circuit board by adopting a straight-through filtering method to be processedCreating a kd-tree by the point cloud data P, establishing a topological relation between discrete points by using the kd-tree, and establishing a kd-tree topological structure and a point cloud index vector C for storing a clustering result; the method can directly process the acquired point cloud data, avoids the inconvenience of mapping from a three-dimensional image to a two-dimensional image, improves the efficiency of extracting the key outline characteristics of the point cloud of the PCB, and avoids the loss of data in the conversion process.
Has better robustness. In the invention, because the components of the printed circuit board are all welded on a flat surface with the largest area, the characteristic of RANSAC (random sample consensus) algorithm is utilized to finish the independent extraction of the planar area with the largest area in the printed circuit board, realize the separation of key features on space and provide point cloud data which can be used for feature extraction for the subsequent algorithm. Under the theoretical condition, after the sampling interval of the three-dimensional laser scanner is set, the acquired data point interval values of the same target object are all the sampling interval, namely point clouds of the same target object are always close to each other, and no obvious fracture or dispersion phenomenon appears visually, and the original Euclidean clustering algorithm utilizes a distance threshold value so as to realize the clustering of fractured or dispersed data points, but the point clustering cannot be realized on the point cloud data in a continuous area, so that the key features of the printed circuit board need to be firstly separated into discrete clusters, and then the method provided by the text is used. The regular planar structures on the printed circuit board are more, and RANSAC (random sample consensus) as a model-based segmentation method can estimate parameters in a planar model, so that the corresponding planar model is segmented, and the method has better robustness to external points and noise points.
The method comprises three major parts, namely point cloud data acquisition, data preprocessing and key contour feature point extraction, can acquire target structure parameters which cannot be acquired or are not suitable for acquisition by a traditional measuring instrument and digital image analysis by introducing measurement into a three-dimensional space, converts edge folding points into boundary points, and extracts key contour features by using a boundary extraction algorithm; finally, the extracted key outline information of the printed circuit board is neat, the total time consumption is less, and the efficiency of extracting the key outline characteristics of the point cloud of the printed circuit board is improved. By introducing measurement into a three-dimensional space, target structure parameters which cannot be obtained or are not suitable for being obtained by a traditional measuring instrument and digital image analysis can be obtained, so that the obtained phenotype information of a Printed Circuit Board (PCB) is more comprehensive, the workload of respectively collecting images from multiple angles in the traditional method is reduced to a certain extent, meanwhile, tasks such as collecting size information of typical characteristics of a target, reconstructing the surface and the like can be completed only by one complete model, and the processing speed of subsequent related work is improved.
Drawings
FIG. 1 is a basic process for extracting characteristic contour lines of a printed circuit board according to the present invention.
FIG. 2 is a diagram of raw point cloud data of a printed circuit board.
Fig. 3 shows the result after printed circuit board filtering.
Fig. 4 shows the result of extracting the maximum plane using the RANSAC method.
FIG. 5 is a schematic diagram of a normal vector for calculating a fitting plane using a covariance matrix.
FIG. 6 is a schematic diagram of the boundary properties of a point when the point is on the boundary contour.
FIG. 7 is a schematic diagram of the boundary attribute of a point when the point is not on the boundary contour.
Fig. 8 illustrates key outline features of a printed circuit board extracted using the present invention.
The invention is described in further detail below with reference to the figures and the detailed description.
Detailed Description
See fig. 1. According to the invention, point cloud data of the printed circuit board PCB is acquired by scanning and collecting the printed circuit board PCB from multiple angles around the printed circuit board PCB, the point cloud data is projected onto a plane, and a query point and k neighbor points thereof on the projection plane are utilized
Figure BDA0002357071990000055
Angle theta of connecting lineiDefining boundary points, indexing a variable i, preprocessing PCB point cloud data by adopting a direct filtering method, establishing a kd-tree topological structure and storing the point cloud of a clustering resultIndexing the vector C to realize an algorithm for quickly and accurately generating a point cloud contour curve model in a subsequent module; random sampling consistency, detecting the maximum plane area of the PCB by using RANSAC algorithm, and searching the distance P by using kd-treeqNearest k points, performing k neighbor search on each query point, and calculating each k neighbor point
Figure BDA0002357071990000051
