CN103954275A - Lane line detection and GIS map information development-based vision navigation method - Google Patents

Abstract

The invention discloses a lane line detection and GIS map information development-based vision navigation method. The method comprises the following steps: acquiring GIS map information, preprocessing, carrying out real-time image acquisition and lane line detection to generate a GIS local map in real-time, carrying out coupling verification on a track line detection result and the GIS local map through an ICP algorithm, and generating a high-credibility track line result according to the coupling result. The method can realize the credibility evaluation of the lane line detection through a geographic information system, corrects when false detection or missing inspection appears, expands the application range of the lane detection algorithm, and also can improve the map positioning precision through visual perception information, so the performances of a vision navigation system in a whole intelligent driving and aided driving system are improved, thereby the system can adapt to more complex environments.

Description

Vision navigation method based on lane detection and the exploitation of GIS cartographic information

Technical field

The invention belongs to unmanned field of intelligent control, relate to the vision navigation method that a kind of Vehicular intelligent is driven, especially a kind of based on lane detection and Geographic Information System (Geographic InformationSystem, GIS) vision navigation method of cartographic information combined together exploitation, for the reliability of Real-Time Evaluation and verification traditional vehicle diatom testing result, improve the precision of intelligent driving vision guided navigation.

Background technology

Vision guided navigation is the gordian technique step in intelligent driving and auxiliary driving, is by the processing of vision data (being generally view data), and Useful Information in extraction environment, for Driving Decision-making provides reliable basis.Lane detection is one of them importance, generally by forward sight collected by camera image, carries out image processing, obtains the information of road surface situation, is included in quantity, position, width and the bifurcated etc. that crosses of lane line under the different scenes such as city, rural area and high speed.Lane detection is mainly divided into following module: image pre-service, feature extraction, track models fitting, the conversion of time domain association and image and world coordinates.Current lane detection system can meet the requirement under basic scene, but method based on vision can be subject to the impact of many factors, and such as road multi-obstacle avoidance, pavement markers are unclear, weather effect and illumination variation etc.These factors can cause system to occur flase drop and undetected, and system lacks the evaluation to the verification of testing result and confidence level.On the other hand, Geographic Information System (GIS), GPS and Inertial Measurement Unit (IMU) also start to be widely used in automatic Pilot and auxiliary location of driving and navigation.The precision of its measurement and location is key issue.The GPS of current commercial use can reach the accuracy rating of 5-10m, in conjunction with inertial navigation unit, can bring up to 1-2m, and the navigation of still track precision being got off is travelled needs precision to propose higher requirement.How to design to realize lane detection is carried out to verification and evaluation, make detection method adapt to different complex environments, strengthen the robustness of algorithm, improve testing result confidence level, the method that simultaneously improves the precision of location and navigation has become one of automatic driving and auxiliary study hotspot of driving.

Summary of the invention

The object of the invention is to overcome above-mentioned technological deficiency, a kind of vision navigation method based on lane detection and the exploitation of GIS cartographic information is provided, this vision navigation method can carry out trust evaluation to lane detection result by Geographic Information System, there is flase drop or proofreading and correct when undetected, expand the scope of application of lane detection algorithm, can improve by visual information the precision of map location again, thereby improve the performance of whole vision navigation system.

For achieving the above object, the present invention has adopted following technical scheme.

During assumed initial state of the present invention, car body is positioned in the GIS map providing, and obtains after initialization data, and the local map of real-time update, then mates verification with lane detection result.

In order to guarantee stability of the present invention and applicability, the method is based on following several hypothesis: 1. by GIS cartographic information through pre-service, provide road edge point that car will running section and accordingly crossing point as prior imformation; 2. the GPS road edge point providing is sequential storage structure with crossing point, and the section corresponding to Vehicle Driving Cycle and guides vehicle to travel by this path successively; By the error of GPS and inertial navigation cell location and precision in tolerance interval (half about 2m of lane width).

The method of the invention mainly comprises the following steps:

1) gather in advance GIS cartographic information and carry out pre-service;

2) the real-time image acquisition driving diatom of going forward side by side detects;

3) generate in real time the local map of GIS;

4) lane detection result is mated to verification with the local map of GIS;

5) according to matching result, generate lane line result with a high credibility.

In described step 1), by pose measurement equipment, gather in advance the road edge data of global map, and road edge data are carried out to smooth filtering processing, obtain the road edge point set of global map.

