CN102779345A - Point cloud precise registering method based on gravity center Euclidean distance - Google Patents

Abstract

The invention discloses a point cloud precise registering method based on a gravity center Euclidean distance. By calculating the gravity center of an effective point set of a point cloud overlaying portion and the Euclidean distance of all points in each point set relative to the gravity center, optimization screening is conducted, and overall identical points are solved. By judging a hazard ball, the stable overall identical points are obtained, conversion parameters among point cloud coordinate systems are obtained, and precise registering of the point cloud of two measuring stations is achieved. The method adopts the gravity center Euclidean distance to register the point cloud, due to the fact that the gravity center Euclidean distance has the characteristics of uniqueness and quantity limit and the like, point cloud registering speed and precision are improved. By judging the hazard ball, an equal Euclidean distance distinguishing problem is solved, and point cloud matching obscureness is avoided. Point gaps are used as error margins for judging identical Euclidean distance, and registering is more precise. By removing error points in the point cloud continuously, a purpose that abnormal value disturbance is not affected is achieved. A step by step process of first rough registering and then precise registering enables point cloud registering to be easy.

Description

A kind of accurate method for registering of some cloud based on the center of gravity Euclidean distance

Technical field

The present invention relates to fields such as computer graphics, mapping science and reverse Engineering Technology, specially refer to the site cloud registration problems of surveying more, relate in particular to a kind of accurate method for registering of some cloud based on the center of gravity Euclidean distance.

Background technology

Data acquisition phase in that object dimensional is rebuild need use three-dimensional laser scanner that object is carried out comprehensive scanning; But, because the complicacy of object, can not one-off scanning obtain all cloud datas of object; Need substation scanning, this just is faced with the site cloud registration problems of surveying more.The registration of multi-site cloud is exactly all to be surveyed site cloud coordinate systems be planned for the process under the same public coordinate system.The site cloud registrations of surveying can be exchanged into the registration problems in twos between adjacent two stations more, and the method for registering based on a cloud can be divided into two types at present: common point method and ICP algorithm.The common point method requires to have 3 couples of artificial targets or unique point between adjacent two stations at least; Utilize least square method to find the solution conversion parameter, this method precision is high and reliable, but this method needs scanning separately each sign and identification center; The record period carries out registration, is semi-automatic method; ICP (Iterative Closest Point) algorithm is the 20th century famous algorithms that are suggested of the nineties; This algorithm at first supposes to obtain an initial attitude transformation parameter; From a site cloud, choose the point of some, and the neighbor point of in another cloud, choosing these points is as corresponding point, with point to distance minimization be that condition is carried out iterative computation; Till satisfying the condition of convergence, but the ICP algorithm exists in practical application and is subject to that exceptional value is disturbed, arithmetic speed waits deficiency slowly.Therefore how effectively, autoregistration point cloud fast, accurately, be problem demanding prompt solution in the present object dimensional process of reconstruction.

Summary of the invention

Goal of the invention: the purpose of this invention is to provide a kind of efficiently, accurately based on the accurate method for registering of some cloud of center of gravity Euclidean distance.

Technical scheme: in order to realize the foregoing invention purpose, a kind of accurate method for registering of some cloud based on the center of gravity Euclidean distance of the present invention comprises the steps:

(1), obtains definite initial attitude transformation parameter (R based on the discrete point characteristic for reference point clouds P and registration point cloud M ₀, t ₀), R ₀Be rotation parameter, t ₀Be translation parameters;

(2) according to initial attitude transformation parameter (R ₀, t ₀) point among the registration point cloud M all is transformed into carries out thick registration under the P coordinate system, thick registration equation is:

P=R ₀M+t ₀；

(3) search and the some M that puts among the cloud M in a cloud P _i(i=1,2 ..., m) the some P of correspondence _j(j=1,2 ..., n), based on a P _jWith a M _iEuclidean distance minimum value screening, obtain effective point set (PS, MS);

(4) obtain the center of gravity PS of each effective point set _GAnd MS _G, and each point is concentrated Euclidean distance storehouse PD and the MD of have a few with respect to its center of gravity:

${\mathrm{PS}}_G=\frac{1}{h}\underset{i=1}{\overset{h}{\mathrm{\Σ}}}{\mathrm{PS}}_i,$

