CN105957342A - Lane-level road mapping method and system based on crowdsourcing space-time big data - Google Patents

Abstract

The invention provides a lane-level road mapping method based on crowdsourcing space-time big data, and the method comprises the steps: building a similarity estimation model of a track vector, carrying out the track optimization based on a growth clustering method integrated with experiential knowledge, building a Gaussian constraint mixed model, and employing an EM algorithm to solve model parameters; detecting lane information, and obtaining an initial detection result of the number of road lanes; correcting the initial detection result based on a road construction rule; and correcting the central lines of the lanes according to the corrected number of lanes and the conditions of the surroundings. The method reduces the cost in obtaining the urban precise road information, is simple, and is easy to implement.

Description

Tracks based on the big data of mass-rent space-time level road plotting method and system

Technical field

The present invention relates to the high accuracy track level road mapping of the big data of mass-rent space-time, belong to GIS-Geographic Information System and grind with intelligent transportation Study carefully field.

Background technology

Road-map is blood and the soul of following automatic Pilot in high precision, and level fine road information in track is to build high accuracy road The key components of road map.At present, researcher proposes view-based access control model method and extracts lane line from high-resolution remote sensing image, Also or utilize high-precision laser cloud data obtain lane specific information, and from measure car gather a large amount of high-precision GPS rails Mark extracting data road pavement information (lane boundary line, track quantity, lane center).Rogerset al. (1999) is Attempt the earliest utilizing space-time DGPS (DGPS) track data to extract grinding of road axis and lane line One of the person of studying carefully.Subsequently on the basis of Rogers et al. studies, space-time GPS track data acquisition road information is utilized gradually to develop Become a kind of end-to-end pattern.The road information of this end-to-end pattern obtains can be summarized as following several process: the most right DGPS track data is optimized, then by DGPS track data and existing map data matching, in spline curve fitting road Heart line, extracts lane information and the geometry of refinement crossing finally by clustering method.It is de-that John Krumm proposes one From the road information obtaining mode of original map, this pattern is classified initially with track and fusion method is from a large amount of DGPS track datas Middle extraction road grade information, then utilizes gauss hybrid models to extract car from a large amount of track datas belonging to each section Road information.But, these approach obtaining track fine road informations of level all exist data acquisition cost height, acquisition time length, The shortcomings such as renewal speed is slow, data processing complex.

Along with developing rapidly of sensor technology, radio communication and network technology, " everybody is sensor ", people go out guild Produce a large amount of big data of space-time track, contain abundant fine road information and behavior of men action message.Adopting of track data Collection gradually by specialized department measure car or professional gather develop into by layman the most voluntary record its trip track Form, the collection of data starts to be changed into mass-rent pattern.Vehicle-mounted track data (the big data of mass-rent) under mass-rent pattern is undoubtedly It it is the optimum data source that track level road information can be provided at present to extract.Compared with existing taxi data, by mass-rent pattern The vehicle-mounted track data gathered belongs to big data, and (big data refer to enter with conventional software instrument in the time range that can bear Row catches, the data acquisition system that manages and process, be need new tupe just can have higher decision edge, see clearly discovery power and Process optimization ability adapts to magnanimity, high growth rate and diversified information assets).Current domestic scholars Tang stove is bright et al. (2015, 2016) propose to utilize low precision GPS track data to extract track, city level road information, turn to including track quantity, track, car Road centrage, but how to utilize the big data of mass-rent, carrying out track level fine road mapping is the difficulty that whole world scientists faces Topic.

Summary of the invention

The present invention is on the basis of above research, it is proposed that a kind of high accuracy tracks based on the big data of mass-rent space-time level road mapping The new solution of (high-quality track data filtering and high accuracy road information extract).

Technical solution of the present invention provides a kind of tracks based on the big data of mass-rent space-time level road plotting method, comprises the following steps,

Step 1, sets up the similarity evaluating model of track vector, if v_aAnd v_bIt is two different track vector, described similarity Evaluation model is as follows,

${\mathrm{sim}}_{(v_a,v_b)}={\mathrm{\ω}}_1{e}^{-{\mathrm{diff}}_{Hd}}+{\mathrm{\ω}}_2{e}^{-{\mathrm{diff}}_{\mathrm{\θ}ab}}$

Wherein,Representing the Similarity value between vector, e is the nature truth of a matter, ω₁And ω₂Represent distance factor diff respectively_Hd With angular factors diff_θabWeighted value, and ω₁+ω₂=1；Distance factor diff_HdWith angular factors diff_θabRepresent vector v respectively_a And v_bDistance difference and angle difference；

Step 2, carries out track based on the growth clustering method merging Heuristics preferred, including according to existing high-precision GPS rail Mark data and synchronization low precision GPS track data, determine the weighted value ω of similarity evaluation model₁And ω₂, extract track preferred Priori, using growth cluster mode based on the similarity between mass-rent track data, to carry out data preferred；

Step 3, builds Gauss and retrains mixed model, and use EM Algorithm for Solving model parameter；Described Gauss retrains hybrid guided mode Type is defined as follows,

$p\left(x\right)=\underset{j=1}{\overset{k}{\Σ}}{\mathrm{\ω}}_j\frac{1}{\sqrt{2{\mathrm{\π\σ}}^2}}\exp (-\frac{{(x-{\mathrm{\μ}}_j)}^2}{2{\mathrm{\σ}}^2})$

Wherein, p (x) is expressed as Gauss and retrains the combined chance value of mixed model, and x represents sample value to be calculated, is carrying out track meter During calculation, x represents and judges tracing point ordinate value of upright projection on its vertical section in window；K is the quantity of gauss component, often One corresponding track of gauss component；ω_jIt is the weight of jth gauss component, the traffic flow in corresponding track；Parameter μ₁…μ_k It is the meansigma methods of track in each gauss component, equal to the centrage in each track, μ_jRepresent μ₁…μ_kAny one in parameter Individual value, j=1,2 ..., k；σ is the standard deviation of track in each gauss component；

It is that the value of computation structure risk model, with knot that described Gauss retrains the quantity k acquisition mode of gauss component in mixed model The minimum principle of structure risk model value determines k；

Step 4, according to step 3 acquired results, detects lane information, obtains the road section track first result of detection of quantity； Implementation is as follows,

Using all tracks of being on same section as an extraction unit, if given one group from crossing Intersection₁ To crossing Intersection₂Track set A^T, from track set A^TOne end start, build mobile rectangular window, wherein The long limit of mobile rectangular window is parallel to the centrage currently covering all tracks, and the broadside of mobile rectangular window is then perpendicular to currently Covering the centrage of all tracks, the central axis on the long limit of rectangular window is in the centrage of its all track datas covered, square The centrage of the track data that the center line of shape window broadside covers with it overlaps；

Rectangular window will be moved from track set A^TOne end start, according to the long edge lengths of rectangular window start translation, successively profit Track quantity and the lane center in the section covered is retrained in mixed model detects each rectangular window, including root by Gauss According to mobile rectangular window, project to, on the long limit centrage of rectangular window, be thrown by all tracing points in mobile rectangular window The track data collection X=(x of movie queen₁,x₂,…,x_N), t=1,2,3 ..., N, wherein, x_tRepresent the vertical seat of t tracing point after projecting Scale value, N is the number participating in projected footprint point；Track data collection X substitution Gauss is retrained mixed model, extracts rectangular window The track quantity in interior section and lane center；Assume from crossing Intersection₁To crossing Intersection₂Track collection Close A^T, rectangular window has carried out altogether l translation, has translated the track quantity determined each time and be designated as Nlane_f, f=1,2 ..., l, As the road section track first result of detection of quantity；

Step 5, the road section track first result of detection of quantity obtained according to step 4, based on road construction rule, to just Secondary result of detection is modified；

Step 6, the revised track quantity obtained according to step 5, according to adjacencies, lane center is modified.

