CN100466010C - Different species traffic information real time integrating method - Google Patents

Abstract

This invention discloses one method for information process in traffic technique by use of abnormal traffic information for real time traffic status merging estimation, which comprises the following steps: sensor test information property extracting; b, reliability matrix W determining; c, reliability establishing; d, property merging module; e, final property deciding.

Description

Different species traffic information real time integrating method

Technical field

The present invention relates to the method for information processing in a kind of technical field of traffic transportation, specifically is a kind of different species traffic information real time integrating method.

Background technology

Along with the development of intelligent transport technology, the dynamic estimation of urban traffic flow has been subjected to paying close attention to all the more widely.And traffic behavior estimates to be unable to do without transport information accurately and reliably, and the precision that therefore detects information approach has determined the accuracy that traffic behavior is estimated.City road network traffic flow detection information approach has multiple, mainly contains magnetic inductive coil detection method frequently, GPS probe vehicles detection method, Video Detection method etc.These methods have obtained success in some aspects and have had its practical value, but have limitation separately aspect the estimation of city road net traffic state.The inductive coil detecting device can obtain multiple traffic flow parameter, lacks the inaccurate effect of estimating that also influenced greatly of state representation amount (average velocity, running time) reduction formula in a large number but its high failure rate causes detecting data; The information detecting method of applying GPS (Global Positioning System (GPS)) probe vehicles, by moving vehicle is monitored in real time, vehicle locating information such as dynamic acquisition time, position, speed, has the precision height, widely distributed in the city scope, be subjected to advantages such as weather effect is little, shortcoming also causes estimated result inaccurate but its sampling rate is low, and the map-matching algorithm error is big etc.; Video Detection is had relatively high expectations to hardware device, is subjected to weather effect big, also can't accomplish the traffic network state is carried out comprehensive and accurate estimation.In view of the situation, there is the method that proposes the introducing information fusion to improve the accuracy that traffic behavior is estimated, that is to say, merging, traffic behavior is carried out more accurate comprehensive estimation by the complementary information between the multiple detecting device with detected transport information of multiple foreign peoples's detecting device such as inductive coil detecting device, GPS probe vehicles detecting devices.

Find by prior art documents, the paper that people such as R.L.Cheu delivered on " IEEE IntelligentTransportation Systems Conference Proceedings " (IEEE intelligent transportation system meeting files) in calendar year 2001 " An arterial speed estimation model fusing data from stationary and mobilesensors ", (being used for merging a major trunk roads velocity estimation model of static and dynamic pickup data) paper fusion that has realized detection ring data and GPS probe vehicles data based on the model of neural network, and verified the effect of model with emulated data by emulation platform, yet, neural network needs the lot of data true value to train, and this is to be difficult to realize in actual applications; The paper that people such as K.Choi delivered on Intelligent TransportationSystems (intelligent transportation system) in 2002 " A data fusion algorithm for estimating linktravel time ", (highway section running time estimate data anastomosing algorithm) wherein, proposed to merge this two kinds of data based on the algorithm of fuzzy theory, obtained reasonable effect, but this algorithm has very high complicacy, and portable difference is difficult to adapt to the real time fusion environment of transport information; People such as E.Faouzi were published in " Multisensor in 2006 at them, Multisource Information Fusion:Architecture, Algorithm, Application, Processing of SPIE " (SPIE meeting---multisensor, Multi-source Information Fusion: framework, algorithm and application) on paper " Classifiers and distance-based evidential fusion for road travel timeestimation " (being used for estimating merging of highway section running time) based on sorter with based on the evidence of distance a kind of Multiple Classifier Fusion method based on evidence theory has been proposed, for traffic behavior estimates to provide a new thinking, yet, said method is owing to be difficult to realize the real time fusion estimation, need a large amount of priori datas, problems such as portable difference are used so limited its engineering greatly.

