CN102135620A - Geometric statistical characteristic-based global scan matching method - Google Patents

Abstract

The invention discloses a geometric statistical characteristic-based global scan matching method, which is provided and implemented in the research of synchronously positioning a mobile robot, drawing an environmental outline, determining an environmental change area and the like. The method can be used for environmental modeling of the mobile robot and serves as auxiliary means for disaster relief, mine survey and other specific tasks. In the method, statistical result-defined characteristics of geometric information on laser scanning data are utilized; and the method comprises the following steps of: defining segmentation characteristics and simplifying whole scanning description through segmentation scanning; describing local scanning property by adopting scanning spot characteristics, and obtaining a relative posture between scanning through a trial-verification-solution strategy, namely obtaining a posture trial solution at a matching segmentation, verifying accuracy of the solution, and solving the relative posture by utilizing corresponding point characteristics and the least square method when the solution passes the verification; and taking the optimal verified solution as a final solution. Therefore, by the method, the environmental outline is described, the posture of the robot is corrected, and the environmental change is identified according to the characteristic matching result.

Description

A kind of whole scan matching process based on how much statistical natures

One, technical field

The present invention relates to the intelligent mobile robot field, be mainly used in the mobile robot and in unknown dynamic environment, locate and modeling, but method itself is not limited thereto.The research and the application of scanning matching process relate to numerous areas such as medical treatment, archaeology, rescue, military affairs, virtual reality, mobile robot.Reappear the original appearance of destructuring circumstances not known, be very helpful to finishing difficult dangerous tasks such as the disaster relief, prospecting undoubtedly.And from scanning coupling research, derive for static, dynamically, the research of non-rigid object identification, also have bigger using value.Be different from the needs that reproduce for the object model high precision in the researchs such as medical treatment, archaeology, mobile robot's research is had higher requirement to the real-time of scanning coupling, to guarantee that robot can finish set task quickly and accurately, as rescuing after the calamity.The quality of scanning matching result has directly embodied the accuracy that robot locatees in three-dimensional environment.

Two, background technology

The scanning coupling is the corresponding relation of seeking between different pose lower sensor scan-datas, estimates the relative pose conversion, and scan-data is combined, and recovers the general name of class methods of object or environment original appearance.The process object of scanning matching technique is the analyzing spot set that utilizes stadimeter to gather.Because the scanning coupling has the ability of setting up environmental model simultaneously and estimating relative pose, it is introduced in localization for Mobile Robot and builds in the research of figure, and become and, and based on the localization method of bayesian theory (as Markov and Monte Carlo) the important method for positioning mobile robot of a class arranged side by side based on the localization method of wave filter (as EKF).This technology is directly utilized the raw data of scanning, builds figure result and more can embody the environment original appearance.At present the robot field is deep day by day to the research of scanning coupling, has obtained a series of valuable achievements, as based on figure the time location with build drawing method (Graph-SLAM) etc.

According to the initial estimation that whether relies on the scanning room relative pose, the scanning coupling is divided into local coupling and global registration.The pose of local matching process is estimated to come from odometer usually, seeks corresponding analyzing spot on this basis iteratively, proofreaies and correct relative pose and until convergence scanning is overlaped.ICP (IterativeClosest/Corresponding Point) and NDT (Normal Distribution Transformation) are the representatives of these class methods.Local matching process hour is obtained desired result in static environment and pose evaluated error, then can not guarantee to converge to globally optimal solution when environmental change or pose evaluated error are big.Then need the global registration method in this case.The global registration method is mostly extracted feature from scan-data, find the solution pose by characteristic matching.Set up local coordinate system such as the method that has for each analyzing spot, utilize Hough transformation around analyzing spot, to extract Hough point constitutive characteristic.This feature is vulnerable to the influence of noise analyzing spot, observation pose and moving object.The method that perhaps has adopts two kinds of histograms to describe whole scanning, utilizes cross-correlation operation to obtain the scanning room relative pose, but is subjected to the influence of overlapping region size and environmental change.Also have method to utilize the section of cutting apart of cumulative angle function representation scanning, and for cutting apart the intersegmental location association of setting up.This method more can embody the local characteristics of scanning, but location association still is vulnerable to observe the influence of pose and environmental change.