And, first result of detection is modified by step 5, it is achieved mode is as follows,

The first step, in step 4, certain translates track quantity Nlane determined_f, compare Nlane_f+1And Nlane_f、 Nlane_f+2If, Nlane_fAnd Nlane_f+1It is different, then uses Nlane_fReplace Nlane_f+1, f=1,2 ..., l-2；

Second step, according to Nlane_fValue and distribution the result of the first step is classified, if there is s class, be designated as C_g=< Nl_g,nc_g>, Nl_gIt is class bunch C_gTrack quantity, nc_gIt is Nlane_gIn belong to C_g, g=1,2 ..., the total quantity of s；

3rd step, compares C_g+1And C_gIf, Nl_g+1It is different from Nl_g, and nc_g+1< cv makes C_gNl_gReplace C_g+1's Nl_g+1, g=1,2 ..., s, complete the final optimization pass of track quantity result, wherein, cv is default threshold value.

And, lane center is modified by step 6, it is achieved mode is as follows,

If the track quantity of a certain section of section La is corrected, if La adjacent segments Lb and Lc meets has phase with La simultaneously Do not change with revised track quantity before same track quantity, and correction, then just by the lane center of Lb Yu Lc Line connects, and obtains the final lane center of La；If with correction before the track quantity correction of La adjacent segments Lb or Lc After also change, then just according to La revise before extract lane center line position, calculate based on La revise before lane center Line position calculates the road axis in La section, after after revising according to La, lane width and track quantity redefine La correction Lane center line position.

The present invention provides a kind of tracks based on the big data of mass-rent space-time level road mapping system, including with lower module,

First module, for setting up the similarity evaluating model of track vector, if v_aAnd v_bIt is two different track vector, described Similarity evaluation model is as follows,

${\mathrm{sim}}_{(v_a,v_b)}={\mathrm{\&omega;}}_1{e}^{-{\mathrm{diff}}_{Hd}}+{\mathrm{\&omega;}}_2{e}^{-{\mathrm{diff}}_{\mathrm{\&theta;}ab}}$

Wherein,Representing the Similarity value between vector, e is the nature truth of a matter, ω₁And ω₂Represent distance factor diff respectively_Hd With angular factors diff_θabWeighted value, and ω₁+ω₂=1；Distance factor diff_HdWith angular factors diff_θabRepresent vector v respectively_a And v_bDistance difference and angle difference；

Second module, preferred, including according to existing high-precision for carrying out track based on the growth clustering method merging Heuristics Degree GPS track data and synchronization low precision GPS track data, determine the weighted value ω of similarity evaluation model₁And ω₂, extract The preferred priori of track, uses growth cluster mode to carry out data based on the similarity between mass-rent track data preferred；

Three module, is used for building Gauss and retrains mixed model, and use EM Algorithm for Solving model parameter；Described Gauss retrains Mixed model is defined as follows,

$p\left(x\right)=\underset{j=1}{\overset{k}{\&Sigma;}}{\mathrm{\&omega;}}_j\frac{1}{\sqrt{2{\mathrm{\&pi;\&sigma;}}^2}}\exp (-\frac{{(x-{\mathrm{\&mu;}}_j)}^2}{2{\mathrm{\&sigma;}}^2})$

Wherein, p (x) is expressed as Gauss and retrains the combined chance value of mixed model, and x represents sample value to be calculated, is carrying out track meter During calculation, x represents and judges tracing point ordinate value of upright projection on its vertical section in window；K is the quantity of gauss component, often One corresponding track of gauss component；ω_jIt is the weight of jth gauss component, the traffic flow in corresponding track；Parameter μ₁…μ_k It is the meansigma methods of track in each gauss component, equal to the centrage in each track, μ_jRepresent μ₁…μ_kAny one in parameter Individual value, j=1,2 ..., k；σ is the standard deviation of track in each gauss component；

It is that the value of computation structure risk model, with knot that described Gauss retrains the quantity k acquisition mode of gauss component in mixed model The minimum principle of structure risk model value determines k；

4th module, for according to three module acquired results, detects lane information, obtains road section track quantity and visit for the first time Survey result；Implementation is as follows,

Using all tracks of being on same section as an extraction unit, if given one group from crossing Intersection₁ To crossing Intersection₂Track set A^T, from track set A^TOne end start, build mobile rectangular window, wherein The long limit of mobile rectangular window is parallel to the centrage currently covering all tracks, and the broadside of mobile rectangular window is then perpendicular to currently Covering the centrage of all tracks, the central axis on the long limit of rectangular window is in the centrage of its all track datas covered, square The centrage of the track data that the center line of shape window broadside covers with it overlaps；

Rectangular window will be moved from track set A^TOne end start, according to the long edge lengths of rectangular window start translation, successively profit Track quantity and the lane center in the section covered is retrained in mixed model detects each rectangular window, including root by Gauss According to mobile rectangular window, project to, on the long limit centrage of rectangular window, be thrown by all tracing points in mobile rectangular window The track data collection X=(x of movie queen₁,x₂,…,x_N), t=1,2,3 ..., N, wherein, x_tRepresent the vertical seat of t tracing point after projecting Scale value, N is the number participating in projected footprint point；Track data collection X substitution Gauss is retrained mixed model, extracts rectangular window The track quantity in interior section and lane center；Assume from crossing Intersection₁To crossing Intersection₂Track collection Close A^T, rectangular window has carried out altogether l translation, has translated the track quantity determined each time and be designated as Nlane_f, f=1,2 ..., l, As the road section track first result of detection of quantity；

5th module, for the road section track first result of detection of quantity obtained according to the 4th module, advises based on road construction Then, first result of detection is modified；

6th module, for the revised track quantity obtained according to the 5th module, enters lane center according to adjacencies Row is revised.