And the seed point PqConverting the edge folding points into boundary points, extracting key outline features by using a boundary extraction algorithm, and extracting a key outline feature boundary 1 of the point cloud data of the printed circuit board; on the basis of finishing the maximum plane separation of the printed circuit board, utilizing an Euclidean clustering algorithm based on a normal vector included angle to divide point cloud of the residual point cloud data according to the Euclidean distance of discrete data points in space by adopting the point cloud clustering algorithm, separating key features on the space, extracting a key outline feature boundary 2 of the point cloud data of the printed circuit board, introducing the normal vector included angle to limit clustering conditions, and combining the boundary 1 and the boundary 2; according to the sequence Q [ j]Has seed point PqJudging whether the cluster meets the set distance threshold condition, if so, continuing to compare the k neighbor points
Figure BDA0002357071990000052
Normal to
Figure BDA0002357071990000053
And the seed point PqNormal to
Figure BDA0002357071990000054
The included angle between the data points still meets the condition, the data points at different positions are judged
Figure BDA0002357071990000061
And the seed point PqIs the same cluster class and new k neighbor points are arranged
Figure BDA0002357071990000062
Join to current sequenceQ[j]In the method, the normal vector of the fitting plane is calculated by using the covariance matrix
Figure BDA0002357071990000063
Separately find the query point pqAnd its k neighbor points
Figure BDA0002357071990000064
Angle of line of (a) thetaiDetermining a sequence of angles theta from small to large as { theta ═ theta1,…,θiAnd merging the boundary points 1 and 2 of the two parts together, and extracting the point cloud key outline characteristics of the Printed Circuit Board (PCB) to obtain the result of the key outline of the PCB.
The extraction of the key outline characteristics of the point cloud of the printed circuit board comprises the following steps: the method comprises the following steps of point cloud data acquisition, data preprocessing and key contour feature point extraction.
Step 1: collecting point cloud data of a Printed Circuit Board (PCB): before scanning, spraying a developer on the PCB to ensure that data information obtained by scanning is complete; during scanning, the PCB is fixed and can be scanned around the PCB from a plurality of angles respectively to obtain a processable point cloud. The scanning equipment used for collecting the point cloud data can adopt a FreeScan X5 laser handheld three-dimensional scanner of Beijing Tianyuan three-dimensional science and technology company, a corresponding scanning strategy is formulated according to the actual situation before scanning, and a developer is sprayed on a PCB in advance to ensure that the data information obtained by scanning is complete as much as possible; during scanning, the PCB is fixed and is scanned around the PCB from a plurality of angles, so that a processable point cloud without a shielding problem is obtained, and the visual effect of the point cloud is shown in fig. 2.
Step 2: preprocessing point cloud data of a Printed Circuit Board (PCB):
(2.1) through filtering: and filtering according to a certain dimension by adopting a straight-through filtering method, and removing points in a specified range so as to improve the extraction speed and the extraction precision of the key contour line characteristic points in the later period.
(2.2) establishment of a kd-Tree topology: and establishing a topological relation between the discrete points by using the kd-tree, thereby realizing the quick search based on the adjacent points.
And step 3: extracting key contour feature points:
(3.1) detection of the maximum plane: the RANSAC algorithm is used for independently extracting the plane area with the largest area in the printed circuit board, so that the separation of key features on the space is realized, and point cloud data which can be used for feature extraction is provided for the subsequent algorithm.
(3.2) Euclidean clustering based on normal vector included angles:
on the basis of finishing the maximum plane separation of the printed circuit board, according to the Euclidean distance of discrete data points in the space, the point cloud is divided by adopting a point cloud clustering algorithm, and the condition that the data points at different positions are judged to be in the same cluster is as follows: k neighborhood points piTo a query point pqDistance L (p) ofq,pi) Certain threshold conditions are met, which can be described as: l (p)q,pi)≤dth (1)
Wherein d isthIs a threshold value. Meanwhile, in order to enable the same cluster point set to have more similar geometric properties, namely to enable the boundary point information to be more complete, a normal vector included angle is introduced to limit the clustering condition; first, for each query point p, using k-nearest neighbor searchq∈PkPerforming fast search of k adjacent points; wherein, Pk＝{p1,…,piFor each pi∈PkDistance diThe expression of (a) is:
Figure BDA0002357071990000071
Figure BDA0002357071990000072
fitting the plane by least squares, using the constraint d i0, a covariance matrix c ∈ R is obtained3×3The expression is
Figure BDA0002357071990000073
In the formula:
Figure BDA0002357071990000074
represents piThe weight of the point is 1;
Figure BDA0002357071990000075
Figure BDA0002357071990000076
representing the three-dimensional centroid determined by the k neighbor points, we can then derive:
Figure BDA0002357071990000077
in the formula: lambda [ alpha ]mIs a characteristic value;
Figure BDA0002357071990000078
as a characteristic value λmA corresponding feature vector; the feature vector corresponding to the minimum feature value obtained from the constraint condition is the query point pq∈PkNormal vector of (A)
Figure BDA0002357071990000079
According to the calculated normal vector, by formula
Figure BDA00023570719900000710
Obtaining the threshold value alpha of the included angle of the normal vectorthFor use in the calculation of the next part of the algorithm.
(3.3) boundary line extraction:
calculating normal vector of fitting plane by using covariance matrix
Figure BDA00023570719900000711
P for projection planekPoints of a set of points
Figure BDA00023570719900000712
And the normal vector of the plane
Figure BDA00023570719900000713
Represents, so that the point p is queriedq∈PkThe equation of the projection plane is:
A(x-xq)+B(y-yq)+C(z-zq)＝0 (7)
boundary point using query point p on projection planeq＝(xq,yq,zq) And its k neighbor points
Figure BDA00023570719900000714
Angle theta of connecting lineiThe definition is carried out, namely the expression:
Figure BDA00023570719900000715
wherein k is a neighboring point
Figure BDA00023570719900000716
The projection point on the projection surface is pi′＝(xi′,yi′,zi'), η is a parameter; the projection point p can be obtained by the equation (8)iThe coordinates of' are:
Figure BDA00023570719900000717
separately find the query point pqAnd its k neighbor points
Figure BDA0002357071990000081
Angle theta of connecting lineiSo as to determine a set of angles theta arranged from small to large { theta ═ theta1,…,θiAt this point, define query point pqThe condition for being a boundary point is: max (α ═ θ)i+1-θi)≥αth (10)
Where α isthThe maximum threshold angle set for extracting the boundary contour line is set; and finally, combining the boundary points 1 and 2 of the two parts together to obtain the result of the PCB key outline.
Further, the european style clustering algorithm based on the normal vector included angle specifically includes:
a: based on the Euclidean clustering algorithm of normal vector included angle, point cloud data P to be processed is utilized, and P is { P ═ P }1,…,pnAnd (4) creating a kd-tree according to the total number n of point cloud data points, and initializing: number of adjacent points k, minimum cluster number, maximum cluster number, and distance threshold dth。
B: newly building an empty point cloud index vector C for storing a clustering result; newly building a vector Q, initializing the vector Q to be empty, and storing a single clustering result; newly building a pool type vector pr for judging whether a certain point is processed or not, initializing the certain point to false, and indicating that the point is not processed at the current stage;
c: newly creating a point index variable i, initializing the point index variable i to be 0, and aiming at any point pi∈PkThe following steps are carried out:
a) first, p is judgediIf pr is processed, if true, i is i +1, and the execution is continued; if pr ═ false, p is substitutediIndex of points into the current sequence Q [ j ]]In this case, pjI.e. the first seed point pqAnd marks that the point has been clustered into a cluster class, i.e., pr ═ true.
b) According to the sequence Q [ j]Has seed point pqBy using kd-tree to search for a distance pqNearest k points, calculating each k neighbor point
Figure BDA0002357071990000082
And seed point pqAccording to the distance threshold d set in AthUsing the formula (1) as a judgment point
Figure BDA0002357071990000083
Whether the distance limiting condition is met or not, if so, continuing to compare the k neighbor points
Figure BDA0002357071990000084
Normal to
Figure BDA0002357071990000085
And seed point pqNormal to
Figure BDA0002357071990000086
If the following are satisfied:
Figure BDA0002357071990000087
then judge the point
Figure BDA0002357071990000088
And the seed point pqClassifying into the same cluster, and grouping new k neighbor points
Figure BDA0002357071990000089
Is added to Q [ j ]]In this case, the serial number j is j +1 and marked simultaneously
Figure BDA00023570719900000810
The pr of a point is true.
c) Adding newly added point pjAs new seed points pqRepeat b).
d) Until all points in the vector Q have been processed, at which time, according to the equation Min ≦ Q [ size ] ≦ Max
And judging whether the clustering meets the requirement, if so, putting the point index in the Q into the C, emptying the Q, if not, discarding and emptying the Q, and selecting the next seed point, wherein i is i + 1.
Further, the scanning device used for collecting the point cloud data in the step 1 is a FreeScan X5 laser handheld three-dimensional scanner.
Further, the initialization values in step a are as follows: k is 20; min-100; max 100000; dth＝0.5mm。
2.1 straight-through filtering