Described step 2), in, by image capture device (comprising vehicle-mounted digital camera and fixed focus lens etc.), adopt based on detection method monocular vision, that carry out feature extraction and track models fitting under time domain association and detect in real time road surface lane line.

In described step 3), by pose measurement equipment, (comprise inertia combined navigation system, optical fibre gyro, vehicle-mounted odometer and front wheel angle meter etc.), obtain when front vehicle body position in real time, and locate in GIS global map, then according to graph model partly, generate local environment map.

In described step 4), integrating step 2) with the result of step 3), under same scale, use iterative closest point (Iterative Closest Point, ICP) algorithm carries out shiding matching verification to lane detection result and GIS localized road edge, obtain best match position and maximum matching error, during coupling, according to lane line model and graph model group match partly, select optimum one group of coupling.

In described step 5), the matching result obtaining according to step 4), by maximum matching error and threshold under best match position, determines the confidence level of lane detection result, then revises the positional information of translation distance and location.

If testing result is insincere, according to road edge line generating virtual lane line.

By lane line position constraint, improve correctness and the stability of testing result, finally select corresponding scheme to generate the high lane line result of degree of accuracy.

Beneficial effect of the present invention is embodied in:

First the present invention is provided the road edge information of the global map of vehicle institute running region by GIS, then the local lane line result that obtains the local cartographic information of GIS and detect, next GIS information is mated verification mutually with lane detection result, by GIS information, revised trend and the trend of lane line, the road edge being provided by vision lane detection modified result GIS and the position relationship between lane line, finally send the lane line result that confidence level and degree of accuracy are high simultaneously.The object that adds GIS cartographic information is that the road edge by take on map is priori, and the matching degree of moving towards trend by curve is proofreaied and correct flase drop that lane line exists and undetected.On the overall trend of curve, GIS cartographic information has higher confidence level, therefore can proofread and correct thus lane line; But for positional precision, due to instability and the accuracy limitations of GPS, it is with a low credibility in the result of visually-perceptible, therefore can carry out correction position precision by the testing result of visually-perceptible.

The present invention has following characteristics:

1. the present invention can revise lane line flase drop or the inaccurate situation of testing result causing due to factors such as terrestrial reference, light or deep cambers;

2. the present invention can, to out position virtual vehicle diatom accurately, not guarantee that car has continuous perception data in the process of moving when there is no lane detection result;

3. due to the adding of GIS prior imformation, greatly improved the reliability of lane detection result;

4. make system can adapt to complex environment more, for example light variation, multi-obstacle avoidance and night running etc.

Accompanying drawing explanation

Fig. 1 is system construction drawing of the present invention.

Fig. 2 is outline flowchart of the present invention.

Fig. 3 is detailed design process flow diagram of the present invention.

Fig. 4 is GIS map road edge model figure of the present invention.

Fig. 5 is lane detection process flow diagram of the present invention.

Fig. 6 is track basic model figure of the present invention.

Fig. 7 is car body coordinate schematic diagram of the present invention.

Fig. 8 is lane line of the present invention and road edge matching algorithm process flow diagram.

Embodiment

Below in conjunction with accompanying drawing, the present invention is elaborated.

Referring to Fig. 1 and Fig. 2, described method mainly comprises the following steps:

1) gather in advance GIS cartographic information and carry out pre-service;

2) the real-time image acquisition driving diatom of going forward side by side detects;

3) generate in real time the local map of GIS;

4) lane detection result is mated verification with the local map of GIS;

5) according to matching result, generate lane line result with a high credibility.

In algorithm will with GIS map road marginal information by gps data collection on the spot and map, demarcate to realize.Timing signal marks respectively road left hand edge and road right hand edge, and the road marking is unidirectional, and sequentially arrange according to predetermined travel route in each section.The data mode of cartographic information provides in the mode of discrete sampling point, data content is the gps coordinate of each sampled point, data are carried out the GIS global map raw data that smothing filtering obtains meeting road edge information model the most at last, and model is referring to Fig. 4, and EL and ER represent respectively the left and right edge of road.

