CN103559507B - The method for traffic sign detection being combined based on color with shape facility - Google Patents
The method for traffic sign detection being combined based on color with shape facility Download PDFInfo
- Publication number
- CN103559507B CN103559507B CN201310582174.9A CN201310582174A CN103559507B CN 103559507 B CN103559507 B CN 103559507B CN 201310582174 A CN201310582174 A CN 201310582174A CN 103559507 B CN103559507 B CN 103559507B
- Authority
- CN
- China
- Prior art keywords
- image
- color
- shape
- traffic sign
- extraction
- Prior art date
- Legal status (The legal status 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 status listed.)
- Active
Links
Abstract
The present invention proposes a kind of method for traffic sign detection being combined based on color and shape feature, hsv color feature is used first, extract the suspicious region of traffic sign in entire image, then shape facility is used, suspicious region is divided again, the region that color and shape all meets condition is extracted, filters off other regions.The inventive method operand is small, and robustness is good, insensitive to extraneous environmental change, can detect the traffic sign under the complex background of road both sides accurately, in real time.
Description
Technical field
The present invention relates to a kind of detection method of road signs, more particularly to a kind of friendship based under complex background
Logical marker detection method.
Background technology
Traffic Sign Recognition（Traffic Sign Recognition, TSR）System refers to people by advanced information skill
Art, data transmission technology, control technology and computer processing technology etc. effectively integrated use in whole transportation management system,
Make one, car, road and environment close fit, so as to set up it is a kind of in a wide range of, it is comprehensive play a role it is real-time, accurate,
Efficient traffic information prompting system.It is an important component of intelligent transportation system, due to containing in traffic sign
Many important transport information, such as speed prompt, the change of road ahead situation, the driving behavior currently driven a vehicle are restricted
Deng, therefore in this accessory system, how quickly, accurately and efficiently detect the traffic sign in road and feed back it
To human pilot or control system, for ensureing driving safety, avoiding traffic accident has highly important research meaning
Justice, therefore, paid attention to by more and more experts and scholars.
It is more famous in terms of intelligent transportation system research to have：European Prometheus (Program for
European Traffic with Highest Efficiency and Unprecedented Safety,PROMETHEUS)
Plan, the intelligent vehicle roadnet that the U.S. proposes（Intelligent Vehicle and Highway Systems,
IVHS), and Japan advanced safety vehicle（Advanced Safety Vehicle, ASV) etc..At present, in intelligent vehicle machine
The UBM for mainly having Germany in forward position in the research of device visual field（Universitat der Bundeswehr Munchen）
University, the seminar of Italian professor Broggi leader, U.S.'s International Technology are using company etc..Wherein, German UBM universities
The intelligent vehicle research group of professor Dick-manns leader is directed to the research of dynamic machine visual field always, development
EMS-Vision visions can preferably simulate human eye functions.In past more than 10 years, some countries successfully develop
The road Identification and tracking system of view-based access control model.Wherein, representative system has：LOIS systems, GOLD systems, RALPH
System, SCARF systems and ALVINN systems.
Pilotless automobile project is just proposed before Google's two-and-a-half years, wherein, Traffic Sign Recognition System is also be unable to do without,
More than the ten of the said firm is just announced early in the leader Chris E Musen of the 8Yue Fen Googles pilotless automobile project of 2011
Pilotless automobile has driven safely 480,000 kilometers under computer.The automobile factorys such as subsequent BMW, Volvo, Audi
Business also passes in succession through and is equipped with intelligence software, the visual field aids in and EMS is proposed unmanned concept car.Major design
For city traffic driving, by the judgement to roadside traffic sign and other correlation techniques, automobile independently can accelerate and brake.State
In terms of interior unmanned vehicle research and development, the He Hangen professors of the National University of Defense technology and professor Dai Bin lead since the mid-90 in last century
" autonomous driving technology " innovation team start from scratch, by autonomous innovation realize series of key techniques break through, in 2001
Year succeeds in developing first of China pilotless automobile, has broken the external block to the unmanned technology in China.In addition, military hand over
Attend a school by taking daily trips Xu Youchun professors, professor Deng Zhidong etc. of Tsing-Hua University of institute, also fruitful in terms of unmanned vehicle research and development.It is domestic
Other colleges and universities（Such as Beijing Institute of Technology, Tsing-Hua University, Xi'an Communications University, Wuhan University）, research institution and some vapour
Car enterprise also expands the exploratory development and preliminary experiment of some correlation techniques in succession.