Three, summary of the invention

The present invention from characterizing definition, that the matching strategy equal angles has proposed robustness is higher, adapts to the global registration method of dynamic environment.Whether the analyzing spot basis takes from same body surface is classified as a class, promptly regards the section of cutting apart as, and the analyzing spot set just is made of some sections of cutting apart.New method utilizes the statistical information of analyzing spot normal direction to be described from the section of cutting apart and two level defined features of analyzing spot, improves the robustness of characterizing definition.The section of cutting apart feature is the rough description to scan-data integral body.The analyzing spot characterizing definition is not disturbed by moving object on every side on the section of cutting apart at its place, can portray the section of cutting apart local properties meticulously.Final whole scan matching process adopts the strategy of " souning out-verify-find the solution ".But discrimination degree and similarity according to the section of cutting apart are set up corresponding relation, utilize the analyzing spot characteristic similarity to mate and obtain trial solution; Correctness according to evaluation rule checking trial solution; On the trial solution basis, find the section of cutting apart that all can be corresponding and find the solution the relative pose of scanning room.On the true laser data collection of a plurality of indoor circumstances not knowns, verified the validity of method by experiment.New method can correctly be finished coupling, proofreaies and correct pose, and dynamic object and region of variation are then demarcated in the process of characteristic matching.New method can be applicable to static state or dynamic environment, can assist the discovery closed loop when the positioning error accumulation is big.Though global registration method efficient is a little less than local matching process, the application on dynamic environment and closed loop is irreplaceable.

Four, description of drawings

(a) back (b) synoptic diagram before the pre-service of Fig. 1 scan-data.Gather 180 ° of scopes totally 361 points in the each laser scanning of this data centralization.(a) figure is the situation before the pre-service, and the analyzing spot after the pre-service on (b) same section of cutting apart links together by line segment, and normal direction is represented by short line segment.

Fig. 2 section of cutting apart (feature circular arc) but the discrimination degree synoptic diagram.This figure has reflected the irregular section of cutting apart of use characteristic arc representation, but the discrimination degree size of the section of cutting apart is represented in the shadow region.

Fig. 3 analyzing spot feature extraction synoptic diagram.How the feature that this Figure illustrates analyzing spot P defines, and α represents its normal direction, and L represents the length of the related little section division of analyzing spot, and l and r represent the border, the left and right sides of projection respectively, and the section of cutting apart goes up the normal direction of analyzing spot and uses directed line segment to represent.

The result of odometer (a) and new method (b) coupling is used in twice scanning of Fig. 4 respectively, and the figure orbicular spot is represented laser scanning point.Position 1 corresponding scanning 1, position 2 corresponding scannings 2.(a) figure reflects that the odometer error is bigger, scans inconsistent; (b) figure reflects and utilizes new method can finish coupling in good condition, corrects pose.(c) figure has amplified the result that utilization new method in place, corner is mated.

Mate 149 scannings on Fig. 5 Minguez J. data set continuously and set up the environment planimetric map.Be the dynamic area of mark in the rectangle wherein.Marked the track of robot operation among the figure, and zone that may dynamic change in the environment of finding by new method.

Mate 1525 scannings on Fig. 6 Fr079 data set continuously and set up the environment planimetric map.1 and 2 have marked two corners among the figure.

Mate 401 scannings on Fig. 7 Albert-b-laser data set continuously and set up the environment planimetric map.Be irregular area in the rectangle frame among the figure, mark the dynamic area in the environment.