Referring to Fig. 5, lane detection process is by off-line calibration, to obtain internal reference and the outer ginseng of camera, to image image vertical view before line projection changes rear acquisition car.Based on lane line parts of images pixel, higher than the basic assumption in region, road surface, extracting possible lane line information is binary segmentation image, and it is carried out connected domain detection and carries out matching according to broken line.All connected domains are completed after segmented fitting, according to its length and angle restriction, carry out broken line and be connected and obtain track candidate's line.The time domain of utilizing width with collimation, candidate's lane line to be carried out screening and filtering and utilizes pose data to carry out between multiple image is associated, obtains final lane detection result.Track basic model, referring to Fig. 6, is taked three track models, and L1 and R1 represent respectively the left and right lane line in current track, and L2 and R2 represent respectively the second from left and right two lane highways line.

The generation of real-time generation, lane line and the verification of GIS information matches and the net result of the local map of GIS is realized by following detailed step, referring to Fig. 3:

1) GIS information point and translation distance initialization

To the processing of GIS cartographic information based on following hypothesis: car wants the GIS cartographic information of running region by road edge point, to be provided; GIS cartographic information is expressed as road left hand edge and road right hand edge is put paired sequential organization; Car body enable position is in GIS body of a map or chart.First system loads GIS cartographic information file, then positioning car body position in map.Initial alignment is by the road edge point of global search GIS information and search for nearest GIS road edge point according to car body pose.Find out i,

$s.t.\underset{i}{\min }\mathrm{dis}(P_v,G_i)$

P wherein _vrepresent car body position coordinates, G _irepresent GIS information point coordinate, dis (P _v, G _i) represent the distance of point-to-point transmission.

Consideration need to provide the local map under car body coordinate, so distance is calculated, is that GIS point is transformed under bodywork reference frame; In real road situation, road right hand edge often there will be the situation of widening or dwindling a track in addition, can affect the judgement of car body location, and road left hand edge is relatively stable, thus when location, only consider road left hand edge point, that is:

$\underset{i}{\min }\mathrm{dis}(P_v,{\mathrm{GL}}_i)$

GL wherein _irepresent road left hand edge point in GIS information.

After initial alignment, start the local map under initialization car body coordinate.The scope of the local map of car body coordinate is (10m, the 60m) of car body position.In global map behind location, according to the information in the local body of a map or chart of its sequential storage structure real-time loading, then according to graph model partly, be transformed in corresponding structure.

Local map creates after merit, the translation distance at initialization lane line and GIS map road edge (during practical operation, the position of car body in the middle of certain track starts, in order to obtain correct initial translational movement).Initial translation distance is defaulted as the left-lane line L1 in current road and the distance between road left hand edge EL.If initially there is lane detection result, calculate left-lane line and road left hand edge translation distance; If without lane detection result, suppose that the current position of car body is track mid point, according to lane width, estimate the position of left-lane line L1, calculate translation distance.That is:

$\mathrm{IniTransDis}=\left\{\begin{array}{c}\mathrm{dis}(P_v-\mathrm{LW}/2,{\mathrm{EL}}_{i,x=0}),\mathrm{if}(\mathrm{DetectLane}.\mathrm{IsOK}=0)\\ \mathrm{dis}({L1}_{x=0},{\mathrm{EL}}_{i,x=0}),\mathrm{if}(\mathrm{DetectLane}.\mathrm{IsOK}=1)\end{array}\right.$

Wherein InitTransDis represents initial translation distance, and LW represents lane width, and subscript x=0 represents car body coordinate upper/lower positions, i.e. the position at car body place, and DetectLane.IsOK is the zone bit whether lane detection result is correct.Referring to Fig. 7, it is car body coordinate schematic diagram.Initial translation distance provides verification foundation for following situation: 1. initial frame coupling verification; 2. without lane detection result, when lane line information dropout or historical context loss of data.

2) local GIS information map upgrades and matching

Behind initial frame car body location, later every frame no longer carries out global search location, but according to the local map of car body position real-time update.Specific practice is:

Step1: the local map of history is transformed under front vehicle body coordinate;

Step2: remove the point that exceeds current local body of a map or chart in historical local map;

Step3: read in order subsequent point in GIS information map until exceed local body of a map or chart.

In addition, for the GIS map (starting point that is map file is identical with terminating point) that forms loop, near car body drives to this position, need to carry out the judgement of end of file to process, and then be written into from file is initial.