${\mathrm{MS}}_G=\frac{1}{h}\underset{i=1}{\overset{h}{\mathrm{\Σ}}}{\mathrm{MS}}_i,$

d(PS _i,PS _G)=|PS _i-PS _G|，(i=1,2,…,h)，

d(MS _i,MS _G)=|MS _i-MS _G|，(i=1,2,…,h)，

In the formula, d (PS _i, PS _G) and d (MS _i, MS _G) represent respectively the Euclidean distance of point set PS and MS mid point i and center of gravity to form Euclidean distance storehouse PD and MD;

(5) i Euclidean distance MD among the pair set MD _i(i=1,2 ..., h) search minimum euclidean distance PD in set PD _j(j=1,2 ..., h), try to achieve the h that satisfies under the certain distance threshold value to same place (MS _i, PS _j), h representes the number of effective point set, perhaps the number of Euclidean distance storehouse PD or MD;

(6) utilize same place (MS _i, PS _j) ask a conversion parameter of cloud coordinate system, carry out the coordinate conversion of a cloud, realize the accurate registration of some cloud.

Said step is worked as PD in (5) _j-MD _iDuring≤2 α, said PD _jWith MD _iBe corresponding Euclidean distance, corresponding (MS _i, PS _j) be same place, if PD _j-MD _i＞2 α, description of step (4) effectively point set search contains the error point, rejects i current Euclidean distance MD _iWith a PS _i, getting back to step (4) and search for next Euclidean distance, α representes the dot spacing of a cloud.

The same place that said step (5) is tried to achieve carries out dangerous ball judgement and distinguishes same place: in point set MS or PS, be the centre of sphere with center of gravity o, and a certain Euclidean distance MD _iDraw a spheroid for radius, have individual some q of x (x>=2) on the spheroid _x(x=1,2 ... X) time, be defined as dangerous ball, the same place on the dangerous ball can't be distinguished, and in order to address this problem, from the same place of having obtained, selects two some D _kQ is calculated in (k=1,2) _xWith two some D _kEuclidean distance dq with centre of sphere o _Xy(y=1,2,3), in the space respectively with o and two D _kBe the centre of sphere, with the Euclidean distance dq of correspondence _XyFor radius is made ball, obtain to intersect and unique some q _x, in like manner to the unique definite some q ' of another point set _x, obtain to distinguish same place (q _x, q ' _x), thereby realize differentiation to a plurality of same places on the dangerous ball; And then get back to said step (4) iterative computation, up to the same place that obtains no dangerous ball.

Confirm initial attitude transformation parameter (R in the said step (1) ₀, t ₀) concrete steps be:

Set up transformation equation: P _i=R ₀M _i+ t ₀, (i=1,2 ..., m), in the formula, P _iFor among the reference point clouds P a bit, M _iFor among the registration point cloud M a bit, R ₀Be rotation parameter, t ₀Be translation parameters, m for select public count into, m (m>=3) adopts linear least square to find the solution conversion parameter (R ₀, t ₀), based on some space Euclidean distance error is defined as:

$e=\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{|P_i-R_0{M}_i-t_0|}^2,$ T is asked local derviation, $\frac{\∂e}{\∂t}=2\underset{i=1}{\overset{m}{\mathrm{\Σ}}}(P_i-R_0{M}_i-t_0)=0,$ Draw:

t ₀=P _G-R ₀M _G，

Wherein, $P_G=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{P}_i,$ $M_G=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{M}_i$ Be respectively the center of gravity of P and M;

Make R ₀'=P-P _G, M _G'=M-M _GBe the point after the center of gravityization, then

$e=\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{|P_i-R_0{M}_i-t_0|}^2=\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{|P_G^{\′}-R_0{M}_G^{\′}|}^2,$

Utilize unit quaternion to minimize e and calculate initial transformation parameter (R ₀, t ₀).

Said step is utilized same place (MS in (6) _i, PS _j) ask a conversion parameter of cloud coordinate system, carry out the coordinate conversion of a cloud, be based on the overall same place (MS that step (5) is tried to achieve _i, PS _j), confirm initial attitude transformation parameter (R in utilization and the step (1) ₀, t ₀) identical method find the solution conversion parameter between the cloud coordinate system (R t), changes a cloud M then, realizes smart registration:

M"=RM+t，

M in the formula is " for the some cloud after the conversion, under the coordinate of cloud P.After this step, promptly accomplished the smart registration of reference point clouds P and registration point cloud M.

