CN105590087A

CN105590087A - Road recognition method and device

Info

Publication number: CN105590087A
Application number: CN201510255539.6A
Authority: CN
Inventors: 吴涛; 史美萍; 李焱
Original assignee: National University of Defense Technology
Current assignee: National University of Defense Technology
Priority date: 2015-05-19
Filing date: 2015-05-19
Publication date: 2016-05-18
Anticipated expiration: 2035-05-19
Also published as: CN105590087B

Abstract

An embodiment of the invention discloses a road recognition method and a road recognition device. According to the road recognition method and the road recognition device, a driving trajectory of a vehicle is obtained through acquiring a reference image in front of the motion of the vehicle and based on a virtual control command input by a user; the driving trajectory is superposed onto the reference image, and a plurality of pixel points reflecting road features are selected from the reference image according to the driving trajectory, each pixel point is regarded as a road sample, a road sample set containing a plurality of the road samples is obtained, and a road recognition model is obtained by utilizing all the road samples in the road sample set, so that the vehicle driving on a road can recognize the road region and proceeds along the road according to the road recognition model.

Description

A kind of roads recognition method and device

Technical field

The present invention relates to unmanned vehicle technical field, particularly relate to a kind of roads recognition method and device.

Background technology

Unmanned vehicle be one can be under different kinds of roads and wild environment paleocinetic intelligent mobile robot, military andIn daily life, have broad application prospects. Conventionally, obtain vehicle movement front by the camera carrying on car bodyImage, and utilize the image recognition road of vehicle front, and then can be along moving according to the definite direction of road.

At present, unmanned vehicle adopts the image recognition road of following methods based on vehicle movement front more, first, at a width orIn the image of several vehicle fronts, gather multiple road samples and non-road sample, reflect the pixel of roadway characteristic and non-The pixel of roadway characteristic, and, road sample and non-road Sample Establishing road model utilized; Then, along with unmannedThe image in the motion real-time update vehicle movement front of car, chooses in image automatically corresponding to vehicle movement front FXInterior sample, as current road sample, utilizes described current road sample to judge whether road model needs to revise, in roadRoad model provides corrected parameter and carries out road model correction operation need to revise time, make discrimination model can adapt to continuous changeThe image of changing; Finally, make road model act on the image of vehicle front, thereby utilize the image recognition road constantly changingRoad.

But, in the time utilizing according to the method described above the image recognition road in unmanned vehicle vehicle movement front, as road sampleThe choosing directly affecting the differentiation accuracy rate of road model of car forefoot area, too small if car forefoot area is chosen, Ke NengwuMethod covers complete road information, and part reflects the not collected use of sample of roadway characteristic, therefore, and according to before this carThe road sample that region gathers can not reflect whole features of road, and road model is easily judged to be segment path non-road;If it is excessive that car forefoot area is chosen, unmanned vehicle may comprise non-road sample in the time that bend moves in this region, make roadModel is easily judged to be road by non-part road. Therefore, in the time that unmanned vehicle moves on the road of situation complexity, existingThe roads recognition method that is applied to unmanned vehicle is difficult to accurately identify road area and the non-road area in the front image of car, and nothingNon-road area is identified as road area by people Che Ruo mistake, moves to possibly outside road, more seriously may be because ofThe situations such as this overturns, falls, make unmanned vehicle suffer crushing damage.

Summary of the invention

A kind of roads recognition method and device are provided in the embodiment of the present invention, can not have continued accurate knowledge to solve prior artThe problem of road area in other image.

In order to solve the problems of the technologies described above, the embodiment of the invention discloses following technical scheme:

A kind of roads recognition method, described method comprises:

Obtain the reference picture in vehicle movement front;

Receive the virtual controlling instruction of user input, described virtual controlling instruction at least comprises: front-wheel pivot angle, throttle amount andOne in brake amount;

According to the driving trace line of described virtual controlling instruction prediction vehicle;

Described driving trace line is added on described reference picture;

In described reference picture, choose the pixel of multiple reflection roadway characteristics according to described driving trace line;

Each pixel of choosing, all as a road sample, is obtained including the road sample of multiple road samplesCollection;

Utilize the road area in reference picture described in all road specimen discernings in described road sample set.

Alternatively,

Described driving trace line comprises two wheel trajectories lines;

In the interval region of two described wheel trajectories lines, choose the first predeterminable area;

Get two the second predeterminable areas according to two described wheel trajectories line selections, corresponding one of each described track of vehicle lineThe second predeterminable area, and each described the second predeterminable area line centered by the described wheel trajectories line of correspondence;

In described reference picture, choose described road sample according to described the first predeterminable area and/or described the second predeterminable areaThis.

Alternatively, the roadway area in reference picture described in the described all road specimen discernings that utilize in described road sample setTerritory, comprising:

Utilize the default road model of all road Sample Establishings in described road sample set;

Utilize described default road model to differentiate the attribute of each pixel in described reference picture, described attribute is roadOr non-road;

Obtain the pixel of all reflection roadway characteristics in described reference picture according to the differentiation result of described default road modelPoint;

Utilize the pixel of all reflection roadway characteristics in described reference picture to identify the road area in described reference picture.

Alternatively, described method also comprises:

In the non-road predeterminable area of described reference picture, choose the pixel of the non-roadway characteristic of multiple reflections;

Each pixel of choosing, all as a non-road sample, is obtained including the non-road of multiple non-road samplesSample set;

In described reference picture, choose multiple pixels undetermined according to preset strategy;

Determine respectively the attribute of pixel undetermined described in each according to described road sample set and described non-road sample set,Described attribute is road or non-road;

Utilize described default road model to differentiate respectively in described reference picture the institute of pixel reflection undetermined described in eachState attribute;

For pixel undetermined described in each, if the differentiation result of described default road model and described pixel undeterminedAttribute consistent, determine that the differentiation result of described default road model is correct;

If the pixel quantity described undetermined that correct described differentiation result is corresponding, with the ratio of described pixel quantity undeterminedValue is less than predetermined threshold value, utilizes default road described in pixel correction described undetermined corresponding to incorrect described differentiation resultModel.

Alternatively,

According to all non-road sample in all road samples in described road sample set, described non-road sample set andAll described pixels undetermined are set up described default road model.