For the local map obtaining, according to the model structure of lane line, represent that mode carries out matching to GIS cartographic information point wherein, to carry out coupling verification below.During practical operation, according to multinomial model, represent road curve.During matching, the left and right edge of road adopts respectively adaptive matching mode.Concrete steps are:

Step1: the road edge point of local map is started to matching from single order, obtain after fitting parameter, then carry out discrete sampling, calculate sample point and the actual error that detects sample point.If error surpasses threshold value, illustrate that precision does not meet the demands under current matching order, need to improve matching order, order adds 1 to be repeated this step and (considers the actual form of road until meet fitting precision afterwards, can't there is the curve of overbend, therefore matching order is set, be no more than 3 times);

Step2: if precision does not still meet the demands after 3 rank matchings, need to dwindle the scope of local map institute matching marginal point, reject distal-most end sample point from the sample point of matching, again according to second-order fit, repeat this step until meet fitting precision.

3) lane detection result and the verification of GIS information matches

Algorithm, when coupling calibration vehicle diatom, is only considered the left and right lane line (being L1 and R1) in current track, car body place.During coupling respectively by the left and right edge (EL of the road of the GIS information after L1, R1 and matching, ER) match, carry out altogether four groups of couplings: left-lane line and road left hand edge L1EL, left-lane line and road right hand edge L1ER, right lane line and road left hand edge R1EL, right lane line and road right hand edge R1ER, four groups of coupling brief notes are LL, LR, RL, RR.Record wherein best matching result, and record the matching error of Optimum Matching under this coupling group.Threshold value by comparison match error and setting judge detect to such an extent that whether lane line result correct.

In matching process, local environment map road edge is as the model point set E of institute's reference, and the lane line of detection is as shape data point set L.The object of coupling is the optimal transformation T finding between two two-dimentional point sets, make shape data point concentrate all points to concentrate the otherness of corresponding point to reach minimum with model points after conversion, the point set T (L) after conversion is minimum with the similarity measurement of E.That is:

$\underset{T,C}{\min }J(T\left(L\right),E)$

Wherein, J (T (L), E) represents the similarity measurement pattern between two point sets, if the mapping relations between two point sets are expressed as C:L → E, above formula repeat into:

$\underset{T,C}{\min }J(T\left(L\right),C\left(L\right))$

The part of coupling verification has adopted the Discrete Method for Solving of ICP algorithm, and the T in above formula represents rotation and translation transformation, and similarity measurement J adopts minimum squared distance, that is:

$J(T\left(L\right),E)=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\left|\right|T{\stackrel{\→}{\left(l_i\right)}-\stackrel{\→}{e}}_{j\left(i\right)}|{|}_2^2$

In formula, l represents the point in shape data point set L, the point in e representative model point set E, and j (i) represents that i point in point set L is corresponding with j point in point set E, N _lthe number that represents shape data point set L mid point.

In coupling calibration vehicle road line process, assess the trend of moving towards of lane line, do not consider rotational transform, so T (L) only comprises translation transformation, use represent translational movement, the lane line requiring and the minimum difference of road edge line are:

$\min J\left(\stackrel{\→}{t}\right)=\underset{i=1}{\overset{N_l}{\mathrm{\Σ}}}\left|\right|{{\stackrel{\→}{l}}_i+\stackrel{\→}{t}-\stackrel{\→}{e}}_{j\left(i\right)}|{|}_2^2$

When above formula objective function is got minimum value, for:

$\stackrel{\→}{t}=\frac{1}{N}\underset{j=1}{\overset{N}{\mathrm{\Σ}}}\stackrel{\→}{e_{j\left(i\right)}}-\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{\stackrel{\→}{l}}_i$

When lane line point set aligns with road edge line point set center, objective function is got minimum value, and namely otherness is minimum.Obtain the translational movement under minimum difference, also just obtained the position of Optimum Matching.Then the shape data point set of lane line is moved to the model point set place of road edge line, computation of match errors according to translational movement.In order to judge that whether two point sets meet the coupling requirement of practical application, need to obtain the maximum matching error under Optimum Matching position, that is:

$\underset{i}{\max }{w}_i=\left|\right|{\stackrel{\→}{l}}_i+\stackrel{\→}{t}-{\stackrel{\→}{e}}_{j\left(i\right)}|{|}_2^2$

W wherein _ithe maximum matching error of single-point that represents the concentrated corresponding point of point.