Five, embodiment

5.1 the section of cutting apart feature description (Segment feature description)

5.1.1 pre-service

Before the extraction section of cutting apart feature, need laser data is carried out pre-service, comprise and cut apart and estimate two steps of normal direction.Cut apart is that analyzing spot according to the position distribution of analyzing spot may be gathered from same object is divided into the section of cutting apart.The section of cutting apart has been described the overall distribution of analyzing spot; Analyzing spot relies on the section of cutting apart and has just had more information.

Dividing method is taken all factors into consideration the distance between the laser spots, and the relation of angle between laser spots line and the laser beam, when reducing angle too small or hypertelorism because adopt a little sparse influence that brings.The very few section of cutting apart of number of scan points is filtered in this step.The environmental information of these section of cutting apart representatives is few, and some then is wrong laser readings or dynamic object.

For the analyzing spot in each analyzing spot P both sides certain limit, adopt the method for linear regression to estimate regression straight line, after cutting apart with the normal direction of vertical line direction as this analyzing spot P.Normal direction is equal orientation sensor after conversion.The normal direction of hereinafter mentioning is consistent therewith.Fig. 1 has shown the result before and after the pre-service, and the analyzing spot on the same section of cutting apart links together by line segment.

5.1.2 the section of cutting apart feature

The section of cutting apart is different, in order to unify to describe and compare, the feature F of this paper definition section of cutting apart S _SAs follows:

$F_S=(L_S,{\mathrm{\δ}}_S^{\mathrm{\α}},C_S)---\left(1\right)$

L _SOrder of representation connects the total length that the section of cutting apart goes up all analyzing spots, δ _S ^αThe variance that expression analyzing spot normal direction changes, C _S{ 1,0,1} represents the concavity and convexity (numerical value is represented recessed, flat, protruding successively) that the section of cutting apart is whole to ∈.Because zones of different analyzing spot density degree difference is at statistics δ _S ^αThe time according to the Density Weighted of analyzing spot both sides certain limit interscan point.The section of cutting apart can present the form of repeatedly concavo-convex variation, in this average normal direction of adding up the section of cutting apart integral body and left half side zone respectively, knows the concavity and convexity C that the section of cutting apart is whole in view of the above _S

Above characterizing definition reality is reduced to line segment or circular arc with the complicated dividing section.Suppose that normal direction satisfies consistent the distribution, when

The time section of cutting apart a corresponding segment length L _SCircular arc, the central angle scope

C _SThe decision opening direction.When the corresponding circumference of the section of cutting apart, central angle obtain maximum magnitude (π, π), corresponding δ _S ^αGet extreme value When

The time, the corresponding length L of the section of cutting apart _SLine segment.

But 5.1.3 the section of cutting apart discrimination degree

Usually the long more more irregular section of cutting apart is easy more to be distinguished mutually with other sections of cutting apart, big more to the contribution of coupling.In order to describe the degree that the difference section of cutting apart can be distinguished, but the discrimination degree D of this paper definition section of cutting apart S _SFor:

1) if the section of cutting apart character representation is a line segment, D _SBe 0;

2) if the section of cutting apart character representation is a circular arc, D _SExpression shaded area as shown in Figure 2 is about to the feature circular arc center of circle and places the longitudinal axis and tangent with transverse axis, the area that circular arc and transverse axis and boundary rectangle surrounded.

D _SBut the big more expression section of cutting apart discrimination degree is high more.Because there is error in laser readings, the δ of the section of cutting apart of true line segment correspondence _S ^αUsually be not 0, be expressed as the very little circular arc of central angle.Suppose that the laser readings error satisfies white Gaussian noise, then true long more easy more the distinguishing of line segment.