TSR accessory systems are the multi-disciplinary cross-applications such as computer vision, artificial intelligence, image procossing, pattern-recognition,
With the development of video technique and computer technology, identified using the video detection of image processing method intelligent extraction traffic sign
Have become possibility.Mainly include two basic links in TSR systems：It is the detection of traffic sign first（Segmentation）, including it is right
The pretreatment of the image collected and the segmentation positioning of traffic sign；Next to that the identification of traffic sign（Classification）, including traffic
Feature extraction, signature analysis and final identification of mark etc..And detection is the key link of TSR systems, the quality of its result is straight
Connecing influences the quality of recognition result.
The content of the invention
Goal of the invention
In order to which preferably traffic sign is identified, the present invention is proposed one kind and is combined based on color with shape facility
Method for traffic sign detection.
Technical scheme
The present invention is implemented by the following technical programs：
A kind of method for traffic sign detection being combined based on color with shape facility, it is characterised in that：This method step
It is as follows：
（1）, extraction color characteristic
Using HSV colorimetry models, suspicious object color is extracted to entire image by assigning corresponding value to H, S, V.
（2）, extraction shape facility
To entire image after color extraction only be left with target traffic sign color identical region, now, by right
Regional carries out the shape extraction based on region, by asking circularity, rectangular degree, range of extension to sentence the shape of regional
It is disconnected, and only leave and take circular, rectangle and delta-shaped region.
（3）Morphologic filter
The noise at the boundary of image is removed as smoothing filter using two-dimensional Gaussian function so that target image is more flat
It is sliding；By expanding the burr reduced with corrosion in image, fuzzy noise and removing the dark-coloured details smaller than structural element, keep
Image overall gray value and big dark areas are basically unchanged.
（4）Suspicious object is extracted
By the shape facility extracted after morphologic filter, with original image progress and computing, colored mesh is extracted
Logo image, i.e. Traffic Sign Images.
Advantage and effect
Because the single features of image describe the content of image unavoidable excessively unilateral, it is only capable of giving the part attribute of image
With expression, extracted for real-time traffic sign under nature, the retrieval of single features is usually ineffective, and is based on color
The retrieval being combined with shape facility, the expression to picture material more comprehensively, so as to maximize favourable factors and minimize unfavourable ones, obtain more preferable retrieval
Effect.Color is just used alone below, shape, other colors and shape collective effect method is used alone and this patent is adopted
Color shape is combined to be contrasted to extract traffic sign icon.
Brief description of the drawings
Fig. 1 is that color-shape of road traffic sign detection matches model and schematic diagram of just classifying；
Fig. 2 is extraction target minimized profile schematic diagram；
Fig. 3 is each shape length and width schematic diagram of different traffic signs；
Fig. 4 is original image；
Fig. 5 extracts schematic diagram individually to carry out blue logo using RGB；
Fig. 6 extracts schematic diagram individually to carry out blue logo using RGB；
Fig. 7 is that color detects with shape collective effect；
Fig. 8 is to carry out circle marker using this patent method to extract schematic diagram；
Fig. 9 is the original graph of embodiment 1；
Figure 10 is embodiment 1HSV color model；
Figure 11 is the design sketch that embodiment 1 extracts red area to entire image；
Figure 12 is embodiment 1 with morphologic filtering and carries out the design sketch of area filling；
Figure 13 is that embodiment 1 extracts target shape figure；
Figure 14 is the Objective extraction final result figure of embodiment 1.
Embodiment
The present invention is specifically described below in conjunction with the accompanying drawings：
The invention provides a kind of method for traffic sign detection being combined based on color with shape facility, this method step
It is as follows：
（1）, extraction color characteristic
Using HSV colorimetry models, suspicious object color is extracted to entire image by assigning corresponding value to H, S, V；Example
Such as, blueness, red, yellow etc..
（2）, extraction shape facility
To entire image after color extraction only be left with target traffic sign color identical region, now, by right
Regional carries out the shape extraction based on region, by asking circularity, rectangular degree, range of extension to sentence the shape of regional
It is disconnected, and only leave and take circular, rectangle and delta-shaped region.
（3）Morphologic filter
The noise at the boundary of image is removed as smoothing filter using two-dimensional Gaussian function, so that target image is more
Add smooth；By expanding and corroding the noise such as the burr reduced in image, fuzzy and remove the dead color smaller than structural element carefully
Section, image overall gray value and big dark areas is kept to be basically unchanged.
（4）Suspicious object is extracted
By the shape facility extracted after morphologic filter, with original image carry out and computing, so as to extract colour
Target image, i.e. Traffic Sign Images.
Traffic sign is exactly a kind of very lively language for driver, and the driving procedure of driver is almost entirely
This traffic language information is handled by vision system.Want to accomplish to abide by known or unfamiliar environment
Traffic rules are kept, it is exactly crucial premise accurately, timely to identify traffic sign.