To each group in four groups of couplings (LL, LR, RL, RR), all pass through the alternately mode of iteration, obtain respectively corresponding relation and the translation transformation of lane line point set L and road edge point set E, and the maximum matching error of single-point, referring to Fig. 8.In addition in order to guarantee the degree of priority of vision lane line precision, as long as think in four groups of couplings have to meet a coupling requirement, just think that lane detection result meets local cartographic information, testing result is correct and have a higher confidence level.Therefore four groups of matching degrees are compared.Matching degree concentrates the cumulative errors of all corresponding point to determine by two points.The coupling group of cumulative errors minimum is Optimum Matching.That is:

$\underset{\mathrm{LL},\mathrm{LR},\mathrm{RL},\mathrm{RR}}{\min }W=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{w}_i=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\left|\right|{\stackrel{\→}{l}}_i+\stackrel{\→}{t}-{\stackrel{\→}{e}}_{j\left(i\right)}|{|}_2^2$

The maximum matching error of single-point under final Optimum Matching is w _i ^m, M ∈ (LL, LR, RL, RR), subscript M has represented Optimum Matching group.This error has just determined with the relation of threshold value whether lane line point set meets and mate requirement with road edge line point set, and whether the trend of moving towards of the lane line integral body namely detecting under local map meets the trend of moving towards of road edge line.

In order to carry out above algorithm, first to carry out discrete sampling to the lane line after matching and road edge, obtain shape data point set and model point set.In discrete sampling, according to shiding matching mode, determine two corresponding relations between point set.Calculate the translation distance of translation transformation, the maximum distance error that similarity measurement and corresponding point occur.Specifically carry out according to the following steps:

Step1: to four coupling groups, carry out respectively Step2 to Step6:

Step2: the lane line detecting and GIS road edge line shiding matching, the value of sliding distance d in the scope [D, D] and step-length be D _step, circulation is carried out Step3 to Step5;

Step3: lane line point concentrates the coordinate xl of sampled point to sample by distance h from lane line origin-to-destination, and presses lane line fitting parameter and calculate each point y ^lvalue be y _i ^l=f ^l(x _i ^l), the point on the lane line detecting is designated as l _i(x _i ^l, y _i ^l); According to current shiding matching, apart from d, calculate the coordinate x of the sampled point on road edge line ^e, i.e. x ^e=x ^l+ d, and calculate each point y according to road edge fitting parameter ^evalue,

Y _j ^e=f ^e(x _j ^e)=f ^e(x _i ^l+ d), the point set of road edge line is designated as e _j(x _j ^e, y _j ^e); Add up the total N of some centrostigma simultaneously _p;

Step4: calculate the mean place (being center) of lane line and road edge line up-sampling point, two points concentrate the x coordinate of all points and y coordinate figure to add up respectively, concentrate the number N of the point of statistics divided by point _p, obtain with

Step5: calculate translation distance

Step6: all corresponding point that two points are concentrated are calculated single-point matching error record maximum single-point matching error MAXw, simultaneously the total matching error W=W+w of accumulation calculating _i;

Step7: find out total matching error minimum in four groups of couplings, corresponding maximum single-point matching error under Optimum Matching position, returns to MAXw now, translation vector and the group matching (being LL, LR, RL or RR).

Error amount MAXw and the threshold value returned are compared, if be less than threshold value, think that detected lane line and GIS road edge trend matches, increased the confidence level of lane line, think that lane detection is correct; If be greater than threshold value, the lane line that detects of explanation and road edge move towards trend and there are differences, and think lane detection mistake, according to road edge line generating virtual lane line.

4) generating virtual lane line

Several situations all need GIS information to generate virtual vehicle diatom below: 1. detected lane line is too short, now think that requirement or confidence level that lane line information can not satisfaction primary data are not high; 2. detected lane line and GIS road edge line matching error surpass threshold value, now think that lane detection result is inaccurate; 3. fail to detect lane line, now directly by GIS information, provide virtual vehicle diatom, guarantee perception data continuity.

The generation of virtual vehicle diatom is based on following hypothesis: step 3) can be upgraded match information and translation transformation according to lane detection result in real time, and this match information and translation transformation are applicable within one period of continuous time, one section continuous time inside lane line there is identical translational movement with road edge coupling, until next update.Coupling group and translation distance while correctly mating for the last time in system log (SYSLOG) historical frames, when needs generating virtual lane line,, according to coupling group and the translation distance of record, corresponding road left or right edge fitting line is moved to lane line position by translational movement.Then the fitting parameter of road edge line is assigned to lane line fitting parameter, determines the information such as starting and terminal point coordinate, road width, lane line attribute of lane line simultaneously, be virtual vehicle diatom.After generation, need to empty wrong lane detection result, avoid time domain association to cause multiframe flase drop.

