CN102831766A - Multi-source traffic data fusion method based on multiple sensors - Google Patents

Abstract

The invention discloses a multi-source traffic data fusion method based on multiple sensors. The method comprises the following steps: unifying a space-time standard of multi-source traffic data acquired by the multiple sensors; adding space-time tags to the multi-source traffic data acquired by the multiple sensors, and converting the multi-source traffic data into a uniform format; fusing the multi-source traffic data in uniform format after carrying out space-time matching, and matching with a map so as to obtain fusion data; and combining the fusion data and subsequently outputting the fusion data. The invention provides a method for fusing multi-source traffic data information obtained through the observation of various sensors, and the method can be applied to road side systems in intelligent traffic fields.

Description

Multi-source traffic data fusion method based on multisensor

Technical field

The invention belongs to the intelligent transport technology field, relate in particular to a kind of multi-source traffic data fusion method based on multisensor.

Background technology

Data fusion (data fusion) is also referred to as information fusion; Be the process of an informix and processing, general definition is: the information process that utilizes computer technology that the observation information of some sensors of obtaining is chronologically in addition analyzed automatically, comprehensively carried out to accomplish required decision-making and estimation task under certain criterion.Data fusion is multi-level, a many-sided data handling procedure; Its ultimate principle makes full use of each sensor resource exactly; Observation information through these sensors are obtained is rationally utilized; Combine multiple sensors redundant with complementary in the space or on the time according to certain principle, the consistance of measurand is described obtaining, improve the validity of sensor.The internal of the information fusion of multisensor is that it can occur on different dimensions, different levels, different time sections, has more complicated character and more near the Intelligent Calculation of human brain.

Data fusion is divided into by the abstraction hierarchy that merges object: data level merges, the characteristic level merges and decision level fusion.

(1) data level merges

Sensing data to the most original merges, and generally only is applied between the sensor of the same race, and its advantage is to preserve more field data, and trickle information is provided, and shortcoming is that data volume is big, and it is high to handle cost, and real-time is poor.

(2) the characteristic level merges

The characteristic information that utilization is extracted from sensing data fully utilizes and handles, and its advantage is to have implemented considerable data compression, and has extracted the characteristic information relevant with decision-making, and shortcoming is that Feature Extraction and method for expressing are various, does not have unified pattern.

(3) decision level fusion

In the data process analyzing and processing of each sensor, and made the more higher leveled fusion of making on the basis of basic judgement and decision-making.Its advantage is to have higher fault-tolerance, when certain sensor breaks down or be wrong, still can obtain correct result through rational fusion structure.Before the fusion of this level, carried out a series of processes such as pre-service, feature extraction, Target Recognition, decision-making judgement, made preliminary judgement, for fusion provides more reliable data.Shortcoming is that the processing cost of fusion front end is higher.

In the intelligent transport technology field; Utilize polytype sensor obtain more multi-source, more accurately, site traffic information is the developing direction of intelligent transportation more reliably, so need the multi-source traffic data information that various kinds of sensors observation obtains be merged.

Summary of the invention

The purpose of this invention is to provide a kind of multi-source traffic data fusion method based on multisensor, the inventive method has realized that the multi-source space-time data that multisensor is observed carries out robotization and merges.

In order to achieve the above object, the present invention adopts following technical scheme:

A kind of multi-source traffic data fusion method based on multisensor comprises step:

The space-time benchmark of the multi-source traffic data that unified multisensor is gathered is to realize the synchronous recording of multisensor institute image data;

The multi-source traffic data that arrives to multi-sensor collection increases the space-time label, and is converted into consolidation form;

Multi-source traffic data to uniform format carries out merging after the space-time coupling, obtains fused data with map match;

Fused data is merged back output.

In the space-time benchmark of the multi-source traffic data that above-mentioned unified multisensor is gathered, the space-time benchmark that is adopted is WGS-84 coordinate system and split-second precision benchmark.Described split-second precision benchmark is 10 ^-12Second.

The above-mentioned employing principle of least square merges the multi-source traffic data after the space-time coupling.

The multi-source traffic data fusion method based on multisensor that the present invention proposes is: at first data are carried out pre-service, unify the space-time benchmark of various data, and be translated into the data of consolidation form; Then the data after the conversion are merged and map match; At last the data after handling are merged and output.The inventive method can be used for the driver test system in the intelligent transportation system.Adopt the various traffic events of the present invention's computing in real time, telecommunication flow information, traffic detection information etc., handled information is used for the trackside system, can solve the safety and efficiency problem in vehicle and the pedestrian's driving process timely.