Alternatively, the described pixel of choosing the non-roadway characteristic of multiple reflections in the non-road predeterminable area of described reference picturePoint, comprising:

Choose described non-road according to described virtual controlling instruction in top, the upper left corner and/or the upper right corner of described reference pictureRoad predeterminable area;

Choose multiple pixels in described non-road predeterminable area pixel as the non-roadway characteristic of reflection.

Alternatively, describedly determine respectively described in each and treat fixation according to described road sample set and described non-road sample setThe attribute of vegetarian refreshments, comprising:

Obtain in the pixel characteristic of road sample described in each in described road sample set, described non-road sample setThe pixel characteristic of non-road sample described in each, and, the pixel characteristic of pixel undetermined described in each;

By a similar set of pixel composition similar pixel characteristic, described similar set has multiple, described in any twoThe pixel characteristic dissmilarity of pixel in similar set, includes in each described similar set: undetermined described at least onePixel, and, at least one pixel in described road sample set and described non-road sample set;

Judge respectively whether the quantity of road sample described in similar set described in each is greater than described non-road sampleQuantity;

If the quantity of described road sample is greater than the quantity of described non-road sample, determine owning in described similar setDescribed pixel undetermined is the pixel of reflection roadway characteristic; If the quantity of described road sample is less than described non-roadThe quantity of sample, determines that all described pixel undetermined in described similar set is the pixel of the non-roadway characteristic of reflection.

Alternatively,

Obtain the parameters of the video camera of taking described reference picture;

Utilize the parameters of video camera to obtain the picture position of subject in described reference picture with empty in realityBetween in locus between matching relationship;

According to described matching relationship, described driving trace line is superimposed upon on described reference picture.

A kind of road Identification device, is applied to Vehicular body front and installs the vehicle of camera, described device comprise acquiring unit,Trajectory unit, superpositing unit, sample unit and recognition unit:

The control appliance of the virtual controlling instruction that described acquiring unit is inputted with described camera and transmission user is respectively connected,For obtaining the reference picture in vehicle movement front, and receive described virtual controlling instruction, described virtual controlling instruction at leastComprise: the one in front-wheel pivot angle and throttle amount, brake amount;

Described trajectory unit is connected with described acquiring unit, for predict the rail that travels of vehicle according to described virtual controlling instructionTrace;

Described superpositing unit is connected with described trajectory unit and described acquiring unit respectively, for described driving trace line is foldedBe added on described reference picture;

Described sample unit is connected with described superpositing unit and described acquiring unit respectively, for according to described driving trace lineIn described reference picture, choose the pixel of multiple reflection roadway characteristics;

Described recognition unit is connected with described sample unit, knows for all road samples that utilize described road sample setRoad area in not described reference picture.

Alternatively,

Described trajectory unit comprises wheel trajectories unit, and described wheel trajectories unit is used for obtaining two wheel trajectories lines;

Described sample unit comprises that the first predeterminable area chooses unit, and described the first predeterminable area is chosen unit at twoIn the interval region of described wheel trajectories line, choose the first predeterminable area;

Described sample unit also comprises that the second predeterminable area chooses unit, and described the second predeterminable area is chosen unit for basisArticle two, two the second predeterminable areas are got in described wheel trajectories line selection, and corresponding one second of each described track of vehicle line is presetRegion, and each described the second predeterminable area line centered by the described wheel trajectories line of correspondence;

Described sample unit also comprises chooses unit with described the first predeterminable area respectively and the second predeterminable area is chosen unitConnect road sample choose unit, described road sample choose unit for according to described the first predeterminable area and/or described inThe second predeterminable area is chosen described road sample in described reference picture.

Alternatively, described recognition unit comprises model unit, template(-let), road pixel acquiring unit and road IdentificationUnit;

Described model unit is for utilizing the default road model of all road Sample Establishings of described road sample set;

Described template(-let) is for utilizing described default road model to differentiate the genus of described each pixel of reference pictureProperty, described attribute is road or non-road;

Described road pixel acquiring unit is connected with described template(-let), for according to the differentiation of described default road modelResult is obtained the pixel of all reflection roadway characteristics in described reference picture;

Described road Identification unit is connected with described road pixel acquiring unit, for utilizing described reference picture allThe pixel of reflection roadway characteristic is identified the road area in described reference picture.

Alternatively,

Described sample unit comprises non-road sample unit, non-at described reference picture of described non-road sample unitIn road predeterminable area, choose the pixel of the non-roadway characteristic of multiple reflections; Each pixel of choosing is all non-as oneRoad sample, obtains including the non-road sample set of multiple non-road samples;

Described recognition unit comprises unit undetermined, determining unit, judgement unit, differentiation result determining unit and amending unit;

Described unit undetermined is for choosing multiple pixels undetermined according to preset strategy at described reference picture;

Described determining unit is connected with described non-road sample unit and described unit undetermined respectively, for according to described roadSample set and described non-road sample set are determined respectively the attribute of pixel undetermined described in each, described attribute be road orNon-road;

Described judgement unit is connected with described template(-let), for utilizing described default road model to differentiate respectively described referenceThe described attribute of pixel reflection undetermined described in each in image;

Described differentiation result unit is connected with described determining unit and described judgement unit respectively, for for described in eachPixel undetermined, when consistent, determines institute at the differentiation result of described default road model and the attribute of described pixel undeterminedThe differentiation result of stating default road model is correct;

Described amending unit is connected with described differentiation result determining unit, in institute corresponding to correct described differentiation resultState pixel quantity undetermined, while being less than predetermined threshold value with the ratio of described pixel quantity undetermined, utilize incorrect described inDifferentiate default road model described in pixel correction described undetermined corresponding to result.

Alternatively,

Described sample unit comprises non-road sample unit, non-at described reference picture of described non-road sample unitIn road predeterminable area, choose the pixel of the non-roadway characteristic of multiple reflections;

Described recognition unit comprises unit undetermined, determining unit and road model unit;

Described road model unit is connected with described non-road sample unit and described determining unit, for according to described roadAll non-road sample and described all pixels undetermined in all road samples, described non-road sample set in sample setPoint is set up described default road model.

Alternatively, described non-road sample unit comprises that non-road area chooses unit;

Described non-road area choose unit for according to described virtual controlling instruction at described road image reference pictureDescribed non-road predeterminable area is chosen in top, the upper left corner and/or the upper right corner; Choose multiple in described non-road predeterminable areaPixel is as the pixel of the non-roadway characteristic of reflection.