Beneficial effect: this gordian technique of using gravity-center Euclidean distance of the present invention is come the registration point cloud, because the center of gravity Euclidean distance has characteristics such as uniqueness and limited amount property, has improved some cloud registration speed and precision; Judgement through dangerous ball has solved equal Euclidean distance differentiation problem, avoids a cloud coupling fuzzy; As the limits of error of judging Euclidean distance of the same name, make registration more accurate through dot spacing; The present invention reaches the purpose that not disturbed by exceptional value through constantly rejecting the error point in the some cloud; The incremental process of smart registration is more prone to a cloud registration behind the earlier thick registration.

Description of drawings

Fig. 1 is the smart registration process flow diagram of some cloud that the present invention is based on the center of gravity Euclidean distance;

Fig. 2 is the plane figure of the dangerous ball situation of point set MS in the specific embodiment of the invention.

Embodiment

Below in conjunction with accompanying drawing the present invention is done explanation further.

As shown in Figure 1, for to given reference point clouds P that certain degree of overlapping is arranged and match point cloud M, a cloud P and M are carried out thick registration according to the initial attitude transformation parameter of obtaining based on the discrete point characteristic; Through the screening point, obtain the effective corresponding point set of consecutive point cloud lap; Obtain the center of gravity and the each point of each effective point set and concentrate the Euclidean distance of have a few with respect to its center of gravity; For confirming the foundation of same place, find the solution overall same place with the Euclidean distance approximately equal; Effective point set center of gravity of double counting and corresponding Euclidean distance, until the overall same place that obtains count stable till.Utilize same place to try to achieve the conversion parameter between a cloud coordinate system, and carry out the coordinate conversion of a cloud, realize that two survey the smart registration of site cloud.

A kind of accurate method for registering of some cloud based on the center of gravity Euclidean distance of the present invention comprises the steps:

S100:, obtain definite initial attitude transformation parameter (R based on the discrete point characteristic for reference point clouds P and registration point cloud M ₀, t ₀):

Set up transformation equation:

P _i=R ₀M _i+t ₀,(i=1,2,…,m)，

In the formula, P _iFor among the reference point clouds P a bit, M _iFor among the registration point cloud M a bit, R ₀Be rotation parameter, t ₀Be translation parameters, m for select public count into, m (m>=3) adopts linear least square to find the solution conversion parameter (R ₀, t ₀), based on some space Euclidean distance error is defined as:

$e=\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{|P_i-R_0{M}_i-t_0|}^2,$

T is asked local derviation,

$\frac{\∂e}{\∂t}=2\underset{i=1}{\overset{m}{\mathrm{\Σ}}}(P_i-R_0{M}_i-t_0)=0,$

Draw:

t ₀=P _G-R ₀M _G，

Wherein, $P_G=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{P}_i,$ $M_G=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{M}_i$ Be respectively the center of gravity of P and M; Make P ' G=P-PG, M ' G=M-MG $P_G^{\′}=P-P_G,$ $M_G^{\′}=M-M_G$ Be the point after the center of gravityization, then

$e=\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{|P_i-R_0{M}_i-t_0|}^2=\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{|P_G^{\′}-R_0{M}_G^{\′}|}^2,$

Utilize unit quaternion to minimize e and calculate initial transformation parameter (R ₀, t ₀).

S200: according to initial attitude transformation parameter (R ₀, t ₀) point among the registration point cloud M all is transformed into carries out thick registration under the P coordinate system, thick registration equation is:

P=R ₀M+t ₀；

S300: search and the some M that puts among the cloud M in a cloud P _i(i=1,2 ..., m) the some P of correspondence _j(j=1,2 ..., n), calculate each point and some M among the P _iEuclidean distance:

d(M _i,P _j)=|P _j-M _i|，

In the formula, d (M _i, P _j) be a some P _jWith a M _iEuclidean distance, search for the minimum value of all Euclidean distances then:

d _min(M _i,P _k)=Min(d(M _i,P _j)),(j=1,2,…,n)

If d _Min(M _i, P _k) less than a threshold value and M _iAnd P _kThe method vector approaching, show that then some cloud P exists and a M _iCorresponding point, retention point M _i, if minor increment or method vector do not meet the demands, retention point M not then _iOther point among the double counting point cloud M is removed ineligible point, form at last effective point set (PS, MS).