5.1.4 the section of cutting apart similarity

Mate in order to seek the similar section of cutting apart on different scanning, definition is the section of cutting apart S arbitrarily ₁And S ₂Between similarity S _FFor:

$S_F=R_L\·R_{\mathrm{\δ}}\·\frac{1}{1+|C_{S_1}{\mathrm{\δ}}_{S_1}^{\mathrm{\α}}-C_{S_2}{\mathrm{\δ}}_{S_2}^{\mathrm{\α}}|}---\left(2\right)$

Wherein

Similarity is actual has described the similarity degree on the geometric configuration between feature line segment or the feature circular arc.R _LSegment length ratio, R are cut apart in expression _δThe overlapping ratio of expression central angle scope,

The total difference of expression concavity and convexity and central angle.According to definition, S _F∈ [0,1] and S _FBig more similarity is high more.

5.2 analyzing spot feature description (Scan point feature description)

5.2.1 analyzing spot feature

The section of cutting apart feature is the rough description to scan-data integral body, and accurately coupling also needs to portray better the scan-data local properties, and this paper introduces the feature F that the section of cutting apart S goes up analyzing spot P for this reason _PAs shown in Figure 3, be initial point with a P, normal direction α _PFor transverse axis is set up local coordinate system.Along putting P place tangent line to the equidistant successively division in both sides, every section is designated as S _iBe called " related little section of analyzing spot ", the point on the S is projected to corresponding little section S successively _iIn until the end points of S.S _iIn statistics be that S goes up continuous one section, if in the process of i.e. S projection because before crooked having projected on little section of the statistics end then the premature termination projection process.Put the feature F of P thus _PBe designated as:

L is a partition length, U _P ^B(l, r) expression projector distance point of origin P border, the left and right sides farthest, Be all little section S _iThe set of feature, little hop count order of analyzing spot association is According to definition, the analyzing spot feature does not change with the observation pose and changes.

5.2.2 related little section feature of analyzing spot

The related little section S of analyzing spot _iActual is the section of cutting apart with local coordinate system information, its feature

Be designated as:

$F_{S_i}^{\′}=(L_{S_i},U_{S_i}^{\mathrm{\α}}({\mathrm{\δ}}_{S_i}^{\mathrm{\α}}-\sqrt{3}{\mathrm{\δ}}_{S_i}^{\mathrm{\α}}),C_{S_i},U_{S_i}^H(t,d))---\left(4\right)$

Represent little segment length,

Represent little section overall concave convexity, The distribution of expression analyzing spot projection vertical range,

Represent that little section interscan put relative α _PAverage drift angle,

The variance of expression drift angle.Consistent with the section of cutting apart feature, suppose that going up normal direction for little section satisfies consistent the distribution, related little section is reduced to feature line segment or circular arc,

Expression central angle scope, its position is by S _iWith

The decision, opening direction by

Decision.

The section of cutting apart feature F _SCan regard related little section feature as

Special shape:

5.2.3 related little section similarity of analyzing spot

Similar with the section of cutting apart similarity, little section similarity S ' _FBe to be used for when the coupling analyzing spot, describing different little section S ₁And S ₂Between similarity degree:

$S_F^{\′}=R_H\·R_L\·R_{\mathrm{\α}}\·\frac{1}{1+|C_{S_1}{\mathrm{\δ}}_{S_1}-C_{S_2}{\mathrm{\δ}}_{S_2}|}---\left(6\right)$

R _LRepresent little segment length ratio, R _δThe overlapping ratio of expression central angle scope, R _HThe overlapping ratio that the expression projection distributes.According to definition, S ' _F∈ [0,1] and S ' _FBig more similarity is high more.

5.2.4 analyzing spot matching degree

In this paper proposition method, accurately coupling is finished by the analyzing spot coupling.Arbitrary scan point P ₁And P ₂, its local coordinate system is overlaped little section correspondence of partial association then, the degree M that the two is complementary _PBe expressed as follows:

M _P＝(η ₁，η ₂，η ₃，η ₄)＝(S _F，O _B，S′ _F，O′ _B)(7)

