CN102589557B - Intersection map matching method based on driver behavior characteristics and logit model - Google Patents

Abstract

The invention provides an intersection map matching method based on driver behavior characteristics and a logit model, relating to the field of vehicle positioning and map matching. The method comprises four phases of storing input data, mapping an initial site map, matching a map on a road section and matching an intersection map, so as to finish high-precision and real-time positioning of a vehicle driving track. The intersection map matching method disclosed by the invention solves the problem that the matching precision of the intersection map is low and overcomes the defect of the traditional intersection map matching method, so that the high-precision and real-time positioning of an intersection is realized.

Description

A kind of crossing's map-matching method based on driver behavior pattern and decilog model

Technical field

The present invention relates to vehicle location and map match field, is location technology and the map-matching method of a kind of vehicle in crossing, is specifically related to a kind of crossing's map-matching method based on driver behavior pattern and decilog model.

Background technology

Map match is to improve a kind of effective means of vehicle location precision.It is mainly longitude and the dimension coordinate of the vehicle location point that provides according to GPS (Global Position System, be called for short GPS), finds the location point on this section, location point place and section on Digital map.Urban intersection is the crucial hinge that vehicle changes travel direction, conversion running section, and the map match at urban intersection place is core and the difficult point of map-matching method research.Traditional map-matching method is mainly to start with from the space geometry relationship analysis between vehicle location point and section, road network geometric data, topological relation data and GPS locator data based in Digital map produces the optimum estimate of vehicle location point, be applicable to the map match on common section, seem not fully up to expectations in crossing, often the vehicle location anchor point of Turning travel is matched on craspedodrome section mistakenly.

Wherein, cause the lower basic reason of crossing's map match precision: 1) crossing connects many densely distributed sections among a small circle simultaneously; 2) left turn lane, Through Lane and right-turn lane is tangent within the scope of crossing, phase mutual interference in each entrance driveway direction; 3) when Vehicular turn travels, there is inertia.At present, a lot of vehicle positioning equipment on market are owing to cannot realizing the high precision of crossing's vehicle location point, the vehicle location location that real-time online map match has all been omitted crossing, or the location of first skipping the vehicle location point of crossing, treat the anchor point on follow-up running section coupling section determine after, according to the consistency principle, the vehicle location point of crossing is carried out to off-line location again.

Summary of the invention

The problem such as not high for current crossing map match precision, real-time is not strong, the present invention proposes a kind of crossing's map-matching method based on driver behavior pattern and decilog model, has realized crossing's high precision, location in real time.Vehicle GPS receiver receives GPS locator data at a certain time interval, by map match and crossing's map match four-stage on the storage of input data, initial point map match, section, completes the high precision of vehicle driving trace, locates in real time.

Technical scheme of the present invention is the generation sequential according to vehicle location point, and vehicle location is divided into map match on input data storage, initial point map match, section and the map match four-stage of crossing.

First stage: input data storage

The storage of input data is before map match, needed data to be stored in top set database table, and input data comprise the section attribute data in Digital map, GPS locator data and the intersection turning restricting data of vehicle location.

Subordinate phase: initial point map match

Initial point map match comprises initial alignment point candidate matches section identification module, the modeling of initial alignment point map match and weights coefficient estimation module.Initial alignment point candidate matches section identification module is determined the set of candidate matches section by circular error territory, and the modeling of initial alignment point map match and weights coefficient estimation module are responsible for selecting optimum matching section from the set of candidate matches section.

Phase III: the map match on section

Map match on section comprises dead ship condition discrimination module and vehicle relative position discrimination module, dead ship condition discrimination module is whether to differentiate vehicle in dead ship condition, if, the matching result of current anchor point is identical with the matching result of previous anchor point, if not, carry out the differentiation of vehicle relative position.Vehicle relative position discrimination module is the coupling section whether differentiation vehicle has left previous anchor point place, if not, determine that the coupling section of current anchor point is identical with the coupling section of previous location, if so, enters the map match of crossing.

Fourth stage: the map match of crossing

The map match of crossing comprises the crucial anchor point identification module in crossing, crucial anchor point candidate matches section, crossing mark module, the modeling of decilog map match and weights coefficient estimation module.The crucial anchor point identification module in crossing is for determining the crucial anchor point set that affects crossing's map match precision, section and last relativeness of mating section that crucial anchor point candidate matches section, crossing mark module connects according to crossing, the candidate matches section of crucial anchor point is defined as candidate matches section, craspedodrome candidate matches section and right-hand rotation candidate matches section, and rejected and limited current candidate matches section according to intersection turning restricting data, make effective candidate matches section of the crucial anchor point in crossing be no more than at most 3, the modeling of decilog map match and weights coefficient estimation module, start with from three kinds of corresponding driver behavior pattern's variance analyses in different candidate matches section, propose the estimation problem in crossing's optimum matching section to be transformed into the selection problem of vehicle to follow-up running section, using the crucial weight of considering that the weight (symbol and numerical values recited) of driver's driving behavior characteristic is selected follow-up running section as differentiation driver, from three candidate road section, select optimum matching section according to decilog model, and adopt maximum likelihood estimate to estimate weights coefficient.

By adopting such scheme, solve the low problem of restriction crossing's map match precision, overcome the drawback of existing crossing map-matching method, realize crossing's high precision, real-time map coupling.