But often the notice of people is not in complete set, easily by external interference object in driving procedure（Than
Such as say phone, big small advertisement, passerby, vehicle）Influence, and ignore traffic mark board.Therefore from the 1980s
Mid-term starts, and various traffic sign recognition methods are just constantly suggested, for reminding driver to reach the purpose of safe driving.But
Due to diversity and complexity that scene is arbitrarily brought, the also high request to accuracy, real-time and robustness etc.,
It result in and up to the present go back the relatively good system of the real effect of neither one.Because research object is the unforeseen complicated back of the body
Traffic sign under scape, therefore the present invention is in illumination variation, color incompleteness, shooting angle, blurring or distortion and mark
Aspect of blocking all done comprehensive consideration, to reach exclusive PCR, automatically, timely find out traffic mark board, detection mark
Will information, and driver is promptly fed back to, driving safety is farthest improved, the possibility that accident occurs is reduced, makes to drive
Sailing becomes more safe and light.
China's traffic sign directly relevant with traffic safety is divided into 3 major classes, total 131 kinds (not include deriving from
Mark）.Wherein, prohibitory sign 48, Warning Mark 36, caution sign 47.The research object of the present invention is this 131
Traffic sign.
1）Warning Mark：Vehicle and the traveling of pedestrian are used to refer to, the color characterized by blueness, is shaped as circular or rectangular
Shape.
2）Caution sign:Danger for alerting vehicle and pedestrian place, the color characterized by yellow, except " emergency lane "
Outside for rectangle, shape is the equilateral triangle on angular.
3）Prohibitory sign:For being prevented or restricted from the traveling of vehicle and pedestrian, in addition to " ban releasing " characterized by red
Color, except " Stop and give way " and " giving way " outer shape is circle.
Analyzed more than, traffic sign frame there are blue, black, red three kinds of colors, and shape has circular, triangle and square
Shape, therefore, region that can be according to where the color of traffic sign frame detects traffic sign in traffic image with shape localization
Position.So the extraction by first carrying out yellow, red, blueness and black portions to entire image, is carried by scanning afterwards
Take shape facility such as triangle, circle and the rectangle in region.The suspicious region of traffic sign can probably be extracted.Fig. 1 is friendship
The color of logical Mark Detection-shape pairing model and just classification.
Because the single features of image describe the content of image unavoidable excessively unilateral, it is only capable of giving the part attribute of image
With expression, for the different demands of different user, the retrieval of single features is usually ineffective, and is based on color and shape feature
The retrieval being combined, the expression to picture material more comprehensively, so as to maximize favourable factors and minimize unfavourable ones, obtain more preferable retrieval effectiveness.
（1）Extract color characteristic
The color space mainly used in terms of color extraction has RGB, HSI, LAB, HSV etc..Wherein, RGB color
It is most basic, most widely used, almost all of other spaces can be changed by rgb space.But it reacts
The information of image is not directly perceived enough, does not meet the perception of human vision, and three variables have certain phase in RGB color
Guan Xing so that image is easier to be influenceed by illumination.In HSI models, although tri- component relevance very littles of H, S, I, compared with
For independence, but the unstable of tone can be caused when saturation degree and relatively low brightness, be not used to the detection of traffic sign, only have
The tone that standby certain saturation degree and brightness value can just be stablized, therefore, real-time color detection is not applied to.LAB colors
Space is a kind of color mode formulated by the International Commission on Illumination of tailor measurement color standard.Color can directly be used
The geometric distance in space carries out the comparison of different colours, therefore is effectively used for measuring small aberration, but because it is non-thread
Property conversion, therefore amount of calculation is larger, and color space has singular point.
Hsv color space is the reduced form in Munsell colour space, is the tone with color（H）, saturation degree（S）, it is bright
Degree（V）It is three elements come what is represented, is that non-linear color represents system.Wherein, tone is the attribute for describing pure color, and saturation degree is
The measurement for the degree that description pure color is diluted by white light, brightness are subjective description, embody colourless brightness concept, are
The key parameter of color perception is described.Hsv color space is consistent with perception of the people to color, and in HSV space, people is to color
The perception of difference is more uniform, is the color space for being adapted to human vision property.We are compact from visual consistency, integrality below
Property, several aspects such as naturality are compared to above-mentioned color space, as shown in table 1.
The comparison of 1 several chrominance spaces of table
As can be seen from the above table in hsv color space, be advantageous to the processing of image, such as the segmentation of rim detection, image
With target identification etc..Therefore, this patent carries out the color extraction of the first step using hsv color space to entire image.
（2）Extract shape facility
Shape facility is one of core feature of image, and the shape information of image does not change with the change of color of image,
It is the invariant feature of object.Picture shape feature can distinguish the object of identical category, and color and texture usually do not have
There is this feature.It is and very directly perceived with shape facility difference object.Therefore, retrieval can be improved using shape facility retrieval image
Accuracy and efficiency.