Alternatively, described determining unit comprises that pixel characteristic unit, set component units, judging unit and feature are definite singleUnit;

Described pixel characteristic unit is connected with described sample unit and described unit undetermined respectively, for obtaining described road sampleOriginally non-road sample described in each described in each that concentrate in pixel characteristic, the described non-road sample set of road sampleThis pixel characteristic, and, the pixel characteristic of pixel undetermined described in each;

Described set component units is connected with described pixel characteristic unit, for pixel similar pixel characteristic is formed to oneIndividual similar set, described similar set has multiple, and the pixel characteristic dissmilarity of pixel in any two similar set is everyIn individual described similar set, include: pixel undetermined described at least one, and, described road sample set and described non-At least one pixel in road sample set;

Described judging unit is connected with described set component units, for judging respectively described in each described in similar setWhether the quantity of road sample is greater than the quantity of described non-road sample;

Described feature determining unit is connected with described judging unit, for be greater than described non-road in the quantity of described road sampleWhen the quantity of road sample, determine that all described pixel undetermined in described similar set is the pixel of reflection roadway characteristicThe described attributive character of point reflection is roadway characteristic; Be less than the quantity of described non-road sample in the quantity of described road sampleTime, determine that all described pixel undetermined in described similar set is the pixel of the non-roadway characteristic of reflection.

Alternatively, described superpositing unit comprises camera parameter unit, matching relationship unit and image superpositing unit;

Described camera parameter unit is for obtaining the parameters of the video camera of taking described reference picture;

Described matching relationship unit is connected with described camera parameter unit, is clapped for utilizing the parameters of video camera to obtainTake the photograph the matching relationship between the picture position of object object in described reference picture and the locus in real space;

Described image superpositing unit is connected with described matching relationship unit, for according to described matching relationship will described in the rail that travelsTrace is superimposed upon on described reference picture.

From above technical scheme, a kind of roads recognition method and device that the embodiment of the present invention provides, by obtaining carFront reference picture, and virtual vehicle target front-wheel pivot angle and vehicle target acceleration, obtain the dummy row of vehicleSail trajectory, and driving trace line is superimposed upon on reference picture, intuitively know vehicle future taking reference picture as backgroundDriving trace. The pixel of choosing multiple reflection roadway characteristics according to driving trace line in reference picture is road sample,Road in the road specimen discerning reference picture that utilization is chosen, the vehicle that makes to travel on road is identified with reference to reference pictureThe road going out continues to advance along road, and the embodiment of the present invention can exist according to the virtual vehicle driving trace line of real-time updateIn the reference picture of constantly updating, effectively choose reliable road sample, to continue in the reference picture after identification is upgradedRoad. Because road sample is that driving trace line and the vehicle front real-time reference picture real-time according to vehicle chosen,Therefore, the road sample of continuous renewal can adapt to the road that vehicle front constantly changes, thereby in the reference of vehicle frontIn image, accurately identify road, advance according to correct direction with guiding vehicle.

Brief description of the drawings

In order to be illustrated more clearly in the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existingIn technical description, the accompanying drawing of required use is briefly described, apparently, for those of ordinary skill in the art andSpeech, is not paying under the prerequisite of creative work, can also obtain according to these accompanying drawings other accompanying drawing.

The schematic flow sheet of a kind of roads recognition method that Fig. 1 provides for the embodiment of the present invention;

The schematic diagram of a kind of driving trace line that Fig. 2 provides for the embodiment of the present invention;

A kind of the first predeterminable area that Fig. 3 provides for the embodiment of the present invention and the schematic diagram of the second predeterminable area;

The travel schematic diagram of trajectory of a kind of reference picture stack that Fig. 4 provides for the embodiment of the present invention;

A kind of schematic flow sheet that utilizes default road model identification road that Fig. 5 provides for the embodiment of the present invention;

A kind of schematic flow sheet of setting up default road model that Fig. 6 provides for the embodiment of the present invention;

The schematic diagram of a kind of non-road predeterminable area that Fig. 7 provides for the embodiment of the present invention;

A kind of schematic flow sheet of determining pixel attribute undetermined that Fig. 8 provides for the embodiment of the present invention;

A kind of schematic flow sheet of checking default road model differentiation result that Fig. 9 provides for the embodiment of the present invention;

The structural representation of a kind of road Identification device that Figure 10 provides for the embodiment of the present invention.

Detailed description of the invention

In order to make those skilled in the art person understand better the technical scheme in the present invention, below in conjunction with the invention processAccompanying drawing in example, is clearly and completely described the technical scheme in the embodiment of the present invention, obviously, and described realityExecuting example is only the present invention's part embodiment, instead of whole embodiment. Based on the embodiment in the present invention, abilityTerritory those of ordinary skill is not being made the every other embodiment obtaining under creative work prerequisite, all should belong to thisThe scope of invention protection.

Fig. 1 is a kind of roads recognition method schematic flow sheet that the embodiment of the present invention provides, and mainly comprises the following steps:

Step S101: obtain the reference picture of vehicle front, and, receive the virtual controlling instruction of user's input, wherein,Virtual controlling instruction at least comprises the one in front-wheel pivot angle, throttle amount and brake amount, and certainly, virtual controlling instruction also canTo comprise other instructions.

The embodiment of the present invention is applied to car body front end sets up the vehicle of video camera, for example unmanned vehicle, in following embodiment allTaking unmanned vehicle as example. Obtain the reference picture in the unmanned vehicle direction of motion front of shot by camera.

Reference picture in the whole embodiment of the present invention is the data of reference picture, by the pixel group of reflection body characteristicsBecome, each pixel has pixel characteristic, and for example, the pixel characteristic of each pixel is by colourity, brightness or other numerical valueRepresent. In an embodiment of the present invention, the pixel in reference picture can be divided into the pixel and the reflection that reflect roadway characteristicThe pixel of non-roadway characteristic, wherein, the attribute of pixel of reflection roadway characteristic is road, reflects non-roadway characteristicThe attribute of pixel is non-road.