And, first result of detection is modified by the 5th module, it is achieved mode is as follows,

The first step, in the 4th module, certain translates track quantity Nlane determined_f, compare Nlane_f+1And Nlane_f、 Nlane_f+2If, Nlane_fAnd Nlane_f+1It is different, then uses Nlane_fReplace Nlane_f+1, f=1,2 ..., l-2；

Second step, according to Nlane_fValue and distribution the result of the first step is classified, if there is s class, be designated as C_g=< Nl_g,nc_g>, Nl_gIt is class bunch C_gTrack quantity, nc_gIt is Nlane_gIn belong to C_g, g=1,2 ..., the total quantity of s；

3rd step, compares C_g+1And C_gIf, Nl_g+1It is different from Nl_g, and nc_g+1< cv makes C_gNl_gReplace C_g+1's Nl_g+1, g=1,2 ..., s, complete the final optimization pass of track quantity result, wherein, cv is default threshold value.

And, lane center is modified by the 6th module, it is achieved mode is as follows,

If the track quantity of a certain section of section La is corrected, if La adjacent segments Lb and Lc meets has phase with La simultaneously Do not change with revised track quantity before same track quantity, and correction, then just by the lane center of Lb Yu Lc Line connects, and obtains the final lane center of La；If with correction before the track quantity correction of La adjacent segments Lb or Lc After also change, then just according to La revise before extract lane center line position, calculate based on La revise before lane center Line position calculates the road axis in La section, after after revising according to La, lane width and track quantity redefine La correction Lane center line position.

The present invention has constructed the high accuracy track level road surveying & mapping scheme of a kind of big data of mass-rent space-time, reduces acquisition city The cost of fine road information, and detection method is simply, easily realize.Technical scheme provided by the present invention includes: first, logical Cross the space similarity between the contrast high accuracy synchronous low precision GPS track data of vehicle-mounted GPS track, use based on warp Test the growth clustering method of knowledge, from mass-rent track data, pick out the data that positioning precision is of a relatively high；Secondly, structure hangs down Straight in the moving window of track data；Then, the institute's rail to being in road section of the gauss hybrid models method after optimizing is used Mark carries out longitudinal probing, obtains the track quantity in detection window；Further, utilize road construction rule, that is, same Section only there will be the situation setting up track close to position, crossing, and mid portion track, section quantity generally remains constant, Propose track quantity optimisation strategy, track quantity information is modified；Finally, utilize revised track quantity information, right The lane center extracted is modified, and completes the extraction of respective stretch track level road information.The number of track-lines that the present invention obtains Amount judgment accuracy is 85%, and the positioning precision of lane center, at about 0.35m, reduces the fine road information in acquisition city Cost, and detection method is simply, easily realize.

Accompanying drawing illustrates:

Fig. 1 is the method flow diagram of the embodiment of the present invention；

Fig. 2 is the track vector similarity schematic diagram of the embodiment of the present invention；

Fig. 3 is growth based on the Heuristics cluster schematic diagram of the embodiment of the present invention；

Fig. 4 is the growth clustering method track preferred result schematic diagram based on Heuristics of the embodiment of the present invention, wherein Fig. 4 a For mass-rent track data Experimental Area schematic diagram, Fig. 4 b track preferred result schematic diagram；

Fig. 5 is gauss hybrid models and the lane center position sensing schematic diagram of the embodiment of the present invention, and wherein Fig. 5 a is that Gauss mixes Matched moulds type optimum results, Fig. 5 b is lane center position sensing result；

Fig. 6 is the structure rectangular window detection lane information schematic diagram of the embodiment of the present invention；

Fig. 7 is the track quantity optimization schematic diagram of the embodiment of the present invention；

Fig. 8 is the track quantity result of detection schematic diagram of the embodiment of the present invention；

Fig. 9 is the track quantity optimum results schematic diagram of the embodiment of the present invention.

Detailed description of the invention

Technical solution of the present invention is described in detail below in conjunction with drawings and Examples.

The present invention provides the high accuracy track level road plotting method of a kind of big data of mass-rent space-time, sees Fig. 1, and embodiment includes Following steps:

Step 1, sets up the similarity evaluating model of track vector.Travelling feature according to vehicle, general driver is according to driving rule Can travel along lane center.Therefore, under the conditions of not considering of short duration change track behavior, can truly portray driver and drive The high-precision track data sailing track is generally focused near lane center, and the spacing of adjacent track is less than lane width, Course angle angle between track levels off to about 0 °.In order to assess between this track belonged to around same lane center Similarity, the present invention establishes a kind of new track vector similarity evaluating model, and this similarity model is between track vector Vertical dimension and angle two aspect carry out measuring similarity, during wherein track vector refers to a track by each tracing point and The unit track vector that its course angle is constituted, namely the mould of these track vector is the most identical, and can arbitrarily define.Two tracks The difference of vector is measured in terms of direction and distance two.As in figure 2 it is shown, N represents direct north, v_a<(x_a,y_a),(x_a+1,y_a+1)> And v_b<(x_b,y_b),(x_b+1,y_b+1) > it is two different track vector, vector v_aAnd v_bAzimuth be θ respectively_aAnd θ_b, similarity is commented Valency model is as follows:

Wherein,Represent the Similarity value between vector, andTwo are represented when Similarity value is 1 Track vector is identical, and Similarity value is to represent when 0 that two track vector are the most dissimilar；E is the nature truth of a matter；ω₁And ω₂Point Biao Shi the distance factor (diff_Hd) and angular factors (diff_θab) weighted value, and ω₁+ω₂=1；diff_HdAnd diff_θabTable respectively Show vector v_aAnd v_bDistance difference and angle difference.Formula 8 defines distance difference diff of two track vector_Hd, formula 9 defines Angle difference diff of two track vector_θab:

Dis in formula 8_confineDetermined by the width in track, be a constant, be used for retraining on identical track near each track The similarity of the GPS track of centrage.Hd_abVector v_aStarting point is to v_bThe vertical dimension of starting point, computing formula such as formula 11 institute Show；Hd_baIt it is vector v_bStarting point is to vector v_aThe vertical dimension of starting point, computing formula is as shown in Equation 12；△ θ is vector v_a With vector v_bCourse heading difference, wherein vector v_aAnd v_bAzimuth be θ respectively_aAnd θ_b, computing formula is as follows:

Δ θ=| θ_a-θ_b| formula 10

Step 2, carries out track based on the growth clustering method merging Heuristics preferred.Its track vector of high-precision tracing point Between Similarity value the most of a relatively high.When carrying out track and being preferred, it is necessary first to existing high-precision GPS track data and its Synchronize low precision GPS track data, extract the preferred priori of track.In embodiment, DGPS track and synchronization in high precision The precision of low precision track be respectively 0.5m and 10-15m, sample frequency is 1s.The similarity set up according to step one is estimated Calculate model and calculate low precision track and the similarity of DGPS track, carry out similarity evaluation Model Weight value and determine and Heuristics Obtain.

