CN105550665B - A kind of pilotless automobile based on binocular vision can lead to method for detecting area - Google Patents

A kind of pilotless automobile based on binocular vision can lead to method for detecting area Download PDF

Info

Publication number
CN105550665B
CN105550665B CN201610027922.0A CN201610027922A CN105550665B CN 105550665 B CN105550665 B CN 105550665B CN 201610027922 A CN201610027922 A CN 201610027922A CN 105550665 B CN105550665 B CN 105550665B
Authority
CN
China
Prior art keywords
disparity map
dense
carried
rejecting
edge
Prior art date
Application number
CN201610027922.0A
Other languages
Chinese (zh)
Other versions
CN105550665A (en
Inventor
付梦印
宋文杰
杨毅
汪稚力
邱凡
Original Assignee
北京理工大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 北京理工大学 filed Critical 北京理工大学
Priority to CN201610027922.0A priority Critical patent/CN105550665B/en
Publication of CN105550665A publication Critical patent/CN105550665A/en
Application granted granted Critical
Publication of CN105550665B publication Critical patent/CN105550665B/en

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
    • G06K9/00Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
    • G06K9/00624Recognising scenes, i.e. recognition of a whole field of perception; recognising scene-specific objects
    • G06K9/00791Recognising scenes perceived from the perspective of a land vehicle, e.g. recognising lanes, obstacles or traffic signs on road scenes
    • G06K9/00798Recognition of lanes or road borders, e.g. of lane markings, or recognition of driver's driving pattern in relation to lanes perceived from the vehicle; Analysis of car trajectory relative to detected road
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01CMEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
    • G01C11/00Photogrammetry or videogrammetry, e.g. stereogrammetry; Photographic surveying
    • G01C11/02Picture taking arrangements specially adapted for photogrammetry or photographic surveying, e.g. controlling overlapping of pictures
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
    • G06K9/00Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
    • G06K9/62Methods or arrangements for recognition using electronic means
    • G06K9/6201Matching; Proximity measures
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20024Filtering details
    • G06T2207/20032Median filtering
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/30Subject of image; Context of image processing
    • G06T2207/30248Vehicle exterior or interior
    • G06T2207/30252Vehicle exterior; Vicinity of vehicle
    • G06T2207/30256Lane; Road marking

Abstract

The present invention, which provides a kind of pilotless automobile based on binocular vision, can lead to method for detecting area, process are as follows: obtain the vehicle front of vehicle-mounted binocular camera acquisition being mounted on automatic driving car or so mesh image as original identification image;Left and right mesh image is pre-processed, treated dense disparity map is obtained;Corresponding U disparity map is obtained for dense disparity map;For U disparity map acquired disturbance object figure;Use barriers object figure carries out barrier rejecting to dense disparity map, obtains the disparity map after rejecting a large amount of barriers;For the disparity map after a large amount of barriers of rejecting, the V disparity map of its standardization is obtained;For V disparity map, the upper edge of road area is obtained;The upper edge above section of road area in the dense disparity map is rejected, the disparity map for rejecting non-rice habitats region is obtained;For the disparity map for rejecting non-rice habitats region, further barrier rejecting is carried out, gray inversion then is carried out to the obstructions chart of acquisition, acquisition can traffic areas figure.

Description

A kind of pilotless automobile based on binocular vision can lead to method for detecting area

Technical field

The invention belongs to technical field of image processing, and in particular to a kind of pilotless automobile based on binocular vision can lead to Row method for detecting area.

Background technique

With the development of society, automobile has become the irreplaceable vehicles of mankind's daily life.However, therewith Being safety problem that its bring becomes increasingly conspicuous.And the fast development of vehicle intellectualized technology then provides to solve this problem Powerful measure.In recent years, the major well-known Automobile Enterprises in the world play an active part in vehicle intellectualized Industrial Revolution, so that nothing People's driving has no longer been a concept, and many mature intellectualized technologies have been applied to automobile industry, and is achieved significant Economic and social benefit.Meanwhile the unmanned ground vehicle technical research in the fields such as military affairs, security also makes a breakthrough, The equipment such as unmanned explosive-removal robot play great function in major areas such as people's security protection, national security.

Obstacle detection method is mainly based upon laser radar, millimetre-wave radar or super in pilotless automobile technology at present The initiatives sensor such as detector of sound, and this kind of sensor typically cost is higher, power consumption is larger, is easy to interfere with each other.Vapour at present Garage's industry mainly has automatic cruising system, Lane Keeping System, autonomous parking system using mature pilotless automobile technology Deng these systems work general with radar and image information.Specific method is by radar detection barrier, video camera Lane line or other road informations are detected, then the two is subjected to fusion treatment.If only realizing that institute is functional using video camera, no Equipment cost can be only reduced, biosensor power consumption is more reduced, increases the service life of system.Therefore, binocular is relied on merely Camera realizes that the job requirements such as lane detection, obstacle detection, driving recording have a vast market foreground.