Brief description of the drawings

Fig. 1 is block diagram of the present invention;

Fig. 2 is initial point map match flow process;

Fig. 3 is the map match flow process on section;

Fig. 4 is crossing's map match flow process;

Fig. 5 is the crossing decussation mouths of 90 degree;

Fig. 6 is less than the crossing decussation mouth of 90 degree;

Fig. 7 is single order oriented difference diagram in travel direction angle between the adjacent positioned point based on GN-79N model GPS receiver (30 meters of left and right of positioning precision);

Fig. 8 is single order oriented difference diagram in travel direction angle between the adjacent positioned point based on OEMV model GPS receiver (5 meters of left and right of positioning precision);

Fig. 9 is the intention of section attribute data shown in Fig. 6 in excel data table stores Digital map;

Figure 10 is the GPS locator data schematic diagram of excel data table stores vehicle location;

Figure 11 is that shown in excel data table stores Fig. 6, the each entrance driveway in crossing turns to restricting data schematic diagram.

Specific implementation method

First stage: input data storage

The input data of map match comprise the section attribute data in Digital map, GPS locator data and the intersection turning restricting data of vehicle location.Specific as follows:

1) the section attribute data in Digital map: the numbering of the numbering in section, the starting point in section and terminal and coordinate (Digital map adopts WGS 84 coordinate systems, identical with the coordinate system of GPS);

2) the GPS locator data of vehicle location: the longitude of the vehicle location that vehicle GPS receiver receives and dimension coordinate, vehicle heading, Vehicle Speed, data are accepted the time (taking 1 second time interval as data receiver);

3) intersection turning restricting data: crossing respectively connects section restricting data that turns to about left-hand rotation, craspedodrome and right-hand rotation in entrance driveway direction, if the entrance driveway direction in the i.e. a certain connection in crossing section restriction vehicle turns left, the direction of turning left restricting data value is 1, do not limit, value is 0.

The data table stores method of Various types of data is shown in Fig. 9～Figure 11.

Subordinate phase: initial alignment point map match

Initial alignment point map match refers to that first anchor point of the vehicle location that gps receiver is received carries out map match, identify its correct coupling section and the location point on coupling section, map match flow process is shown in Fig. 2, mainly comprises initial alignment point candidate matches section identification module, the modeling of initial alignment point map match and two modules of weights coefficient estimation module.

1, initial alignment point candidate matches section identification module

Initial alignment point has material impact to overall map matching precision of the present invention.Initial alignment point candidate matches section identification is by taking initial alignment point as the center of circle, taking the positioning error of gps receiver as radius, build circular error territory, all be defined as candidate matches section by all with the section in this circle error territory crossing or tangent (having common intersection), and form candidate matches section set omega.

2, the modeling of initial alignment point map match and weights coefficient estimation module

1) initial alignment point map match modeling

After candidate matches section is determined, the present invention is according to the conforming weight of travel direction and the weights W of consideration anchor point and subpoint proximity _das initial alignment point map match weight.To each the candidate matches section L in the set omega of candidate matches section, do to calculate as follows:

The coupling section of initial alignment point is:

$L_b=\left\{L\right|\underset{L\∈\mathrm{\Ω}}{\max }{Z}_L\}---\left(2\right)$

2) initial alignment point map match weight calculation

1. consider the conforming weight of travel direction

First, regulation direct north is 0 °, and deflection is by increasing progressively clockwise.For ease of calculate, 360 ° of deflection unifications are changed to 0 ° of deflection, the span of deflection be [0 °, 360 °).Making α is vehicle heading angle, θ _lfor the deflection of candidate matches section L, (x _s, y _s), (x _f, y _f) be respectively the starting point of candidate matches section L and the coordinate of terminal,, in the time that candidate matches section is straight line section, have

In Digital map, conventionally represent curve section with the broken line section that multiple shorter straight line sections connect, in broken line section, the intersection point of adjacent straight part of path is flex point.Therefore, in the time that candidate matches section is broken line section, if subpoint between two adjacent flex points, θ _lfor the deflection in the straight line section between these two flex points, if subpoint drops in flex point, θ _lfor the deflection in the straight line section of angle absolute value minimum between flex point two ends and vehicle heading.

Consider the conforming weight of travel direction computing formula is:

In formula, for corresponding weights coefficient, value try to achieve according to weights coefficients calculation block; for the angle absolute value between vehicle heading and section.

Meanwhile, regulation larger, candidate matches section be coupling section possibility larger, otherwise, less.

2. consider the weight of anchor point and subpoint proximity

Make (x _q, y _q), (x _m, y _m) be respectively the coordinate of GPS anchor point q and its subpoint m on the L of candidate matches section, $\mathrm{\η}=\frac{y_F-y_S}{x_F-x_S},$ $\mathrm{\μ}=-\frac{1}{\mathrm{\η}},$ ?

$\left\{\begin{array}{c}{x}_m=\frac{y_q-y_S+\mathrm{\η}\×x_S-\mathrm{\μ}\×x_q}{\mathrm{\η}-\mathrm{\μ}}\\ y_m=y_S+\mathrm{\η}\×(x_m-x_S)\end{array}\right.---\left(6\right)$

Anchor point to the vertical projection in candidate matches section apart from d is:

$d=\sqrt{{(x_q-x_m)}^2+{(y_q-y_m)}^2}---\left(7\right)$

When on the straight line section between two adjacent flex points of subpoint in broken line section, in formula (7), the starting point in section and terminal point coordinate become the coordinate of two flex points; In the time that subpoint is in flex point, subpoint coordinate is flex point coordinate; In the time that subpoint is on the extended line in straight line section, directly reject this candidate matches section.