To based on the Type division that shape image is retrieved, into two kinds, a kind of is the characteristic key based on image-region, another
Kind is the retrieval based on edge feature.Shape description based on edge is the description to surrounding the profile of target area, and it is not examined
Consider the information of shaped interior.This kind of description typically has two kinds of forms of continuous type and discrete type.The description method of continuous type is also referred to as
Global type, it is to extract characteristic vector from the entirety of profile；The description method of discrete type is also referred to as structural type, and it is to take turns
Exterior feature extracts corresponding feature after being divided into many fragments.In the image retrieval based on edge feature, we are through commonly used side
The method of boundary's extraction operator mainly has：Canny operators, Laplacian operators, Sobel operators and Roberts operators etc..It is based on
The description method in region is to regard shape area as an entirety, using all pixels in region, by noise and change in shape
Influence it is relatively small, the method for expressing is divided into global type and local type two types.The provincial characteristics of shape mainly has region
Area, Euler's numbers, dispersion, eccentricity, area invariant moment, region framework, set not bending moment, Zernike squares, angular radius become
The methods of changing.
For retive boundary extraction, extracted region accuracy is higher, and due to the color extraction of previous step, has made whole
Width image binaryzation, complex background is had no, therefore be based on extracted region algorithm, carrying for target area suspicious can be realized completely
Take.In summary, this patent is using global type region description extraction suspicious object.
（3）Morphologic filter
1）Smoothed image
In order to remove the noise at the boundary of image, we implement smoothing processing by gaussian filtering method to target image.
Among the various noises of image, Gaussian noise be it is most, for filter Normal Distribution in image edge noise I
Carried out generally by gaussian filtering method.Therefore, it is to use the Gaussian function with two dimension that we, which handle image, herein
Number is carried out as smoothing filter.The expression formula of the Gaussian function of two dimension is as follows：
This patent can be eliminated the noise at the boundary of image using gaussian filtering method, so that target image is more flat
Sliding, accurate extraction of these pretreatment operations for ensuring picture shape feature plays a good role.
2）Expansion and corrosion
Mathematical morphology can be extracted from image for expression and the useful picture content of description region shape, in two-value
In image, the set of all black picture elements is the complete morphology description of image.Morphologic basic operation has expansion and corrosion.
First corrode to expand afterwards and be called out operation, first expand post-etching and be called closed operation.Opening operation can make target area profile become smooth,
Disconnect narrow interruption and eliminate thin protrusion, meanwhile, it will not substantially change the area of target area again.It is defined as：
A ο B=(A Θ B) ⊕ B
Closed operation can diminish narrow interruption and long thin wide gap, eliminate small hole, and the fracture filled up in contour line,
The effect on its smooth border in the case where unobvious change area.It is defined as：
AB=(A ⊕ B) Θ B
By opening operation and closed operation, the point isolated in image and the fine grained chippings of some interference formation can be eliminated, while
Object boundary point can be eliminated, the objective area in image area after corrosion can also compensated.Then with 3 × 3 rectangular window
Mouth carries out medium filtering to image, can play the work for further eliminating picture noise, spot, further segmentation object and background
With.
（4）Suspicious object is extracted
After morphologic filter, the contour of object in image just becomes apparent from clear, needs to extract suspicious area first
X, y direction coordinate starting point in domain, as shown in Fig. 2 obtaining the zone length in direction, and area is less than to the region of certain given threshold
Remove.
Traffic sign of different shapes has different attributes, such as barycenter Edge Distance figure, circularity, rectangular degree, elongation
Degree etc..As shown in figure 3, using in scheming at dotted arrow meaning as starting point, rotate counterclockwise one week, ask barycenter and each edge successively
The distance between point, the second row is the barycenter Edge Distance figure of 3 kinds of different shape traffic signs in figure, and wherein abscissa represents
The pixel number that traffic sign edge image includes, ordinate represent barycenter to the distance value of edge pixel point.
The characteristics of integrating circular, triangle and rectangle each, this patent choose circularity, rectangular degree and elongation as shape
The major parameter of shape feature extraction.Wherein, circularity refers mainly to object close to the degree of theoretical circle；Rectangular degree reflect object with
The similar degree of rectangle；And elongation then that refer to material because external force effect be deformed after extend degree.By asking
The minimum, maximum and average value for going out barycenter to Edge Distance obtain circularity, rectangular degree and the elongation in region.
Specific formula is as follows：
The circularity in region：Cratio=4*pi*S/P^2；
The rectangular degree in region：Rratio=S/(W*H)；
The elongation in region：Eratio=min(W,H)/max(W,H)；
Wherein, the girth in P- regions, S- region areas, W-x directions zone length, H-y directions zone length.