S400: the center of gravity PS that obtains each effective point set _GAnd MS _G, and each point is concentrated Euclidean distance storehouse PD and the MD of have a few with respect to its center of gravity:

${\mathrm{PS}}_G=\frac{1}{h}\underset{i=1}{\overset{h}{\mathrm{\Σ}}}{\mathrm{PS}}_i,$

${\mathrm{MS}}_G=\frac{1}{h}\underset{i=1}{\overset{h}{\mathrm{\Σ}}}{\mathrm{MS}}_i,$

d(PS _i,PS _G)=|PS _i-PS _G|，(i=1,2,…,h)，

d(MS _i,MS _G)=|MS _i-MS _G|，(i=1,2,…,h)，

In the formula, PS _GAnd MS _GBe respectively the center of gravity of point set PS and MS, d (PS _i, PS _G) and d (MS _i, MS _G) represent respectively the Euclidean distance of point set PS and MS mid point i and center of gravity to form Euclidean distance storehouse PD and MD;

S500: this step obtains effective point set through calculating the minimum Eustachian distance row filter of going forward side by side.

I Euclidean distance MD among the pair set MD _i(i=1,2 ... H, search minimum euclidean distance PD in set PD _j(j=1,2 ..., h tries to achieve the h that satisfies under the certain distance threshold value to same place (MS _i, PS _j), h representes the number of effective point set, perhaps the number of Euclidean distance storehouse PD or MD.

The certain distance threshold value here i.e. 2 α, and α representes the dot spacing of a cloud, when such, and said PD _jWith MD _iBe corresponding Euclidean distance, corresponding (MS _i, PS _j) be same place, if PD _j-MD _i＞2 α, the effective point set search of description of step S400 contains the error point, rejects i current Euclidean distance MD _i, get back to step S400 and search for next Euclidean distance.

S600: in order to solve the problem that equates that Euclidean distance can't be distinguished, avoid a cloud coupling fuzzy, the same place that step S500 is tried to achieve carries out dangerous ball judgement, in point set MS or PS, is the centre of sphere with center of gravity o, a certain Euclidean distance MD _iDraw a spheroid for radius, have individual some q of x (x>=2) on the spheroid _x(x=1,2 ... X) time, be defined as dangerous ball, the same place on the dangerous ball can't be distinguished.

Please refer to Fig. 2, is example with point set MS here, (x>=2) the individual some q that suppose that x is arranged among the MS _x(x=1,2 ..., x) drop on the dangerous ball, from the same place of having obtained, select two same place D ₁And D ₂And, calculate q then as pseudo-center of gravity _xEuclidean distance with two pseudo-centers of gravity and centre of sphere o:

${\mathrm{dq}}_x=|o-q_{x_1}|,$

${\mathrm{dq}}_x=|D_1-q_{x_2}|,$

${\mathrm{dq}}_{x3}=|D_2-q_{x_3}|,$

In the space respectively with o and D ₁, D ₂Be the centre of sphere, with

For radius is made 3 balls, three balls intersect unique intersection point q _x

In like manner, in point set PS, carrying out similar operation can uniquely confirm and q _xCorresponding some q ' _x, obtain same place (q _x, q ' _x), distinguished a plurality of points on the dangerous ball.

And then get back to said step S400 iterative computation, up to the same place that obtains no dangerous ball.

S700: utilize same place (MS _i, PS _j) ask a conversion parameter of cloud coordinate system, carry out the coordinate conversion of a cloud, realize the accurate registration of some cloud.

Overall same place (the MS that tries to achieve based on S600 _i, PS _j), confirm initial attitude transformation parameter (R among utilization and the S100 ₀, t ₀) identical method find the solution conversion parameter between the cloud coordinate system (R t), changes a cloud M then, realizes smart registration:

M"=RM+t

M in the formula is " for the some cloud after the conversion, under the coordinate of cloud P.After this step, promptly accomplished the smart registration of reference point clouds P and registration point cloud M.

The above only is a preferred implementation of the present invention; Be noted that for those skilled in the art; Under the prerequisite that does not break away from the principle of the invention, can also make some improvement and retouching, these improvement and retouching also should be regarded as protection scope of the present invention.

Claims (4)

1. the accurate method for registering of some cloud based on the center of gravity Euclidean distance is characterized in that comprising the steps:

(1), obtains definite initial attitude transformation parameter (R based on the discrete point characteristic for reference point clouds P and registration point cloud M ₀, t ₀), R ₀Be rotation parameter, t ₀Be translation parameters;

(2) according to initial attitude transformation parameter (R ₀, t ₀) point among the registration point cloud M all is transformed into carries out thick registration under the P coordinate system, thick registration equation is:

P=R ₀M+t ₀；

(3) search and the some M that puts among the cloud M in a cloud P _i(i=1,2 ..., m) the some P of correspondence _j(j=1,2 ..., n), based on a P _jWith a M _iEuclidean distance minimum value screening, obtain effective point set (PS, MS);