Obstacle detection is logical in the safety of automobile as the key technology in Unmanned Systems and DAS (Driver Assistant System) Row, comfortable driving etc. are played a great role.It is how accurate, efficiently detect barrier, acquisition can traffic areas to nobody Driving and DAS (Driver Assistant System) suffer from vital influence.Carrying out generally, based on image can traffic areas detection master It is divided into based on textural characteristics, color characteristic and depth characteristic three categories.Currently, some scholars obtain colour using monocular camera Image, using Texture Segmentation or color segmentation identify can traffic areas, but its effect is often unsatisfactory, it is affected by environment compared with Greatly, it and is mainly used in the urban traffic environment more standardized, in unstructured moving grids and is not suitable for.And believed based on depth Breath mainly obtains environment disparity map by binocular or more mesh cameras, and disparity map is recycled to obtain dense or sparse depth letter Breath, estimates ground model.Such methods are applicable to complex road surface environment, but often computationally intensive, and real-time is poor, can not Preferably it is applied to unmanned ground vehicle.It can be seen that based on image information efficiently and in real time detect can traffic areas be one Huge challenge, while there is very high application value to the development of unmanned ground vehicle and DAS (Driver Assistant System).

Summary of the invention

To solve the above problems, the present invention, which provides a kind of pilotless automobile based on binocular vision, can lead to region detection side Method, this method strong applicability, can under the complicated various roads environment such as weather condition and field, city such as sleet steady operation, And real-time is preferable, can be widely applied to unmanned ground vehicle and DAS (Driver Assistant System).

Realize that technical scheme is as follows:

A kind of pilotless automobile based on binocular vision can lead to method for detecting area comprising:

Step 1, the vehicle front of vehicle-mounted binocular camera acquisition being mounted on automatic driving car or so mesh image is obtained As original identification image;

Step 2, left and right mesh image is pre-processed: color is carried out to left and right mesh image first with color constancy method Coloured silk enhancing, is secondly converted to gray level image for the image after color enhancement;Binocular is carried out to gray level image using SGM method again Stereo matching obtains dense disparity map of the disparity range between 0-128;Intermediate value filter finally is carried out to the dense disparity map Wave, expansion and corrosion treatment obtain treated dense disparity map;

Step 3, treated that each column all pixels of dense disparity map carry out gray-scale statistical to acquired in step 2, Corresponding U disparity map is obtained, standardization U disparity map will be obtained between the disparity range standardization of the U disparity map to 0 to 255;

Step 4, Boundary Extraction is carried out to the standardization U disparity map obtained in step 3 using Canny operator, obtains two The boundary U disparity map of value;It traverses boundary U disparity map and traverses the dense parallax if the pixel value of pixel (i, j) is not 0 The all pixels of figure jth column find or difference identical as the pixel value of pixel (i, j) and are less than given threshold δ1Pixel, and will It is set as obstacle pixel, to obtain rough obstructions chart, then carries out at median filtering, expansion and corrosion to the obstructions chart Reason obtains treated obstructions chart;

Step 5, using step 4 treated obstructions chart obtained, to step 2, treated that dense disparity map carries out Barrier is rejected, and the disparity map of (or containing a small amount of barrier) is obtained after rejecting a large amount of barriers;

Step 6, ash is carried out to every a line all pixels of the disparity map after a large amount of barriers of rejecting obtained in step 5 Degree statistics (statistical parallax range is 0-128), obtains corresponding V disparity map, by the disparity range standardization of the V disparity map to 0 The V disparity map to be standardized between to 255;

Step 7, Boundary Extraction is carried out using V disparity map of the Canny operator to the standardization obtained in step 6, obtained The boundary V disparity map of binaryzation;Using Hough transformation, line segment is detected in the boundary V disparity map, from the line segment detected Selection corresponds to the line segment of road area in treated dense disparity map in step 2, by itself and boundary V disparity map longitudinal axis intersection point Upper edge of the horizontal line as road area;

Step 8, according to the road area upper edge obtained in step 7, the dense view that will be obtained that treated in step 2 The upper edge above section of road area is rejected in poor figure, obtains the disparity map for rejecting non-rice habitats region;

Step 9, for the disparity map for rejecting non-rice habitats region, final barrier is obtained in the way of step 3-4, wherein In the implementation procedure in the way of step 3, the pixel for the road area above section being no longer removed in statistical parallax figure, and In the way of step 4 when implementation procedure, the road area upper edge above section in obstacle figure obtained is all set to Barrier (the road area upper edge in step 7 by obtaining);

Step 10, gray inversion is carried out to the obstructions chart that step 9 obtains, acquisition can traffic areas figure.