Scholar Miguel A.Garc ' 1a-Garrido are in the paper on road traffic sign detection delivered in 2012, together
Sample using color by the way of shape facility is combined, but its shape extract part, enter using discrete candy operators
Row bound is extracted, and recognition failures can be caused due to breakpoint in extraction process.Such as shown in figure below.This is two adjacent circles
Shape traffic sign, due to the breakpoint of Discrete Operator, cause mark of two circle markers by flase drop into a connection.
From experimental result as can be seen that the effect extracted using color or shape merely is all undesirable, color is utilized
When, especially in blue logo is extracted, because most car plates are also blueness, so be easy to car plate carrying out extraction by mistake, so as to
Influence experimental result.And when being extracted using shape facility, due to the complex background under natural environment so that it extracts effect
Fruit is worse.To sum up, based on color and shape collective effect combine road traffic sign detection, the expression to picture material more comprehensively,
So as to maximize favourable factors and minimize unfavourable ones, more preferable retrieval effectiveness is obtained.Table 2 is carried out pair by quoting recall ratio and precision ratio to three kinds of methods
Than.
Wherein, the sample image of definition datum inquiry is set as N, and the image set for retrieving image is set as M, can obtain：
The number of the associated picture retrieved is：
The number of flase drop is：
The number of missing inspection is：
The number of nd irrelevant image is：
Herein, VmOn the basis of retrieve the correlation of the image that image and ordinal number are m, Vm∈(0,l)；K is cutoff.Look into
The criterion of the two retrieval Performance Evaluations of quasi- rate and recall ratio, we several more than can be worth.
（1）Precision ratio：Precision ratio is commonly used to assess the ability that image indexing system rejects irrelevant image, and what it was obtained is
The ratio shared by image correctly searched among retrieval set.
Precision ratio：
（2）Recall ratio：Recall ratio is used for determining the ability that image indexing system retrieves associated picture, and what it was obtained is to have
Imitate the ratio of the quantity of all similar images in image data base shared by image.
Recall ratio：
The contrast of 2 three kinds of algorithm recall ratios of table and precision ratio
In summary, the road traffic sign detection algorithm being combined based on color with shape facility that this patent proposes, no matter
It is in recall ratio or in terms of precision ratio, better than individually using colored or shape detection algorithm.Color and shape are used with same
The method that shape combines is compared, and the inventive method can more highlight the traffic sign to be extracted, and it is small to extract error.And by
Entire image is filtered for the first time using color in first, this eliminates the substantial amounts of complicated back of the body for the Shape Feature Extraction of next step
Scape and interference, greatly save the time of Shape Feature Extraction and improve its Detection accuracy, have in the process of the present invention very well
Robustness, real-time, suitable for practical application.
The present invention is described further with specific embodiment below in conjunction with the accompanying drawings,
Embodiment 1：
Original image is obtained in vehicle-mounted vidicon first, as shown in Figure 9.
（1）Color extraction
In the water chestnuts of HSV six bore colour model, form and aspect（H）In the color plane parallel to six water chestnut vertex of a cone faces, they are surrounded
Central shaft V rotates and change, and red, yellow, and green, green grass or young crops, indigo plant, pinkish red six reference colours are separated by 60 degree respectively.Along six water chestnuts bore central shaft V from
Up to lower to change, central shaft top is white（V=1）, bottom is in black（V=0）, they represent the greyscale color of netrual colour system.
Color saturation（S）Change in the horizontal direction, closer to the color of six water chestnuts cone central shaft, its saturation degree is lower, hexagon center
The color saturation of the heart is zero（S=0）, coincided with V=1 of highest lightness, the color of highest saturation is then in outside hexagon
On the edge line of frame（S=1）.
Yellow, blueness and red of the present invention in color extraction section uses hsv color model extraction entire image
Part.Wherein,
The items of blue region detection are entered as：
H(i,j)>=240&&H(i,j)<=255))||((H(i,j)>=0&&H(i,j)<=10)；
The items of red area detection are entered as：
(H(i,j)>=240&&H(i,j)<=255))||((H(i,j)>=0&&H(i,j)<=10)；
The items of yellow area detection are entered as：
H(i,j)>=18&&H(i,j)<=45。
By taking red as an example, the design sketch extracted using this patent method is illustrated in fig. 11 shown below.
