CN112085822A - Point cloud model descriptor construction method and point cloud model matching method - Google Patents

Point cloud model descriptor construction method and point cloud model matching method Download PDF

Info

Publication number
CN112085822A
CN112085822A CN202010885849.7A CN202010885849A CN112085822A CN 112085822 A CN112085822 A CN 112085822A CN 202010885849 A CN202010885849 A CN 202010885849A CN 112085822 A CN112085822 A CN 112085822A
Authority
CN
China
Prior art keywords
point cloud
layer
point
models
vector
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.)
Granted
Application number
CN202010885849.7A
Other languages
Chinese (zh)
Other versions
CN112085822B (en
Inventor
刘喆
李纯辉
张雨禾
耿国华
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.)
Northwestern University
Original Assignee
Northwestern University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Northwestern University filed Critical Northwestern University
Priority to CN202010885849.7A priority Critical patent/CN112085822B/en
Publication of CN112085822A publication Critical patent/CN112085822A/en
Application granted granted Critical
Publication of CN112085822B publication Critical patent/CN112085822B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T15/00—3D [Three Dimensional] image rendering
    • G06T15/005—General purpose rendering architectures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects

Abstract

The invention discloses a point cloud model descriptor construction method and a point cloud model matching method. The disclosed solution includes first layering the model; then, calculating the center distance of each layer, and constructing a layered center distance matrix of the model; then, obtaining a Fourier coefficient matrix of each model by using discrete Fourier transform; and finally, calculating the similarity of the two Fourier coefficient matrixes, and judging a matching result according to a threshold value. The descriptor of the invention has rotational-translational scaling invariance, is sensitive to the details of the model, and can directly determine the similarity of the model compared with other traditional descriptors.