S _FThe average similarity that expression is corresponding little section, O _BThe span U of expression analyzing spot feature _P ^BOverlapping ratio.With corresponding little section coupling seeing perform, the then S ' of similarity greater than certain threshold value _FAnd O ' _BRepresent to mate little section average similarity and proportion respectively.S _FAnd O _BPortrayed the average match condition of analyzing spot feature, and S ' _FAnd O ' _BIn the time of then can avoiding causing average portrayal relatively poor, think that mistakenly analyzing spot does not match because of the section of cutting apart localized variation.Best matching degree is described as M _Best=(1,1,1,1), the numeric representation of matching degree is as follows:

$V_{M_P}=\cos (\frac{\mathrm{\π}}{2}\·\underset{i=1}{\overset{4}{\mathrm{\Σ}}}{\mathrm{\ω}}_i{({\mathrm{\η}}_i-1)}^2),\underset{i=1}{\overset{4}{\mathrm{\Σ}}}{\mathrm{\ω}}_i=1---\left(8\right)$

And with M _BestApproaching more Big more matching degree is high more.ω in the formula _iBe expressed as the weighted value of matching degree vector.

F5.3 adopts the whole scan matching process (Global Scan MatchingMethod Using Probe-Verify-Solve Strategy) of exploration-checking-solution strategies

The main flow process of F5.3.1 algorithm

The scan-data LS that the different observation of input pose is gathered ₁And LS ₂, the algorithm purpose be obtain relative pose T=(x, y, θ), with LS ₂Be transformed into LS ₁Local coordinate system under, correct LS ₂Pose.Algorithm adopts the strategy of " souning out-verify-find the solution ", and main flow process is as follows:

1) pre-service: scan-data is cut apart estimation scan point normal direction.

2) initialization: all section of cutting apart features of initialization and analyzing spot feature, and the section of cutting apart corresponding lists L _Cor ^S

3) select: from L _Cor ^SIn choose the correspondence section of cutting apart successively

Carry out step 4)-6), finish until choosing, algorithm finishes.

4) sound out: from P _SIn choose analyzing spot, according to matching degree initialization scan point corresponding lists L _Cor ^PThe filtration corresponding point are right, if the too small step 3) that then goes to of remaining proportion; Otherwise, obtain the section of cutting apart relative pose T among the substitution solved function E (T) _S

5) checking: according to T _S, judge LS ₁And LS ₂Can be by correct coupling checking, by then proceeding to step 5); Otherwise forward step 3) to.

6) find the solution: according to T _SThe acquisition section of cutting apart corresponding lists L _Cor ^S', therefrom select the correspondence section of cutting apart successively, according to step 4) and 5) find the solution pose and checking, authentication failed is then from L _Cor ^S' remove, otherwise checking whether with T _SUnanimity, unanimity then continues, otherwise from L _Cor ^S' remove.Last L _Cor ^S' preserve P at least _S, filter corresponding point, find the solution scanning relative pose T.Empirical tests is if T separates excellent existing the separating that then substitute than having now; Go to step 3) then.

5.3.2 initialization with select

Usually the initialized time complexity of analyzing spot feature is low more for a long time as the Duan Yue of cutting apart of scan-data.The section of cutting apart corresponding lists L _Cor ^SComprised LS ₁And LS ₂It is right to go up all sections of cutting apart, initialization as follows: note LS ₁And LS ₂In the section of cutting apart number be respectively N ₁And N ₂L _Cor ^SMiddle element is divided into N ₁Piece, every N ₂To the section of cutting apart is according to LS ₂But in the section of cutting apart discrimination degree D _SSort from high to low, and the every element of going up same index is according to the section of cutting apart similarity S _FSort on earth by height.Initialization order guarantees at first to select the easiest distinguishing and the most similar section of cutting apart, and sounds out all sections of cutting apart as early as possible.

5.3.3 sound out

Analyzing spot corresponding lists L _Cor ^PWhat comprise is all analyzing spots and the corresponding analyzing spot the highest with its matching degree.The corresponding point of adjacent analyzing spot should be closed on, and the overlapping ratio that correct corresponding its projection of some feature distributes is relatively large.Utilize this constraint initialization L _Cor ^PCan avoid a large amount of unnecessary matching degrees to calculate.