Summary of the invention

The objective of the invention is to overcome deficiency of the prior art, a kind of different species traffic information real time integrating method is provided, make its actual needs at traffic engineering, on the feature level, carry out real-time traffic behavior and merge and estimate, finally for the control of entire city traffic with induce and provide in real time, detect comprehensively and accurately information material.

The present invention is achieved by the following technical solutions, and the present invention specifically may further comprise the steps:

(1) feature extraction of sensor detection information;

(2) the reliability matrix W determines;

(3) foundation of reliability;

(4) feature level Fusion Model;

(5) decision-making of final feature.

The feature extraction of described sensor detection information is meant: the inductive coil data are converted into road-section average speed or journey time by traffic ripple theory; The feature extraction that the GPS probe vehicles detects information becomes road-section average speed to GPS probe vehicles data conversion through coordinate transform, map match, modeling match, and purpose is the characteristic quantity that obtains to become with traffic behavior dull corresponding relation, as the input of emerging system.

Foreign peoples's sensor information is carried out feature extraction can adopt any extracting method that is fit to this type of sensor according to the concrete characteristics of variety classes sensor, and for example: the inductive coil data can be converted into road-section average speed or journey time.

Determining of described reliability matrix W, realize by following steps:

(1) determine the reliability matrix of every kind of sensor according to historical data: promptly based on the method for correct judgment rate, or based on the method for Euclidean distance, or based on the method for evidence distance;

(2) the reliability matrix W is expressed as:

Wherein, w _{I, j}∈ [0,1], (i=1,2 ..., M; J=1,2 ..., N); S _jRepresent j kind state independently to be identified; E _iRepresent independently evidence of i kind.

Because different sensors has different reliabilities for the detection of different traffic, therefore, before fusion process begins, should at first determine the reliability matrix of every kind of sensor according to historical data.

The foundation of described reliability, realize by following steps:

(1) classification of determining traffic behavior to be identified is counted N;

(2) determine the center of gravity of each status categories in each sensor characteristic data, form center of gravity vector G=[g ₁, g ₂..., g _N], can carry out cluster calculation and two kinds of methods of artificial setting with historical data;

(3) serve as that object is discussed with a highway section, after detecting device i detects data, obtain characteristic quantity through feature extraction;

(4) calculate with all centers of gravity apart from d _i=[d ₁, d ₂..., d _N];

(5), set up each elementary probability that detects data and distribute according to following formula:

m _i(A)＝esp(-γ _id _i ^β)

Wherein, parameter beta and γ _iAdjust according to practical application, they are determining the discrimination that elementary probability distributes, and afterwards, need carry out normalized, and the probability sum that all states are assigned with equals 1.

Described feature level Fusion Model comprises:

(1) M following sub-emerging system:

$m_i\left(B_t\right)=m\left(C_{t-1}\right)\⊕m_i\left(A_{i,t}\right)$

$=\frac{{\mathrm{\Σ}}_{C_{t-1}\∩A_{i,t}=B_t}g\left(m\left(C_{t-1}\right)\right)\×m_i\left(A_{i,t}\right)}{1-{\mathrm{\Σ}}_{C_{t-1}\∩A_{i,t}=\mathrm{\Φ}}g\left(m\left(C_{t-1}\right)\right)\×m_i\left(A_{i,t}\right)}$

$\left\{\begin{array}{c}g\left(m\left(C_{t-1}\right)\right)=\mathrm{\λm}\left(C_{t-1}\right)\\ g\left(m\left(\mathrm{\Ω}\right)\right)=1-{\mathrm{\Σ}}_{C_{t-1}\⋐\mathrm{\Ω}}\mathrm{\λm}\left(C_{t-1}\right)\end{array}\right.$

Wherein: m _i(A _{I, t}), i=1,2 ..., M is illustrated in t and constantly the elementary probability that i sensor extracts is distributed; M (C _T-1) the main emerging system of expression is in the fusion results of moment t-1; λ is a variable that is defined between 0 and 1, characterizes m (C _T-1) weakened degree, be called the weakening parameter, adjust according to practical application;

