CN101021954A - Three-dimensional scanning point cloud compressing method - Google Patents

Abstract

This invention relates to a method for reducing dot clouds of 3-D scanning, which carries out grid division to the dot-clouds, and the grid is the smallest exterior square body of a dot-cloud, the side length is L, then divides it into sub-grids with the side length of DalphaL3/n along the directions of three coordinate axises, and alpha is the dilution ratio decided by the design, n is the dot number of the cloud, then dot set of the neighbour domain of Pi in the adjacent sub-grid of each dot KNB9Pi)={X1, X2apostrophe, Xn} is taken,takes the least square approximation plane of KNB9Pi) as the micro-cut plane S(Pi), takes Pi as the core of KNB(Pi), sets the distance from jth adjacent point of Pi to S(Pi) dj, and the distance to Pi lambdaj, then the function is fj(Pi)=dj/lambdaj, the curvature function of Pi is f(pi)=1/kSigmafj(Pi) so as to obtain the curvature of every data point Pi, finally acquires the mean curvature in every small grid and gets the difference of the curvature of a point and the mean curvature and points smaller than the needed accuracy are selected as the points of the grid and others are deleted.

Description

The point cloud compressing method of 3-D scanning

Technical field

The present invention relates to a kind of method of the point cloud compressing to 3-D scanning, relate in particular to the method for a kind of 3 d grid method based on curvature point cloud compressing.

Background technology

The precision of 3D scanning device is more and more higher, and the acquisition speed of data sharply increases, and adopts contactless some cloud mapping method acquisition point speed of movement of cloud degree very fast, measurement data points is many, per second can produce thousands of data points, can clearly express object, has obtained application more and more widely.Though the non-contact optical scanner can obtain the some cloud of very dense, not all data point can both be used for surface reconstruction.These redundant cloud datas can cause computer run, the poor efficiency of storage and operation, generate surface model and need consume more time, the robustness variation of programmed algorithm, thereby its use also has been subjected to challenge, and too intensive some cloud can influence the fairness of reconstruct curved surface, and fairing is very important requirement in the product design design, simultaneously by repeatedly splicing, the closeness of some cloud uprises, and overlay region especially is for fear of the problems referred to above, just need the deletion data point, promptly point cloud data is simplified processing.Therefore, measurement data is simplified help to improve modeling efficiency and modeling quality.The present invention mainly relates to a kind of method that the product point cloud model that 3 D scanning system (Fig. 1) obtains is simplified.

People have done many researchs about the point cloud data compressing method in recent years, it mainly is the method that adopts bounding box that traditional cloud data is simplified, this method adopts bounding box to come the obligatory point cloud, then big bounding box is resolved into several evenly little bounding boxs of size, the point of choosing the most close bounding box center in each bounding box replaces the point in the whole bounding box.This method is simple, and is efficient, but owing to the size of bounding box is set arbitrarily by the user, therefore can't guarantee constructed model and the precision between the original point cloud data.Calendar year 2001, Sun W, Bradley etc. determine the size of bounding box automatically by using local surface interpolation, improve this method, but not enough be this method can only be applied to simple curved surface data and efficiency ratio lower, applicability is not high.Based on binary tensor product Haar wavelet decomposition, the curved surface data that instrument error drives is simplified algorithm.This algorithm is driven by error, need not prior specific data and count, can directly simplify the measurement data of unknown surface equation, but this algorithm does not have the self-organization characteristic, can't handle inter characteristic points, the data of simplifying of last gained also are some scattered data being but not regular grid data.A kind of usefulness uniform grid (uniform grid) reduces the method for data, is widely used in the medium filtering of image processing process.Its principle is: at first set up a series of uniform lattices on the plane perpendicular to the direction of scanning, the point that each scanning obtains all is assigned to some grids, calculate the distance of this point to grid, all are assigned to the data point of same grid by the big minispread of distance, get the data point of representing all to be allocated in this grid apart from the data point that is arranged in intermediate value, other points are then deleted.See Fig. 2, the E point is selected, and all the other A, B, C, D, F, G point are then deleted.