The weight of considering anchor point and subpoint proximity is:

$W_d={\mathrm{\β}}_d\left(\frac{C_d-d}{C_d}\right)---\left(8\right)$

In formula, β _dfor corresponding weights coefficient, β _dvalue try to achieve according to weights coefficients calculation block; C _dfor constant, value is the positioning error of GPS receiver conventionally.

Meanwhile, regulation Wd larger, candidate matches section be coupling section possibility larger, otherwise, less.

3) weights coefficient is estimated

Choose at random the individual anchor point of u (u > 100) from multiple sections, first, only to consider the conforming weight of travel direction as the weight of differentiating coupling section, mate section order represent the anchor point sum of correct coupling in u anchor point.Again only to consider that anchor point and the weights W d of subpoint proximity mate the weight in section as differentiation, mate section make u _drepresent the anchor point sum of correct coupling in u anchor point, weights coefficient and β _destimated value with for

Phase III: map match on section

The present invention is mainly the map match for crossing, therefore the map-matching method on section is simplified, and map match flow process is shown in Fig. 3, mainly comprises dead ship condition discrimination module and vehicle relative position discrimination module.

1, dead ship condition discrimination module

If the speed of a motor vehicle is less than 2km/h, think that vehicle is in dead ship condition, match point is identical with previous moment, otherwise match point is the vertical projection of anchor point on coupling section.

2, vehicle relative position discrimination module

After initial alignment point coupling, determine Vehicle Driving Cycle section, because vehicle travelling on section has continuity, only just can convert running section in crossing, therefore, if it (is current anchor point to the subpoint in last coupling section on last coupling section that vehicle does not leave this section, but not on extended line), the coupling section of current anchor point is identical with a upper coupling section.

Fourth stage: crossing's map match

Crossing is the hinge that vehicle changes travel direction, conversion running section.The flow process of crossing's map-matching method that the present invention proposes is shown in Fig. 4, mainly comprises that crossing's map match comprises the crucial anchor point identification module in crossing, crucial anchor point candidate matches section, crossing mark module, the modeling of decilog map match and three modules of weights coefficient estimation module.

1, the crucial anchor point identification module in crossing

Many experimental results shows, when vehicle turns left or turns right in crossing, due to left turn lane, right-turn lane and Through Lane tangent relation and the Vehicle Driving Cycle inertia among a small circle, vehicle sails the error-prone coupling of front 5～10 anchor points behind next section into, also be to cause the low main cause of crossing's map match precision, these points are defined as the crucial anchor point in crossing by the present invention.Taking Fig. 5 as example, the crucial anchor point of crossing is q ₅～q ₁₀.Concrete recognition methods is as follows:

First, judge that current anchor point q is when the subpoint in the coupling section of a upper anchor point is whether on the extended line of the terminal direction in this coupling section, if, think that anchor point q is first anchor point in crucial anchor point set, and jointly form the crucial anchor point set in crossing E with the individual anchor point of n-1 (experimental test show that n is generally 6～8) thereafter.

2, the crucial anchor point candidate matches in crossing section mark module

Because Vehicle Driving Cycle has continuity, the coupling section of the crucial anchor point in crossing is inevitable crossing with last coupling section, and two sections have connectedness.Therefore, this crossing all sections crossing with last coupling section are all considered as to candidate matches section.Simultaneously, because the travel behavior of vehicle when by crossing comprises left-hand rotation, keeps straight on and turn right, crucial anchor point candidate matches section, crossing mark is exactly to mark respectively left-hand rotation direction candidate matches section, craspedodrome direction candidate matches section and the right-hand rotation direction candidate matches section with respect to last coupling section in the candidate matches section of the crucial anchor point in crossing.

Make θ _lwith be respectively candidate matches section L and last coupling section L _fdeflection, candidate matches section L and last coupling section L _fbetween angle Δ θ be:

Candidate matches section L and last coupling section L _frelativeness be:

1) if | Δ θ | 20 ° of < or 180 °>=| Δ θ | 160 ° of >, think that candidate matches section L is last coupling section L _frelative craspedodrome section;

2) if Δ θ≤-20 ° think that candidate matches section L is last coupling section L _frelative left-hand rotation section;

3) if Δ θ>=20 ° think that candidate matches section L is last coupling section L _frelative right-hand rotation section.

If with respect to L ₁there is one or more left-hand rotation or right-hand rotation section, think that with immediate that section of GPS anchor point be relative left-hand rotation or right-hand rotation section, therefore, these anchor points remain unique relative left-hand rotation, craspedodrome and right-hand rotation candidate matches section, are reduced to maximum three by candidate matches section.Meanwhile, according to intersection turning restricting data, in the candidate matches section from mark, reject the current candidate matches section of restriction.Taking crossing shown in Fig. 5 as example, candidate matches section, crossing mark refers to when Vehicle Driving Cycle is in section _2656-1440when upper, according to the angle between intersection leg, mark is with respect to three follow-up running sections in last coupling section, left-hand rotation candidate matches section _1440-1438, craspedodrome candidate matches section _1440-1441and right-hand rotation candidate matches section _1440-1444; Taking crossing shown in Fig. 6 as example, candidate matches section, crossing mark refers to when Vehicle Driving Cycle is in section _1410-1408when upper, according to the angle between intersection leg, mark is with respect to three follow-up running sections in last coupling section, left-hand rotation candidate matches section _1408-1407with craspedodrome candidate matches section _1408-1413.