Description

Point cloud model descriptor construction method and point cloud model matching method
Technical Field
The invention relates to a three-dimensional model processing technology, in particular to a point cloud shape description and three-dimensional model matching method based on a hierarchical center descriptor.
Background
The identification and matching of the three-dimensional model (point cloud model) refers to identifying the heterogeneity of two objects under a certain threshold range by collecting and integrating the characteristics of the model and using a corresponding algorithm. Recognition mainly includes two main tasks of shape description and shape similarity measurement:
first, the shape description refers to how to make the feature information of the model relatively complete, and at the same time, the shape descriptor has scale invariance to translation, rotation and scaling. The feature information of the same three-dimensional object is complex, needs to be considered from a certain angle, and also needs to be controlled for feature dimension, the higher the dimension information is, although the included features are more, the processing is more difficult, and the high-dimensional feature is only part of useful information, and the high-dimensional feature extraction also generates information redundancy;
the similarity measurement of the shape needs to be characterized according to the characteristics of the three-dimensional model, and the matching relation of the three-dimensional model is determined based on the similarity measurement of the shape.
Some of the existing three-dimensional model matching methods are considered from local information, and the global characteristics of the model are lost; or local detail parts are not depicted in place in consideration of global information; still other prior art projects a three-dimensional model onto a two-dimensional plane for processing, which may lose depth information of the three-dimensional model.
Disclosure of Invention
Aiming at the defects or shortcomings of the prior art, the invention firstly provides a point cloud model descriptor construction method.
Therefore, the point cloud model descriptor construction method provided by the invention comprises the following steps:
step 1, dividing a point cloud model into M layers along a fixed direction, wherein M is more than or equal to 2;
step 2, determining the center point of each layer, and calculating the center distance between each point cloud in each layer and the center point of the layer to which the point cloud belongs, wherein the center distances corresponding to all the point clouds in each layer form the center distance vector of the corresponding layer;
step 3, carrying out normalization processing on the central distance vector of each layer;
step 4, constructing a hierarchical central distance matrix of M rows by N columns, and taking the numerical value N with the maximum element number in all central matrix vectorsmaxThe row vector of the hierarchical center distance matrix is the center distance vector of each layer, and the center distance vector with the vector element number smaller than N is filled to N by taking 0 as an elementmaxAll the central distance vector rows are arranged in the hierarchical order in the step 1; constructed byThe hierarchical center distance matrix of (2) is a descriptor of the point cloud model.
Preferably, the central distance vector of each layer is normalized by local normalization.
Preferably, each layer uses the maximum value of the corresponding center distances of all point clouds in the layer as a normalization criterion.
Further, the invention provides a point cloud model matching method. Therefore, the point cloud model matching method provided by the invention comprises the following steps:
calibrating two point cloud models to be matched to enable outline orientations of the two point cloud models to be consistent;
step two, respectively carrying out the processing of the method of claim 1,2 or 3 on the two point cloud models along a fixed direction on the outline azimuth, and respectively constructing descriptors of the two point cloud models, wherein the layering layers of the two point cloud models are the same, the column numbers of the two descriptors are the same, and the element number of the vector with the maximum vector element number in all the central distance vectors of the two point cloud models is taken;
step three, respectively acquiring discrete Fourier coefficient matrixes of the two descriptors;
and step four, calculating the similarity of the two discrete Fourier coefficient matrixes, and matching the two models when the obtained similarity value is smaller than or equal to a threshold value, wherein the value range of the threshold value is 0-1.
Preferably, the calibration is four-point calibration.
Further, the two point cloud models are simplified before the step one, so that the point cloud number of the two point cloud models is in the same order of magnitude.
The method takes global shape information and local characteristic information of the three-dimensional model into consideration, and has the effect of integrating global information; in addition, the method uses the mean value of the Fourier coefficient difference as the criterion of similarity measurement, can accurately convert the similarity of two models into a numerical value, and judges the clustering result through a threshold value.
Drawings
FIG. 1 is a schematic diagram of a four-point calibration method and a layering process of a point cloud model according to the present invention;
fig. 2 is a schematic diagram of a matching object in an embodiment.
Detailed Description
Unless otherwise indicated, the terms described herein are to be understood in accordance with their ordinary knowledge in the art.
The point cloud model of the present invention is a concept conventionally understood in the art, and is a common three-dimensional model data format, for example, three-dimensional models with suffix of. The feature points on the point cloud model generally refer to points with large-scale curved surface gradients, such as common points with large curvature, and can be obtained by using a feature point extraction method. The corresponding characteristic points on the two point cloud models mean that the curvatures of the two characteristic points on the two models are close to each other and are located at the same or close positions in the overall contour of each model when viewed from the overall appearance. For example, the outlines of the two point cloud models to be matched as shown in fig. 1 are in the shape of a circular cap, and the feature points corresponding to the highest points on the two point cloud models are corresponding feature points. The contour orientations are consistent, that is, the height and width directions of the two point cloud models are consistent, or the axial direction and the radial direction are consistent, or the angle directions of the corresponding feature points or the relative position relations between the corresponding feature points are consistent, for example, the contours of the two point cloud models to be matched are in a circular hat shape as shown in fig. 1, the highest points of the outer contours of the two point cloud models are located at the tops, the contour planes are parallel to the horizontal plane and located at the bottoms, that is, the contour directions of the two point cloud models are consistent, of course, the highest points of the outer contours of the two point cloud models can also be located at the bottoms, and the contour planes are parallel to the horizontal plane and located at the tops, so that the two point cloud. The fixed direction along the outline azimuth refers to the height, width, axial direction, radial direction or other directions which can represent the shape characteristics of the point cloud models according to the height, width, axial direction, radial direction or other consistent outline azimuth of the two point cloud models.