（2）Morphologic filtering
Due to the complexity of street background, have noise after color extraction unavoidably, and by mistake extraction with blip face
Color identical zonule, for such case, it is possible, firstly, to compensate uneven background luminance by modified opening operator, select
Take suitable structural element to carry out opening operation to image, produce the estimation to whole image background.Modified opening operator is designated as I ° of b,
Structural element b (i', j') is to image I（I, j) opening operation be defined as I ° of b=(I Θ b) ⊕ b；
In formula：I is the image after color extraction；Symbol " Θ " and " ⊕ " are respectively structural element b (i ', j ') to image I
（I, j) expansion and erosion operation, formula it is as follows：
(I ⊕ b) (i, j)=max { I (i-i', j-j')+b (i', j')/(i', j') ∈ Db}
(I Θ b) (i, j)=min { I (i+i', j+j')-b (i', j')/(i', j') ∈ Db}
In formula：DbIt is structural element b (i ', j ') domain.
Opening operation can remove the bright detail smaller than structural element, keep image overall gray value and big bright area base
This is constant, and uniform background estimating can be obtained by opening operation, then can generate one from the background of artwork image subtraction estimation
Width has the image of homogeneous background, eliminates the influence of noise and interference, this process is the top cap in morphology（Top-h
At）Conversion, formula are as follows：
G=I- (I ο b)
In formula：G is output image.
After being converted by top-hat, the dash area of image can be capped substantially, and be combined together with original background, from
And target area can be extracted in partitioning portion, be not in the phenomenon for coming out shade as Target Segmentation.The opposing party
Face, due to the phenomenon that vehicle-mounted camera jolts, shaken, cause that the universal contrast of image is not high, and noise pollution is serious, edge mould
Paste, therefore the present invention performs top cap-bottom cap after carrying out top cap computing and removing shade（Bottom-hat）It is combined
Computing carries out sonar image enhancing, and bottom cap operation definition is：
In formula, Ib is closed operation, and structural element b (i ', j ') to image I（i,j）Closed operation formula it is as follows：
Ib=(I ⊕ b) Θ b
Closed operation can remove the dark-coloured details smaller than structural element, keep image overall gray value and big dark areas base
This is constant.Morphology preprocessing part is this completes, as shown in figure 12.
（3）Shape is extracted
After morphologic filtering, noise unnecessary in image and interference are eliminated, now can be carried out the mesh based on shape
Mark extraction, by test of many times, author is had found when circularity, rectangular degree and elongation choose following scope, target area extraction
Best results, as shown in figure 13.
Circle marker：C>0.85, R>0.70, E>0.85；
Triangle mark：0.35<C<0.70,0.4<R<0.65, E>0.8；
Rectangular：0.60<C<0.85, R>0.70, E>0.85.
（4）Objective extraction
The target shape extracted and original image are done and computing, can obtain final target identification result.Objective extraction
Final result is as shown in figure 14.
（5）Method accuracy rate
In this experiment, each 40 of four kinds of representative type of signs are employed altogether, are handed over using the inventive method
The detection case of logical mark is as shown in table 3 below：
The all kinds road traffic sign detection situation of table 3
。
Claims (1)
- A kind of 1. method for traffic sign detection being combined based on color with shape facility, it is characterised in that：This method step is such as Under：(1) color extraction；In the water chestnuts of HSV six bore colour model, form and aspect (H) are in parallel in the color plane in six water chestnut vertex of a cone faces, and they surround center Axle V rotates and change, and red, yellow, and green, green grass or young crops, indigo plant, pinkish red six reference colours are separated by 60 degree respectively；Central shaft V is bored from up to along six water chestnuts Lower change, central shaft top is white (V=1), and bottom is in black (V=0), and they represent the greyscale color of netrual colour system；Color Color saturation degree (S) changes in the horizontal direction, and closer to the color of six water chestnuts cone central shaft, its saturation degree is lower, hexagon center Color saturation be zero (S=0), coincided with the V=1 of highest lightness, the color of highest saturation is then outside hexagon On the edge line of frame (S=1)；Yellow, blueness and RED sector in color extraction section uses hsv color model extraction entire image；(2) morphologic filtering；It is possible, firstly, to compensate uneven background luminance by modified opening operator, choose suitable structural element and image is entered Row opening operation, produce the estimation to whole image background；Modified opening operator is designated as IB, structural element b (i', j') are