1) similarity evaluation Model Weight value determines

Similarity is applied not only in DGPS and low precision gps data extract priori, but also for many source data clusters with excellent Choosing, therefore present invention further propose that and uses vertical dimension, the angle difference dependency with measurement error to estimate ω₁And ω₂Value.

For GPS track set T=< Trace₁,Trace₂,…,Trace_s>, it comprises s bar GPS track: Trace₁,Trace₂,…, Trace_s, the synchronization high accuracy DGPS track of track set T is expressed as DT=< Dt₁,Dt₂,…,Dt_s>, Dt₁,Dt₂,…,Dt_sIt it is collection Close the track data in DT.The precision of track set DT and track set T is 0.5m and 10-15m respectively.Assume i-th GPS rail Mark Trace_i=< p₁,p₂,…,p_n>, p₁,p₂,…,p_nRepresent track Trace respectively_iTracing point, total n tracing point；Dt_i=< rp₁, rp₂,…,rp_n>, rp₁,rp₂,…,rp_nIt is then track Trace_iSynchronization high accuracy track Dt_iTracing point；Trace_i∈T,Dt_i∈ DT, i=1,2 ..., s.By track Trace_iHigh accuracy track Dt is synchronized with it_iInterior tracing point, the track vector of composition is expressed as: Tv_i=< v₁,v₂,…,v_n-1>, v₁,v₂,…,v_n-1It is expressed as by track Trace_iTracing point p₁With p₂,…,p_n-1With p_nConstitute Track vector；Dv_i=< rv₁,rv₂,…,rv_n-1>, rv₁,rv₂,…,rv_n-1It is expressed as by track Trace_iSynchronization high accuracy rail Mark Dt_iTracing point rp₁With rp₂,…,rp_n-1With rp_nThe track vector constituted.Tv_iAnd Dv_iDistance and angle be expressed as: D_i=< d₁,d₂..., d_n-1>,d₁,d₂..., d_n-1Represent track vector set Tv_iAnd Dv_iInterior corresponding track vector v₁,v₂,…, v_n-1To rv₁,rv₂,…,rv_n-1Distance；A_i=< a₁,a₂..., a_n-1>,a₁,a₂..., a_n-1Represent track vector set Tv_iWith Dv_iInterior corresponding track vector v₁,v₂,…,v_n-1With rv₁,rv₂,…,rv_n-1Angle difference, i=1,2 ..., s.Track set T The position error of interior all of track data can be expressed as E_i=< ε₁,ε₂,…,ε_n>, wherein E_iRepresent the track in track set T Trace_iThe synchronization high accuracy track Dt that all tracing points are corresponding_iThe space length of all tracing points, ε_jIt is set E_iInterior appoints Anticipate error amount, wherein a ε_j=| p_j-rp_j|, p_jFor Trace_iAny one tracing point interior, rp_jFor p_jCorresponding high accuracy tracing point, I=1,2 ..., s, j=1,2 ..., n.In similarity evaluation model, weights ω₁And ω₂Computing formula respectively following (r_DεAnd r_AεPoint Do not represent D_iAnd E_iCorrelation coefficient, can be based on covariance matrix A_iAnd E_iValue):

ω₂=1-ω₁Formula 14

2) priori is extracted

The first step: according to step 1 propose vector similarity evaluation model and the determination of weighted value, calculate high accuracy DGPS with It synchronizes the similarity between low precision GPS, and position and its synchronization by contrasting low precision GPS track data are high-precision simultaneously The position of degree DGPS data, calculates the measurement error of low precision gps data；

Second step, the similarity obtained according to the first step and the GPS measurement error of its correspondence, build and belong to same GPS The attribute description pair of track data, namely (Similarity value, GPS measurement error) be the attribute description pair of some track.

3rd step, internal from the attribute description of gps data, according to similarity threshold, namely from the beginning of Similarity value is 0.5, Select Similarity value more than 0.5, more than 0.6, more than 0.7 ..., all gps datas more than 0.9, and add up and belong to each phase Like the GPS measurement error of all gps datas in degree threshold value, and calculate the meansigma methods of these GPS measurement error and meet this The number of the data of a little threshold values accounts for the ratio of conceptual data.

4th step, the similarity threshold that the 3rd step is set (such as: similarity threshold is more than 0.5, similarity threshold is more than 0.6 ... .) It is defined as: Ts_h, h=1,2,3,4,5；It is Ts by the mean value definition meeting the GPS measurement error of these threshold values_h, h=1,2,3,4,5； The ratio that the GPS measurement error meeting threshold value accounts for conceptual data is defined as: Per_h, h=1,2,3,4,5；Complete priori to obtain Take.

Wherein, ST_hRepresent and meet Ts_hData set, ST_h∈ T, T are the conceptual data collection of the experimental data extracted for experience, ST_hPercentage calculation formula as follows:

Per_hRepresent ST_hPercentage ratio, N (ST_h) and N (T) be to be ST respectively_hWith the quantity of T tracing point, ST_hComputing formula be:

Tε_hIt is Ts_hMeasurement error, ∑ ε is ST_hIn the sum of all tracing point errors, h=1,2 ..., 5, Ts_h,Tε_h, Per_hIt is designated as RST_h=< Ts_h,Tε_h,Per_h>,RST_hFor priori set, as the priori of growth clustering method.

According to the priori extracted, the distance weights in similarity evaluation model and angle weights are calculated, is then based on It is preferred that similarity between mass-rent track data uses growth clustering method to carry out data, and wherein the T ε in priori is cluster Threshold value, Per is used for choosing the ratio of high-precision data from whole track bunch, sees Fig. 3 (v in Fig. 3_sRepresent seed to Amount, v_snExpression and Seeding vector carry out any one vector of Similarity Measure, Cluster₁, Cluster₂, Cluster₃Represent logical The several track vector classes bunch obtained after crossing cluster；In Fig. 3 in (d) part ' Selected data ' then represents the rail being finally selected Mark data), the key step of growth clustering method is as follows:

The first step: initialize all of track vector, is labeled as not clustering；Initialize the numbering (CC of current class bunch_L), remember CC_L =1；

Second step: if there is the track vector not clustered, then randomly choose a track from remaining cluster track vector Vector is as seed track vector v_s, the cluster labelling of seed track vector is CL (v_s), and CL (v_s)=CC_L, in Fig. 3 A () part, enters the 3rd step；If all tracks have been clustered, such as (c) part in Fig. 3, then enter the 5th step；

3rd step: search v_sThe track vector closed on, is designated as v_snIf: meet Sim (v_s,v_sn) > T ε, by v_sAnd v_snPermeate Individual track bunch, v_snCluster labelling be designated as CL (v_sn), and CL (v_sn)=CC_L, enter the 4th step；If can not find with currently Seeding vector v_sWhen meeting the track vector of similarity threshold, make CC_L=CC_L+ 1, return second step；(Sim(v_s,v_sn) it is v_sWith v_snSimilarity value, T ε is similarity threshold)

4th step: make track vector v_snAs kind of a sub-trajectory v_s, return the 3rd step, such as (b) part in Fig. 3；

5th step: according to the operation result of the first step to the 4th step, namely final cluster class bunch, calculate track in all classes bunch Point quantity accounts for the ratio participating in cluster tracing point sum, and then counting all of track bunch according to its track accounts for overall track and count Ratio size inverted order arranges, and wherein Per represents the preferred selectance of data, starts accumulation from the tracing point proportional of first class bunch Summation is until meet Per, and these classes bunch participating in accumulation summation are preferably chosen as high accuracy data, such as (d) part in Fig. 3, The Experimental Area mass-rent track data such as Fig. 4 a chosen, the preferred result of track such as Fig. 4 b.