In a specific embodiment of the present invention, virtual controlling instruction can obtain by human-computer interaction device or other equipmentGetting, is the virtual controlling instruction that operator arranges according to trend of road, and unmanned vehicle is not directly controlled in this virtual controlling instructionMotion state, and only for realizing the driving trace line of predicting vehicle in following embodiment. Wherein, human-computer interaction device canTo be steering wheel, gas pedal and the brake pedal of emulation, can be also other controlling equipments taking mouse-keyboard as basis,Thereby can input by user, obtain the virtual controlling instructions such as such as front-wheel pivot angle, throttle amount and brake amount.

Step S102: according to the driving trace line of virtual controlling instruction prediction vehicle.

Obtain at step S101 on the basis of virtual controlling instruction, in conjunction with parameter presets such as the distances between each wheel,Obtain the multiple tracing points of unmanned vehicle within following a period of time according to vehicle dynamic model, and by suitable the plurality of tracing pointOrder connects and composes the driving trace line of unmanned vehicle.

Step S103: driving trace line is superimposed upon on reference picture.

The driving trace line that step S102 is obtained is superimposed upon on the reference picture of step S101 acquisition, so that follow-upIn step, choose road sample, and provide reference basis for operator manipulates human-computer interaction device.

In a specific embodiment of the present invention, first, obtain the parameters of the video camera of taking reference picture, exampleAs the parameter such as setting height(from bottom), field range of video camera, the parameters of the video camera obtaining can be the reality of video cameraBorder parameter can be also the parameter of user's input; Then, utilize the parameters of video camera to obtain subject in ginsengExamine the matching relationship between picture position and the locus of subject in real space in image.

Finally, according to matching relationship, driving trace line is superimposed upon on reference picture.

After obtaining matching relationship, according to matching relationship, the conversion of driving trace line is superimposed upon on reference picture. For example, asShown in dotted line in Fig. 2 on road, when utilizing display show reference picture and be superimposed upon the rail that travels on reference pictureWhen trace, on reference picture, just can intuitively demonstrate the driving trace line of unmanned vehicle, reflect following a period of time of unmanned vehicleWithin running orbit on corresponding reference picture.

In a specific embodiment of the present invention, the travel reference picture of trajectory of stack transmits by wireless communication systemTo Remote computer demonstration. Operating personnel are towards the superimposed image of real-time reception, according to the road in reference pictureThe vehicle driving trace line of environment and current prediction, learn current driving trace line whether with reference picture in roadway areaTerritory is consistent, and sends virtual controlling instruction by human-computer interaction device in the time that both are inconsistent, makes based on virtual controllingThe driving trace line that instruction generates always can be consistent with the road area in reference picture, and the trajectory of travelling always canEnough be superimposed upon on the road area in reference picture.

Step S104: choose the pixel of multiple reflection roadway characteristics according to driving trace line in reference picture, and will selectEach pixel of getting, all as a road sample, obtains including the road sample set of multiple road samples.

After the stack of driving trace line and reference picture, the picture position of driving trace line in reference picture conventionally with ginsengExamine and in image, reflect that the pixel of roadway characteristic intersects, the present embodiment is chosen multiple according to driving trace line in reference pictureReflection roadway characteristic pixel, and using each pixel of choosing all as a road sample, obtain including manyThe road sample set of individual road sample.

In a specific embodiment of the present invention, as shown in Figure 3, driving trace line comprises wheel trajectories line (car in figureTwo front solid lines), wheel trajectories line is two lines of front-wheel movement locus on reflection vehicle. At two wheel trajectories linesIn interval region, choose the first predeterminable area (in figure shown in A), and, choose respectively second according to two wheel trajectories linesPredeterminable area (in figure shown in B), the second predeterminable area is line centered by corresponding wheel trajectories line, every wheel trajectories lineAll corresponding second predeterminable area.

The region that the first predeterminable area and the second predeterminable area all obtain for physical location by object, according to above-mentioned objectThe matching relationship that picture position and locus calculate, by the physical location of the first predeterminable area and the second predeterminable areaConvert the picture position in reference picture to, and the first predeterminable area after stack conversion and second is preset on reference pictureRegion, this obtains the first predeterminable area and the corresponding region of the second predeterminable area in reference picture (as A ' in figure and B 'Shown in), the pixel of reference picture in this corresponding region is the pixel of reflection roadway characteristic substantially, therefore, right at thisAnswering selected pixels point in region is reliably, accurately as the mode of road sample.

In addition, in the time that practical application the present embodiment is identified road area in reference picture, can be according to the first predeterminable area andThe second predeterminable area is chosen road sample, or only chooses road sample according to the first predeterminable area or the second predeterminable area, ifThe road that unmanned vehicle is about to operation process is two narrow passages that are a bit larger tham wheel width, as the transfer bridge of building temporarily etc., orPerson's road mid portion when thering is no for a long time wheel to roll to grow the road of a large amount of weeds, as shown in Figure 4, clearlyThe mode that is road sample by selected pixels point in FX in reference picture will inevitably be chosen for a large amount of non-road samplesRoad sample, causes identification road to occur mistake, therefore, in a specific embodiment of the present invention, can be by only selectingThe mode of getting the second predeterminable area and do not choose the first predeterminable area effectively avoids occurring above-mentioned identification error, thereby ensuresThe accuracy of road sample.

Step S105: utilize the road area in all road specimen discerning reference pictures in road sample set.

The road sample set that utilizes step S104 to obtain, in all pixels in reference picture, determines and road sampleThis similar pixel of pixel characteristic, these pixels similar to the pixel characteristic of road sample are in reference pictureThe pixel of reflection roadway characteristic, after determining the pixel of all reflection roadway characteristics in reference picture, just identifiesRoad area in reference picture, unmanned vehicle can be according to identified road area adjust operation parameter, and continues motion.

In a specific embodiment of the present invention, operator can, according to the road identifying, utilize human-computer interaction device to adjustWhole virtual controlling instruction, is always consistent with the road in reference picture with the driving trace line that ensures vehicle. For example, asFruit operator sees the straight distant place that extends to of road in reference picture, and operator's front-wheel pivot angle is set to 0 degree increasingOpen out amount, is that straight line and extension are far away according to the driving trace line of front-wheel pivot angle and the acquisition of throttle amount, makes to be like this added toTrend of road in driving trace line and reference picture on reference picture better coincide, and is convenient to follow-up road sample collectionCorrectness.