Existing information fusion method, what have exists the big problem of freight volume, and the method that has is not suitable for practical application.The inventive method merges based on data Layer, and operand is moderate, is convenient to practical application, and miscue fast can obtain more information accurately through the optimal combination of available data effectively, and can carry out the secondary combination to a certain extent, and is convenient, flexible.

Description of drawings

Fig. 1 is the process flow diagram of the inventive method.

Embodiment

The inventive method is mainly used in and adopts various kinds of sensors to obtain the trackside system of transport information, and described various kinds of sensors can be time service sensors such as passive and synchronous sensor such as active sensors such as CCD camera, range finder using laser, temperature sensor and laser scanner.

To combine embodiment that the inventive method is described further below.

Multi-source traffic data fusion method based on multisensor provided by the invention comprises step:

1) adopt integrated satnav chip and time service chip, the space-time benchmark of the multi-source traffic data that unified multisensor is gathered to realize the synchronous recording of multisensor institute image data, lets the data of all collections all be unified in on the time shaft.

The space-time benchmark is specially time reference and space reference.This step specifically comprises following substep (can be CN101949715A referring to publication number, the integrated synchronisation control means of multisensor that open day to be on January 19th, 2011, denomination of invention obtain for the high precision space-time data and the Chinese patent of system):

1-1 sets up the space-time reference circuit time reference and space reference is provided, and the time reference that is provided is 10 ^-12The split-second precision benchmark of second, the space reference that is provided is the WGS-84 geodetic coordinate system;

During enforcement, the foundation of time reference is specially: adopt CPLD (CPLD) that the 10MHz pulse frequency division is produced millisecond pulse and pulse per second (PPS); Filtering and bit synchronization through to the PPS pulse signal of GPS realize the gps time calibration, after being about to pulse per second (PPS) that CPLD produces and the pulse per second (PPS) of GPS being alignd, and the operation of driving internal clock.

During enforcement, the foundation of space reference is specially: signal of sensor such as comprehensive GPS, speed pickup (or range sensor) and micromechanical gyro, (DR Dead-Reckoning) sets up space reference to adopt the dead reckoning algorithm.

1-2 sets up the linear reference coordinate system through above-mentioned time reference and space reference, and realizes related between linear reference coordinate system and earth coordinates and change;

Setting up the linear reference coordinate system is specially:

Information through the space-time reference circuit is gathered gps coordinate, gps time, vehicle ' distance and multisensor is carried out synchronous processing, carries out related through the information that time mark is gathered locator data and multisensor; The distance that the space-time reference circuit is periodically exported vehicle current GPS coordinate and gone, during the information of gathering as sensor input, the time information that takes place of the output transducer information of gathering immediately; Be the linear reference coordinate of current location according to the distance of vehicle ' and the linear reference coordinate sum of vehicle operating starting point.

Conversion between linear reference coordinate system and earth coordinates is specially:

Linear reference coordinate P according to impact point; In route sheet, confirm the highway section and the straight-line segment S at impact point place fast; And according to this straight-line segment S the direction that next step calculates terrestrial coordinate is confirmed in the ordering in all straight-line segments in this highway section; When this straight-line segment sequence number smaller or equal to two/for the moment of total hop count, begin to calculate from the starting point of this section; When this straight-line segment sequence number greater than two/for the moment of total hop count, begin to calculate from the terminal point of this section; Then, with the starting point P in this highway section _sOr terminal point P _eBe transformed into rectangular space coordinate from terrestrial coordinate, then according to the order of straight-line segment, the rectangular space coordinate of calculating the terminal point of each straight-line segment successively according to the length and the direction of this straight-line segment; For the straight-line segment at impact point place, it is poor that its length equals the starting point of linear reference coordinate and this straight-line segment of impact point when calculating.

1-3 is to the synchro control of active sensor:

According to preset parameters, active sensor is sent the pulse simulation control signal, realize synchro control to active sensors such as CCD camera, range finder using lasers, active sensor can send the synchronization point data to host computer through serial ports or USB simultaneously.

1-4 is to the synchro control of passive sensor:

Receive the external event pulse signal, and respond the interruption that this external event pulse signal produces, obtain the moment and positional information that this pulse signal produces, and send the synchronization point data to host computer, realize synchro control passive sensor through serial ports or USB.