to image I The opening operation of (i, j) is defined as：IB=(I Θ b) ⊕ b；In formula：I is the image after color extraction；Symbol " Θ " and " ⊕ " are respectively structural element b (i ', j ') to image I (i, j) Expansion and erosion operation, formula it is as follows：(I ⊕ b) (i, j)=max { I (i-i', j-j')+b (i', j')/(i', j') ∈ Db}；(I Θ b) (i, j)=min { I (i+i', j+j')-b (i', j')/(i', j') ∈ Db}；In formula：DbIt is structural element b (i ', j ') domain；Opening operation can remove the bright detail smaller than structural element, keep image overall gray value and big bright area substantially not Become, uniform background estimating can be obtained by opening operation, then can generate width tool from the background of artwork image subtraction estimation There is the image of homogeneous background, eliminate the influence of noise and interference, this process is the top cap (top-hat) in morphology Conversion, formula are as follows：In formula：G is output image；After being converted by top-hat, the dash area of image can be capped substantially, and be combined together with original background, so that Partitioning portion can extract target area, be not in the phenomenon for coming out shade as Target Segmentation；On the other hand, by In the phenomenon that vehicle-mounted camera jolts, shaken, cause that the universal contrast of image is not high, and noise pollution is serious, edge blurry, therefore The present invention performs the computing that is combined of top cap-bottom cap (bottom-hat) to enter after carrying out top cap computing and removing shade Row sonar image strengthens, and bottom cap operation definition is：In formula, Ib is closed operation, and the closed operation formula of structural element b (i ', j ') to image I (i, j) is as follows：Ib=(I ⊕ b) Θ b；Closed operation can remove the dark-coloured details smaller than structural element, keep image overall gray value and big dark areas substantially not Become；It this completes morphology preprocessing part；(3) shape is extracted；After morphologic filtering, noise unnecessary in image and interference are eliminated, the target based on shape is now can be carried out and carries Take, by test of many times, it is found that the effect of target area extraction is most when circularity, rectangular degree and elongation choose following scope It is good；Circle marker：C>0.85, R>0.70, E>0.85；Triangle mark：0.35<C<0.70,0.4<R<0.65, E>0.8；Rectangular：0.60<C<0.85, R>0.70, E>0.85；(4) Objective extraction；The target shape extracted and original image are done and computing, can obtain final target identification result；(5) method accuracy rate；Each 40 of four kinds of representative type of signs are employed altogether in the method, using the detection of this method traffic sign Situation is as shown in table 1 below：The all kinds road traffic sign detection situation of table 1；
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310582174.9A CN103559507B (en) | 2013-11-18 | 2013-11-18 | The method for traffic sign detection being combined based on color with shape facility |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310582174.9A CN103559507B (en) | 2013-11-18 | 2013-11-18 | The method for traffic sign detection being combined based on color with shape facility |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103559507A CN103559507A (en) | 2014-02-05 |
| CN103559507B true CN103559507B (en) | 2017-12-05 |
Family
ID=50013748
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310582174.9A Active CN103559507B (en) | 2013-11-18 | 2013-11-18 | The method for traffic sign detection being combined based on color with shape facility |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103559507B (en) |
Families Citing this family (27)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104182760B (en) * | 2014-04-16 | 2017-12-26 | 南京通用电器有限公司 | A kind of transport information read method based on coding type traffic sign |
| CN103971128B (en) * | 2014-05-23 | 2017-04-05 | 北京理工大学 | A kind of traffic sign recognition method towards automatic driving car |
| CN106503607B (en) * | 2015-09-07 | 2020-02-07 | 北京易车互联信息技术有限公司 | Road traffic sign identification method and device |
| CN105335708A (en) * | 2015-10-22 | 2016-02-17 | 四川膨旭科技有限公司 | System for identifying road signs and guideboards during process of vehicle moving |
| CN106845324B (en) * | 2015-12-03 | 2020-07-24 | 阿里巴巴（中国）有限公司 | Method and device for processing guideboard information |
| CN105787475A (en) * | 2016-03-29 | 2016-07-20 | 西南交通大学 | Traffic sign detection and identification method under complex environment |
| JP6625475B2 (en) * | 2016-04-22 | 2019-12-25 | 株式会社 東京ウエルズ | Defect inspection method and defect inspection system |
| CN106205077B (en) * | 2016-06-29 | 2018-07-06 | 南京南工大安全科技有限公司 | A kind of alarm system quickly identified |
| CN105921886B (en) * | 2016-06-29 | 2018-06-26 | 江苏大金激光科技有限公司 | A kind of laser process machine of high recognition performance |
| CN106077971B (en) * | 2016-06-29 | 2018-03-20 | 中江机电科技江苏有限公司 | A kind of laser cutting device for being capable of autonomous classification |