The uniform grid method that proposes with people such as Martin is the same, Lee also adopts the uniform grid perpendicular to direction of scanning (Z axle) to come to extract from cloud data a little, this method at first create one identical and by size perpendicular to the grid plan of direction of scanning, the size of size of mesh opening has determined the size of data compaction ratio, the size of grid is more little, and many more sampled points are just arranged, after grid plan is created, to grid plan, each grid all is assigned to respective point with all spot projections.Distance according to the grid plan of having a few in each grid sorts to these points, the point that chosen position mediates.If the point in the same grid has n, then when n is odd number, will choose (n+1)/2 point; When n is even number, will choose a n/2 point or (n+2)/2 point.Fujimoto and Kariye thought that googol has produced a lot of problems according to amount to use in the subsequent manufacturing processes, proposed a kind of improvement alphabetic data compressing method towards 2D digitised points cloud in 1993.This method has guaranteed that the error range of simplifying data is within the given angle and distance tolerance.Thereby Chen and Ng have proposed a kind of method that reduces triangle gridding number deletion data point, the direct triangle gridding of the point cloud data that scanning is obtained at first, generate stl file, thereafter the contiguous tri patch normal vector of place, comparand strong point tri patch, according to a kind of vectorial weighting algorithm, on the plane or almost plane replace little tri patch than flat site with big tri patch, delete unnecessary point, thereby realize data compaction.Hamann has proposed the choice that a kind of curvature value according to the triangular facet place decides this triangular facet, and the method that fits again then is applicable to the automatic generation of stl file.Hamann and Chen are making up the Different Plane curve, and compression 2D image and visual entity aspect have proposed to simplify the method for data point.Estimate to come selected point according to the local absolute curvature that piecewise linear curve is approached, the control of counting that the degree of therefore simplifying not only is selected also is subjected to the control of error level.

Above majority is all concentrated the effort of cloud data compressing method research and is to operate the multiaspect model, in the application process of reality, with the some cloud master pattern that normally has magnanimity point at random that 3 D scanning system obtains, it is crucial that these points at random are simplified.For this reason, use a kind of new algorithm that a cloud is simplified among the present invention.It is difficult point in this type of algorithm that the point that how a cloud is carried out rasterizing and extract the most approaching little grid mean curvature replaces point in the whole little grid.What propose among the present invention can guarantee to reach the more rational effect of simplifying under the different ratios of simplifying and precision based on the 3 d grid method of the curvature algorithm to point cloud compressing.

Summary of the invention

The invention provides the method for a kind of 3 d grid method based on curvature to point cloud compressing, the present invention has the simple advantage of method.

The present invention adopts following technical scheme:

A kind of 3 d grid method based on curvature is to the method for point cloud compressing:

The first step: three-dimensional point cloud is carried out grid division, this grid division is the minimum external square of a cloud data, its 3 limits perpendicular to each other are parallel with 3 coordinate axis of Cartesian coordinates respectively, and being divided into the length of side along three change in coordinate axis direction is D space hexahedron cubic grid.

Maximum coordinates (x according to a cloud mid point _Max, y _Max, z _Max) and min coordinates (x _Min, y _Min, z _Min), obtain the minimum external square length of side L of cloud data:

L＝max((x _max-x _min)，(y _max-y _min)，(z _max-z _min))

The big or small D of each little grid then is set at so:

$D=\sqrt[3]{\mathrm{\α}{L}^3/n}$

α is a dilution ratio in the formula, the ratio of number of spots and definite before and after dilution ratio α is meant and simplifies by designing requirement, and n is the whole some data point number in the cloud;

Second step: at each data point P of a cloud _i(i=1 ... n) about, about, its k neighbor point is taken out in totally 27 sub-grids of vicinity in front and back, obtains P _iNeighborhood point set KNB (P _i), and be designated as KNB (P _i)={ X ₁, X ₂..., X _n, after this, with neighborhood point set KNB (P _i) the least square approximation plane and with this as P _iLittle section S (P _i), order

Be KNB (P _i) the centre of form,

Be called data point set P _iCentral point:

${\stackrel{\_}{P}}_i=\frac{1}{k+1}(\underset{X\∈\mathrm{KNB}\left(P_i\right)}{\mathrm{\Σ}}X+P_i)$

If data point set P _iJ neighbor point to little section S (P _i) distance be d _j, arrive