Before sending final lane line result, also need the integral position of lane line result further to retrain, judge whether that all lane lines position is all between the left and right edge line of GIS road.Meet following constraint:

ER _Y<Lane _L2,L1,R1,R2<EL _Y

Lane wherein _{l2, L1, R1, R2}represent four lane line L2 in lane line model, L1, R1, R2, EL _y, ER _ythe coordinate figure that represents respectively the left and right edge of road Y-axis under bodywork reference frame.

When algorithm is carried out, choose starting point and terminal point coordinate and the comparison of road edge line position of every lane line, judge whether to meet constraint.

If R2 or L2 lane line exceed the restriction range of road edge line, remove this lane line; If R1 or L1 A-road line exceed the restriction range of road edge line, according to the LL coupling group of acquiescence and initial translation distance InitTransDis, at the current position of car body generating virtual lane line.

Final lane detection result merges mutually with other perception informations of road (comprising crossing, barrier etc.), form with UDP bag (User Datagram Protocol) is passed through switch communication, and complete visually-perceptible data are sent to follow-up planning performance element.

First this method gathers GIS cartographic information and carries out pre-service, then the real-time image acquisition driving diatom of going forward side by side detects, generate in real time the local map of GIS simultaneously, lane detection result is mated to verification with the local map of GIS by ICP algorithm, finally according to matching result, generate lane line result with a high credibility.The method can be carried out trust evaluation to lane detection result by Geographic Information System, there is flase drop or proofreading and correct when undetected, expand the scope of application of lane detection algorithm, can improve by visually-perceptible information the precision of map location again, thereby improve the performance of vision navigation system in whole intelligent driving and DAS (Driver Assistant System), make system can adapt to complex environment more.

Above content is in conjunction with concrete preferred implementation further description made for the present invention; can not assert that the specific embodiment of the present invention only limits to this; for general technical staff of the technical field of the invention; without departing from the inventive concept of the premise; can also make some simple deduction or replace, all should be considered as belonging to the definite scope of patent protection of claims of the present invention.

Claims (8)

1. the vision navigation method based on the exploitation of lane detection and GIS cartographic information, is characterized in that: comprise the following steps:

1) gather in advance GIS cartographic information and carry out pre-service;

2) the real-time image acquisition driving diatom of going forward side by side detects;

3) generate in real time GIS local environment map;

4) lane detection result is mated to verification with the local map of GIS;

5) according to matching result, generate lane line result with a high credibility.

2. a kind of vision navigation method based on the exploitation of lane detection and GIS cartographic information according to claim 1, it is characterized in that: in described step 1), by pose measurement equipment, gather in advance the road edge data of global map, and road edge data are carried out to smooth filtering processing, obtain the road edge point set of global map.

3. a kind of vision navigation method based on the exploitation of lane detection and GIS cartographic information according to claim 1, it is characterized in that: described step 2), by image capture device, adopt based on detection method monocular vision, that carry out feature extraction and track models fitting under time domain association and detect in real time road surface lane line.

4. a kind of vision navigation method based on the exploitation of lane detection and GIS cartographic information according to claim 1, it is characterized in that: in described step 3), by pose measurement equipment, obtain in real time when front vehicle body position, and locate in GIS global map, then according to graph model partly, generate local environment map.

5. a kind of vision navigation method based on the exploitation of lane detection and GIS cartographic information according to claim 1, it is characterized in that: in described step 4), integrating step 2) with the result of step 3), under same scale, use iterative closest point algorithms to carry out shiding matching verification to lane detection result and GIS localized road edge, obtain best match position and maximum matching error, during coupling, according to lane line model and graph model group match partly, select optimum one group of coupling.

6. a kind of vision navigation method based on the exploitation of lane detection and GIS cartographic information according to claim 1, it is characterized in that: in described step 5), the matching result obtaining according to step 4), by maximum matching error and threshold under best match position, determine the confidence level of lane detection result, then revise the positional information of translation distance and location.

7. a kind of vision navigation method based on the exploitation of lane detection and GIS cartographic information according to claim 6, is characterized in that: if testing result is insincere, according to road edge line generating virtual lane line.

8. a kind of vision navigation method based on the exploitation of lane detection and GIS cartographic information according to claim 6, is characterized in that: the correctness and the stability that by lane line position constraint, improve testing result.