3, the modeling of decilog map match and weights coefficient estimation module

The present invention estimates to regard as the selection of driver to follow-up running section by the coupling section of the crucial anchor point in crossing, according to the difference of vehicle driving behavior characteristic of driver in the time turning left, keep straight on and turn right, from the set of candidate matches section, select optimum matching section.

1) decilog map match modeling

Decilog model (being K-Logit model) can select to have with maximum probability selection limb from K kind the selection limb of maximum utility.The present invention utilizes this modeling feature of decilog model, regard three candidate matches sections of the crucial anchor point in crossing as three and select limb, effectiveness using the weight sum in each candidate matches section as this candidate matches section, calculate the matching probability in each candidate matches section according to K-Logit computing formula, and to there is the candidate matches section of maximum matching probability as the optimum matching section of current anchor point.The modeling process of K-Logit model is as follows:

Suppose that S is the selection limb set of the crucial anchor point q in crossing,

S={ left-hand rotation candidate matches section, craspedodrome candidate matches section, right-hand rotation candidate matches section } (11)

For each candidate matches section, if coupling section, the selection result of this selection limb is 1, if not, value is 0.

For example, the left-hand rotation candidate matches section of a crucial anchor point of certain crossing is coupling section, and this anchor point respectively selects the selection result of limb to be

δ＝{1，0，0}(12)

Make V _lfor the weight sum of candidate matches section L,

$V_L=\underset{r=1}{\overset{R}{\mathrm{\&Sigma;}}}{W}_{L,r}+{\mathrm{\&epsiv;}}_L---\left(14\right)$

In formula: W _{l, r}for r the weight of candidate matches section L, ε _lfor the stochastic error of candidate matches section L, suppose ε _lwith separate, and the stochastic error in each candidate matches section is separate and all obey double exponential distribution.

The probability selection formula of K-Logit model is:

$P_L=\frac{\exp \left(\underset{r=1}{\overset{R}{\mathrm{\&Sigma;}}}{W}_{L,r}\right)}{\underset{L}{\mathrm{\&Sigma;}}\exp \left(\underset{r=1}{\overset{R}{\mathrm{\&Sigma;}}}{W}_{L,r}\right)}---\left(15\right)$

The optimum matching section L of anchor point q _bfor:

$L_b=\left\{L\right|\underset{L\&Element;S}{\max }{P}_L\}---\left(17\right)$

Meanwhile, anchor point q is at optimum matching section L _bon vertical projection point be the optimal match point of vehicle.

2) map match weight calculation

Ride characteristic according to vehicle in crossing, the present invention chooses the map match weight of four kinds of weights as the crucial anchor point in crossing, consider the weight of weight, the conforming weight of consideration journey time, the consideration conforming weight of travel direction and consideration anchor point and the subpoint proximity of driver's driving behavior characteristic, wherein, consider the conforming weight of travel direction and consider anchor point and the weighing computation method of subpoint proximity and identical above.Consider the weight of driver's driving behavior characteristic and consider that the computing method of the conforming weight of journey time are as follows:

1. consider the weight of driver's driving behavior characteristic

Map match weight is the principal element that affects map match precision, test findings shows: due to left turn lane, right-turn lane and Through Lane tangent relation and the Vehicle Driving Cycle inertia among a small circle, only rely on and consider the conforming weight of travel direction and consider that the weight etc. of anchor point and subpoint proximity still cannot effectively improve the map match precision of crossing key anchor point, as shown in Fig. 5～Fig. 6, rely on above-mentioned two weights all by anchor point q ₅, q ₆, q ₇match mistakenly craspedodrome candidate road section _1440-1441and section _1408-1413on.By to bus, private car, finding with car investigation of the polytype such as taxi and automobile carrier vehicle: driver crossing to turn to driving behavior to have angular setting amplitude than the random direction adjustment on straight line section large, direction is clear and definite and move continuous characteristic, can by the size of the oriented difference value of single order at vehicle heading angle while keeping straight on (when turning left and turning right with have a bigger difference) and positive and negative (right-hand rotation be for just, turn left for negative) embody, Fig. 7～Fig. 8 is the oriented difference value of single order at the vehicle heading angle added up in the time that four Adjacent Intersections turn left continuously of vehicle, as seen from the figure, when vehicle turns left in crossing, the oriented difference of single order at vehicle heading angle is for negative, and the oriented difference of travel direction angle single order when being obviously greater than vehicle and travelling on section.

Therefore, the present invention proposes using oriented the single order at vehicle heading angle difference as characterizing the direct acting factor of the driver of crossing driving behavior to vehicle driving trace, with continuous three sampling period (n-2, n-1, the weighting of the oriented difference of single order at vehicle heading angle n) is as the weight of considering driver driving behavior characteristic, and thinks that the oriented differential pair of single order at vehicle heading angle relatively turns left, keeps straight on and three sections of turning right show different weights.

Make Δ θ _nfor GPS anchor point q _nthe oriented difference of single order at vehicle heading angle,

Δθ _n＝θ _n-θ _n-1(18)

Meanwhile, for Efficient Characterization vehicle is the turning to and corner amplitude of crossing, to Δ θ _nvalue do following correction:

If vehicle within the scope of crossing, Δ θ in formula (19) _nwhen value is negative, think that vehicle heading is to counterclockwise rotation, if Δ θ _nexceed certain threshold value, as 5 °s/sec, think that vehicle may turn left to travel; If Δ θ _nvalue is timing, thinks that vehicle heading rotates to clockwise direction, if Δ θ _nexceed certain threshold value, as 8 °s/sec, think that vehicle may travel turning right.