The present embodiment utilizes the virtual controlling instruction that user inputs to predict the driving trace line of vehicle, and driving trace line is foldedBe added on the reference picture in vehicle operating front, on reference picture, choose road sample set according to driving trace line, finalAccording to the road area in each the road specimen discerning reference picture in road sample set. In the present embodiment scheme, chooseRoad sample set always choose according to the actual conditions of road area in reference picture, no matter vehicle is to travel at straight lineOn road or on crankcase ventilaton, all can choose flexibly, exactly road sample set, thereby effectively avoid occurring adopting choosingGet in reference picture in FX sample as the prior art scheme of road sample set easily non-road sample is falsely dropped intoThe bad result of road sample.

In another embodiment of the present invention, as shown in Figure 5, realize step S105 in above-described embodiment and utilize roadRoad area in all road specimen discerning reference pictures in sample set, comprises following step:

Step S501: utilize the default road model of all road Sample Establishings in road sample set.

Because all road samples in road sample set are the sample that reflects roadway characteristic, therefore, with these road samplesThis pixel is characterized as benchmark, the pixel that has a similar pixel feature can be regarded as in image to road sampleThe pixel of reflection roadway characteristic, will regard as instead with the larger pixel of the pixel feature difference of road sample in imageReflect the pixel of non-roadway characteristic.

In conjunction with existing Images Classification model, and using all road samples in road sample set as classification reference basis,The default road model in the present embodiment can be obtained, by this default road model, all reflections in image can be identifiedThe pixel of the pixel of roadway characteristic and the non-roadway characteristic of reflection.

Step S502: utilize default road model to differentiate the attribute of each pixel in reference picture, attribute be road orNon-road.

Utilizing predefined road model to differentiate each pixel in reference picture is reflection roadway characteristic or reflectionNon-roadway characteristic, if certain pixel is the pixel of reflection roadway characteristic, the attribute of this pixel is road; IfCertain pixel is the pixel of the non-roadway characteristic of reflection, and the attribute of this pixel is non-road.

Step S503: obtain the pixel that in reference picture, all properties is road according to the differentiation result of default road modelPoint.

Step S504: utilize the road area in the pixel identification reference picture that in reference picture, all properties is road.

The set that the pixel that in reference picture, all properties is road forms is the road area in reference picture.

In another embodiment of the present invention, as shown in Figure 6, the default road model in above-described embodiment can be by followingSeveral steps obtain.

Step S1051: in the non-road predeterminable area of reference picture, choose the pixel of the non-roadway characteristic of multiple reflections,And using each pixel of choosing all as a non-road sample, obtain the non-road sample that comprises multiple non-road samplesThis collection.

In a specific embodiment of the present invention, the upper left corner and/or upper right according to virtual controlling instruction at reference pictureNon-road predeterminable area is chosen at angle. For example, as shown in Figure 7, when the installation site of vehicle-mounted vidicon and attitude fixing after,Because the field range of video camera is limited, the upper part in reference picture may be sky, ground at a distance or above ground levelObject etc. Therefore, can in reference picture, choose on image pixel in part, the upper left corner and region, the upper right corner non-Road sample, as shown in Figure 7, the non-road predeterminable area of the region of filling taking dotted line in figure as selecting, wherein w_m、w_l、w_rBe respectively height, the reference picture upper left corner and the region, the reference picture upper right corner of selected reference picture upper areaWidth, l_l、l_rBe respectively the length in the reference picture upper left corner and region, the reference picture upper right corner.

And, when vehicle turn left to time, because the translational speed of right side wheels is greater than left side wheel, therefore make l_lReduce, l_rIncrease; When vehicle turn right to time, because the translational speed of right side wheels is greater than left side wheel, therefore make l_lIncrease,l_rReduce, to meet the actual conditions of Vehicle Driving Cycle.

Above-mentioned selected region, as the non-road predeterminable area with reference to image, is superimposed upon to reference by non-road predeterminable areaOn image, concrete stacked system when choosing road sample by the first predeterminable area and the second predeterminable area and reference pictureThe mode of stack is similar, repeats no more herein, chooses the non-road of reflection taking non-road predeterminable area as basis on reference pictureThe pixel of road feature is as non-road sample.

Step S1052: choose multiple pixels undetermined according to preset strategy in reference picture;

For example, can be to reference picture interval sampling, gather a pixel every several pixels, and, with everyCentered by the pixel gathering, obtain the pixel characteristic of this pixel periphery predetermined number pixel, according to neighboring pixelThe pixel characteristic of point is determined the pixel characteristic of this pixel, and using this pixel as representing that himself and periphery are pre-If the pixel undetermined of a quantity pixel. In reference picture, choose in the manner described above multiple pixels undetermined, eachPixel undetermined all can represent its pixel characteristic of a predetermined number pixel around. At the attribute of determining certain pixel undeterminedAfterwards, can think that the attribute of all pixels of this pixel undetermined representative is all consistent with the attribute of this pixel.

Step S1053: according to each the non-road in each road sample and non-road sample set in road sample setRoad sample is determined respectively the attribute of each pixel reflection undetermined, and attribute is road or non-road.

In a specific embodiment of the present invention, as shown in Figure 8, by following steps completing steps S1053:

Step S531: obtain every in the pixel characteristic of each the road sample in road sample set, non-road sample setThe pixel characteristic of a non-road sample, and, the pixel characteristic of each pixel undetermined.

Step S532: by a similar set of pixel composition similar pixel characteristic, similar set has multiple, any twoThe pixel characteristic dissmilarity of pixel in similar set, includes in each similar set: at least one pixel undetermined, and,At least one pixel in road sample set and non-road sample set. Analyze each the road sample in road sample setPixel characteristic, non-road sample set in the pixel characteristic of each non-road sample, and each pixel undeterminedPixel characteristic, pixel characteristic at least comprises the parameter such as brightness, colourity of pixel, by pixel similar pixel characteristicForm an independently similar set, and the dissimilar pixel of pixel characteristic belongs to different similar set. For example,A similar set comprises 71 pixels that pixel characteristic is similar, wherein comprises 50 pixels undetermined, 20Ge roadRoad sample and 1 non-road sample.

Step S533: judge respectively whether the quantity of road sample in each similar set is greater than the number of non-road sampleAmount.