(2) main emerging systems that have feedback arrangement are as follows:

$m\left(C_t\right)=m_1\left(B_t\right)\⊕m_2\left(B_t\right)\⊕\·\·\·\⊕m_M\left(B_t\right)$

$=\frac{{\mathrm{\Σ}}_{\underset{i=1}{\overset{M}{\∩}}{B}_t=C_t}\left(\underset{i=1}{\overset{M}{\mathrm{\Π}}}{m}_i\left(B_t\right)\right)}{1-{\mathrm{\Σ}}_{\underset{i=1}{\overset{M}{\∩}}{B}_t=\mathrm{\Φ}}\left(\underset{i=1}{\overset{M}{\mathrm{\Π}}}{m}_i\left(B_t\right)\right)}$

Wherein: m (C _t) main emerging system is in the fusion results of moment t, also is t final fusion results constantly; m _i(B _t), i=1,2 ..., M represents that subsystem i is in t fusion results constantly.

This fusion structure that has feedback can effectively overcome the shortcoming that is difficult to correctly handle conflicting evidence, makes it be fit to real-time integrated environment more.

The decision-making of described final feature, use maximum reliability rule: after each merges end constantly, the elementary probability that obtains is distributed fusion results judgements of making a strategic decision, find out state that probability counts maximum as this state fusion results constantly, promptly emerging system is in the output in this moment.

The present invention's remarkable result compared with prior art is: do not need a large amount of priori traffic data information and traffic behavior true value information, can real time fusion foreign peoples detection resources information, overcome effectively that traffic detection information is incomplete, inaccurate, noise pollution is serious, be subjected to shortcomings such as concrete detector characteristic influences, for the traffic control and the inducible system of entire city provides more in real time, transport information state estimation comprehensively and accurately, advantages such as but also it is few to have an algorithm parameter, and computing time is short.In the continuous 2 hours estimation effect demonstration test of carrying out with true traffic detection data, the rate of accuracy reached of this method is to more than 95%, and, the time that per 5 minutes data fusion estimation is handled to whole Shanghai City inner ring road net can be satisfied the requirement of real time fusion report traffic state information fully less than 10 seconds.

The present invention is directed to the actual needs of traffic engineering, on the feature level, carry out real-time traffic behavior and merge estimation, have advantages such as computing is easy, reliability is high, practical, particularly can realize foreign peoples's traffic sensor information (inductive coil data, the probe vehicles data, Video Detection data etc.) real-time online merges, and this is that the conventional art method all is difficult to the difficult point that overcomes, also is to obstruct them to realize the major obstacle that real engineering is used.The present invention uses advanced intelligent method, realizes organically blending of different species traffic information, finally for the control of entire city traffic with induce and provide in real time, detect comprehensively and accurately information material.

Description of drawings

Fig. 1 the inventive method step block diagram

The traffic of Fig. 2 present embodiment detects data and curves figure

Be intraday SCATS in Xuhui District of Shanghai highway section and gps data among the figure, solid line is represented the SCATS data, and dotted line is represented gps data.

Fig. 3 present embodiment carries out the experimental result picture that real time fusion is estimated to data

Embodiment

Below in conjunction with accompanying drawing embodiments of the invention are elaborated: present embodiment has provided detailed embodiment and process being to implement under the prerequisite with the technical solution of the present invention, but protection scope of the present invention is not limited to following embodiment.

In order to understand the method that present embodiment proposes better, choose the Xuhui District of Shanghai highway section and carry out traffic behavior convergence analysis example, can be applied to the different road networks of different cities, highway section.Present embodiment requires to provide the detection data of SCATS inductive coil detecting device in this highway section in a day, comprises information such as the magnitude of traffic flow, coil free time; Also have the detection data of GPS probe vehicles detecting device, comprise information such as coordinate, the speed of a motor vehicle, headstock direction, state, time; Also has the GIS-T digital map information.