Step 3, builds Gauss and retrains mixed model, and use EM Algorithm for Solving model parameter.Retrain mixed in existing Gauss On the basis of hop algorithm asks for the algorithm of lane information, it is optimized by the present invention.Gauss retrains mixed model and is defined as follows:

Wherein, p (x) is expressed as Gauss and retrains the combined chance value of mixed model, and x represents that sample value to be calculated (is carrying out track meter During calculation, x represents and judges tracing point ordinate value of upright projection on its vertical section in window)；K is the quantity of gauss component, Namely the number of Gaussian peak in constraint gauss hybrid models, represent the quantity in track, the corresponding track of each gauss component； ω₁…ω_kBeing the weight of each composition, the traffic flow in corresponding each track, wherein weighted value is just and to be standardized, i.e. ω_jIt is The weight of jth gauss component, ω_j> 0, j=1,2 ..., k, ω₁+ω₂+…ω_k=1；Parameter μ₁…μ_kIt is in each gauss component The meansigma methods of track, equal to the centrage in each track, μ_jRepresent μ₁…μ_kAny one value in parameter, j=1,2 ..., k；σ It is the standard deviation of track in each gauss component, and owing to each track is the most identical, therefore σ with the width in neighbouring track Being set to a constant, (according to domestic road construction standard, lane width is generally about 3.75m, at tool to be typically set at 1.75 In body implementation process, those skilled in the art can reset according to selected areas road construction standard).Use EM method Solving model parameter: θ_j ^(m)(ω_j ^(m),μ_j ^(m),σ^(m)), j=1,2 ..., k, wherein m is iterations.How to utilize EM algorithm at present The parameters solving gauss hybrid models has had a lot of ripe method, and in specific implementation process, technical staff is referred to Existing method, it will not go into details for the present invention.

Gauss retrain mixed model it is crucial that obtain the quantity of gauss component, namely calculate under each k value correspondence, structure risk The value of model, selects k corresponding during structure risk model value minimum the most from which as its track quantity.Structure risk model Construction method as follows:

K=min (R_srm(p(x_i|θ_k))) formula 19

L(x_i,p(x_i|θ_k))=-log (p (x_i|θ_k)) formula 20

R in formula 2_srm(p(x_i|θ_k)) it is structure risk model, L (x_i,p(x_i|θ_k)) it is the empiric risk model for assessing fitness, J(p(x_i|θ_k)) it is regular terms, it is used for indicating model complexity, namely is expressed as: J_TSW(p(x_i|θ_k)), λ > 0 is regular parameter, p(x_i|θ_k) represent sample value x_iIn model parameter θ_kUnder the conditions of gaussian probability value, wherein model parameter θ_kIt is represented by: θ_k(ω_k, μ_k, σ), n represents the number of sample, i=1,2 ..., n；Computing formula is as follows:

Wherein D_wIt is the tiling width at pavement of road of the track after optimizing, such as in Fig. 5, Fig. 5 a ' the 1 of long dotted line sign^st Gaussian component ' represents first composition of gauss hybrid models, ' 2 that short dash line indicates^ndGaussian component’ Represent second composition of gauss hybrid models；Wherein it is positioned at the parameter μ in Fig. 5 b₁、μ₂Correspond in Fig. 5 a first the most respectively The average of individual gauss component and the average of second gauss component, be also the lane center line position in first track and second The lane center line position in track.(how obtaining track has had a lot of method to propose at the tiling width of pavement of road, tool In body implementation process, those skilled in the art can choose voluntarily, and it will not go into details for the present invention), wherein k express possibility exist track Quantity, according to current domestic road construction standard, track, city quantity generally comprises two tracks, three lanes, Four-Lane Road, five cars Road, namely k=2,3,4,5.Δμ^kIt is two average value mu closing on Gaussian peak in gauss component corresponding for k_jChanging value, its become Change value also reflects detection lane width, j=1, and 2 ..., k.Δμ^kComputational methods such as formula 6, κ, η, γ in formula 6_ijFor The hyper parameter of EM algorithm based on MAP estimation；X represents that (when carrying out track and calculating, x represents judgement window to sample value Interior tracing point is the ordinate value of upright projection on its vertical section)；N represents the number of sample data, ω_j+1Represent that jth+1 is high The weighted value of this composition, j=1,2 ..., k-1, k are gauss component number, k=2,3,4,5；Existing highly developed parameter in existing research Recommending, the numerical value being referred to when being embodied as in existing method case be given calculates, and specifically repeats no more.

Step 4, according to the gauss hybrid models method after the optimization described in step 3, completes high accuracy track level road mapping, Realize detection lane information, obtain the road section track first result of detection of quantity.It is being embodied as high accuracy level road, track drive test During figure, utilize the existing method (the most how will be with as extraction unit using all tracks of being on same section Article one, all tracks on section are classified as an extraction unit, have had a lot of maturation method, ability in specific implementation process Field technique personnel may refer to existing method, and it will not go into details for the present invention).

Assume to give one group from crossing Intersection₁To crossing Intersection₂Track set A^T, from track set A^T One end start, build mobile rectangular window, as shown in Figure 6.The length and width of mobile rectangular window is respectively rh and rw (length of rectangular window and wide recommendation are defined as 10m and 30m, and when being embodied as, those skilled in the art can preset value voluntarily), The long limit wherein moving rectangular window is parallel to currently cover the centrage of all tracks, and the broadside of mobile rectangular window is then perpendicular to The current centrage covering all tracks, the central axis on the long limit of rectangular window in the centrage of its all track datas covered, The centrage of the track data that the center line of rectangular window broadside covers with it overlaps, and how to obtain the centrage of one section of track data, Having had a lot of method at present, in specific implementation process, those skilled in the art may refer to existing method, and it will not go into details for the present invention.