Order and be respectively the weight function vector of considering driver's driving behavior characteristic, and be respectively corresponding weights coefficient vector, and have:

In formula, C _{Δ θ, a left side}, C _{Δ θ is straight}and C _{Δ θ, the right side}be respectively constant, conventionally get the maximal value that in the sampling period, vehicle angles changes.

While considering driver's driving behavior characteristic, the weights W in the section of turning left in candidate matches section _{Δ θ, a left side}, craspedodrome section weights W _{Δ θ is straight}and the weights W in right-hand rotation section _{Δ θ, the right side}for:

2. consider the conforming weight of journey time

Consider journey time conforming weight refer to for last match point be optimum matching section to the candidate matches section of the difference minimum in the vehicle travel time between subpoint on candidate matches section and sampling period.Weighing computation method is as follows:

Make v _n-1and v _nrepresent respectively adjacent positioned point q _n-1and q _nvehicle Speed, to vehicle from anchor point q _n-1match point travel to anchor point q _nthe travel time estimation of the subpoint on certain candidate matches section for:

$\hat{T}=l/\left(\frac{v_{n-1}+v_n}{2}\right)---\left(22\right)$

In formula, l is Vehicle Driving Cycle distance.

Journey time with the difference of sampling period Γ be:

$\mathrm{\&Delta;T}=|\hat{T}-\mathrm{\&Gamma;}|---\left(23\right)$

Consider the conforming weights W of journey time _{Δ T}for:

$W_{\mathrm{\&Delta;T}}={\stackrel{V}{\mathrm{\&beta;}}}_{\mathrm{\&Delta;T}}f\left(x_{\mathrm{\&Delta;T}}\right)={\stackrel{V}{\mathrm{\&beta;}}}_{\mathrm{\&Delta;T}}\frac{\mathrm{\&Delta;T}}{C_{\mathrm{\&Delta;T}}}---\left(24\right)$

In formula, f (x _{Δ T}) for considering the conforming weight function of journey time, for corresponding weights coefficient, C _{Δ T}for constant, be generally the maximal value of Δ T in the sampling period.

If candidate matches section sum is less than 3, anchor point as crucial in a certain crossing only has left-hand rotation candidate matches section and craspedodrome candidate matches section, makes the entitlement weight average value in right-hand rotation candidate matches section be 0.

3) maximal possibility estimation of weights coefficient

In cartographic model, the value of weights coefficient directly affects map match model accuracy, but in practical implementation, for reducing the calculated amount of map match model, weights coefficient is mostly by manually determining, do not carry out detailed theoretical model checking, in recent years also very rare about the optimization method research of weights coefficient in various cartographic models.The present invention is directed to the feature of K-Logit model, adopt maximum likelihood estimate to estimate the weights factor beta of K-Logit _{i, j}, computing method are as follows:

According to maximum likelihood estimate, suppose the crucial anchor point set in crossing Q={i|i=1 in road network, 2, L, in I}, i anchor point is δ to the selection result of each selection limb _{1, i}, L, δ _{k, i}, L, δ _{k, I}, the joint probability density of each selection result is:

$\begin{array}{ccccc}{P}_{{1,i,\mathrm{\&delta;}}_{1,i}}{P}_{{2,i,\mathrm{\&delta;}}_{2,i}}& L& P_{{k,i,\mathrm{\&delta;}}_{k,i}}& L& P_{K,i,{\mathrm{\&delta;}}_{K,i}}=\underset{k=1}{\overset{K}{\mathrm{\&Pi;}}}{P}_{{k,i,\mathrm{\&delta;}}_{k,i}}\end{array}---\left(25\right)$

Therefore, anchor point 1, L, i, L, the probability of the selection result of I to candidate matches section, namely likelihood function is:

$L^{*}=\underset{i=1}{\overset{I}{\mathrm{\&Pi;}}}\underset{k=1}{\overset{K}{\mathrm{\&Pi;}}}{P}_{{k,i,\mathrm{\&delta;}}_{k,i}}---\left(26\right)$

To L ^*take the logarithm to such an extent that log-likelihood function is:

$L^{*}=\underset{i=1}{\overset{I}{\mathrm{\&Sigma;}}}\underset{k=1}{\overset{K}{\mathrm{\&Sigma;}}}{\mathrm{\&delta;}}_{k,i}\ln \left(P_{k,i}\right)$ (27)

$=\underset{i=1}{\overset{I}{\mathrm{\&Sigma;}}}\underset{k=1}{\overset{K}{\mathrm{\&Sigma;}}}{\mathrm{\&delta;}}_{k,i}[\underset{j=1}{\overset{J}{\mathrm{\&Sigma;}}}{\mathrm{\&beta;}}_{j}f\left(x_{k,j}\right)-\ln \underset{k=1}{\overset{K}{\mathrm{\&Sigma;}}}{e}^{\underset{j=1}{\overset{J}{\mathrm{\&Sigma;}}}{\mathrm{\&beta;}}_{j}f\left(x_{k,j}\right)}]$

Make L reach peaked maximum likelihood function estimate vector should be the solution of following partial differential simultaneous equations:

$\frac{\&PartialD;L}{{\&PartialD;\mathrm{\&beta;}}_j}=0,j=\mathrm{1,2},L,J---\left(28\right)$

?