The present invention further includes executing following steps after executing the step 10:

Step 11, to step 10 obtain can traffic areas figure carry out outermost contour detection, obtain all outermost contours, often One profile be one it is potential can traffic areas;

Step 12, the profile obtained in step 11 is screened, is selected near vehicle front-wheel, and the maximum wheel of area Exterior feature is used as finally can traffic areas.

Beneficial effect

First, the present invention by obtain dense disparity map, directly in disparity map using statistical information cognitive disorders object with Can traffic areas, processing speed can be improved, and guarantee identification accuracy.

It is more demanding to recognition rate since the present invention serves Unmanned Ground Vehicle platform.Typically, in nothing People's platform with the speed of 40km/h move ahead situation under, it is desirable that can traffic areas identification frame per second in 8Hz or so.This method exists It is per second to handle 5 to 7 frames under the conditions of Intel double-core 2.6GHzCPU processor, have compared to the other methods of similar function bright Aobvious rate advantage, if being transplanted to the Embedded Hardware Platforms such as DSP, processing speed is sufficient for unmanned vehicle requirement.

Why recognition rate is very fast by the present invention, and reason is that this method is for statistical analysis to know directly on disparity map Not can traffic areas recycle 3D information to carry out ground face mould unlike dense parallax information is converted to 3D point cloud by other methods Type fitting, to reduce a large amount of nonlinear operations.It is sensitive to noise information meanwhile because the present invention is to be based on statistical information Spend it is lower, can get it is stable can traffic areas testing result.

Second, the present invention obtains U disparity map and V disparity map by carrying out statistics to disparity map, will be hindered using UV disparity map Object feature and roadway characteristic is hindered to be mapped as linear feature.

Since U disparity map is each column of image to be carried out with gray-scale statistical, and V disparity map carries out gray scale system to each row of image Meter.And the parallax of barrier is characterized in that gray scale is more concentrated in a column direction, and barrier is higher, then gray value is more concentrated, and The parallax of road is characterized in that gray scale is more concentrated in the row direction, and its parallax is only gradually increased from the near to the distant from headstock, concentrates It spends relatively high always.Therefore, lateral line correspondences exactly barrier one by one in original disparity map in U disparity map, and V parallax It is exactly road area in original disparity map that angled straight lines, which correspond to, in figure.It therefore, then can more accurately, fastly using UV disparity map Barrier and road are differentiated fastly.

Third, the present invention carry out edge detection to U disparity map and V disparity map using Canny operator, U disparity map and V are regarded Difference figure carries out binaryzation, and the noise that non-barrier generates is reduced in U disparity map, reduces what barrier generated in V disparity map Noise.

The method of traditional UV disparity map binaryzation be using fixed binarization threshold, according to common impairments object height into Row setting, there is no adaptivitys.The present invention first standardizes UV disparity map between 0 to 255, is then calculated using Canny Son carries out binaryzation, can adaptive varying environment and different height barrier.

4th, the present invention uses U disparity map first, rejects interference information of a large amount of barriers in V disparity map, improves V The Detection accuracy of road area mapping straight line in disparity map, at the same according to V disparity map detect road area approximate range, into One step obtains accurate U disparity map, to obtain accurate obstructions chart.

Since barrier can generate interference in V disparity map, the detection difficulty of road mapping straight line in V disparity map is caused to add Greatly.This method obtains rough obstructions chart according to U disparity map first, is then rejected using the obstructions chart a large amount of in disparity map Barrier is interfered, recycles the disparity map for eliminating a large amount of barriers to obtain V disparity map and carries out road area estimation.To reduce Influence of the barrier to V disparity map.Simultaneously as the invalid barrier of road upper edge above section can be in U disparity map Interference is generated to effective barrier in road boundary or road, this method estimates rough road area using V disparity map, Then it is counted just for estimated road area range to obtain fine U disparity map, to further obtain fine Obstructions chart.

5th, the present invention to acquisition can traffic areas figure carry out outermost contour detection, can efficiently reject invalid lead to Row region, select to vehicle effectively can traffic areas, while eliminate can in traffic areas small barrier influence, into one Step increases identification stability.