Filter corresponding point to adopting following criterion:

1) filter according to the drift angle of normal direction: the corresponding point of filter false are right.

Suppose that correctly corresponding analyzing spot satisfies Gaussian distribution to the drift angle Δ α of normal direction.Usually the drift angle deviation that erroneous point is right is bigger.Statistics L _Cor ^PMiddle corresponding point are to the drift angle average μ of normal direction _{Δ α}With variance δ _{Δ α}, corresponding point are to basis Weighting.ω represents that matching degree is high more, but the high more then contribution of the high more discrimination degree of the similarity of the correspondence section of cutting apart is big more.The corresponding point that the drift angle do not satisfied certain degree of confidence are to removing.

2) just filter according to matching degree: the corresponding point of filtering the localized variation zone are right.

The corresponding point matching degree in localized variation zone is relatively low.Suppose that the same section of cutting apart goes up the right matching degree of corresponding point and satisfies Gaussian distribution, statistical match degree average and variance, it is right to keep the high corresponding point of matching degree, and the corresponding point that matching degree are lower than certain degree of confidence are to removing.

The above process of iteration is found the solution relative pose with left point to the following least squares equation of substitution until not having corresponding point to being removed:

$E\left(T\right)=\mathrm{\Σ}{\mathrm{\ω}}_i{\left[{\stackrel{\→}{n}}_{2i}(R_{\mathrm{\θ}}{P}_{1i}+t-P_{2i})\right]}^2---\left(9\right)$

T _S＝arg?min?E(T)(10)

P wherein _1iAnd P _2iRepresent i to corresponding point,

Expression point P _2iThe unit vector of normal direction,

And t=(x y) is P _1iTransform to P _2iThe rotation matrix of position and translation vector, ω _iConsistent with weighting in the aforementioned filter criteria.The distance of all analyzing spots to the corresponding point tangent line dwindled in the actual expression of E (T), and method for solving can adopt singular value decomposition, normal matrix or quaternions etc.Adopt point to allow analyzing spot along the tangential direction translation to tangent distance.When the section of cutting apart is irregular, separate unique, if line segment then can have countless separating.Because there is error in true laser, the tangential direction of all corresponding point is different, separates when abundant uniquely under the polyteny constraint when corresponding point, and matching result is that line segment overlaps.

5.3.4 checking

Above exploration process is to obtain relative pose by the section of cutting apart coupling, and position and attitude error is bigger if the corresponding mistake of the section of cutting apart then is bound to, and needs below to verify whether coupling can be accepted under the new pose:

1) analyzing spot to average match error ET less than certain threshold value;

2) the maximum ratio VT of the mutual visibility region of scanning room is greater than certain threshold value;

3) in the visibility region overlapping region ratio OT greater than certain threshold value.

As the maximum 0.5mm of average match error, at least one side has 40% zone to be seen by the opposing party, sees in the zone that at least 60% is overlapping.

The checking result

Numeric representation is as follows, And the big more coupling that shows is good more:

$V_{T_S}=\cos \left({\stackrel{\‾}{E}}_T\right)\sin ({\stackrel{\‾}{V}}_T\·\frac{\mathrm{\π}}{2})\sin ({\stackrel{\‾}{O}}_T\·\frac{\mathrm{\π}}{2})---\left(11\right)$

5.3.5 find the solution

If trial solution T _SBy checking, then the angular range that covers according to the section of cutting apart on its basis obtains the section of cutting apart corresponding lists L _Cor ^S'.Find the solution L successively _Cor ^SThe relative pose of ' middle correspondence the section of cutting apart more then thinks inconsistent if find the solution the angle and the trial solution deviation of pose.These sections of cutting apart have been represented wrong corresponding or mobile object usually.Corresponding point in the residue section of cutting apart are filtered according to the matching degree size, and guarantee point is not to existing one-to-many or many-to-one situation.Then left point is found the solution to substitution E (T) that the scanning room relative pose changes T and checking is separated.Can not overlapping areas then be the zone that the part changes on the section of cutting apart continuously in the checking, the Duan Ze of cutting apart that can not mate be the object that moves.Getting the conduct of separating of checking end value maximum finally separates.Because the limited amount of the section of cutting apart and orderly, the exploration process is limited, if the complete difference of scanning is then mated always failure, otherwise method always can find correct matched position.