The quantity of road sample and the quantity of non-road sample in the similar set that statistics above-mentioned steps S532 obtains, pointDo not judge whether the quantity of road sample in each similar set is greater than the quantity of non-road sample.

Step S534: if the quantity of road sample is greater than the quantity of non-road sample, determine undetermined in this similar setThe attribute of pixel is road.

Step S535: if the quantity of road sample is less than the quantity of non-road sample, determine the fixation for the treatment of in similar setThe attribute of vegetarian refreshments is non-road. In this way, determine the attribute of each pixel undetermined, determine each treatFixation vegetarian refreshments is reflection roadway characteristic or reflects non-roadway characteristic.

Wherein, if the quantity of road sample equals the quantity of non-road sample, be not the pixel undetermined in similar setThe definite attribute of point, gives up the pixel undetermined in this similar set.

Step S1054: according to each the non-road in each road sample, non-road sample set in road sample setRoad sample and each pixel undetermined are set up default road model.

According to having determined pixel in each the non-road sample in non-road sample set in step S1051, step S1053Each road sample in the pixel undetermined of feature and above-described embodiment in road sample set, sets up default road mouldType, for example, default road model can be for adopting the classifier algorithm that SVMs is representative, can be also to adopt to mixClose the disaggregated model of Gauss model to road Sample Establishing.

In another embodiment of the present invention, consider the operation along with unmanned vehicle, the condition of road surface that unmanned vehicle movesMay change, for example, unmanned vehicle runs to dirt road by asphalt road, now, and the road being obtained by former reference pictureThe default road model of sample and non-road Sample Establishing may be not suitable for identifying road in current reference picture, therefore,The present embodiment provides the scheme of the default road model of a kind of real-time update, and as shown in Figure 9, this scheme mainly comprises followingSeveral steps:

Step S201: in the non-road predeterminable area of reference picture, choose the pixel of the non-roadway characteristic of multiple reflections, andEach pixel of choosing, all as a non-road sample, is obtained to the non-road sample that comprises multiple non-road samplesCollection.

Step S202: choose multiple pixels undetermined according to preset strategy in reference picture.

Step S203: determine respectively the genus that each pixel undetermined reflects according to road sample set and non-road sample setProperty, attribute is road or non-road.

Wherein, above-mentioned steps S201, step S202 and step S203 respectively with above-described embodiment in step S5031, stepRapid S5032 and step S5033 are similar, repeat no more herein.

Step S204: utilize default road model to differentiate respectively the attribute of each pixel undetermined.

Utilize the default road model to judge respectively the attribute of each pixel undetermined, judge each pixel undeterminedAttribute is that road is also non-road, obtains the differentiation result of default road model to each pixel undetermined.

Step S205: whether the default differentiation result of road model of judgement and the attribute of fixed pixel undetermined be consistent.

Step S206: for each pixel undetermined, if the differentiation result of above-mentioned default road model and above-mentioned enforcementThe attribute of the pixel undetermined of determining in step S203 in example is consistent, determines that the differentiation result of default road model is correct. ExampleAs, the attribute of some pixels undetermined is road, the attribute that default road model is differentiated this pixel undetermined is also road,The differentiation result that default road model is described is consistent with the attribute of pixel reality undetermined, can determine for this pixel undetermined,The differentiation result of default road model is correct.

Step S207: for each pixel undetermined, if the differentiation result of above-mentioned default road model and above-mentioned enforcementThe attribute of the pixel undetermined of determining in step S203 in example is inconsistent, is not just determining the differentiation result of default road modelReally. For example, the attribute of certain pixel undetermined is road, and default road model by the attribute differentiation of this pixel undetermined isNon-road, therefore, the differentiation result of default road model is incorrect.

Step S208: the pixel quantity undetermined that judicious differentiation result is corresponding, with the ratio of pixel quantity undeterminedWhether be less than predetermined threshold value.

Step S209: if the quantity of pixel undetermined corresponding to correct differentiation result, with all pixel quantity undeterminedRatio be less than predetermined threshold value, utilize pixel correction undetermined corresponding to incorrect differentiation result to preset road model. ExampleAs, if pixel quantity undetermined is 1000, and utilize default road model to differentiate each pixel undeterminedAfter, the differentiation result that defines 600 pixels is correct, differentiates the pixel quantity 600 that result is correct and ownsThe ratio of pixel quantity 1000 undetermined is 0.6, is 0.99 if set predetermined threshold value, obtains default road model and differentiatesIncorrect 400 pixels undetermined of result, and utilize these 400 pixel corrections undetermined to preset road model, for example,Adopt the default road model of the mode correction such as attribute of changing basic pixel or revising former basic pixel.

Step S210: if the quantity of pixel undetermined corresponding to correct differentiation result, with all pixel quantity undeterminedRatio be not less than predetermined threshold value, do not revise default road model.

The general principle of default road model is to analyze successively each pixel in reference picture, and attribute is determinedPixel as basic pixel, obtain the attribute of each pixel in reference picture, that is, obtain with attribute and beThe pixel on road has the pixel of similar pixel characteristic, and determines that the attribute of the pixel with similar pixel characteristic is sameFor road.

Therefore,, if the attribute of certain basic pixel is wrong in default road model, can cause default road modelCan not correctly judge the attribute of the undetermined pixel similar to the pixel characteristic of this basis pixel, default road mould occursThe differentiation result of type is incorrect situation.

In the present embodiment, can utilize the default road of the incorrect pixel correction undetermined of differentiation result of default road modelModel, and only correction causes pixel undetermined to differentiate the incorrect basic pixel of result, comprises and changes basic pixelOr revise the attribute etc. of former basic pixel, can avoid, to not needing the basic pixel of revising to revise, effectively reducingBuild computational resource and computing time that default road model takies. For example, in a specific embodiment, obtain and causeThe attribute misjudgement of the pixel undetermined that is road by certain attribute is for off-highroad basic pixel, taking this basis pixel as basePlinth builds single Gauss model, is divided into the distribution that certain original model of representative is beyond expression to express this mistake, on this basis,The existing mixed Gauss model about road and several new single Gauss models are combined, re-use E-MAlgorithm builds new mixed Gauss model to upgrade default road model.