This method take full advantage of can traffic areas general features, dexterously eliminated using outermost contour detection means In vain can the noises such as traffic areas, really effectively can traffic areas to unmanned vehicle driving to obtain.And profile is handled into one Step reduce can traffic areas edge noise, increase the stability of identification.Can traffic areas profile stored in the form of point set, For carry out later can traffic areas project (such as IPM image projection) and be further processed and provide convenience, follow-up work can be reduced Complexity.

Detailed description of the invention

Fig. 1 is that the pilotless automobile of the invention based on binocular vision can lead to method for detecting area entirety identification process;

Fig. 2 is that the pilotless automobile of the invention based on binocular vision can lead to the original identification figure in method for detecting area Picture;

Fig. 3 is that the pilotless automobile of the invention based on binocular vision can lead to the original dense view in method for detecting area Difference figure (under) and for the first time acquisition U disparity map (on);

Fig. 4 can lead to for the pilotless automobile of the invention based on binocular vision handles U for the first time in method for detecting area The boundary disparity map that disparity map obtains;

Fig. 5 is that the pilotless automobile of the invention based on binocular vision can lead in method for detecting area for the first time by boundary U disparity map obtains rough obstructions chart for the first time from original disparity map;

Fig. 6 can lead to for the pilotless automobile of the invention based on binocular vision rejects major obstacle in method for detecting area Dense disparity map (left side) after object and the V disparity map (right side) using disparity map acquisition;

Fig. 7 can lead to for the pilotless automobile of the invention based on binocular vision utilizes V disparity map in method for detecting area Acquisition boundary V disparity map (in) and detect that road corresponds to straight line as road area upper edge (left side) in V disparity map;

Fig. 8 can lead to for the pilotless automobile of the invention based on binocular vision only counts dense view in method for detecting area Poor figure road area upper edge once part U disparity map obtained and boundary U disparity map;

Fig. 9 is that the pilotless automobile of the invention based on binocular vision can lead in method for detecting area according to second of U Disparity map obtain can traffic areas figure (left side) and to this can traffic areas figure progress contour detecting result figure (right side);

Figure 10 can lead in method for detecting area for the pilotless automobile of the invention based on binocular vision can traffic areas Testing result figure.

Specific embodiment

The following describes the present invention in detail with reference to the accompanying drawings and specific embodiments.

As shown in Figure 1, the present invention is based on the pilotless automobiles of binocular vision can lead to method for detecting area, including following several A step:

Step 1: the vehicle front of vehicle-mounted binocular camera acquisition being mounted on automatic driving car or so mesh image is obtained As original identification image.

Step 2: pretreatment is carried out to the left and right mesh image acquired in step 1 and Stereo matching obtains dense disparity map:

(201) left and right mesh image is pre-processed using color constancy method, keeps image color feature more obvious, Then it is translated into gray scale GREY image;

(202) matching of SGM binocular solid is carried out to image obtained by step (201), selecting maximum disparity here is 128, is obtained Obtain dense disparity map;

(203) it to step (202) obtained dense disparity map progress median filtering, carries out once expanding and primary corruption later again Erosion processing, disparity map after being handled.

Step 3: obtain that treated that disparity map acquires U disparity map using step 2:

(301) each column all pixels of dense disparity map are counted, are 0 to 128 due to selecting disparity range before, institute U disparity map is longitudinally 128 rows, lateral columns is identical as dense disparity map columns, obtained U disparity map each pixel (i, J) pixel value Val (i, j) is the number of pixels that jth arranges that corresponding parallax is i in dense disparity map;

(302) standardization processing is carried out to the obtained U disparity map of step (301), especially by subtracted image all elements Middle minimum value, divided by the difference of maxima and minima, then multiplied by 255 method, by image all elements range standardize Between to 0 to 255, standardization U disparity map is obtained.

Step 4: it next goes to obtain rough obstructions chart using step 3 standardization U disparity map obtained, specifically Are as follows:

(401) border detection is carried out to standardization U disparity map using Canny operator, to obtain the boundary U view of binaryzation Difference figure;

(402) traversal (401) boundary U disparity map obtained traverses dense if the pixel value of pixel (i, j) is not 0 Disparity map jth column all pixels, searching pixel value is identical as the pixel value of pixel (i, j) or differs given threshold δ1Picture These pixels are set as barrier pixel by element, so that rough obstructions chart is obtained, given threshold δ described in the present embodiment1For 7;

(403) step (402) obstructions chart obtained is carried out carrying out after median filtering once expanding again and primary rotten Erosion processing, obtains treated obstructions chart.