When point cloud data is acquired, due to influences caused by equipment precision, operator experience, environmental factors and the like, and influences of electromagnetic wave diffraction characteristics and surface properties of a measured object, some noise points inevitably appear in the point cloud data. By observing the actual collected printed circuit board data, the noise points are mainly outliers, i.e. noise points that are free outside the surface of the printed circuit board. Since the point cloud data are directly processed by the method, the outliers can interfere with the later algorithm result, influence the calculation accuracy and even generate wrong results. Therefore, in the process of extracting the key contour line characteristic points of the printed circuit board, the filtering process is required to be used as the first step of the preprocessing. The outliers in the data collected in this document are mainly concentrated in the extending direction of the back surface of the printed circuit board, i.e. filtering needs to be performed for a certain dimension. Therefore, the filtering is convenient by adopting the straight-through filtering method, the points in the specified range can be removed, and the extraction speed and the extraction precision of the key contour line characteristic points in the later period are improved. Fig. 3 shows the visualization result of the Printed Circuit Board (PCB) after the through filtering process.
2.2kd-Tree (k-dimensional tree) topology establishment kd-Tree (short for k-dimensional tree) is a data structure for partitioning k-dimensional data space. The method is mainly applied to searching of multidimensional space key data (such as range searching and nearest neighbor searching). As the point cloud data mainly represents a large number of point sets of the target surface, the redundancy is high, and topological relation information of a two-dimensional image or a traditional entity grid data set is not provided, in order to estimate a normal vector plane, the topological relation between discrete points is established by using a kd-tree, and therefore the fast search based on k adjacent points is realized.
3. Extraction of key contour feature points
3.1 detection of maximum plane
Fig. 4 shows the result of extracting the maximum plane of the Printed Circuit Board (PCB) by using the RANSAC algorithm and performing visualization.
3.2 European clustering based on normal vector included angles
In the field of three-dimensional point cloud, the contour characteristic line mainly comprises a boundary line and a folded edge, wherein the boundary line is mainly a non-connection area line, and the folded edge mainly comprises a line or a fold line formed by connection areas with large curvature change.
In order to convert the folding point into the boundary point, the processing idea is as follows: on the basis of finishing the maximum plane separation of the printed circuit board, according to the Euclidean distance of discrete data points in the space, the point cloud is divided by adopting a point cloud clustering algorithm, and the condition that the data points at different positions are judged to be in the same cluster is as follows: k neighborhood points piTo a query point pqDistance L (p) ofq,pi) Certain threshold conditions are met, which can be described as: l (p)q,pi)≤dth (1)
Wherein the threshold value dthThe setting of (2) is related to the point cloud resolution of the printed circuit board acquired by the experiment. The original resolution of point cloud acquired by equipment used in the method is 0.02 mm, the previous experiment verifies that the cluster set acquired under 10 times of point cloud resolution completely contains all key electronic components on a PCB and still contains partial unnecessary missing edge components, and through multiple experimental tests, the method finally selects a threshold value dthIs 0.5 mm. Meanwhile, in order to enable the same cluster point set to have more similar geometric properties, namely, the boundary point information is more complete, a normal vector included angle is introduced to limit the clustering condition. First, for each query point p, a k-nearest neighbor search (k-NNS) is usedq∈PkAnd performing fast search of k adjacent points. Wherein, Pk＝{p1,…,piFor each pi∈PkDefining a distance diThe expression of (a) is:
Figure BDA0002357071990000101
fitting the plane by least squares, using the constraint d i0, a covariance matrix c ∈ R can be obtained3×3Of the formula
Figure BDA0002357071990000102
In the formula:
Figure BDA0002357071990000103
represents piThe weight of a point, typically 1,
Figure BDA0002357071990000104
Figure BDA0002357071990000105
Figure BDA0002357071990000106
representing the three-dimensional centroid determined by the k neighbor points, and hence:
Figure BDA0002357071990000107
in the formula: lambda [ alpha ]mIs a characteristic value;
Figure BDA0002357071990000108
as a characteristic value λmThe corresponding feature vector. The feature vector corresponding to the minimum feature value obtained from the constraint condition is the query point pq∈PkNormal vector of (c)
Figure BDA0002357071990000109
At this time, according to the normal vector obtained by calculation, the formula
Figure BDA00023570719900001010
Obtaining the threshold value alpha of the included angle of the normal vectorthFor use in the calculation of the next part of the algorithm.
3.3 boundary line extraction
As shown in FIG. 5, the covariance matrix is used to calculate the normal vector of the fitting plane
Figure BDA00023570719900001011
The projection plane can be used for PkPoints of a set of points
Figure BDA00023570719900001012
And the normal vector of the plane
Figure BDA00023570719900001013
And (4) showing. So query point pq∈PkThe equation of the projection plane is:
A(x-xq)+B(y-yq)+C(z-zq)＝0 (7)
the boundary point may be a query point p on the projection planeq＝(xq,yq,zq) And k neighbor points thereof
Figure BDA00023570719900001014
Angle theta of connecting lineiThe definition is carried out, namely the expression:
Figure BDA00023570719900001015
wherein k is a neighboring point
Figure BDA00023570719900001016