Claims (10)

1. the whole scan matching process based on how much statistical natures is characterized in that step comprises the pre-service of (1) laser sensor data; (2) feature of the data section of cutting apart after the extraction pre-service; (3) feature of laser scanning point after the extraction pre-service; (4) adopt the strategy of " souning out-verify-find the solution " to obtain relative pose between different scanning.Wherein step (1) in step (2) (3) (4) before; Step (2) (3) order is not limit, and can put upside down, can walk abreast; Step (4) in the end.

2. the whole scan matching process based on how much statistical natures as claimed in claim 1 is characterized in that comprising in the pre-service of its step (1) laser sensor data (1.1) laser data and cuts apart and two steps of (1.2) laser scanning point normal direction estimation.

3. the whole scan matching process based on how much statistical natures as claimed in claim 1 is characterized in that the definition of the data section of cutting apart feature in its step (2) meets following tlv triple:

$F_S=(L_S,{\mathrm{\δ}}_S^{\mathrm{\α}},C_S)$

L wherein _SThe total length of the expression section of cutting apart, δ _S ^αThe variance that expression analyzing spot normal direction changes, C _SThe whole concavity and convexity (being expressed as recessed, flat, protruding) of the expression section of cutting apart.At F _STlv triple represent L _S, δ _S ^α, C _SThere is no sequencing, the data section of cutting apart characterizing definition is all satisfied in its random order.

4. the whole scan matching process based on how much statistical natures as claimed in claim 1 is characterized in that the definition of laser scanning point feature in its step (3) meets following tlv triple:

Normal direction and tangential direction with analyzing spot P are set up local coordinate system, and wherein L is for dividing the length of the section of cutting apart, U along the P tangential direction _P ^B(l, r) the expression section of cutting apart goes up normal direction border, the left and right sides l and the r farthest of other analyzing spot distance P,

Be the set of the related little section Si feature of all analyzing spots, the number that analyzing spot is related little section is

F _PThe tlv triple that is comprised there is no sequencing, and the laser scanning point characterizing definition is all satisfied in random order.

5. the whole scan matching process based on how much statistical natures as claimed in claim 4 is characterized in that in the laser scanning point characterizing definition tlv triple

Be the set of the related little section Si feature of all analyzing spots, wherein the definition of the related little section feature of analyzing spot meets following four-tuple:

$F_{S_i}^{\′}=(L_{S_i},U_{S_i}^{\mathrm{\α}}({\mathrm{\μ}}_{S_i}^{\mathrm{\α}}-\sqrt{3}{\mathrm{\δ}}_{S_i}^{\mathrm{\α}},{\mathrm{\μ}}_{S_i}^{\mathrm{\α}}+\sqrt{3}{\mathrm{\δ}}_{S_i}^{\mathrm{\α}}),C_{S_i},U_{S_i}^H(t,d))$

Wherein

Represent little segment length,

Represent little section overall concave convexity,

The distribution of expression analyzing spot projection vertical range,

Represent that little section interscan put the average drift angle of relative local coordinate system,

The variance of expression drift angle.Four-tuple there is no sequencing, and the related little section characterizing definition of analyzing spot all satisfied in random order.