$\frac{\&PartialD;L}{{\&PartialD;\mathrm{\&beta;}}_j}=\underset{i=1}{\overset{I}{\mathrm{\&Sigma;}}}\underset{k=1}{\overset{K}{\mathrm{\&Sigma;}}}{\mathrm{\&delta;}}_{k,i}[f\left(x_{k,i,j}\right)-\frac{\underset{k=1}{\overset{K}{\mathrm{\&Sigma;}}}f\left(x_{k,i,j}\right)e^{{\mathrm{\&beta;}}_{j}f(x_{k,i},j)}}{\underset{k=1}{\overset{K}{\mathrm{\&Sigma;}}}{e}^{{\mathrm{\&beta;}}_{j}f\left(x_{k,i,j}\right)}}]=0,(j=\mathrm{1,2},L,J)---\left(29\right)$

Due to abbreviation formula (29) to make it be 0,

$\underset{i=1}{\overset{I}{\mathrm{\&Sigma;}}}\underset{k=1}{\overset{K}{\mathrm{\&Sigma;}}}({\mathrm{\&delta;}}_{k,i}-P_{k,i}){\mathrm{\&beta;}}_{j}f\left(x_{k,i,j}\right)=0---\left(30\right)$

Formula (30) is also the J unit Nonlinear System of Equations of β.Let it be to the greatest extent, and solution not necessarily exists, but in the time having solution, solution will be unique.

Adopt analytical method cannot solve the Nonlinear System of Equations shown in formula (30), the present invention adopts Newton Algorithm ? iterative computation formula be:

${\widehat{\mathrm{\&beta;}}}^{(m+1)}={\widehat{\mathrm{\&beta;}}}^{\left(m\right)}-{\left[{\&dtri;}^{2}L\left({\widehat{\mathrm{\&beta;}}}^{\left(m\right)}\right)\right]}^{-1}\&dtri;L\left({\widehat{\mathrm{\&beta;}}}^{\left(m\right)}\right)---\left(31\right)$

In formula, be respectively m time and m+1 estimate vector, respectively gradient function and the Hessian inverse of a matrix matrix of L.

Formula (31) stopping criterion for iteration is:

$\frac{1}{J}{\left[\underset{j=1}{\overset{J}{\mathrm{\&Sigma;}}}{({\mathrm{\&beta;}}_j^{(m+1)}-{\mathrm{\&beta;}}_j^{\left(m\right)})}^2\right]}^{1/2}<{\mathrm{\&tau;}}_1---\left(32\right)$

$\left|\frac{{\mathrm{\&beta;}}_j^{(m+1)}-{\mathrm{\&beta;}}_j^{\left(m\right)}}{{\mathrm{\&beta;}}_j^{\left(m\right)}}\right|<{\mathrm{\&tau;}}_2---\left(33\right)$

In formula, τ ₁, τ ₂be respectively predefined allowable error value, conventionally get τ ₁=10 ^-4, τ ₂=10 ^-3.Parameter vector in the t inspection computing formula of each parameter be:

$t_j={\mathrm{\&beta;}}_j/\sqrt{H_j}(j=\mathrm{1,2},L,J)---\left(34\right)$

In formula, t _jfor weights factor beta _jt statistical value, H _jfor matrix k element on middle diagonal line.

Claims (2)

1. the crossing's map-matching method based on driver behavior pattern and decilog model, is characterized in that being realized by following steps:

[1] input data storage

Be before map match, needed data to be stored in top set database table, input data comprise the section attribute data in Digital map, GPS locator data and the intersection turning restricting data of vehicle location;

[2] initial point map match

Comprise initial alignment point candidate matches section identification module, the modeling of initial alignment point map match and weights coefficient estimation module, initial alignment point candidate matches section identification module is determined the set of candidate matches section by circular error territory, and the modeling of initial alignment point map match and weights coefficient estimation module are responsible for selecting optimum matching section from the set of candidate matches section;

[3] map match on section

Comprise dead ship condition discrimination module and vehicle relative position discrimination module, dead ship condition discrimination module is whether to differentiate vehicle in dead ship condition, if, the matching result of current anchor point is identical with the matching result of previous anchor point, if not, carry out the differentiation of vehicle relative position, vehicle relative position discrimination module is the coupling section whether differentiation vehicle has left previous anchor point place, if not, determine that the coupling section of current anchor point is identical with the coupling section of previous location, if so, enter the map match of crossing;

[4] map match of crossing

Comprise the crucial anchor point identification module in crossing, crucial anchor point candidate matches section, crossing mark module, the modeling of decilog map match and weights coefficient estimation module, the crucial anchor point identification module in crossing is for determining the crucial anchor point set that affects crossing's map match precision; Section and last relativeness of mating section that crucial anchor point candidate matches section, crossing mark module connects according to crossing, the candidate matches section of crucial anchor point is defined as candidate matches section, craspedodrome candidate matches section and right-hand rotation candidate matches section, and rejected and limited current candidate matches section according to intersection turning restricting data, make effective candidate matches section of the crucial anchor point in crossing be no more than at most 3; The modeling of decilog map match and weights coefficient estimation module, start with from three kinds of corresponding driver behavior pattern's variance analyses in different candidate matches section, propose the estimation problem in crossing's optimum matching section to be transformed into the selection problem of vehicle to follow-up running section, using the crucial weight of considering that the weight of driver's driving behavior characteristic is selected follow-up running section as differentiation driver, from three candidate road section, select optimum matching section according to decilog model, and adopt maximum likelihood estimate to estimate weights coefficient;

Wherein initial alignment point map match refers to that first anchor point of the vehicle location that gps receiver is received carries out map match, identify its correct coupling section and the location point on coupling section, mainly comprise initial alignment point candidate matches section identification module, the modeling of initial alignment point map match and two modules of weights coefficient estimation module, wherein