The projection point on the projection surface is pi′＝(xi′,yi′,zi') η is a parameter. The projection point p can be obtained by the equation (8)iThe coordinates of' are:
Figure BDA00023570719900001017
separately find the query point pqAnd its k neighbor points
Figure BDA0002357071990000111
Angle of line of (a) thetaiSo as to determine a set of angles theta arranged from small to large { theta ═ theta1,…,θiAt this point, define query point pqThe condition for being a boundary point is: max (α ═ θ)i+1-θi)≥αth (10)
The schematic diagram of the boundary attribute of the point is shown in fig. 6 and 7. Where α isthThe maximum threshold angle set for extracting the boundary contour line is obtained
Figure BDA0002357071990000112
In many cases, the effect is more excellent. Finally, the boundary points of the different parts are merged together, and the result of the final Printed Circuit Board (PCB) key outline is obtained. The key profile features of the Printed Circuit Board (PCB) extracted by the method are shown in fig. 8.
The Euclidean clustering algorithm based on the normal vector included angle condition is described as follows:
a: creating a kd-tree for point cloud data P to be processed, wherein the point cloud data P comprises n data points, namely P ═ { P1,…,pn}. And n is the total number of point cloud data points. Initialization: the number of neighboring points is k; the minimum cluster number is Min; the maximum clustering number is Max; distance threshold value dth. Wherein the number k of adjacent points is determined by a distance threshold dthDetermining that the number k of the adjacent points floats up and down at 20 under the condition that the distance threshold is 0.5 mm, and therefore, determining that the value of the number k of the adjacent points is 20; in order to more effectively set the minimum clustering number Min, loading point cloud data of a Printed Circuit Board (PCB) into visual software in advance, manually selecting an electronic module with the minimum volume on the PCB and reading the selected points, and finally setting the minimum clustering number Min as 100; before the clustering algorithm is carried out, the maximum planes of the printed circuit boards are separated, and the minimum distance between every two of the remaining discrete point cloud clusters to be processed is far larger than the distance threshold value dthTherefore, the maximum clustering number Max is set to be a numerical value larger than the total point number of the point clouds of the printed circuit board, and the maximum clustering number Max is set to be 100000 by the method.
B: newly building an empty point cloud index vector C for storing a clustering result; newly building a vector Q, initializing the vector Q to be empty, and storing a single clustering result; a pool type vector pr is newly established for judging whether a certain point is processed or not, and is initialized to false, which indicates that the point is not processed at the current stage.
C: newly creating a point index variable i, initializing the point index variable i to be 0, and aiming at any point pi∈PkThe following steps are carried out:
a) first, p is judgediWhether or not it has been processed. If pr is true, i is i +1, and step 3 of the pseudo code in table 1 is performed. If pr ═ false, p is substitutediIndex of points into the current sequence Q [ j ]]In this case, pjI.e. the first seed point pqAnd marking that the point is clustered into a cluster class, namely pr ═ true;
b) according to the sequence Q [ j]Has seed point pqBy using kd-tree to search for a distance pqNearest k points, calculating each k neighbor point
Figure BDA0002357071990000113
And the seed point pqAccording to the distance threshold d set in AthUsing the formula (1) to judge the point
Figure BDA0002357071990000114
Whether the distance limit condition is satisfied. If yes, continuing to compare k neighbor points
Figure BDA0002357071990000115
Normal to
Figure BDA0002357071990000116
And seed point pqNormal to
Figure BDA0002357071990000117
If the following are satisfied:
Figure BDA0002357071990000118
Figure BDA0002357071990000119
then judge the point
Figure BDA0002357071990000121
And the seed point pqClassifying into the same cluster, and grouping new k neighbor points
Figure BDA0002357071990000122
Is added to Q [ j ]]In this case, the sequence number j is j +1 and marked at the same time
Figure BDA0002357071990000123
Point pr true;
c) adding newly added point pjAs new seed points pqRepeating b);
d) until all points in vector Q have been processed, at which point the method continues according to the equation Min ≦ Q [ size ] ≦ Max (12)
And judging whether the clustering meets the requirements. And if the seed point is not satisfied, discarding and emptying the Q, and selecting the next seed point, wherein i is i + 1. The pseudo code flow of the above algorithm process is shown in table 1.
TABLE 1 European clustering based on normal vector included angles
Figure BDA0002357071990000124
The foregoing is directed to the preferred embodiment of the present invention and it is noted that the above-mentioned embodiments illustrate rather than limit the invention, and that those skilled in the art will be able to design alternative embodiments without departing from the scope of the appended claims. It will be apparent to those skilled in the art that various modifications and improvements can be made without departing from the spirit and substance of the invention, and these modifications and improvements are also considered to be within the scope of the invention.