The schematic diagram of a kind of road Identification device that Figure 10 provides for the embodiment of the present invention, is applied to Vehicular body front and installs and take the photographThe vehicle of picture head, is characterized in that, device comprises acquiring unit 1, trajectory unit 2, superpositing unit 3, sample unit 4With recognition unit 5:

The control appliance of the virtual controlling instruction that acquiring unit 1 is inputted with camera and transmission user is respectively connected, for obtainingGet the reference picture in vehicle movement front, and receive the virtual controlling instruction of user's input, wherein, virtual controlling instructionAt least comprise: wheel pivot angle, throttle amount and brake amount etc.;

In a specific embodiment of the present invention, control appliance can be human-computer interaction device, and human-computer interaction device canBeing steering wheel, gas pedal and the brake pedal of emulation, can be also other controlling equipments taking mouse-keyboard as basis,Operator obtains virtual controlling instruction by human-computer interaction device.

Trajectory unit 2 is connected with acquiring unit 1, for predict the driving trace line of vehicle according to virtual controlling instruction;

Superpositing unit 3 is connected with trajectory unit 2 and acquiring unit 1 respectively, for reference diagram that driving trace line is added toOn picture;

Sample unit 4 is connected with superpositing unit 3 and acquiring unit 1 respectively, for according to driving trace line at reference pictureIn choose the pixel of multiple reflection roadway characteristics, and using each pixel of choosing all as a road sample, obtainInclude the road set of stereotypes of multiple road samples;

Recognition unit 5 is connected with sample unit 4, for utilizing all road pattern recognition reference diagrams of road sample setRoad area in picture.

In another embodiment of the present invention, the trajectory unit 2 in above-described embodiment comprises vehicle front-wheel trajectory unit 2,Vehicle front-wheel trajectory unit 2 is for obtaining wheel trajectories line;

Sample unit 4 comprises that the first predeterminable area chooses unit, and the first predeterminable area is chosen unit at two wheel railsIn the interval region of trace, choose the first predeterminable area;

Sample unit 4 comprises and also comprises that the second predeterminable area chooses unit, and the second predeterminable area is chosen unit for according to twoTwo the second predeterminable areas are got in the line selection of bar wheel trajectories, corresponding second predeterminable area of each track of vehicle line, andEach the second predeterminable area is line centered by corresponding wheel trajectories line;

Sample unit 4 also comprises chooses unit with the first predeterminable area respectively and the second predeterminable area is chosen the road that unit is connectedRoad sample is chosen unit, road sample choose unit for according to the first predeterminable area and/or the second predeterminable area at reference diagramIn picture, choose road sample.

In another embodiment of the present invention, the recognition unit 5 in above-described embodiment comprise model unit, template(-let),Road pixel acquiring unit and road Identification unit;

Template(-let) is for utilizing default road model to differentiate the attribute of each pixel of reference picture, and attribute is roadOr non-road;

Road pixel acquiring unit is connected with template(-let), for obtaining reference according to the differentiation result of default road modelThe pixel of all reflection roadway characteristics in image;

Road Identification unit is connected with road pixel acquiring unit, for utilizing all reflection roadway characteristics of reference picturePixel identification reference picture in road area.

In another embodiment of the present invention, the sample unit 4 in above-described embodiment comprises non-road sample unit, non-Road sample unit is used for choosing at the non-road predeterminable area of reference picture the pixel of the non-roadway characteristic of multiple reflections,And using each pixel of choosing all as non-road sample, obtain the non-road sample set that comprises multiple non-road samplesClose;

Recognition unit 5 comprises unit undetermined, determining unit, judgement unit, differentiation result determining unit and amending unit;

Unit undetermined is for choosing multiple pixels undetermined according to preset strategy at reference picture;

Determining unit is connected with non-road sample unit and unit undetermined respectively, for according to road sample set and non-road sampleThis collection is determined respectively the attribute of each pixel undetermined, and attribute is road or non-road;

Judgement unit is connected with template(-let), and for utilizing default road model to differentiate reference picture respectively, each is undeterminedThe attribute of pixel;

Differentiate result unit and be connected with determining unit and judgement unit respectively, for for each pixel undetermined, in advanceIf when the attribute of the differentiation result of road model and pixel undetermined is consistent, determine that the differentiation result of default road model is correct;

Amending unit is connected with differentiation result determining unit, for pixel number undetermined corresponding to the differentiation result correctAmount, while being less than predetermined threshold value, utilizes pixel undetermined corresponding to incorrect differentiation result with the ratio of pixel quantity undeterminedPoint is revised default road model.

In another embodiment of the present invention, the sample unit 4 in above-described embodiment comprises non-road sample unit, non-Road sample unit is done for the pixel of choosing the non-roadway characteristic of multiple reflections at the non-road predeterminable area of reference pictureFor non-road sample;

Recognition unit 5 comprises unit undetermined, determining unit and road model unit;

Road model unit is connected with non-road sample unit and determining unit, for according to road sample, non-road sampleSet up default road model with pixel undetermined.

In another embodiment of the present invention, the non-road sample unit in above-described embodiment comprises that non-road area choosesUnit;

Non-road area is chosen unit for top, the upper left corner and/or the upper right at reference picture according to virtual controlling instructionNon-road predeterminable area is chosen at angle, chooses multiple pixels in non-road predeterminable area picture as the non-roadway characteristic of reflectionVegetarian refreshments.

In another embodiment of the present invention, the determining unit in above-described embodiment comprises pixel characteristic unit, set groupBecome unit, judging unit and feature determining unit;

Pixel characteristic unit is connected with sample unit 4 and unit undetermined respectively, for obtaining each of road sample setThe pixel characteristic of each the non-road sample in pixel characteristic, the non-road sample set of road sample, and each is treatedThe pixel characteristic of fixation vegetarian refreshments;

Set component units is connected with pixel characteristic unit, for pixel similar pixel characteristic is formed to a similar collectionClose, similar set has multiple, and the pixel characteristic dissmilarity of pixel in any two similar set is all wrapped in each similar setDraw together: at least one pixel undetermined, and, at least one pixel in road sample set and non-road sample set;

Judging unit is connected with set component units, for the quantity that judges respectively each similar set road sample isThe no quantity that is greater than non-road sample;

Feature determining unit is connected with judging unit, in the time that the quantity of road sample is greater than the quantity of non-road sample,The attribute of determining the pixel undetermined in set is road; In the time that the quantity of road sample is not more than the quantity of non-road sample,The attribute of determining the pixel undetermined in set is non-road.