(1) initial alignment point candidate matches section identification module

Initial alignment point has material impact to overall map matching precision of the present invention; Initial alignment point candidate matches section identification is by taking initial alignment point as the center of circle, taking the positioning error of gps receiver as radius, build circular error territory, all sections crossing or tangent with this circle error territory are all defined as to candidate matches section, and form the set of candidate matches section ;

(2) modeling of initial alignment point map match and weights coefficient estimation module

1) initial alignment point map match modeling

After candidate matches section is determined, the present invention is according to the conforming weight of travel direction and the weight of consideration anchor point and subpoint proximity as initial alignment point map match weight, candidate matches section is gathered in each candidate matches section , do to calculate as follows:

（1）

The coupling section of initial alignment point is:

（2）

2) initial alignment point map match weight calculation

1. consider the conforming weight of travel direction

First, regulation direct north is , deflection is by increasing progressively clockwise; For ease of calculating, will deflection unification changes to deflection, the span of deflection is , order for vehicle heading angle, for candidate matches section ldeflection, ( , ), ( , ) be respectively candidate matches section lstarting point and the coordinate of terminal,, in the time that candidate matches section is straight line section, have

（3）

In Digital map, represent curve section with the broken line section that multiple shorter straight line sections connect, in broken line section, the intersection point of adjacent straight part of path is flex point, therefore, in the time that candidate matches section is broken line section, if subpoint is between two adjacent flex points, for the deflection in the straight line section between these two flex points, if subpoint drops in flex point, for the deflection in the straight line section of angle absolute value minimum between flex point two ends and vehicle heading;

Consider the conforming weight of travel direction computing formula is:

（4）

（5）

In formula, for corresponding weights coefficient, value try to achieve according to weights coefficients calculation block; for the angle absolute value between vehicle heading and section;

Meanwhile, regulation larger, candidate matches section be coupling section possibility larger, otherwise, less;

2. consider the weight of anchor point and subpoint proximity

Make ( , ), ( , ) be respectively GPS anchor point qwith it in candidate matches section on subpoint coordinate, , ,

（6）

Anchor point is to the vertical projection distance in candidate matches section for:

（7）

When on the straight line section between two adjacent flex points of subpoint in broken line section, in formula (7), the starting point in section and terminal point coordinate become the coordinate of two flex points; In the time that subpoint is in flex point, subpoint coordinate is flex point coordinate; In the time that subpoint is on the extended line in straight line section, directly reject this candidate matches section;

The weight of considering anchor point and subpoint proximity is:

（8）

In formula, for corresponding weights coefficient, value try to achieve according to weights coefficients calculation block; for constant, value is the positioning error of GPS receiver;

Meanwhile, regulation larger, candidate matches section be coupling section possibility larger, otherwise, less;

3) weights coefficient is estimated

Choose at random from multiple sections individual anchor point, wherein be greater than 100, first, only to consider the conforming weight of travel direction as the weight of differentiating coupling section, mate section , order represent the anchor point sum of correct coupling in individual anchor point; Again only to consider the weight of anchor point and subpoint proximity as the weight of differentiating coupling section, mate section , order represent the anchor point sum of correct coupling in individual anchor point, weights coefficient with estimated value with for

（9）

。

2. the crossing's map-matching method based on driver behavior pattern and decilog model according to claim 1, it is characterized in that: crossing's map match mainly comprises that crossing's map match comprises the crucial anchor point identification module in crossing, crucial anchor point candidate matches section, crossing mark module, the modeling of decilog map match and three modules of weights coefficient estimation module, wherein

(1) the crucial anchor point identification module in crossing

These points are defined as the crucial anchor point in crossing by the present invention, and the crucial anchor point of crossing is ; Concrete recognition methods is as follows:

First, judge current anchor point when the subpoint in the coupling section of a upper anchor point is whether on the extended line of the terminal direction in this coupling section, if so, think anchor point for first anchor point in crucial anchor point set, and with thereafter n-1 anchor point forms the crucial anchor point set in crossing jointly ;

(2) the crucial anchor point candidate matches in crossing section mark module

All sections that this crossing is crossing with last coupling section are all considered as candidate matches section, meanwhile, the crucial anchor point candidate matches in crossing section mark is exactly to mark respectively left-hand rotation direction candidate matches section, craspedodrome direction candidate matches section and the right-hand rotation direction candidate matches section with respect to last coupling section in the candidate matches section of the crucial anchor point in crossing;

Order with be respectively candidate matches section with last coupling section deflection, candidate matches section with last coupling section between angle for:

（10）

Candidate matches section with last coupling section relativeness be:

1) if or , think candidate matches section for last coupling section relative craspedodrome section;

2) if , think candidate matches section for last coupling section relative left-hand rotation section;

3) if , think candidate matches section for last coupling section relative right-hand rotation section;

If with respect to there is one or more left-hand rotation or right-hand rotation section, think that with immediate that section of GPS anchor point be relative left-hand rotation or right-hand rotation section, therefore, these anchor points remain unique relative left-hand rotation, craspedodrome and right-hand rotation candidate matches section, be reduced to maximum three by candidate matches section, meanwhile, according to intersection turning restricting data, in the candidate matches section from mark, reject the current candidate matches section of restriction;

(3) modeling of decilog map match and weights coefficient estimation module

According to the difference of vehicle driving behavior characteristic of driver in the time turning left, keep straight on and turn right, from the set of candidate matches section, select optimum matching section;