6. the whole scan matching process based on how much statistical natures as claimed in claim 1, the exploration process is for to choose a pair of section of cutting apart from different scanning rooms in it is characterized in that " souning out-verify-find the solution " in its step (4), seek corresponding analyzing spot according to the section of cutting apart similarity and analyzing spot matching degree, utilize least square method or Newton method to find the solution and cut apart intersegmental relative pose as trial solution.

7. the whole scan matching process based on how much statistical natures as claimed in claim 6 is characterized in that the section of cutting apart similarity wherein calculates in such a way:

$S_F=R_L\·R_{\mathrm{\δ}}\·\frac{1}{1+|C_{S_1}{\mathrm{\δ}}_{S_1}^{\mathrm{\α}}-C_{S_2}{\mathrm{\δ}}_{S_2}^{\mathrm{\α}}|}$

Wherein

With

The concavity and convexity of the sections of cutting apart of expression difference respectively,

With

The variation variance of representing normal direction respectively, R _LSegment length ratio, R are cut apart in expression _δThe expression normal direction changes the ratio of variance.

8. the whole scan matching process based on how much statistical natures as claimed in claim 6 is characterized in that analyzing spot matching degree wherein calculates in such a way:

$S_F^{\′}=R_H\·R_L\·R_{\mathrm{\α}}\·\frac{1}{1+|C_{S_1}{\mathrm{\δ}}_{S_1}-C_{S_2}{\mathrm{\δ}}_{S_2}|}$

M _P＝(η ₁，η ₂，η ₃，η ₄)＝(S _F，O _B，S′ _F，O′ _B)

$V_{M_P}=\cos (\frac{\mathrm{\π}}{2}\·\underset{i=1}{\overset{4}{\mathrm{\Σ}}}{\mathrm{\ω}}_i{({\mathrm{\η}}_i-1)}^2),\underset{i=1}{\overset{4}{\mathrm{\Σ}}}{\mathrm{\ω}}_i=1$

S ' wherein _FThe related little section similarity of expression analyzing spot,

With

The concavity and convexity that expression difference respectively is little section,

With

The variance of representing the normal drift angle respectively, R _HThe overlapping ratio that the expression projection distributes, R _LRepresent little segment length ratio, R _δThe overlapping ratio of expression central angle scope.And four-tuple M _PMatching result between different scanning has been described, S _FThe average similarity that expression is corresponding little section, O _BThe span U of expression analyzing spot feature _P ^BOverlapping ratio, S ' _FAnd O ' _BRepresent to mate little section average similarity and proportion respectively. It then is the calculated value of final analyzing spot matching degree.

9. the whole scan matching process based on how much statistical natures as claimed in claim 1, proof procedure comprises in following three any several for judge trial solution correctness and the quality of separating, criterion according to certain criterion in it is characterized in that " souning out-verify-find the solution " in its step (4): (a) analyzing spot is to average match error E _TLess than certain threshold value; (b) the maximum ratio V of the mutual visibility region of scanning room _TGreater than certain threshold value; (c) ratio O in overlapping region in the visibility region _TGreater than certain threshold value.Evaluation of result calculates according to following formula:

$V_{T_S}=\cos \left({\stackrel{\‾}{E}}_T\right)\sin ({\stackrel{\‾}{V}}_T\·\frac{\mathrm{\π}}{2})\sin ({\stackrel{\‾}{O}}_T\·\frac{\mathrm{\π}}{2})$

10. the whole scan matching process based on how much statistical natures as claimed in claim 1, solution procedure is for to be transformed into different scanning under the same coordinate system on the basis of trial solution in it is characterized in that " souning out-verify-find the solution " in its step (4), the angular range that covers according to the section of cutting apart obtains the correspondence section of cutting apart, find the solution the relative pose of all correspondence sections of cutting apart, filter wherein solving result and the inconsistent corresponding section of cutting apart of trial solution, utilize the corresponding analyzing spot on the corresponding section of cutting apart of residue, utilize least square method or Newton method etc. to find the solution the scanning room relative pose.The trial solution that the checking evaluation of result is the highest is as finally separating.

Images