1-5 is to the synchro control of time service sensor:

Send whole second time information through serial ports and USB to host computer, and send pps pulse per second signal, can realize clock synchronization this type of laser scanner time service sensor through Simulation with I/O mouth.

2) the multi-source traffic data that arrives to multi-sensor collection increases the space-time label, and is converted into consolidation form.

In this practical implementation; Multisensor is periodically gathered traffic data; Collection period is 50ms; The multi-source traffic data that is collected has traffic event information, traffic to detect information and telecommunication flow information three major types, and adopting in advance, the agreement of regulation is converted into consolidation form with above-mentioned three major types information:

At first stipulate employed data type in this agreement: numeric type N and character type C, numeric type N is with N (A) expression integer, and A representes the breadth extreme of numerical value; (A representes the breadth extreme of numerical value for A, B) expression decimal, and B representes the figure place behind the radix point with N.Character type C is with C (A) expression, and wherein A representes the breadth extreme of character.

A, traffic event information

According to the traffic event information that sensor transmits, adopt N (2) categorical data to represent the traffic events classification, it is fixed that the traffic events classification can be come according to the traffic-information service standard, adopts different numerical to represent different classes; Adopt N (2) categorical data to represent the traffic events Alert Level, the traffic events Alert Level comes fixed different Alert Level to different traffic events classifications, adopts different numerical to represent different classes; Adopt C (9) categorical data to represent license plate number, C (9) comprises following content: letter, five bit numbers (can be digital, also can be the mixing of numeral and letter) of city codes represented in the Chinese character, that be called for short in a representative province; Adopt N categorical data presentation of events time of origin, the time time of origin in this practical implementation is the information acquisition time, and the time of being obtained is accurate to millisecond, adopt respectively N (2) categorical data represent hour, minute, second and millisecond; The place of adopting N (2) categorical data to come presentation of events to take place, the sensor that is positioned at different highway sections has all been set different ID, so the locale in this practical implementation adopts ID number expression of sensor.

B, traffic detect information:

Adopt N (2) categorical data to represent the detected object type, 1 is the pedestrian, and 2 is vehicle, and 3 is other;

Adopt N (10) categorical data to represent to gather the sensor ID number of this information;

Adopt two N (4) categorical data to represent the detected object position respectively, wherein a N (4) categorical data is the residing longitude of detected object, and another N (4) categorical data is the residing dimension of detected object;

Adopt two N (2) categorical data to represent the speed and the acceleration of detected object respectively.

C, telecommunication flow information

Adopt N (10) categorical data to represent to gather the sensor ID number of this information;

Adopt N (4.2) categorical data to represent roadway occupancy;

Adopt N (4.2) categorical data to represent road traffic;

Adopt N (2) categorical data to represent to detect the vehicle number.

3) the multi-source traffic data to consolidation form carries out the space-time coupling, adopts the principle of least square that the multi-source traffic data after the space-time coupling is merged, and the gained data is carried out map match obtain fused data.

To be elaborated to the process that the multi-source traffic data after the space-time coupling merges to adopting the principle of least square below.

In this process be traffic parameter that all are to be detected as unknown number, the data that each sensor is gathered are as observed reading.Suppose that all traffic parameters to be detected are β ₀, β ₁..., β _m, observed reading is y ₁, y ₂..., y _n, set up multiple linear regression model:

$\left.\begin{array}{c}{y}_1={\mathrm{\β}}_0+{\mathrm{\β}}_1{x}_{11}+{\mathrm{\β}}_2{x}_{12}+...+{\mathrm{\β}}_m{x}_{1m}+{\mathrm{\ϵ}}_1\\ y_2={\mathrm{\β}}_0+{\mathrm{\β}}_1{x}_{21}+{\mathrm{\β}}_2{x}_{22}+...+{\mathrm{\β}}_m{x}_{2m}+{\mathrm{\ϵ}}_2\\ .\\ .\\ .\\ y_n={\mathrm{\β}}_0+{\mathrm{\β}}_1{x}_{n1}+{\mathrm{\β}}_2{x}_{n2}+...+{\mathrm{\β}}_m{x}_{nm}+{\mathrm{\ϵ}}_n\end{array}\right\}---\left(1\right)$

Wherein, x _IjBe to measure and controllable nonrandom variable, be empirical value, ε _iBe stochastic error, and E (ε _i)=0, D (ε _i)=σ ², E (ε _i) expression ε _iExpectation value, D (ε _i) expression ε _iVariance, i=1,2 ..., n, j=1,2 ..., m.