Step 5: using obstructions chart after step 4 processing obtained, to disparity map after step 2 processing obtained Carry out barrier rejecting, specially traversal obstructions chart, if pixel be barrier pixel when if find the disparity map identical bits It sets and its parallax is set as 0, to obtain the disparity map for rejecting a large amount of barriers;

Step 6: V disparity map is acquired using the disparity map after a large amount of barriers of rejecting obtained in step 5:

(601) the every a line all pixels for counting the disparity map, are 0 to 128 due to selecting disparity range before, institute V disparity map is longitudinally 128 column, longitudinal line number is identical as correspondence disparity map line number, obtained V disparity map each pixel (i, J) pixel value Val (i, j) is that the i-th row corresponds to the number of pixels that parallax is j in corresponding disparity map;

(602) standardization processing is carried out to the obtained V disparity map of step (601), especially by subtracted image all elements Middle minimum value, divided by the difference of maxima and minima, then multiplied by 255 method, by image all elements range standardize Between to 0 to 255, standardization V disparity map is obtained.

Step 7: seeking road area upper edge using the standardization V disparity map that step 6 obtains, specifically:

(701) Boundary Extraction is carried out using Canny operator standardization V disparity map obtained to step 6, obtains two-value The boundary V disparity map of change;

(702) all line segments are detected in the V disparity map of this boundary using Hough transformation, wherein there is an inclined line segment pair What is answered is road area in disparity map;

(703) selection meets the line segment of following four condition from all line segments detected, and (one) is tilted to the left setting Angle, (two) and the V disparity map lower left corner are apart no more than given threshold, and (three) line segment lower edge is at a distance from V disparity map bottom Not super set distance, (four) confidence level are higher than given threshold value;If there is multistage line segment to meet the requirements, retain;

(704) straight line obtained in step (703) is calculated, using its horizontal line with longitudinal axis point of intersection as road area Upper edge, if having in step (703) it is a plurality of as a result, if take it with the mean value of longitudinal axis intersection point horizontal line as road area Upper edge;

Step 8: according to step 7 road area upper edge obtained, by the step 2 dense view that obtains that treated Upper edge above section is rejected in poor figure, the disparity map after being rejected;

Step 9: for the disparity map for rejecting non-rice habitats region, select more accurate parameter according to step 3-step 4 Final obstructions chart is obtained, specifically:

(901) by all pixels of part below road area upper edge in each column of disparity map for rejecting non-rice habitats region Gray-scale statistical is carried out, obtains U disparity map, the U disparity map is identical with the U disparity map format of (301) in step 3 etc.;

(902) standardization processing is carried out to the obtained U disparity map of step (901), especially by subtracted image all elements Middle minimum value, divided by the difference of maxima and minima, then multiplied by 255 method, by image all elements range standardize Between to 0 to 255, standardization U disparity map is obtained;

(903) Canny operator border detection is carried out to the standardization U disparity map that step (902) obtain, to obtain two-value The boundary U disparity map of change;

(904) traversal (903) boundary U disparity map obtained traverses rejecting if the pixel value of pixel (i, j) is not 0 The disparity map jth column all pixels in non-rice habitats region, searching pixel value is identical as the pixel value of pixel (i, j) or difference is less than Given threshold δ2Pixel, it should be noted that the selected threshold parameter unlike step (402) is more stringent here, Given threshold δ described in the present embodiment2=3, these pixels are set as barrier pixel, to obtain more more accurate than (402) Obstructions chart;

(905) step (904) obstructions chart obtained is carried out carrying out after median filtering once expanding again and primary rotten Erosion processing, obtains treated obstructions chart;

(906) road area upper edge in step (905) obstructions chart obtained (is obtained the upper edge in step 7 Taking) above section pixel is set as barrier pixel, to obtain final obstructions chart.

Step 10: gray inversion is carried out to the obstructions chart that step 9 obtains, acquisition can traffic areas figure;

Step 11, to step 10 obtain can traffic areas figure carry out outermost contour detection, obtain all outermost contours, Each profile be one it is potential can traffic areas;

Step 12 screens the profile obtained in step 11, selects near vehicle front-wheel, and area is most Big profile is used as finally can traffic areas.