The preferred implementation that embodiment further provides for is as follows:

Rectangular window will be moved from track set A^TOne end start, according to the long edge lengths of rectangular window start translation, successively profit Track quantity and the lane center in the section covered is retrained in mixed model detects each rectangular window by Gauss.Concrete side Method includes: according to the rectangular window built, all tracing points in rectangular window project to the long limit centrage of rectangular window On, the track data collection X=(x after being projected₁,x₂,…,x_N), t=1,2,3 ..., N, wherein, x_tRepresent the t rail after projection The ordinate value of mark point, N is the number participating in projected footprint point.According to described in step 3, track data collection X is substituted into corresponding Computing formula (i.e. Gauss shown in formula 17 retrains mixed model), extracts track quantity and the lane center in section in rectangular window Line.Assume from crossing Intersection₁To crossing Intersection₂Track set A^T, rectangular window has carried out altogether l Secondary translation, translates the track quantity determined each time and is designated as Nlane_f, f=1,2 ..., l, visits for the first time as road section track quantity Survey result.

Step 5, the road section track first result of detection of quantity obtained according to step 4, based on road construction rule, propose Track quantity optimisation strategy, is modified first result of detection, as it is shown in fig. 7, there is section Part₁、Part₂、Part₃, its Middle Part₁And Part₃Exist and set up region, track.In most cases, the section between two crossings is always near handing over Prong vicinity there will be sets up track, and mid portion track, section quantity generally remains constant, and therefore the present invention proposes one The method optimizing track number of extracted result, concrete grammar is as follows:

The first step: in step 4, certain translates track quantity Nlane determined_f, compare Nlane_f+1And Nlane_f、 Nlane_f+2If, Nlane_fAnd Nlane_f+1It is different, then uses Nlane_fReplace Nlane_f+1, f=1,2 ..., l-2.

Second step: according to Nlane_fValue and distribution the result of the first step is classified, if such as Nlane_e,Nlane_e+1, Nlane_e+2,…,Nlane_e+cValue identical, be divided into a class, wherein e < l, e+c < l.Assume to there is s class, be designated as C_g=< Nl_g, nc_g>,Nl_gIt is class bunch C_gTrack quantity, nc_gIt is Nlane_gIn belong to C_g, g=1,2 ..., the total quantity of s.

3rd step: compare C_g+1And C_gIf, Nl_g+1It is different from Nl_g, and nc_g+1< (cv is one and depends on road and repair cv Building the threshold value of rule, its alteration ruler is mainly reflected in urban road when close to crossing, occurs setting up the relief area model in track Enclose, such as: according to present road design code, a length of about the 50m that track is set up, namely from being positioned at the road of crossing Section, a length of 50m in its newly-increased track, the segmentation section therefore judged when track quantity is set to 10 meters, and the present invention recommends cv Being set to 5, when being embodied as, those skilled in the art can preset value voluntarily), make C_gNl_gReplace C_g+1Nl_g+1, g=1,2 ..., S, completes the final optimization pass of track quantity result.Track quantity result of detection such as Fig. 8, track population detection optimizes such as Fig. 9 (wherein The abscissa of Fig. 8 represents that moving window carries out the mobile quantity in slip detection process, and vertical coordinate then represents detection of sliding each time The track quantity result of detection of process).

Step 6, the revised track quantity obtained according to step 5, the lane center that it is corresponding is modified.When After the track quantity in a certain section of section is corrected, the lane center of its correspondence then uses adjacent principle to be also corrected.Concrete grammar Including:

Assume that the track quantity of a certain section of section La is corrected, then find adjacent segments before and after La, if La adjacent segments Lb and Lc meets simultaneously and has identical track quantity with La, and these sections revise before do not send out with revised track quantity Changing, then be just connected by the lane center of Lb with Lc, obtains the final lane center of La；If the adjacent road of La Also change after revising before the track quantity correction of section Lb or Lc, then in the track extracted before just revising according to La Heart line position, calculates the road axis revising position of center line reckoning La section, front track based on La, (how to repair according to La The most front track position of center line calculates the road axis in La section, the most maturation method, is not repeating) according to After La correction, lane width and track quantity redefine La revised lane center line position, namely from according to La section Road axis starts to obtain after La revises each according to the equidistant road axis that is parallel to of track quantity and lane width successively The lane center line position in individual track.

Based on the present invention, the lane information treating urban road can be obtained easily from GPS track data, lead for following intelligence Boat and unmanned offer basis road net data.

When being embodied as, method provided by the present invention can realize automatic operational process based on software engineering, it is possible to uses modularity side Formula realizes corresponding system.

The present invention provides a kind of tracks based on the big data of mass-rent space-time level road mapping system, including with lower module,

First module, for setting up the similarity evaluating model of track vector, if v_aAnd v_bIt is two different track vector, described Similarity evaluation model is as follows,

${\mathrm{sim}}_{(v_a,v_b)}={\mathrm{\&omega;}}_1{e}^{-{\mathrm{diff}}_{Hd}}+{\mathrm{\&omega;}}_2{e}^{-{\mathrm{diff}}_{\mathrm{\&theta;}ab}}$

Wherein,Representing the Similarity value between vector, e is the nature truth of a matter, ω₁And ω₂Represent distance factor diff respectively_Hd With angular factors diff_θabWeighted value, and ω₁+ω₂=1；Distance factor diff_HdWith angular factors diff_θabRepresent vector v respectively_a And v_bDistance difference and angle difference；

Second module, preferred, including according to existing high-precision for carrying out track based on the growth clustering method merging Heuristics Degree GPS track data and synchronization low precision GPS track data, determine the weighted value ω of similarity evaluation model₁And ω₂, extract The preferred priori of track, uses growth cluster mode to carry out data based on the similarity between mass-rent track data preferred；

Three module, is used for building Gauss and retrains mixed model, and use EM Algorithm for Solving model parameter；Described Gauss retrains Mixed model is defined as follows,

$p\left(x\right)=\underset{j=1}{\overset{k}{\&Sigma;}}{\mathrm{\&omega;}}_j\frac{1}{\sqrt{2{\mathrm{\&pi;\&sigma;}}^2}}\exp (-\frac{{(x-{\mathrm{\&mu;}}_j)}^2}{2{\mathrm{\&sigma;}}^2})$

Wherein, p (x) is expressed as Gauss and retrains the combined chance value of mixed model, and x represents sample value to be calculated, is carrying out track meter During calculation, x represents and judges tracing point ordinate value of upright projection on its vertical section in window；K is the quantity of gauss component, often One corresponding track of gauss component；ω_jIt is the weight of jth gauss component, the traffic flow in corresponding track；Parameter μ₁…μ_k It is the meansigma methods of track in each gauss component, equal to the centrage in each track, μ_jRepresent μ₁…μ_kAny one in parameter Individual value, j=1,2 ..., k；σ is the standard deviation of track in each gauss component；

It is that the value of computation structure risk model, with knot that described Gauss retrains the quantity k acquisition mode of gauss component in mixed model The minimum principle of structure risk model value determines k；

4th module, for according to three module acquired results, detects lane information, obtains road section track quantity and visit for the first time Survey result；Implementation is as follows,