As shown in Figure 1, the concrete implementation step of present embodiment is as follows:

(1) SCATS data and gps data are carried out feature extraction

On the one hand, utilize the theoretical modeling of macroscopical traffic ripple, the inductive coil detector data is converted to the input of road-section average speed as emerging system, wherein road-section average speed is satisfied is dull corresponding with traffic behavior; On the other hand, through coordinate transform, map match, three steps of modeling match GPS probe vehicles data conversion is become road-section average speed, as the input of emerging system, road-section average speed is satisfied here is dull corresponding with traffic behavior.

(2) the reliability matrix W determines

At these two kinds reliability matrixes that detect information of True Data training focus utilization evidence distance calculation, result of calculation is:

In the formula, { S ₁}～{ S ₅Represent traffic behavior is divided into 5 kinds of independently states, in Fig. 3, use 1～5 numeral respectively.E ₁And E ₂Represent the evidence that evidence that the SCATS data provide and gps data provide respectively.This reliability matrix representation be static reliability, only relevant with the characteristic of detecting device itself, have nothing to do with environmental change.Sensor has determined that its reliability vector has just been determined, so this process was carried out, can be the parameter of reliability matrix W as system before merging.

(3) foundation of reliability

At first divide 5 kinds of traffic behavior S according to the average velocity feature ₁: 0～10, S ₂: 10～20, S ₃: 20～30, S ₄: 30～45, S ₅: 45～60, unit be km/hour.Then, calculate their center of gravity, the center of gravity is here thought their center, i.e. S ₁: 5, S ₂: 15, S ₃: 25, S ₄: 37.5, S ₅: 52.5.With a highway section serves as that object is discussed, when two kinds of detector data obtain separately average velocity characteristic quantity through feature extraction after, calculate with all state centers of gravity apart from d _i=[d ₁, d ₂..., d _N], then, set up each elementary probability that detects data according to following formula and distribute.

m _i(A)＝esp(-γ _id _i ^β)

Wherein, parameter is set to respectively: β=2, γ _i=20.Afterwards, also will carry out normalized, the probability sum that all states are assigned with equals 1.

(4) real time fusion is calculated

Can merge calculating to two kinds of experimental datas with this Fusion Model now, algorithm parameter is set to: λ=0.8, and the time interval is 5 minutes, blending algorithm is programmed with the Visual C Plus Plus.

(5) decision-making of final feature

To each fusion results constantly---the elementary probability of traffic behavior distributes, with the judgement of making a strategic decision of maximum reliability rule, as this traffic behavior output constantly, simultaneously this elementary probability is distributed the input end of the system that feeds back to, participate in fusion treatment as one of next input constantly.

Experimental result as shown in Figure 3.By comparing with Fig. 2, this algorithm can merge estimation effectively, follow the tracks of traffic behavior on the feature level as can be seen.Simultaneously, adopt the video recording data of in 2:00 that afternoon～4:00 time period this highway section being taken, verified the accuracy of fusion results.The result shows, judges that to making a next state in per 5 minutes in this highway section 2 hours accuracy reaches more than 95%.

Claims (5)

1, a kind of method of different species traffic information real time fusion is characterized in that, specifically may further comprise the steps:

(1) feature extraction of sensor detection information;

(2) the reliability matrix W determines;

(3) foundation of reliability;

(4) feature level Fusion Model;

(5) decision-making of final feature;

The foundation of described reliability, realize by following steps:

(1) classification of determining traffic behavior to be identified is counted N;

(2) determine the center of gravity of each status categories in each sensor characteristic data, form center of gravity vector G=[g ₁, g ₂..., g _N], carry out cluster calculation and two kinds of methods of artificial setting with historical data;

(3) serve as that object is discussed with a highway section, after detecting device i detects data, obtain characteristic quantity through feature extraction;

(4) calculate with all centers of gravity apart from d _i=[d ₁, d ₂..., d _N];

(5), set up each elementary probability that detects data and distribute according to following formula:

m _i(A)＝esp(-γ _id _i ^β)

Wherein, parameter beta and γ _iAdjust according to practical application, they are determining the discrimination that elementary probability distributes, and afterwards, need carry out normalized, and the probability sum that all states are assigned with equals 1.