In another embodiment of the present invention, the superpositing unit 3 in above-described embodiment comprises camera parameter unit, couplingBe related to unit and image superpositing unit;

Camera parameter unit is for obtaining the parameters of the video camera of taking reference picture;

Matching relationship unit is connected with camera parameter unit, exists for utilizing the parameters of video camera to obtain subjectMatching relationship in reference picture between the picture position of object and the locus in real space;

Image superpositing unit 3 is connected with matching relationship unit, for driving trace line being superimposed upon to reference according to matching relationshipOn image.

It should be noted that, in this article, the relational terms such as " first " and " second " etc. be only used for byEntity or operation and another entity or operating space separate, and not necessarily require or imply these entities or behaviourBetween work, there is relation or the order of any this reality. And term " comprises ", " comprising " or it is anyOther variants are intended to contain comprising of nonexcludability, thus make to comprise a series of key elements process, method, article orEquipment not only comprises those key elements, but also comprises other key elements of clearly not listing, or is also included as this mistakeThe key element that journey, method, article or equipment are intrinsic. In the situation that there is no more restrictions, " comprise one by statementIndividual ... " key element that limits, and be also not precluded within process, method, article or the equipment that comprises described key element and existOther identical element.

The above is only the specific embodiment of the present invention, makes those skilled in the art can understand or realize the present invention.To be apparent to one skilled in the art to the multiple amendment of these embodiment, as defined herein oneAs principle can be in the situation that not departing from the spirit or scope of the present invention, realize in other embodiments. Therefore, thisBrightly will can not be restricted to these embodiment shown in this article, but will meet and principle disclosed herein and features of noveltyThe widest consistent scope.

Claims

1. a roads recognition method, is characterized in that, described method comprises:

Obtain the reference picture in vehicle movement front;

Described driving trace line is added on described reference picture;

Each pixel of choosing, all as a road sample, is obtained including the road sample set of multiple road samples;

2. method according to claim 1, is characterized in that,

Described driving trace line comprises two wheel trajectories lines;

3. method according to claim 1, is characterized in that, the described all roads that utilize in described road sample setRoad area described in the specimen discerning of road in reference picture, comprising:

Obtain according to the differentiation result of described default road model the pixel that in described reference picture, all properties is road;

Utilize the pixel that in described reference picture, all properties is road to identify the road area in described reference picture.

4. method according to claim 3, is characterized in that, described method also comprises:

Utilize described default road model to differentiate respectively in described reference picture the attribute of pixel undetermined described in each;

5. method according to claim 3, is characterized in that,

Determine respectively pixel attribute undetermined described in each, institute according to described road sample set and described non-road sample setStating attribute is road or non-road;

6. according to the method described in claim 4 or 5, it is characterized in that, the described non-road at described reference picture is pre-If choose the pixel of the non-roadway characteristic of multiple reflections in region, comprising:

Choose described non-road according to described virtual controlling instruction in top, the upper left corner and/or the upper right corner of described reference picturePredeterminable area;

7. according to the method described in claim 4 or 5, it is characterized in that, described according to described road sample set and described inNon-road sample set is determined respectively the attribute of pixel undetermined described in each, comprising:

By a similar set of similar pixel characteristic pixel composition, described similar set has multiple, any two described similarThe pixel characteristic dissmilarity of pixel in set, includes in each described similar set: pixel undetermined described at least one,And, at least one pixel in described road sample set and described non-road sample set;

Judge respectively whether the quantity of road sample described in similar set described in each is greater than the number of described non-road sampleAmount;

8. method according to claim 1, is characterized in that,

Utilize the parameters of video camera obtain the picture position of subject in described reference picture with at real spaceIn locus between matching relationship;

9. a road Identification device, is applied to Vehicular body front the vehicle of camera is installed, and it is characterized in that described deviceComprise acquiring unit, trajectory unit, superpositing unit, sample unit and recognition unit:

10. device according to claim 9, is characterized in that,

Described sample unit also comprises that respectively choosing unit and the second predeterminable area with described the first predeterminable area chooses unit and connectThe road sample connecing is chosen unit, and described road sample is chosen unit for according to described the first predeterminable area and/or describedTwo predeterminable areas are chosen described road sample in described reference picture.

11. devices according to claim 9, is characterized in that, described recognition unit comprises model unit, attributeUnit, road pixel acquiring unit and road Identification unit;

12. devices according to claim 11, is characterized in that,

Described judgement unit is connected with described template(-let), for utilizing described default road model to differentiate respectively described referenceThe attribute of pixel undetermined described in each in image;

13. methods according to claim 11, is characterized in that,

Described road model unit is connected with described non-road sample unit and described determining unit, for according to described roadDescribed in each non-road sample in each road sample, described non-road sample set in sample set and each, treatFixation vegetarian refreshments is set up described default road model.

14. according to the device described in claim 12 or 13, it is characterized in that, described non-road sample unit comprises non-Road area is chosen unit;

Described non-road area choose unit for according to described virtual controlling instruction at described road image reference pictureDescribed non-road predeterminable area is chosen in portion, the upper left corner and/or the upper right corner; Choose the multiple pictures in described non-road predeterminable areaVegetarian refreshments is as the pixel of the non-roadway characteristic of reflection.

15. according to the device described in claim 12 or 13, it is characterized in that, described determining unit comprises pixel characteristicUnit, set component units, judging unit and feature determining unit;

Described set component units is connected with described pixel characteristic unit, for pixel similar pixel characteristic is formed to oneSimilar set, described similar set has multiple, the pixel characteristic dissmilarity of pixel in any two similar set, each described inIn similar set, include: pixel undetermined described at least one, and, described road sample set and described non-road sample setIn at least one pixel;

Described feature determining unit is connected with described judging unit, for be greater than described non-road in the quantity of described road sampleWhen the quantity of road sample, determine that the attribute of all described pixels undetermined in described similar set is road; DescribedWhen the quantity of road sample is less than the quantity of described non-road sample, determines described in all in described similar set and treat fixationThe attribute of vegetarian refreshments is non-road.

16. devices according to claim 9, is characterized in that, described superpositing unit comprise camera parameter unit,Matching relationship unit and image superpositing unit;