Using all tracks of being on same section as an extraction unit, if given one group from crossing Intersection₁ To crossing Intersection₂Track set A^T, from track set A^TOne end start, build mobile rectangular window, wherein The long limit of mobile rectangular window is parallel to the centrage currently covering all tracks, and the broadside of mobile rectangular window is then perpendicular to currently Covering the centrage of all tracks, the central axis on the long limit of rectangular window is in the centrage of its all track datas covered, square The centrage of the track data that the center line of shape window broadside covers with it overlaps；

Rectangular window will be moved from track set A^TOne end start, according to the long edge lengths of rectangular window start translation, successively profit Track quantity and the lane center in the section covered is retrained in mixed model detects each rectangular window, including root by Gauss According to mobile rectangular window, project to, on the long limit centrage of rectangular window, be thrown by all tracing points in mobile rectangular window The track data collection X=(x of movie queen₁,x₂,…,x_N), t=1,2,3 ..., N, wherein, x_tRepresent the vertical seat of t tracing point after projecting Scale value, N is the number participating in projected footprint point；Track data collection X substitution Gauss is retrained mixed model, extracts rectangular window The track quantity in interior section and lane center；Assume from crossing Intersection₁To crossing Intersection₂Track collection Close A^T, rectangular window has carried out altogether l translation, has translated the track quantity determined each time and be designated as Nlane_f, f=1,2 ..., l, As the road section track first result of detection of quantity；

5th module, for the road section track first result of detection of quantity obtained according to the 4th module, advises based on road construction Then, first result of detection is modified；

6th module, for the revised track quantity obtained according to the 5th module, enters lane center according to adjacencies Row is revised.

Each module implements and can be found in corresponding steps, and it will not go into details for the present invention.

Specific embodiment described herein is only to present invention spirit explanation for example.The skill of the technical field of the invention Described specific embodiment can be made various amendment or supplements or use similar mode to substitute by art personnel, but not The spirit of the present invention can be deviateed or surmount scope defined in appended claims.

Claims (6)

1. tracks based on the big data of a mass-rent space-time level road plotting method, it is characterised in that: comprise the following steps,

Step 1, sets up the similarity evaluating model of track vector, if v_aAnd v_bIt is two different track vector, described similarity Evaluation model is as follows,

${\mathrm{sim}}_{(v_a,v_b)}={\mathrm{\&omega;}}_1{e}^{-{\mathrm{diff}}_{Hd}}+{\mathrm{\&omega;}}_2{e}^{-{\mathrm{diff}}_{\mathrm{\&theta;}ab}}$

Wherein,Representing the Similarity value between vector, e is the nature truth of a matter, ω₁And ω₂Represent distance factor diff respectively_Hd With angular factors diff_θabWeighted value, and ω₁+ω₂=1；Distance factor diff_HdWith angular factors diff_θabRepresent vector v respectively_a And v_bDistance difference and angle difference；

Step 2, carries out track based on the growth clustering method merging Heuristics preferred, including according to existing high-precision GPS rail Mark data and synchronization low precision GPS track data, determine the weighted value ω of similarity evaluation model₁And ω₂, extract track preferred Priori, using growth cluster mode based on the similarity between mass-rent track data, to carry out data preferred；

Step 3, builds Gauss and retrains mixed model, and use EM Algorithm for Solving model parameter；Described Gauss retrains hybrid guided mode Type is defined as follows,

$p\left(x\right)=\underset{j=1}{\overset{k}{\&Sigma;}}{\mathrm{\&omega;}}_j\frac{1}{\sqrt{2{\mathrm{\&pi;\&sigma;}}^2}}\exp (-\frac{{(x-{\mathrm{\&mu;}}_j)}^2}{2{\mathrm{\&sigma;}}^2})$

Wherein, p (x) is expressed as Gauss and retrains the combined chance value of mixed model, and x represents sample value to be calculated, is carrying out track meter During calculation, x represents and judges tracing point ordinate value of upright projection on its vertical section in window；K is the quantity of gauss component, often One corresponding track of gauss component；ω_jIt is the weight of jth gauss component, the traffic flow in corresponding track；Parameter μ₁…μ_k It is the meansigma methods of track in each gauss component, equal to the centrage in each track, μ_jRepresent μ₁…μ_kAny one in parameter Individual value, j=1,2 ..., k；σ is the standard deviation of track in each gauss component；

It is that the value of computation structure risk model, with knot that described Gauss retrains the quantity k acquisition mode of gauss component in mixed model The minimum principle of structure risk model value determines k；

Step 4, according to step 3 acquired results, detects lane information, obtains the road section track first result of detection of quantity； Implementation is as follows,

Using all tracks of being on same section as an extraction unit, if given one group from crossing Intersection₁ To crossing Intersection₂Track set A^T, from track set A^TOne end start, build mobile rectangular window, wherein The long limit of mobile rectangular window is parallel to the centrage currently covering all tracks, and the broadside of mobile rectangular window is then perpendicular to currently Covering the centrage of all tracks, the central axis on the long limit of rectangular window is in the centrage of its all track datas covered, square The centrage of the track data that the center line of shape window broadside covers with it overlaps；

Rectangular window will be moved from track set A^TOne end start, according to the long edge lengths of rectangular window start translation, successively profit Track quantity and the lane center in the section covered is retrained in mixed model detects each rectangular window, including root by Gauss According to mobile rectangular window, project to, on the long limit centrage of rectangular window, be thrown by all tracing points in mobile rectangular window The track data collection X=(x of movie queen₁,x₂,…,x_N), t=1,2,3 ..., N, wherein, x_tRepresent the vertical seat of t tracing point after projecting Scale value, N is the number participating in projected footprint point；Track data collection X substitution Gauss is retrained mixed model, extracts rectangular window The track quantity in interior section and lane center；Assume from crossing Intersection₁To crossing Intersection₂Track collection Close A^T, rectangular window has carried out altogether l translation, has translated the track quantity determined each time and be designated as Nlane_f, f=1,2 ..., l, As the road section track first result of detection of quantity；

Step 5, the road section track first result of detection of quantity obtained according to step 4, based on road construction rule, to just Secondary result of detection is modified；

Step 6, the revised track quantity obtained according to step 5, according to adjacencies, lane center is modified.

Tracks based on the big data of mass-rent space-time level road plotting method the most according to claim 1, it is characterised in that: in step 5 First result of detection is modified, it is achieved mode is as follows,

The first step, in step 4, certain translates track quantity Nlane determined_f, compare Nlane_f+1And Nlane_f、 Nlane_f+2If, Nlane_fAnd Nlane_f+1It is different, then uses Nlane_fReplace Nlane_f+1, f=1,2 ..., l-2；

Second step, according to Nlane_fValue and distribution the result of the first step is classified, if there is s class, be designated as C_g=< Nl_g,nc_g>, Nl_gIt is class bunch C_gTrack quantity, nc_gIt is Nlane_gIn belong to C_g, g=1,2 ..., the total quantity of s；

3rd step, compares C_g+1And C_gIf, Nl_g+1It is different from Nl_g, and nc_g+1< cv makes C_gNl_gReplace C_g+1's Nl_g+1, g=1,2 ..., s, complete the final optimization pass of track quantity result, wherein, cv is default threshold value.