Because of the change of the scaling scale, the model generates inconsistent descriptors due to different scaling scales, in order to overcome the defect, the normalization processing is carried out on the matrix descriptor, the normalization processing of the invention can adopt the existing normalization method, preferably a local normalization method, the local normalization has the advantage of non-interference among layers, and the value of each vector element in the normalized vector is in the interval of [0,1 ]. Further preferably, the normalization criterion of each layer is the maximum value of the corresponding central distances of all point clouds in the layer.
The central distance matrix of the invention is obtained by integrating the central distance vectors of all layers according to the layering sequence, because the sampling density of scattered point clouds is not uniform, and a given point cloud model possibly contains different point scales, the quantity of the point clouds in each layer of model obtained by layering is different, so that the matrixes are not aligned, in order to solve the problem, the method of the invention uses a 0 filling method, namely: setting the number of center distance vector elements of each layer as the number N of elements of the vector with the largest vector element in the center distance vectors of all layersmaxThe spare part of the shortage layer is replaced by 0 distance until the number of elements is NmaxThe point having the padding distance value of 0 indicates that this point is an imaginary point.
In order to enable the obtained hierarchical descriptor to depict the classification result of the model, the discrete Fourier transform is carried out on the obtained descriptor, time domain signals are converted into frequency information through the discrete Fourier transform, corresponding descriptors with the same distance are placed in the same frequency domain information through the hierarchical center descriptor adopted by the invention, a discrete Fourier coefficient matrix is further formed, and the condition that the sliced surface point cloud is far away from the slice center can be better depicted. The method for acquiring the discrete Fourier coefficient matrix adopts the existing discrete Fourier transform method.
The value range of the threshold value is 0-1, and the specific numerical value is related to the shape and the difference (similarity) of the point cloud model and can be determined according to experiments.
The simplifying processing of the invention means that unnecessary points in the point cloud model are removed, the operation can be carried out on the existing tool software, and the point cloud numbers of the two point cloud models are in the same order of magnitude, thereby ensuring the reliability of the matching result. The same order is as thousands, for example 1000 point clouds and 9000 point clouds.
Example (b):
the point cloud model descriptor construction method of the embodiment is as follows:
firstly, layering a point cloud model;
then, calculating the center of each layer of point cloud, and dividing the model S into M layers, wherein the model of the M layer has NmPoint, M ═ 1,2,. ·, M; pmn＝(Xmn,Ymn,Zmn),n＝1,2,...,NmRepresenting all point sets of the m-th layer, and calculating the centers C of all points in the point setsm＝(Xcm,Ycm,Zcm)：
Figure BDA0002655549390000051
Then, calculating the center distance of each point cloud in each layer to obtain the center distance vector of each layer, wherein all the center distance vectors form the center distance vector describing the layer;
Distm′＝(Distm1′,Distm2′,...,Distmn′,...,DistmNm′)
wherein Distm' is the center distance vector of the mth layer,
Figure BDA0002655549390000061
represents the center distance of the nth point in the mth layer from the center of the layer;
then, carrying out normalization processing on each central distance vector by adopting local normalization to obtain a normalized central distance vector;
Figure BDA0002655549390000062
wherein DistmIs the normalized center distance vector of the mth layer,
Figure BDA0002655549390000063
represents the largest center in the m-th layerA distance;
then, the central distance vectors of each layer are integrated into a matrix according to the hierarchical sequence, and the number of elements is less than NmaxIs filled with 0, NmaxTaking the element number of the vector with the maximum vector elements in the central distance vectors of all the layers to form a descriptor SCs (S) of the model S;
Figure BDA0002655549390000064
wherein each row of SCs (S) is the center distance vector, Dist, of each layermnIs the center distance of the nth point in the mth layer from the center of the mth layer, M is 1,2max。
The point cloud model matching method of the embodiment is as follows:
calibrating four point clouds of a point cloud model to be matched by adopting a four-point calibration method shown in FIG. 1, selecting a plane which cannot be positioned at the same position from the four point clouds, and then layering the four point clouds along the direction of the triangular pyramid height determined by the four point clouds;
the method is adopted to obtain the descriptors of the point cloud models to be matched, and the layering directions of the two point cloud models are consistent in the process; the layering layers of the two point cloud models are the same; simultaneously, the column numbers of the two descriptors are the same, and the element number value of the vector with the maximum vector element number in all the central distance vectors of the two point cloud models is taken;
and further transforming each descriptor by adopting discrete Fourier transform to obtain a discrete Fourier matrix. Taking the point cloud model S as an example, a fourier matrix f (S) with a size of U rows × V columns is obtained, where U is M, V is Nmax:
Figure BDA0002655549390000071
Figure BDA0002655549390000072
wherein j2＝-1， m＝1,2,...,M；n＝1,2,...,Nmax；u＝1,2,...,U；v＝1,2,...,V；
Performing integral transformation on each element in the discrete Fourier matrix to obtain a discrete Fourier coefficient matrix;
then calculating the similarity of the discrete Fourier coefficient matrixes of the two models, and matching the two models when the obtained similarity is less than or equal to a threshold value; that is, the larger the similarity value is, the lower the similarity between the two models is, and conversely, the smaller the similarity value is, the higher the similarity between the two models is. Taking A, B two models as an example, assume that the size of the discrete fourier coefficient matrix of the two models is also U rows by V columns:
Figure BDA0002655549390000073
FA(u,v)、FB(u, v) represent the elements of the discrete Fourier matrices of the A, B two models, abs (F), respectivelyA(u,v))、abs(FB(u, v)) elements representing A, B matrices of discrete Fourier coefficients for the two models, respectively, each FA(u,v)、FB(u, v) discrete Fourier coefficients, Diff (A, B) representing A, B similarity of the two models.
Selecting 3 layers, 5 layers and 8 layers, and matching a plurality of groups of Terra cotta warriors and terracotta cotta by the method, as shown in figure 2, wherein symbols in the figure are numbers of each group and in the group. As shown in table 1, wherein Diff-3, Diff-5, and Diff-8 refer to the similarity of 3, 5, and 8 slices, respectively; meanwhile, individual groups in the groups are selected to be matched by adopting the non-method of the invention, wherein the layering layers are 12 layers and 16 layers, and the result is shown in table 2, wherein Diff-12 and Diff-16 respectively refer to the similarity of the slices of 12 layers and 16 layers.
TABLE 1 comparison of diff-3, diff-5, diff-8 results
Figure BDA0002655549390000081
Figure BDA0002655549390000091
TABLE 2 diff-8diff-12diff-16 results comparison
Figure BDA0002655549390000092
And comprehensively comparing, wherein 0.3 can be selected as a threshold value for similarity judgment of the multiple groups of models, and the accuracy is up to 90%.