2, the method for different species traffic information real time fusion according to claim 1 is characterized in that, the feature extraction of described sensor detection information is meant: the inductive coil data are converted into road-section average speed or journey time by traffic ripple theory; The feature extraction that the GPS probe vehicles detects information becomes road-section average speed to GPS probe vehicles data conversion through coordinate transform, map match, modeling match, as the input of emerging system.

3, the method for different species traffic information real time fusion according to claim 1 is characterized in that, determining of described reliability matrix W realizes by following steps:

(1) determine the reliability matrix of every kind of sensor according to historical data, based on the method for correct judgment rate, or based on the method for Euclidean distance, or based on the method for evidence distance;

(2) the reliability matrix W is expressed as:

Wherein, w _{I, j}∈ [0,1], (i=1,2 ..., M; J=1,2 ..., N); S _jRepresent j kind state independently to be identified; E _iRepresent independently evidence of i kind.

4, the method for different species traffic information real time fusion according to claim 1 is characterized in that, described feature level Fusion Model comprises:

(1) M following sub-emerging system:

$m_i\left(B_t\right)=m\left(C_{t-1}\right)\⊕m_i\left(A_{i,t}\right)$

$=\frac{{\mathrm{\Σ}}_{C_{t-1}\∩A_{i,t}=B_t}g\left(m\left(C_{t-1}\right)\right)\×m_i\left(A_{i,t}\right)}{1-{\mathrm{\Σ}}_{C_{t-1}\∩A_{i,t}=\mathrm{\Φ}}g\left(m\left(C_{t-1}\right)\right)\×m_i\left(A_{i,t}\right)}$

$\left\{\begin{array}{c}g\left(m\left(C_{t-1}\right)\right)=\mathrm{\λm}\left(C_{t-1}\right)\\ g\left(m\left(\mathrm{\Ω}\right)\right)=1-{\mathrm{\Σ}}_{C_{t-1}\⋐\mathrm{\Ω}}\mathrm{\λm}\left(C_{t-1}\right)\end{array}\right.$

Wherein: m _i(A _{I, t}), i=1,2 ..., M is illustrated in t and constantly the elementary probability that i sensor extracts is distributed; M (C _T-1) the main emerging system of expression is in the fusion results of moment t-1; λ is a variable that is defined between 0 and 1, characterizes m (C _T-1) weakened degree, be called the weakening parameter, adjust according to practical application;

(2) main emerging systems that have feedback arrangement are as follows:

$m\left(C_t\right)=m_1\left(B_t\right)\⊕m_2\left(B_t\right)\⊕\·\·\·\⊕m_M\left(B_t\right)$

$=\frac{{\mathrm{\Σ}}_{\underset{i=1}{\overset{M}{\mathrm{}\∩\mathrm{}}}{B}_t=C_t}\left(\underset{i=1}{\overset{M}{\mathrm{\Π}}}{m}_i\left(B_t\right)\right)}{1-{\mathrm{\Σ}}_{\underset{i=1}{\overset{M}{\mathrm{}\∩\mathrm{}}}{B}_t=\mathrm{\Φ}}\left(\underset{i=1}{\overset{M}{\mathrm{\Π}}}{m}_i\left(B_t\right)\right)}$

Wherein: m (C _t) main emerging system is in the fusion results of moment t, also is t final fusion results constantly; m _i(B _t), i=1,2 ..., M represents that subsystem i is in t fusion results constantly.

5, the method for different species traffic information real time fusion according to claim 1, it is characterized in that, the decision-making of described final feature, use maximum reliability rule: after each merges end constantly, the elementary probability that obtains is distributed the fusion results judgement of making a strategic decision, find out state that probability counts maximum as this state fusion results constantly, promptly emerging system is in this output constantly.

Images