Embodiment one

The specific steps of region detection can be led to by carrying out pilotless automobile using this method are as follows:

Step 1: it is mounted on the tri- mesh stereoscopic camera of BumblebeeX3 of the Hui Dian company of vehicle front.In the present embodiment, The pixel of the obtained left and right mesh image of vehicle-mounted vidicon is arranged to 800 × 600, color mode RGB.Phase is used in this example The leftmost side of machine and the rightmost side obtain binocular image, and camera baseline length is 0.23998500 meter, and focal length is 1002.912048pixels transmission frame per second is 15Hz, acquisition original image is as shown in Figure 2;

Step 2: carrying out SGM Stereo matching after pre-processing to left and right mesh image, carries out again after obtaining dense disparity map Median filtering, filtering parameter select 3 pixel coverages, expansion process, and expansion parameters are 3 pixels, corrosion treatment, corrosion treatment Parameter is 3 pixels, as a result as shown in lower section figure in Fig. 3;

Step 3: acquiring first time U disparity map using dense disparity map after the resulting processing of step 2, as a result as in Fig. 3 Shown in the figure of top, pixel is 800 × 128;

Step 4: edge extracting is carried out to standardization U disparity map obtained by step 3 using Canny operator and obtains binaryzation Boundary U disparity map, as shown in figure 4, result is as shown in lower section figure in Fig. 4.It is dense being returned using the binaryzation boundary U disparity map Disparity map carries out the screening of barrier pixel, the obstructions chart of first time is obtained, as shown in figure 5, processing result such as Fig. 5 right part of flg institute Show;

Step 5: using obstructions chart obtained in step 4, for the step 2 dense parallax that obtained that treated Scheme to carry out barrier rejecting, obtains the disparity map for rejecting a large amount of barriers (or containing a small amount of barrier), the left side processing result such as Fig. 6 Shown in figure;

Step 6: obtaining V disparity map using the disparity map for rejecting major obstacle object, then normalized, as a result as Fig. 6 is right Shown in the figure of side;

Step 7: Boundary Extraction is carried out to the V disparity map to standardize obtained by step 6 using Canny operator, obtains two-value Change boundary V disparity map, recycles Hough transformation to seek V disparity map middle conductor, filter out the line segment of the condition of satisfaction, ask they and V Disparity map vertical axis intercept takes average conduct road area upper edge, as shown in fig. 7, wherein intermediate is that Hough transformation detects straight line As a result, right side is the V disparity map of standardization, horizontal line is road area upper edge in middle graph, is returned in original disparity map such as In Fig. 7 shown in left hand view;

Step 8: according to upper edge obtained by step 7, rejecting upper edge above section in dense disparity map, then again will Disparity map is in the way of step 3 and step 4 after rejecting, and threshold value is selected as 2 (first when barrier screens in current step 4 In secondary step 4 6) threshold value is selected as, i.e., in dense disparity map pixel value corresponding with white pixel in binaryzation disparity map with it is white Color pixel corresponds to parallax difference then by as barrier pixel, thus acquired disturbance object figure, then to carry out median filtering within 2, Dilation erosion processing, obtains final obstacle figure, parameter is as step 2.Gray inversion is finally carried out again, and acquisition can FOH Domain figure.In the step, second of U disparity map is as shown in figure 8, second of the U parallax obtained as shown in Fig. 8 middle graph, obtains the Secondary boundary U disparity map is as shown in figure above Fig. 8.Finally obtained in the step can traffic areas figure as shown in Fig. 9 left hand view;

Step 9: to step 8 obtain can traffic areas figure carry out outermost contour detection, obtain all outermost contours, often One profile be one it is potential can traffic areas, contour detecting result is as shown in Fig. 9 right part of flg;

Step 10: screening the profile obtained in step 9, selects near vehicle front-wheel, and area is maximum Profile as finally can traffic areas, finally can traffic areas testing result it is as shown in Figure 10, Figure 10 left hand view adds for barrier Effect picture after MASK, in Figure 10 right part of flg be can traffic areas effect picture.

Certainly, the invention may also have other embodiments, without deviating from the spirit and substance of the present invention, ripe It knows those skilled in the art and makes various corresponding changes and modifications, but these corresponding changes and change in accordance with the present invention Shape all should fall within the scope of protection of the appended claims of the present invention.

Claims (3)