Claims (6)

1. A point cloud model descriptor construction method is characterized by comprising the following steps:
step 1, dividing a point cloud model into M layers along a fixed direction, wherein M is more than or equal to 2;
step 2, determining the center point of each layer, and calculating the center distance between each point cloud in each layer and the center point of the layer to which the point cloud belongs, wherein the center distances corresponding to all the point clouds in each layer form the center distance vector of the corresponding layer;
step 3, carrying out normalization processing on the central distance vector of each layer;
step 4, constructing a hierarchical central distance matrix of M rows by N columns, and taking the numerical value N with the maximum element number in all central matrix vectorsmaxThe row vector of the hierarchical center distance matrix is the center distance vector of each layer, and the center distance vector with the vector element number smaller than N is filled to N by taking 0 as an elementmaxAll the central distance vector rows are arranged in the hierarchical order in the step 1; the constructed hierarchical center distance matrix is a descriptor of the point cloud model.
2. The method of constructing point cloud model descriptors of claim 1, wherein the normalization process is performed on the center distance vector of each layer by using local normalization.
3. The method of constructing point cloud model descriptors of claim 2, wherein each layer uses the maximum value among the distances of corresponding centers of all point clouds in the layer as a normalization criterion.
4. A point cloud model matching method is characterized by comprising the following steps:
calibrating two point cloud models to be matched to enable outline orientations of the two point cloud models to be consistent;
step two, respectively carrying out the processing of the method of claim 1,2 or 3 on the two point cloud models along a fixed direction on the outline azimuth, and respectively constructing descriptors of the two point cloud models, wherein the layering layers of the two point cloud models are the same, the column numbers of the two descriptors are the same, and the element number of the vector with the maximum vector element number in all the central distance vectors of the two point cloud models is taken;
step three, respectively acquiring discrete Fourier coefficient matrixes of the two descriptors;
and step four, calculating the similarity of the two discrete Fourier coefficient matrixes, and matching the two models when the obtained similarity value is less than or equal to a threshold value, wherein the value range of the threshold value is 0-1.
5. The point cloud model matching method of claim 4, wherein said calibration employs a four-point calibration.
6. The point cloud model matching method of claim 4, wherein the simplification process for two point cloud models before the first step is performed so that the number of point clouds of the two point cloud models is of the same order of magnitude.
CN202010885849.7A 2020-08-28 2020-08-28 Point cloud model descriptor construction method and point cloud model matching method Active CN112085822B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010885849.7A CN112085822B (en) 2020-08-28 2020-08-28 Point cloud model descriptor construction method and point cloud model matching method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010885849.7A CN112085822B (en) 2020-08-28 2020-08-28 Point cloud model descriptor construction method and point cloud model matching method