Distance be λ _j

So to data point set P _iJ point have a function f _j(P _i):

$f_j\left(P_i\right)=\frac{d_j}{{\mathrm{\λ}}_j}$

Data point set P so _iCurvature function can be expressed as:

$f\left(P_i\right)=\frac{1}{k}\underset{j=1}{\overset{k}{\mathrm{\Σ}}}{f}_j\left(P_i\right)$

Can be according to following formula in the hope of the curvature of each data point in the cloud;

The 3rd step: after some cloud rasterizing is finished, ask for the mean curvature of each little grid mid point

:

$\stackrel{\_}{C}=\frac{1}{m}\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{C}_k$

C in the formula _kCurvature for data point in the little grid, m is the quantity of little grid mid point, and obtain the curvature of each little grid mid point and the difference of mean curvature, this difference is less than the selected point as this grid of naming a person for a particular job of the precision in the designing requirement, remaining names a person for a particular job deleted, final realization simplifying a cloud.

The present invention is mainly used in various application scenarios with point cloud compressing of complex-curved shape in the 3 D scanning system.Utilize 3 d grid method among the present invention can realize right-sizing to a cloud based on curvature.This method mainly contains following advantage:

(1) simplifying of cloud of point helps to improve modeling efficiency and modeling quality, and the optimum efficiency of simplifying is to make the point cloud data after simplifying have less data volume.The algorithm that proposes among the present invention makes the density of data point after simplifying change along with the variation of curvature of curved surface, the curved transition that is curved surface is big more, and data point is more, otherwise curved transition is more little, data point is just lacked, and guarantees the not minutia of lost thing surface.

(2) to simplify mainly be the method that adopts bounding box for traditional cloud data, but because the size of bounding box is set arbitrarily by the user, therefore can't guarantee constructed model and the precision between the original point cloud data, the size of the medium and small grid of 3 d grid algorithm that proposes among the present invention is defined by dilution ratio, more reasonable, applicability is strong.

(3) in the application process of reality, with the some cloud master pattern that normally has magnanimity point at random that 3 D scanning system obtains, it is succinct to propose algorithm among the present invention, and speed is fast, and the efficient height is applicable to a large amount of three-dimensional data points clouds at random, simplifies respond well.

(4) can provide the dilution ratio of a cloud and simplify precision according to actual needs.Dilution ratio has determined initial grid number, precision is defined as the curvature of each little grid mid point and the difference of mean curvature, precision is the principal element of commit point extraction ratio: precision is more little, the point that extracts is few more, vice versa, specifically extracts which point and then determined by the size of initial grid, therefore, if both wanted to guarantee that precision obtained data point as much as possible again, when the extraction of carrying out a little, should select less grid and bigger precision for use so.

(5) operating process is fairly simple, as long as provide dilution ratio and precision, following step all can be finished automatically, and speed is fast.And this method has very strong versatility.

(6) realization of data structure helps the object of management point, the search of point set, and traversal, ordering is dealt with problems fast and effectively, improves the efficient of calculating.

Description of drawings

Fig. 1 is the 3 D scanning system composition diagram.

Fig. 2 is uniform grid method figure.

Fig. 3 is algorithm overall flow figure.

Fig. 4 is people's face point cloud chart.

Fig. 5 deletes the figure that formats.

Fig. 6 is curvature and neighbor point graph of a relation.

Fig. 7 be given dilution ratio be 2 and precision be people's face point cloud chart of 0.15.

Fig. 8 be given dilution ratio be 3 and precision be people's face point cloud chart of 0.1.

Embodiment

A kind of method of point cloud data being simplified based on the 3 d grid method of curvature:

The first step: three-dimensional point cloud is carried out grid division, this grid division is the minimum external square of a cloud data, its 3 limits perpendicular to each other are parallel with 3 coordinate axis of Cartesian coordinates respectively, and being divided into the length of side along three change in coordinate axis direction is D space hexahedron cubic grid.