1. a kind of pilotless automobile based on binocular vision can lead to method for detecting area, characterized in that it comprises:
Step 1, the vehicle front of vehicle-mounted binocular camera acquisition being mounted on automatic driving car or so mesh image conduct is obtained Original identification image;
Step 2, left and right mesh image is pre-processed, obtains treated dense disparity map;
Step 3, gray-scale statistical is carried out to each column all pixels of the dense disparity map, corresponding U disparity map is obtained, by the U The disparity range standardization of disparity map is to obtaining standardization U disparity map between 0 to 255;
Step 4, Boundary Extraction is carried out to the standardization U disparity map obtained in step 3 using Canny operator, obtains binaryzation Boundary U disparity map;Boundary U disparity map is traversed, if the pixel value of pixel (i, j) is not 0, traverses the dense disparity map the The all pixels of j column find or difference identical as the pixel value of pixel (i, j) and are less than given threshold δ1Pixel, and set For obstacle pixel, to obtain rough obstructions chart, then median filtering, expansion and corrosion treatment are carried out to the obstructions chart, Obtain treated obstructions chart;
Step 5, using step 4 treated obstructions chart obtained, to step 2, treated that dense disparity map carries out obstacle Object is rejected, and the disparity map after rejecting a large amount of barriers is obtained;
Step 6, gray scale system is carried out to every a line all pixels of the disparity map after a large amount of barriers of rejecting obtained in step 5 Meter, obtains corresponding V disparity map, the V parallax that will be standardized between the disparity range standardization of the V disparity map to 0 to 255 Figure;
Step 7, the upper edge for determining road area specifically includes following sub-step:
(701) Boundary Extraction is carried out using Canny operator standardization V disparity map obtained to step 6, obtains the side of binaryzation Boundary's V disparity map;
(702) all line segments are detected in the boundary V disparity map using Hough transformation;
(703) selection meets the line segments of following 4 conditions from all line segments detected, and (one) is tilted to the left set angle, (2) the proximal border V disparity map lower left corner is leaned on, close to boundary V disparity map bottom, (four) confidence level is higher than to be set (three) line segment lower edge Determine threshold value;
(704) for detected line segment in step (703), using its horizontal line with V disparity map longitudinal axis point of intersection as road The upper edge in region, if the line segment detected in step (703) have it is a plurality of, by its at V disparity map longitudinal axis intersection point mean value Upper edge of the horizontal line as road area;
Step 8, according to the road area upper edge obtained in step 7, the dense disparity map that will be obtained that treated in step 2 Middle road area upper edge above section is rejected, and the disparity map for rejecting non-rice habitats region is obtained;
Step 9, for the disparity map for rejecting non-rice habitats region, final barrier is obtained in the way of step 3-4, wherein In the way of step 4 in implementation procedure, given threshold δ2, and δ1> δ2
Step 10, gray inversion is carried out to the obstructions chart that step 9 obtains, acquisition can traffic areas figure.
2. the pilotless automobile based on binocular vision can lead to method for detecting area according to claim 1, which is characterized in that The process of the step 2 are as follows: color enhancement is carried out to left and right mesh image first with color constancy method, secondly increases color Image after strong is converted to gray level image;Binocular solid matching is carried out to gray level image using SGM method again, obtains parallax model Enclose the dense disparity map between 0-128;Median filtering, expansion and corrosion treatment finally are carried out to the dense disparity map, obtained The dense disparity map that takes that treated.
3. the pilotless automobile based on binocular vision can lead to method for detecting area according to claim 1, which is characterized in that This method further includes following steps:
Step 11, to step 10 obtain can traffic areas figure carry out outermost contour detection, obtain all outermost contours, each round Exterior feature for one it is potential can traffic areas;
Step 12, the profile obtained in step 11 is screened, selection is near vehicle front-wheel, and the maximum profile of area is made For finally can traffic areas.
CN201610027922.0A 2016-01-15 2016-01-15 A kind of pilotless automobile based on binocular vision can lead to method for detecting area CN105550665B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610027922.0A CN105550665B (en) 2016-01-15 2016-01-15 A kind of pilotless automobile based on binocular vision can lead to method for detecting area

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610027922.0A CN105550665B (en) 2016-01-15 2016-01-15 A kind of pilotless automobile based on binocular vision can lead to method for detecting area

Publications (2)

Publication Number Publication Date
CN105550665A CN105550665A (en) 2016-05-04
CN105550665B true CN105550665B (en) 2019-01-25

Family

ID=55829848

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610027922.0A CN105550665B (en) 2016-01-15 2016-01-15 A kind of pilotless automobile based on binocular vision can lead to method for detecting area

Country Status (1)

Country Link
CN (1) CN105550665B (en)