Publications (2)

Publication Number Publication Date
CN112085822A true CN112085822A (en) 2020-12-15
CN112085822B CN112085822B (en) 2024-04-12

Family

ID=73728851

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010885849.7A Active CN112085822B (en) 2020-08-28 2020-08-28 Point cloud model descriptor construction method and point cloud model matching method

Country Status (1)

Country Link
CN (1) CN112085822B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115170728A (en) * 2022-06-07 2022-10-11 青海师范大学 Similarity measurement method of three-dimensional point cloud model

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120206438A1 (en) * 2011-02-14 2012-08-16 Fatih Porikli Method for Representing Objects with Concentric Ring Signature Descriptors for Detecting 3D Objects in Range Images
US20150003723A1 (en) * 2013-06-27 2015-01-01 Chevron U.S.A. Inc. System and method of detecting objects in scene point cloud
CN109242774A (en) * 2018-08-29 2019-01-18 浙江大学 A kind of flat-type part point cloud method based on hyperspace invariant features

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120206438A1 (en) * 2011-02-14 2012-08-16 Fatih Porikli Method for Representing Objects with Concentric Ring Signature Descriptors for Detecting 3D Objects in Range Images
US20150003723A1 (en) * 2013-06-27 2015-01-01 Chevron U.S.A. Inc. System and method of detecting objects in scene point cloud
CN109242774A (en) * 2018-08-29 2019-01-18 浙江大学 A kind of flat-type part point cloud method based on hyperspace invariant features

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
吴绍根;聂为清;路利军;刘娅琴;: "形状的圆内距离变换", 中国图象图形学报, no. 01 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115170728A (en) * 2022-06-07 2022-10-11 青海师范大学 Similarity measurement method of three-dimensional point cloud model

Also Published As

Publication number Publication date
CN112085822B (en) 2024-04-12

Similar Documents

Publication Publication Date Title
CN107742102B (en) Gesture recognition method based on depth sensor
CN105184250B (en) A kind of terrain classification method of electric power corridor airborne LiDAR point cloud data
CN111696210A (en) Point cloud reconstruction method and system based on three-dimensional point cloud data characteristic lightweight
CN109493372B (en) Rapid global optimization registration method for product point cloud data with large data volume and few characteristics
CN105844600B (en) A kind of extraterrestrial target three-dimensional point cloud fairing denoising method
Li et al. A leaf segmentation and phenotypic feature extraction framework for multiview stereo plant point clouds
CN107368807A (en) A kind of monitor video vehicle type classification method of view-based access control model bag of words
CN113643270B (en) Image registration method and device based on point cloud data
Qian et al. Calculation method of surface shape feature of rice seed based on point cloud
CN105654122B (en) Based on the matched spatial pyramid object identification method of kernel function
CN112837309B (en) Fruit tree canopy target recognition device, method, computing equipment and storage medium
CN112669385A (en) Industrial robot workpiece identification and pose estimation method based on three-dimensional point cloud characteristics
CN111914112A (en) Part CAD model reusing method based on point cloud classification network
CN114299150A (en) Depth 6D pose estimation network model and workpiece pose estimation method
CN112085822A (en) Point cloud model descriptor construction method and point cloud model matching method
CN113961738A (en) Multi-feature casting three-dimensional model retrieval method and device
CN114663373A (en) Point cloud registration method and device for detecting surface quality of part
CN109598261B (en) Three-dimensional face recognition method based on region segmentation
CN116763295B (en) Livestock scale measuring method, electronic equipment and storage medium
CN111914480B (en) Processing feature intelligent recognition method based on point cloud semantic segmentation
CN113744389B (en) Point cloud simplifying method for complex part curved surface feature preservation
CN115131424A (en) Clustering and boundary feature-preserving automobile trunk point cloud volume calculation method
CN115830587A (en) Structural surface rapid automatic identification method based on high-precision point cloud data
CN110781963B (en) K-means clustering-based aerial target clustering method
CN107194998A (en) A kind of method of multi-layer three-dimension point cloud single-layered

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