1) decilog map match modeling

Can be from kplant the selection limb of selecting to have with maximum probability selection in limb maximum utility, utilize this modeling feature of decilog model, regard three candidate matches sections of the crucial anchor point in crossing as three and select limb, effectiveness using the weight sum in each candidate matches section as this candidate matches section, according to k-Logit computing formula is calculated the matching probability in each candidate matches section, and to there is the candidate matches section of maximum matching probability as the optimum matching section of current anchor point, kthe modeling process of-Logit model is as follows:

Suppose sfor the crucial anchor point in crossing qthe set of selection limb,

s={ left-hand rotation candidate matches section, craspedodrome candidate matches section, right-hand rotation candidate matches section } (11)

For each candidate matches section, if coupling section, the selection result of this selection limb is 1, if not, value is 0;

If the left-hand rotation candidate matches section of certain crossing crucial anchor point is coupling section, this anchor point respectively selects the selection result of limb to be

（12）

Order for candidate matches section weight sum,

（14）

In formula: for candidate matches section ? individual weight, for candidate matches section stochastic error, suppose with separate, and the stochastic error in each candidate matches section is separate and all obey double exponential distribution;

kthe probability selection formula of-Logit model is:

（15）

Anchor point qoptimum matching section for:

（17）

Meanwhile, anchor point qin optimum matching section on vertical projection point be the optimal match point of vehicle;

2) map match weight calculation

Ride characteristic according to vehicle in crossing, choose the map match weight of four kinds of weights as the crucial anchor point in crossing, wherein, consider the conforming weight of travel direction and consider anchor point and the weighing computation method of subpoint proximity and identical above, consider the weight of driver's driving behavior characteristic and consider that the computing method of the conforming weight of journey time are as follows:

1. consider the weight of driver's driving behavior characteristic

Using oriented the single order at vehicle heading angle difference as characterizing the direct acting factor of the driver of crossing driving behavior to vehicle driving trace, with continuous three sampling periods the weighting of the oriented difference of single order at vehicle heading angle as the weight of considering driver driving behavior characteristic, and think that the oriented differential pair of single order at vehicle heading angle relatively turns left, keeps straight on and three sections of turning right show different weights;

Order for GPS anchor point the oriented difference of single order at vehicle heading angle,

（18）

Meanwhile, for Efficient Characterization vehicle is the turning to and corner amplitude of crossing, right value do following correction:

（19）

Vehicle is within the scope of crossing, in formula (19) when value is negative, think that vehicle heading is to counterclockwise rotation, when exceed certain threshold value, think that vehicle may turn left to travel; value is timing, thinks that vehicle heading rotates to clockwise direction, when exceed certain threshold value, think that vehicle may travel turning right;

Order , and be respectively the weight function vector of considering driver's driving behavior characteristic, , and be respectively corresponding weights coefficient vector, and have:

（20）

In formula, , and be respectively constant, get the maximal value that in the sampling period, vehicle angles changes;

While considering driver's driving behavior characteristic, the weight in the section of turning left in candidate matches section , craspedodrome section weight and the weight in right-hand rotation section for:

（21）

2. consider the conforming weight of journey time

Consider that journey time conforming weight refers to that weighing computation method is as follows for last match point is optimum matching section to the candidate matches section of the difference minimum in the vehicle travel time between subpoint on candidate matches section and sampling period:

Order with represent respectively adjacent positioned point with vehicle Speed, to vehicle from anchor point match point travel to anchor point the travel time estimation of the subpoint on certain candidate matches section for:

（22）

In formula, lfor Vehicle Driving Cycle distance;

Journey time with the sampling period difference be:

（23）

Consider the conforming weight of journey time for:

（24）

In formula, for considering the conforming weight function of journey time, for corresponding weights coefficient, for constant, within the sampling period maximal value;

If candidate matches section sum is less than 3, anchor point as crucial in a certain crossing only has left-hand rotation candidate matches section and craspedodrome candidate matches section, makes the entitlement weight average value in right-hand rotation candidate matches section be 0;

3) maximal possibility estimation of weights coefficient

Adopt maximum likelihood estimate to estimate kthe weights coefficient of-Logit , computing method are as follows:

According to maximum likelihood estimate, suppose the crucial anchor point set in crossing in road network in iindividual anchor point to the selection result of each selection limb is , the joint probability density of each selection result is:

（25）

Therefore, anchor point the probability of the selection result to candidate matches section, namely likelihood function is:

（26）

Right take the logarithm to such an extent that log-likelihood function is:

（27）

Make lreach peaked maximum likelihood function estimate vector should be the solution of following partial differential simultaneous equations:

（28）

?

（29）

Due to , abbreviation formula (29) to make it be 0,

（30）

Formula (30) is also 's junit's Nonlinear System of Equations; Let it be to the greatest extent, and solution not necessarily exists, but in the time having solution, solution will be unique;

Adopt analytical method cannot solve the Nonlinear System of Equations shown in formula (30), the present invention adopts Newton Algorithm , iterative computation formula be: (31)

In formula, , be respectively 's minferior and m+ 1 estimate vector, , be respectively lgradient function and Hessian inverse of a matrix matrix;

Formula (31) stopping criterion for iteration is:

（32）

（33）

In formula, , be respectively predefined allowable error value, get , ;

Parameter vector in each parameter tinspection computing formula is:

（34）

In formula, for weights coefficient t statistical value, for matrix on middle diagonal line kindividual element.