(4) obtain the center of gravity PS of each effective point set _GAnd MS _G, and each point is concentrated Euclidean distance storehouse PD and the MD of have a few with respect to its center of gravity:

${\mathrm{PS}}_G=\frac{1}{h}\underset{i=1}{\overset{h}{\mathrm{\Σ}}}{\mathrm{PS}}_i,$

${\mathrm{MS}}_G=\frac{1}{h}\underset{i=1}{\overset{h}{\mathrm{\Σ}}}{\mathrm{MS}}_i,$

d(PS _i,PS _G)=|PS _i-PS _G|，(i=1,2,…,h)，

d(MS _i,MS _G)=|MS _i-MS _G|，(i=1,2,…,h)，

In the formula, d (PS _i, PS _G) and d (MS _i, MS _G) represent respectively the Euclidean distance of point set PS and MS mid point i and center of gravity to form Euclidean distance storehouse PD and MD;

(5) i Euclidean distance MD among the pair set MD _i(i=1,2 ..., h) search minimum euclidean distance PD in set PD _j(j=1,2 ..., h), try to achieve the h that satisfies under the certain distance threshold value to same place (MS _i, PS _j), h representes the number of effective point set, perhaps the number of Euclidean distance storehouse PD or MD;

(6) utilize same place (MS _i, PS _j) ask a conversion parameter of cloud coordinate system, carry out the coordinate conversion of a cloud, realize the accurate registration of some cloud.

2. a kind of accurate method for registering of some cloud based on the center of gravity Euclidean distance according to claim 1, it is characterized in that: said step is worked as PD in (5) _j-MD _iDuring≤2 α, said PD _jWith MD _iBe corresponding Euclidean distance, corresponding (MS _i, PS _j) be same place, if PD _j-MD _i＞2 α reject i current Euclidean distance MD _iWith a PS _i, getting back to step (4) and search for next Euclidean distance, α representes the dot spacing of a cloud.

3. a kind of accurate method for registering of some cloud based on the center of gravity Euclidean distance according to claim 1 and 2 is characterized in that: the same place that said step (5) is tried to achieve carries out dangerous ball judgement and distinguishes same place:

In point set MS or PS, be the centre of sphere with center of gravity o, a certain Euclidean distance MD _iDraw a spheroid for radius, have individual some q of x (x>=2) on the spheroid _x(x=1,2 ... During x, be defined as dangerous ball; From the same place of having obtained, select two some D _k(k=1,2, calculate q _xWith two same place D _kEuclidean distance dq with centre of sphere o _Xy(y=1,2,3), in the space respectively with o and two D _kBe the centre of sphere, with the Euclidean distance dq of correspondence _XyFor radius is made ball, obtain to intersect and unique some q _x, in like manner to the unique definite some q ' of another point set _x, obtain to distinguish same place (q _x, q ' _x);

Return said step (4) iterative computation, up to the same place that obtains no dangerous ball.

4. a kind of accurate method for registering of some cloud based on the center of gravity Euclidean distance according to claim 1 is characterized in that: confirm initial attitude transformation parameter (R in the said step (1) ₀, t ₀) concrete steps be:

Set up transformation equation: P _i=R ₀M _i+ t ₀, (i=1,2 ..., m),

In the formula, P _iFor among the reference point clouds P a bit, M _iFor among the registration point cloud M a bit, R ₀Be rotation parameter, t ₀Be translation parameters, m does, m (m>=3) for public the counting of selecting,

Adopt linear least square to find the solution conversion parameter (R ₀, t ₀), based on some space Euclidean distance error is defined as:

$e=\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{|P_i-R_0{M}_i-t_0|}^2,$ T is asked local derviation, $\frac{\∂e}{\∂t}=2\underset{i=1}{\overset{m}{\mathrm{\Σ}}}(P_i-R_0{M}_i-t_0)=0,$ Draw:

t ₀=P _G-R ₀M _G，

Wherein, $P_G=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{P}_i,$ $M_G=\frac{1}{m}\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{M}_i$ Be respectively the center of gravity of P and M;

Make P _G'=P-P _G, M _G'=M-M _GBe the point after the center of gravityization, then

$e=\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{|P_i-R_0{M}_i-t_0|}^2=\underset{i=1}{\overset{m}{\mathrm{\Σ}}}{|P_G^{\′}-R_0{M}_G^{\′}|}^2,$

Utilize unit quaternion to minimize e and calculate initial transformation parameter (R ₀, t ₀).

Images