Tracks based on the big data of mass-rent space-time the most according to claim 1 or claim 2 level road plotting method, it is characterised in that: step 6 Lane center is modified, it is achieved mode is as follows,

If the track quantity of a certain section of section La is corrected, if La adjacent segments Lb and Lc meets has phase with La simultaneously Do not change with revised track quantity before same track quantity, and correction, then just by the lane center of Lb Yu Lc Line connects, and obtains the final lane center of La；If with correction before the track quantity correction of La adjacent segments Lb or Lc After also change, then just according to La revise before extract lane center line position, calculate based on La revise before lane center Line position calculates the road axis in La section, after after revising according to La, lane width and track quantity redefine La correction Lane center line position.

4. tracks based on the big data of a mass-rent space-time level road mapping system, it is characterised in that: include with lower module,

First module, for setting up the similarity evaluating model of track vector, if v_aAnd v_bIt is two different track vector, described Similarity evaluation model is as follows,

${\mathrm{sim}}_{(v_a,v_b)}={\mathrm{\&omega;}}_1{e}^{-{\mathrm{diff}}_{Hd}}+{\mathrm{\&omega;}}_2{e}^{-{\mathrm{diff}}_{\mathrm{\&theta;}ab}}$

Wherein,Representing the Similarity value between vector, e is the nature truth of a matter, ω₁And ω₂Represent distance factor diff respectively_Hd With angular factors diff_θabWeighted value, and ω₁+ω₂=1；Distance factor diff_HdWith angular factors diff_θabRepresent vector v respectively_a And v_bDistance difference and angle difference；

Second module, preferred, including according to existing high-precision for carrying out track based on the growth clustering method merging Heuristics Degree GPS track data and synchronization low precision GPS track data, determine the weighted value ω of similarity evaluation model₁And ω₂, extract The preferred priori of track, uses growth cluster mode to carry out data based on the similarity between mass-rent track data preferred；

Three module, is used for building Gauss and retrains mixed model, and use EM Algorithm for Solving model parameter；Described Gauss retrains Mixed model is defined as follows,

$p\left(x\right)=\underset{j=1}{\overset{k}{\&Sigma;}}{\mathrm{\&omega;}}_j\frac{1}{\sqrt{2{\mathrm{\&pi;\&sigma;}}^2}}\exp (-\frac{{(x-{\mathrm{\&mu;}}_j)}^2}{2{\mathrm{\&sigma;}}^2})$

Wherein, p (x) is expressed as Gauss and retrains the combined chance value of mixed model, and x represents sample value to be calculated, is carrying out track meter During calculation, x represents and judges tracing point ordinate value of upright projection on its vertical section in window；K is the quantity of gauss component, often One corresponding track of gauss component；ω_jIt is the weight of jth gauss component, the traffic flow in corresponding track；Parameter μ₁…μ_k It is the meansigma methods of track in each gauss component, equal to the centrage in each track, μ_jRepresent μ₁…μ_kAny one in parameter Individual value, j=1,2 ..., k；σ is the standard deviation of track in each gauss component；

It is that the value of computation structure risk model, with knot that described Gauss retrains the quantity k acquisition mode of gauss component in mixed model The minimum principle of structure risk model value determines k；

4th module, for according to three module acquired results, detects lane information, obtains road section track quantity and visit for the first time Survey result；Implementation is as follows,

Using all tracks of being on same section as an extraction unit, if given one group from crossing Intersection₁ To crossing Intersection₂Track set A^T, from track set A^TOne end start, build mobile rectangular window, wherein The long limit of mobile rectangular window is parallel to the centrage currently covering all tracks, and the broadside of mobile rectangular window is then perpendicular to currently Covering the centrage of all tracks, the central axis on the long limit of rectangular window is in the centrage of its all track datas covered, square The centrage of the track data that the center line of shape window broadside covers with it overlaps；

Rectangular window will be moved from track set A^TOne end start, according to the long edge lengths of rectangular window start translation, successively profit Track quantity and the lane center in the section covered is retrained in mixed model detects each rectangular window, including root by Gauss According to mobile rectangular window, project to, on the long limit centrage of rectangular window, be thrown by all tracing points in mobile rectangular window The track data collection X=(x of movie queen₁,x₂,…,x_N), t=1,2,3 ..., N, wherein, x_tRepresent the vertical seat of t tracing point after projecting Scale value, N is the number participating in projected footprint point；Track data collection X substitution Gauss is retrained mixed model, extracts rectangular window The track quantity in interior section and lane center；Assume from crossing Intersection₁To crossing Intersection₂Track collection Close A^T, rectangular window has carried out altogether l translation, has translated the track quantity determined each time and be designated as Nlane_f, f=1,2 ..., l, As the road section track first result of detection of quantity；

5th module, for the road section track first result of detection of quantity obtained according to the 4th module, advises based on road construction Then, first result of detection is modified；

6th module, for the revised track quantity obtained according to the 5th module, enters lane center according to adjacencies Row is revised.

Tracks based on the big data of mass-rent space-time level road mapping system the most according to claim 4, it is characterised in that: the 5th module In first result of detection is modified, it is achieved mode is as follows,

The first step, in the 4th module, certain translates track quantity Nlane determined_f, compare Nlane_f+1And Nlane_f、 Nlane_f+2If, Nlane_fAnd Nlane_f+1It is different, then uses Nlane_fReplace Nlane_f+1, f=1,2 ..., l-2；

Second step, according to Nlane_fValue and distribution the result of the first step is classified, if there is s class, be designated as C_g=< Nl_g,nc_g>, Nl_gIt is class bunch C_gTrack quantity, nc_gIt is Nlane_gIn belong to C_g, g=1,2 ..., the total quantity of s；

3rd step, compares C_g+1And C_gIf, Nl_g+1It is different from Nl_g, and nc_g+1< cv makes C_gNl_gReplace C_g+1's Nl_g+1, g=1,2 ..., s, complete the final optimization pass of track quantity result, wherein, cv is default threshold value.

6. according to tracks based on the big data of mass-rent space-time level road mapping system described in claim 4 or 5, it is characterised in that: the 6th Lane center is modified by module, it is achieved mode is as follows,

If the track quantity of a certain section of section La is corrected, if La adjacent segments Lb and Lc meets has phase with La simultaneously Do not change with revised track quantity before same track quantity, and correction, then just by the lane center of Lb Yu Lc Line connects, and obtains the final lane center of La；If with correction before the track quantity correction of La adjacent segments Lb or Lc After also change, then just according to La revise before extract lane center line position, calculate based on La revise before lane center Line position calculates the road axis in La section, after after revising according to La, lane width and track quantity redefine La correction Lane center line position.