Maximum coordinates (x according to a cloud mid point _Max, y _Max, z _Max) and min coordinates (x _Min, y _Min, z _Min), obtain the minimum external square length of side L of cloud data:

L＝max((x _max-x _min)，(y _max-y _min)，(z _max-z _min))

The big or small D of each little grid then is set at so:

$D=\sqrt[3]{\mathrm{\α}{L}^3/n}$

α is a dilution ratio in the formula, the ratio of number of spots and definite before and after dilution ratio α is meant and simplifies by designing requirement, and n is the whole some data point number in the cloud;

Second step: at each data point P of a cloud _i(i=1 ... n) about, about, its k neighbor point is taken out in totally 27 sub-grids of vicinity in front and back, obtains P _iNeighborhood point set KNB (P _i), and be designated as KNB (P _i)={ X ₁, X ₂..., X _n, after this, with neighborhood point set KNB (P _i) the least square approximation plane and with this as P _iLittle section S (P _i), order Be KNB (P _i) the centre of form,

Be called data point set P _iCentral point:

${\stackrel{\_}{P}}_i=\frac{1}{k+1}(\underset{X\∈\mathrm{KNB}\left(P_i\right)}{\mathrm{\Σ}}X+P_i)$

If data point set P _iJ neighbor point to little section S (P _i) distance be d _j, arrive

Distance be λ _j

So to data point set P _iJ point have a function f _j(P _i):

$f_j\left(P_i\right)=\frac{d_j}{{\mathrm{\λ}}_j}$

Data point set P so _iCurvature function can be expressed as:

$f\left(P_i\right)=\frac{1}{k}\underset{j=1}{\overset{k}{\mathrm{\Σ}}}{f}_j\left(P_i\right)$

Can be according to following formula in the hope of the curvature of each data point in the cloud;

The 3rd step: after some cloud rasterizing is finished, ask for the mean curvature of each little grid mid point

:

$\stackrel{\_}{C}=\frac{1}{m}\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{C}_k$

C in the formula _kCurvature for data point in the little grid, m is the quantity of little grid mid point, and obtain the curvature of each little grid mid point and the difference of mean curvature, this difference is less than the selected point as this grid of naming a person for a particular job of the precision in the designing requirement, remaining names a person for a particular job deleted, final realization simplifying a cloud.

With reference to the accompanying drawings, the present invention is described in detail:

In 3 D scanning system, what face is the intensive point random data that disperse like the clouds.Most effort to the research of cloud data compressing method are all concentrated and are the monolithic data of operating the multiaspect model or being applicable to vertical scanning surface, direction of scanning, shape to the seizure part is sensitive inadequately, so the present invention adopts the algorithm of point cloud data being simplified based on the 3 d grid method of curvature.Overall algorithm flow chart is seen Fig. 3.

The present invention relates generally to the content of following three aspects:

1) rasterizing of three-dimensional point cloud

People's face original point cloud (Fig. 4) is carried out grid division, this grid division is the minimum external square of a cloud data, its 3 limits perpendicular to each other are parallel with 3 coordinate axis of Cartesian coordinates respectively, being divided into the length of side along three change in coordinate axis direction is D space hexahedron cubic grid, sees Fig. 5.

Maximum coordinates (x according to a cloud mid point _Max, y _Max, z _Max) and min coordinates (x _Min, y _Min, z _Min), obtain the minimum external square length of side L of cloud data:

L＝max((x _max-x _min)，(y _max-y _min)，(z _max-z _min))

The big or small D of each little grid then is set at so:

$D=\sqrt[3]{\mathrm{\α}{L}^3/n}$

α is a dilution ratio in the formula, the ratio of number of spots and definite before and after dilution ratio α is meant and simplifies by designing requirement, and n is the whole some data point number in the cloud;

2) calculating of the foundation in little section and curvature

At each data point P of a cloud _i(i=1 ... n) about, about, its k neighbor point is taken out in totally 27 sub-grids of vicinity in front and back, obtains P _iNeighborhood point set KNB (P _i), and be designated as KNB (P _i)={ X ₁, X ₂..., X _n, after this, with neighborhood point set KNB (P _i) the least square approximation plane and with this as P _iLittle section S (P _i), order

Be KNB (P _i) the centre of form,

Be called data point set P _iCentral point:

${\stackrel{\_}{P}}_i=\frac{1}{k+1}(\underset{X\∈\mathrm{KNB}\left(P_i\right)}{\mathrm{\Σ}}X+P_i)$

If data point set P _iJ neighbor point to little section S (P _i) distance be d _j, arrive

Distance be λ _j, see Fig. 6.