Claims (10)

1. A method for extracting a key outline feature of a point cloud of a printed circuit board is characterized by comprising the following steps: scanning and acquiring PCB point cloud data p from multiple angles around a PCB (printed Circuit Board), projecting the point cloud data p onto a plane, and utilizing a query point on the projection plane and k neighbor points thereof
Figure FDA0002357071980000011
Angle theta of connecting lineiDefining boundary points and index variables i, preprocessing PCB point cloud data p by adopting a straight-through filtering method, and establishing a kd-tree topological structure and a point cloud index vector C for storing a clustering result; then, the maximum plane area of the printed circuit board is detected by using a random sample consensus (RANSAC) algorithm, and the distance P is searched by using a kd-treeqNearest k points, performing k neighbor search on each query point, and calculating each k neighbor point
Figure FDA0002357071980000012
And seed point PqConverting the edge folding points into boundary points, extracting key contour features by using a boundary extraction algorithm, and extracting a key contour feature boundary 1 of the point cloud data p of the printed circuit board; on the basis of finishing the maximum plane separation of the printed circuit board, utilizing an Euclidean clustering algorithm based on a normal vector included angle to divide the point cloud by adopting the point cloud clustering algorithm according to the Euclidean distance of discrete data points in the space, separating key features on the space, limiting clustering conditions by introducing the normal vector included angle, extracting a key contour feature boundary 2 of the point cloud data of the printed circuit board, and combining the boundary 1 and the boundary 2; according to the sequence Q [ j]Has seed point PqJudging whether the cluster meets the set distance threshold condition, if so, continuing to compare the k neighbor points
Figure FDA0002357071980000013
Normal to
Figure FDA0002357071980000014
And seed point PqNormal to
Figure FDA0002357071980000015
If the included angle between the data points still meets the condition, the data points at different positions are judged
Figure FDA0002357071980000016
And the seed point PqIs the same cluster class and new k neighbor points are arranged
Figure FDA0002357071980000017
Added to the current sequence Q [ j ]]In the method, the normal vector of the fitting plane is calculated by using the covariance matrix
Figure FDA0002357071980000018
Respectively find out the query points pqAnd its k neighbor points
Figure FDA0002357071980000019
Angle of line of (a) thetaiDetermining a sequence of angles theta from small to large as { theta ═ theta1,…,θiAnd merging the boundary points of different parts together, and extracting the point cloud key outline characteristics of the printed circuit board to obtain the result of the PCB key outline.
2. The method for extracting the key outline features of the point cloud of the printed circuit board as claimed in claim 1, wherein the condition for judging that the data points at different positions are in the same cluster is as follows: k neighborhood points piTo a query point pqDistance L (p) ofq,pi) Satisfies a certain threshold dthThe conditions are as follows: l (p)q,pi)≤dth。
3. The method of extracting key outline features of point cloud of printed circuit board as claimed in claim 1, wherein k nearest neighbor search is used to search for each query point pq∈PkPerforming fast search of k adjacent points; for each query point pi∈PkDistance diComprises the following steps:
Figure FDA00023570719800000110
Pk＝{p1,…,pi}。
4. the method of extracting key outline features of point cloud of printed circuit board as claimed in claim 1, wherein fitting the plane by least squares method, using constraint di0, a covariance matrix c ∈ R is obtained3×3The expression is
Figure FDA00023570719800000111
Three-dimensional centroid determined by k neighbor points
Figure FDA00023570719800000217
Figure FDA0002357071980000021
Further, it is possible to obtain:
Figure FDA0002357071980000022
the feature vector corresponding to the minimum feature value obtained from the constraint condition is the query point pq∈PkNormal vector of (c)
Figure FDA0002357071980000023
Wherein R is3×3A 3 x 3 matrix is represented,
Figure FDA0002357071980000024
represents piThe weight of the point, T represents the transpose operation of the matrix,
Figure FDA0002357071980000025
as a characteristic value λmCorresponding feature vector, λmIs the eigenvalue.
5. The method for extracting key outline features of point cloud of printed circuit board as claimed in claim 1, wherein the threshold value alpha of included angle of normal vector is obtained according to the formula of normal vector obtained by calculationth，
Figure FDA0002357071980000026
Wherein the content of the first and second substances,
Figure FDA0002357071980000027
representing the normal vector at the query point q,