Families Citing this family (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105955303A (en) * 2016-07-05 2016-09-21 北京奇虎科技有限公司 UAV autonomous obstacle avoidance method and device
CN106228134A (en) * 2016-07-21 2016-12-14 北京奇虎科技有限公司 Drivable region detection method based on pavement image, Apparatus and system
CN106228138A (en) * 2016-07-26 2016-12-14 国网重庆市电力公司电力科学研究院 A kind of Road Detection algorithm of integration region and marginal information
CN106295607A (en) * 2016-08-19 2017-01-04 北京奇虎科技有限公司 Roads recognition method and device
CN106446785A (en) * 2016-08-30 2017-02-22 电子科技大学 Passable road detection method based on binocular vision
WO2018058355A1 (en) * 2016-09-28 2018-04-05 驭势科技(北京)有限公司 Method and system for detecting vehicle accessible region in real time
CN106651785A (en) * 2016-10-13 2017-05-10 西北工业大学 Color constancy method based on color edge moments and anchoring neighborhood regularized regression
CN106651836B (en) * 2016-11-04 2019-10-22 中国科学院上海微系统与信息技术研究所 A kind of ground level detection method based on binocular vision
CN107564053B (en) * 2017-08-28 2020-06-09 海信集团有限公司 Parallax value correction method and device applied to road scene
CN107729856B (en) * 2017-10-26 2019-08-23 海信集团有限公司 A kind of obstacle detection method and device
CN109741306A (en) * 2018-12-26 2019-05-10 北京石油化工学院 Image processing method applied to hazardous chemical storehouse stacking

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101852609A (en) * 2010-06-02 2010-10-06 北京理工大学 Ground obstacle detection method based on binocular stereo vision of robot
CN102313536A (en) * 2011-07-21 2012-01-11 清华大学 Method for barrier perception based on airborne binocular vision
CN103714532A (en) * 2013-12-09 2014-04-09 广西科技大学 Method for automatically detecting obstacles based on binocular vision
CN104166834A (en) * 2013-05-20 2014-11-26 株式会社理光 Pavement detection method and pavement detection device

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101852609A (en) * 2010-06-02 2010-10-06 北京理工大学 Ground obstacle detection method based on binocular stereo vision of robot
CN102313536A (en) * 2011-07-21 2012-01-11 清华大学 Method for barrier perception based on airborne binocular vision
CN104166834A (en) * 2013-05-20 2014-11-26 株式会社理光 Pavement detection method and pavement detection device
CN103714532A (en) * 2013-12-09 2014-04-09 广西科技大学 Method for automatically detecting obstacles based on binocular vision

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
双目立体视觉稠密视差算法研究;刘小莉;《万方数据》;20130724;第1-62页
基于双目立体视觉的障碍物识别研究;刘絮;《万方数据》;20150817;第1-71页
基于双目视觉的边缘检测算法;陈志波等;《光学与光电技术》;20080831;第71-74页
户外环境下基于视觉的移动机器人可通行区域识别研究;朱效洲;《中国优秀硕士学位论文全文数据库 信息科技辑》;20150115;第15-48页

Also Published As

Publication number Publication date
CN105550665A (en) 2016-05-04

Similar Documents

Publication Publication Date Title
CN104331910B (en) A kind of track obstacle detecting system based on machine vision
Yenikaya et al. Keeping the vehicle on the road: A survey on on-road lane detection systems
CN105261020B (en) A kind of express lane line detecting method
US20160014406A1 (en) Object detection apparatus, object detection method, object detection program, and device control system mountable to moveable apparatus
CN103177246B (en) Based on the dual model Lane detection method of dynami c block division
JP2015207281A (en) Solid detector, solid detection method, solid detection program, and mobile body equipment control system
CN104318258B (en) Time domain fuzzy and kalman filter-based lane detection method
Siogkas et al. Traffic Lights Detection in Adverse Conditions using Color, Symmetry and Spatiotemporal Information.
Aly Real time detection of lane markers in urban streets
CN104573646B (en) Chinese herbaceous peony pedestrian detection method and system based on laser radar and binocular camera
Teoh et al. Symmetry-based monocular vehicle detection system
Kong et al. General road detection from a single image
US8890951B2 (en) Clear path detection with patch smoothing approach
US8670592B2 (en) Clear path detection using segmentation-based method
US8699754B2 (en) Clear path detection through road modeling
He et al. Accurate and robust lane detection based on dual-view convolutional neutral network
CN103123722B (en) Road object detection method and system
EP2811423B1 (en) Method and apparatus for detecting target
Nieto et al. Road environment modeling using robust perspective analysis and recursive Bayesian segmentation
CN102043950B (en) Vehicle outline recognition method based on canny operator and marginal point statistic
CN103714538B (en) road edge detection method, device and vehicle
CN105069415B (en) Method for detecting lane lines and device
US5706355A (en) Method of analyzing sequences of road images, device for implementing it and its application to detecting obstacles
Kühnl et al. Spatial ray features for real-time ego-lane extraction
CN101750049B (en) Monocular vision vehicle distance measuring method based on road and vehicle information

Legal Events

Date Code Title Description
PB01 Publication
C06 Publication
SE01 Entry into force of request for substantive examination
C10 Entry into substantive examination
GR01 Patent grant
GR01 Patent grant