So to data point set P _iJ point have a function f _j(P _i):

$f_j\left(P_i\right)=\frac{d_j}{{\mathrm{\λ}}_j}$

Data point set P so _iCurvature function can be expressed as:

$f\left(P_i\right)=\frac{1}{k}\underset{j=1}{\overset{k}{\mathrm{\Σ}}}{f}_j\left(P_i\right)$

Can be according to following formula in the hope of the curvature of each data point in the cloud;

3) extraction

After some cloud rasterizing is finished, ask for the mean curvature of each little grid mid point

:

$\stackrel{\_}{C}=\frac{1}{m}\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{C}_k$

C in the formula _kBe the curvature of data point in the little grid, m is the quantity of little grid mid point, and obtains the curvature of each little grid mid point and the difference of mean curvature, and this difference is less than the selected point as this grid of naming a person for a particular job of the precision in the designing requirement, and remaining names a person for a particular job deleted.By this strategy point is extracted, the extraction ratio of point depends on dilution ratio and precision a little, and these two variablees are determined by designing requirement.Dilution ratio is 2, and precision is that 0.015 point cloud compressing figure sees Fig. 7, and dilution ratio is 3, precision is that 0.01 point cloud compressing figure sees Fig. 8, as seen the some cloud after simplifying has less data volume, and the dilution ratio of Fig. 8 is big than Fig. 7's, but the tiny characteristics of reproduction point cloud that still can be complete after simplifying.

Claims (1)

1, a kind of point cloud compressing method of 3-D scanning is characterized in that:

The first step: three-dimensional point cloud is carried out grid division, this grid division is the minimum external square of a cloud data, its 3 limits perpendicular to each other are parallel with 3 coordinate axis of Cartesian coordinates respectively, being divided into the length of side along three change in coordinate axis direction is D space hexahedron cubic grid, according to the maximum coordinates (x of a cloud mid point _Max, y _Max, z _Max) and min coordinates (x _Min, y _Min, z _Min), obtain the minimum external square length of side L of cloud data:

L＝max((x _max-x _min)，(y _max-y _min)，(z _max-z _min))

The big or small D of each little grid then is set at so:

$D=\sqrt[3]{\mathrm{\α}{L}^3/n}$

α is a dilution ratio in the formula, the ratio of number of spots and definite before and after dilution ratio α is meant and simplifies by designing requirement, and n is the whole some data point number in the cloud;

Second step: at each data point P of a cloud _i(i=1 ... n) about, about, its k neighbor point is taken out in totally 27 sub-grids of vicinity in front and back, obtains P _iNeighborhood point set KNB (P _i), and be designated as KNB (P _i)={ X ₁, X ₂..., X _n, after this, with neighborhood point set KNB (P _i) the least square approximation plane and with this as P _iLittle section S (P _i), order

Be KNB (P _i) the centre of form,

Be called data point set P _iCentral point:

$\stackrel{-}{P_i}=\frac{1}{(k+1)}(\underset{X\∈\mathrm{KNB}\left(P_i\right)}{\mathrm{\Σ}}X+P_i)$

If data point set P _iJ neighbor point to little section S (P _i) distance be d _j,, arrive

Distance be λ _i, so to data point set P _iJ point have a function f _j(P _i):

$f_j\left(P_i\right)=\frac{d_j}{{\mathrm{\λ}}_j}$

Data point set P so _iCurvature function can be expressed as:

$f\left(P_i\right)=\frac{1}{k}\underset{j=1}{\overset{k}{\mathrm{\Σ}}}{f}_j\left(P_i\right)$

Can be according to following formula in the hope of each data point P in the cloud _iCurvature;

The 3rd step: after some cloud rasterizing is finished, ask for the mean curvature of each little grid mid point

$\stackrel{-}{C}=\frac{1}{m}\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{C}_k$

C in the formula _kCurvature for data point in the little grid, m is the quantity of little grid mid point, and obtain the curvature of each little grid mid point and the difference of mean curvature, this difference is less than the selected point as this grid of naming a person for a particular job of the precision in the designing requirement, remaining names a person for a particular job deleted, final realization simplifying a cloud.

Images