If note

$\underset{n,1}{Y}=\left[\begin{array}{c}{y}_1\\ y_2\\ .\\ .\\ .\\ y_n\end{array}\right],$ $\underset{m+1,1}{\mathrm{\β}}=\left[\begin{array}{c}{\mathrm{\β}}_1\\ {\mathrm{\β}}_2\\ .\\ .\\ .\\ {\mathrm{\β}}_m\end{array}\right],$ $\underset{n,m+1}{X}=\left[\begin{array}{ccccc}1& x_{11}& x_{12}& ...& x_{1m}\\ 1& x_{21}& x_{22}& ...& x_{2m}\\ .& .& .& & .\\ .& .& .& & .\\ .& .& .& & .\\ 1& x_{n1}& x_{n2}& ...& x_{\mathrm{nm}}\end{array}\right],$ $\underset{n,1}{\mathrm{\ϵ}}=\left[\begin{array}{c}{\mathrm{\ϵ}}_1\\ {\mathrm{\ϵ}}_2\\ .\\ .\\ .\\ {\mathrm{\ϵ}}_n\end{array}\right],$

Then have: Y=X β+ε (2)

Try to achieve m+1 unknown regression parameter β by formula (1) ₀, β ₁..., β _mLeast square

Valuation, the gained valuation can be as treating measured value β ₀, β ₁..., β _m

Can adopt following method to judge the degree of accuracy of gained valuation:

Following error equation is formed in least square

valuation:

$V=X\widehat{\mathrm{\β}}-Y---\left(3\right)$

Wherein,, $\mathrm{\β}=\left[\begin{array}{c}{\widehat{\mathrm{\β}}}_0\\ {\widehat{\mathrm{\β}}}_1\\ .\\ .\\ .\\ {\widehat{\mathrm{\β}}}_m\end{array}\right]$

At least-squares estimation V ^TUnder the criterion of V=min, must normal equation be:

$X^{T}X\widehat{\mathrm{\β}}=X^{T}Y---\left(4\right)$

Through type (4) can solve

Set up multiple linear regression equations and residual error V according to

:

$\hat{Y}={\widehat{\mathrm{\β}}}_0+{\widehat{\mathrm{\β}}}_1{x}_1+{\widehat{\mathrm{\β}}}_2{x}_2+\·\·\·+{\widehat{\mathrm{\β}}}_m{x}_m=X\widehat{\mathrm{\β}}---\left(5\right)$

$V=\hat{Y}-Y---\left(6\right)$

Using

The co-factor

and variance

to evaluate

valuation accuracy:

$Q_{\widehat{\mathrm{\β}}\widehat{\mathrm{\β}}}={\left(X^{T}X\right)}^{-1}---\left(7\right)$

$D\left(\widehat{\mathrm{\β}}\right)={\mathrm{\σ}}^2{Q}_{\widehat{\mathrm{\β}}\widehat{\mathrm{\β}}}---\left(8\right)$

The precision of observed reading Y is evaluated in the variance valuation

of employing observed reading Y:

${\widehat{\mathrm{\σ}}}^2=\frac{V^{T}V}{n-(m+1)}---\left(9\right)$

4) combining step 3) gained fused data and output, realize detection of dynamic, location, classification and the issue of traffic events.

Claims (3)

1. the multi-source traffic data fusion method based on multisensor is characterized in that, comprises step:

The space-time benchmark of the multi-source traffic data that unified multisensor is gathered is with the synchronous recording of institute's image data of realizing multisensor;

The multi-source traffic data that arrives to multi-sensor collection increases the space-time label, and is converted into consolidation form;

Multi-source traffic data to uniform format carries out merging after the space-time coupling, obtains fused data with map match;

Fused data is merged back output.

2. the multi-source traffic data fusion method based on multisensor according to claim 1 is characterized in that:

Described space-time benchmark is the WGS-84 earth coordinates split-second precision benchmark of unifying.

3. the multi-source traffic data fusion method based on multisensor according to claim 1 and 2 is characterized in that:

Adopt the principle of least square that the multi-source traffic data after the space-time coupling is merged.

Images