Figure FDA0002357071980000028
representing the normal vector at the ith k neighbor point.
6. The method of claim 1, wherein the projection plane is PkPoints of the point set
Figure FDA0002357071980000029
Sum plane normal vector
Figure FDA00023570719800000210
Representing by computing the normal vector of the fitted plane using the covariance matrix
Figure FDA00023570719800000211
Get the query point pq∈PkEquation of projection plane (c): a (x-x)q)+B(y-yq)+C(z-zq)＝0
In the formula, A, B, C represents three coefficients of a plane equation, x, y and z represent x, y and z coordinates of the point cloud data in space, and x representsq、yq、zqRepresenting the x, y, z coordinates of the query point q.
7. The method for extracting key outline features of point cloud of printed circuit board as claimed in claim 1 or 6, wherein the boundary points utilize query points p on the projection planeq＝(xq,yq,zq) And its k neighbor points
Figure FDA00023570719800000212
Angle theta of connecting lineiDefine according to k neighbor points
Figure FDA00023570719800000213
Projected point p 'on the projection plane'i＝(x′i,y′i,z′i) Parameter η, by expression:
Figure FDA00023570719800000214
obtaining a projected point p'iThe coordinates of (a) are:
Figure FDA00023570719800000215
separately find the query point pqAnd its k neighbor points
Figure FDA00023570719800000216
Angle of line of (a) thetaiDetermining a set of angles theta from small to large as { theta ═ theta1,…,θiAnd the maximum threshold angle alpha set for boundary contour extractionthAt this point, a query point p is definedqThe condition for being a boundary point is: max (α ═ θ)i+1-θi)≥αthAnd finally, combining the boundary points of different parts together to obtain the result of the PCB key outline.
8. The method for extracting key outline features of point cloud of printed circuit board according to claim 1, wherein the Euclidean clustering algorithm based on normal vector included angle utilizes point cloud data P to be processed, P ═ P1,…,pnAnd b, establishing a kd-tree according to the total number n of point cloud data points, and initializing: number of neighboring points k, minimum number of clusters, maximum number of clusters, and distance threshold dth。
9. The method for extracting key outline features of point cloud of printed circuit board according to claim 1, wherein a null point cloud index vector C is newly created for storing clustering results; newly building a vector Q, initializing the vector Q to be empty, and storing a single clustering result; a Boolean type variable pool type vector pr is newly established for judging whether a certain point is processed or not, and is initialized to false, which indicates that the point is not processed at the current stage.
10. The method for extracting key outline features of point cloud of printed circuit board as claimed in claim 1, wherein, a point index variable i is newly created, i is initialized to 0, and for any point pi∈PkThe following steps are carried out:
a) first, p is judgediIf the vector pr of the bool type is processed, if the vector pr of the bool type is true, i is i +1, and the execution is continued; if pr ═ false, p is substitutediIndexing of points into the current sequence Q [ j ]]In this case, pjI.e. the first seed point pqAnd marking that the point is clustered into a cluster class, namely pr ═ true;
b) according to the sequence Q [ j]Has seed point pqBy using kd-tree to search for a distance pqNearest k points, calculating each k neighbor point
Figure FDA0002357071980000031
And the seed point pqAccording to a set distance threshold dthUsing L (p)q,pi)≤dthJudgment point
Figure FDA0002357071980000032
Whether the distance limiting condition is met or not, if so, continuing to compare the k neighbor points
Figure FDA0002357071980000033
Normal to
Figure FDA0002357071980000034
And seed point pqNormal to
Figure FDA0002357071980000035
If the included angle between the two meets the following conditions:
Figure FDA0002357071980000036
then judge the point
Figure FDA0002357071980000037
And seed point pqClassifying into the same cluster, and grouping new k neighbor points
Figure FDA0002357071980000038
Is added to Q [ j ]]In this case, the sequence number j is j +1 and marked at the same time
Figure FDA0002357071980000039
Point pr true;
c) adding newly added point pjAs new seed points pqRepeating b);
d) and until all the points in the vector Q are processed, judging whether the clustering meets the requirement according to the condition that Min is less than or equal to Q [ size ] and less than or equal to Max, if so, placing the point index in Q into C, emptying Q, if not, discarding and emptying Q, and if i is i +1, and selecting the next seed point.
CN202010010772.9A 2020-01-06 2020-01-06 Method for extracting point cloud key outline features of printed circuit board Active CN111222516B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010010772.9A CN111222516B (en) 2020-01-06 2020-01-06 Method for extracting point cloud key outline features of printed circuit board

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010010772.9A CN111222516B (en) 2020-01-06 2020-01-06 Method for extracting point cloud key outline features of printed circuit board

Publications (2)

Publication Number Publication Date
CN111222516A CN111222516A (en) 2020-06-02
CN111222516B true CN111222516B (en) 2022-07-08

Family

ID=70828103

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010010772.9A Active CN111222516B (en) 2020-01-06 2020-01-06 Method for extracting point cloud key outline features of printed circuit board

Country Status (1)

Country Link
CN (1) CN111222516B (en)

Families Citing this family (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111815611B (en) * 2020-07-14 2024-03-22 南京航空航天大学 Round hole feature extraction method for rivet hole measurement point cloud data
CN112529891B (en) * 2020-12-21 2024-03-19 深圳辰视智能科技有限公司 Method and device for identifying hollow holes and detecting contours based on point cloud and storage medium
CN112489207B (en) * 2021-02-07 2021-07-13 深圳大学 Space-constrained dense matching point cloud plane element extraction method
CN112990037A (en) * 2021-03-24 2021-06-18 上海慧姿化妆品有限公司 Method and system for extracting nail outline
CN112949557A (en) * 2021-03-24 2021-06-11 上海慧姿化妆品有限公司 Method and system for extracting nail outline
CN113516695B (en) * 2021-05-25 2023-08-08 中国计量大学 Point cloud registration strategy in laser profiler flatness measurement
CN113963010A (en) * 2021-07-27 2022-01-21 成都睿铂科技有限责任公司 Object contour line extraction method and system
CN113888634B (en) * 2021-09-28 2022-08-23 湖北瑞兴达特种装备科技股份有限公司 Method and system for positioning and detecting target on special-shaped complex surface
CN114511586B (en) * 2022-04-20 2022-08-23 三一筑工科技股份有限公司 Method, device and equipment for determining surface contour of object and storage medium
CN114818591B (en) * 2022-07-01 2022-09-09 南京维拓科技股份有限公司 Method for quickly generating clearance of tool device
CN115082547B (en) * 2022-07-27 2022-11-15 深圳市华汉伟业科技有限公司 Profile measuring method based on point cloud data and storage medium
CN115601565B (en) * 2022-12-15 2023-03-31 安徽大学 Large-span steel structure fixed feature extraction method based on minimum valley distance
CN116912312B (en) * 2023-09-15 2023-12-01 湖南大学 Three-dimensional hole positioning method for complex curved surface component

Non-Patent Citations (6)

* Cited by examiner, † Cited by third party
Title
一种点云特征线提取方法;任前程;《激光与光电子学进展》;20181019(第06期);229-235 *
利用邻近点几何特征实现建筑物点云特征提取;董伟;《激光与光电子学进展》;20180116(第07期);181-188 *
基于3维点云欧氏聚类和RANSAC边界拟合的目标物体尺寸和方位识别;薛连杰等;《机械设计与研究》;20181020(第05期);52-56+61 *
基于激光三维点云的机械工件识别方法;薛珊等;《红外与激光工程》;20190115(第04期);169-176 *
融合改进场力和判定准则的点云特征规则化;刘庆等;《中国激光》;20190125(第04期);200-209 *
采用点云重心距离进行边界检测的点云数据配准;王勇等;《小型微型计算机系统》;20150915(第09期);178-183 *

Also Published As

Publication number Publication date
CN111222516A (en) 2020-06-02

Similar Documents

Publication Publication Date Title
CN111222516B (en) Method for extracting point cloud key outline features of printed circuit board
CN110443836B (en) Point cloud data automatic registration method and device based on plane features
CN103729643B (en) The identification of three dimensional object in multi-mode scene and posture are determined
JP5677798B2 (en) 3D object recognition and position and orientation determination method in 3D scene
CN109887015A (en) A kind of point cloud autoegistration method based on local surface feature histogram
CN110599506B (en) Point cloud segmentation method for three-dimensional measurement of complex special-shaped curved surface robot
CN103727930A (en) Edge-matching-based relative pose calibration method of laser range finder and camera
CN108830888B (en) Coarse matching method based on improved multi-scale covariance matrix characteristic descriptor
CN110930456A (en) Three-dimensional identification and positioning method of sheet metal part based on PCL point cloud library
CN111179321A (en) Point cloud registration method based on template matching
CN110009745B (en) Method for extracting plane from point cloud according to plane element and model drive
CN113093216A (en) Irregular object measurement method based on laser radar and camera fusion
CN109766903B (en) Point cloud model curved surface matching method based on curved surface features
CN116758126A (en) Quick point cloud registration method based on mismatching elimination of similar triangles
CN114463396B (en) Point cloud registration method utilizing plane shape and topological graph voting
Lu et al. Point cloud registration algorithm fusing of super 4pcs and icp based on the key points
Zhang et al. Splitting and merging based multi-model fitting for point cloud segmentation
CN117132630A (en) Point cloud registration method based on second-order spatial compatibility measurement
Yuan et al. 3D point cloud recognition of substation equipment based on plane detection
Zhao et al. Curvature-based registration and segmentation for multisensor coordinate metrology
CN109829502B (en) Image pair efficient dense matching method facing repeated textures and non-rigid deformation
Song Local voxelizer: A shape descriptor for surface registration
CN116310355A (en) Laser point cloud denoising and defect detection method for complex structural member
CN111127667A (en) Point cloud initial registration method based on region curvature binary descriptor
CN116385356A (en) Method and system for extracting regular hexagonal hole features based on laser vision

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