| CN106199610B (en) * | 2016-06-29 | 2018-11-27 | 江苏引拓电力科技有限公司 | A kind of supersonic sounding device with identification function |
| CN106101632A (en) * | 2016-06-29 | 2016-11-09 | 韦醒妃 | The image processing apparatus of view-based access control model feature |
| CN106096587A (en) * | 2016-06-29 | 2016-11-09 | 韦醒妃 | There is the vehicle monitoring system of target recognition function |
| CN105915562A (en) * | 2016-06-29 | 2016-08-31 | 韦醒妃 | Identify verification system based on recognition technology |
| CN106778787A (en) * | 2016-12-29 | 2017-05-31 | 王红军 | A kind of object based on color characteristic assists in identifying method |
| CN108932270B (en) * | 2017-05-27 | 2021-11-30 | 福建省农业科学院果树研究所 | Loquat germplasm resource retrieval contrast method based on Bayes and feedback algorithm |
| CN109255279A (en) * | 2017-07-13 | 2019-01-22 | 深圳市凯立德科技股份有限公司 | A kind of method and system of road traffic sign detection identification |
| CN107679508A (en) * | 2017-10-17 | 2018-02-09 | 广州汽车集团股份有限公司 | Road traffic sign detection recognition methods, apparatus and system |
| CN108256467B (en) * | 2018-01-15 | 2020-06-05 | 河北科技大学 | Traffic sign detection method based on visual attention mechanism and geometric features |
| CN109472781B (en) * | 2018-10-29 | 2022-02-11 | 电子科技大学 | Diabetic retinopathy detection system based on serial structure segmentation |
| CN109859199B (en) * | 2019-02-14 | 2020-10-16 | 浙江科技学院 | Method for detecting quality of freshwater seedless pearls through SD-OCT image |
| CN110110696B (en) * | 2019-05-17 | 2022-07-05 | 百度在线网络技术（北京）有限公司 | Method and apparatus for processing information |
| CN110110148A (en) * | 2019-05-28 | 2019-08-09 | 成都德迈安科技有限公司 | A kind of commodity search method for video image |
| CN110737744B (en) * | 2019-10-14 | 2022-01-25 | 中国地质大学（北京） | Method for manufacturing texture symbols of land utilization classified thematic map |
| CN112906424B (en) * | 2019-11-19 | 2023-10-31 | 上海高德威智能交通系统有限公司 | Image recognition method, device and equipment |
| CN116091795A (en) * | 2022-04-25 | 2023-05-09 | 江苏日盈电子股份有限公司 | Pollution judgment method, pollution judgment system and cleaning method for vehicle-mounted camera |
| CN115488876A (en) * | 2022-06-22 | 2022-12-20 | 湖北商贸学院 | Robot sorting method and device based on machine vision |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2006235752A (en) * | 2005-02-22 | 2006-09-07 | Asahi Engineering Kk | Object recognition device, its control method, computer program and computer readable storage medium |
| CN102542260A (en) * | 2011-12-30 | 2012-07-04 | 中南大学 | Method for recognizing road traffic sign for unmanned vehicle |
-
2013
- 2013-11-18 CN CN201310582174.9A patent/CN103559507B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2006235752A (en) * | 2005-02-22 | 2006-09-07 | Asahi Engineering Kk | Object recognition device, its control method, computer program and computer readable storage medium |
| CN102542260A (en) * | 2011-12-30 | 2012-07-04 | 中南大学 | Method for recognizing road traffic sign for unmanned vehicle |
Non-Patent Citations (4)
| Title |
|---|
| "Support Vector Machines for Traffic Signs Recognition";Min Shi et al.;《2008 IEEE International Joint Conference on Neural Networks (IEEE World Congress on Computational Intelligence)》;20080608;第3820-3827页 * |
| "基于颜色与形状的实景交通标志分类研究";李辰 等;《中国科技论文在线》;20110413;第1页第1段-第6页第2段、表1 * |
| "基于颜色和形态学的交通标志检测技术的研究";赵俊梅 等;《车辆与动力技术》;20091231;第48-51、57页 * |
| "道路交通标志的检测与识别技术研究";刘思平;《中国优秀硕士学位论文全文数据库·工程科技Ⅱ辑》;20120115;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103559507A (en) | 2014-02-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103559507B (en) | The method for traffic sign detection being combined based on color with shape facility | |
| US10635929B2 (en) | Saliency-based method for extracting road target from night vision infrared image | |
| CN102314600B (en) | Shadow removal in image captured by vehicle-based camera for clear path detection | |
| CN106651872B (en) | Pavement crack identification method and system based on Prewitt operator | |
| Broggi et al. | Real time road signs recognition | |
| CN102682292B (en) | Method based on monocular vision for detecting and roughly positioning edge of road | |
| US10074020B2 (en) | Vehicular lane line data processing method, apparatus, storage medium, and device | |
| CN110197589B (en) | Deep learning-based red light violation detection method | |
| CN103020623B (en) | Method for traffic sign detection and road traffic sign detection equipment | |
| CN103824081B (en) | Method for detecting rapid robustness traffic signs on outdoor bad illumination condition | |
| CN102314601B (en) | Shadow removal in image captured by vehicle based camera using non-linear illumination-invariant kernel | |
| CN102314602B (en) | Shadow removal in image captured by vehicle-based camera using optimized oriented linear axis | |
| CN106845408A (en) | A kind of street refuse recognition methods under complex environment | |
| CN102938057B (en) | A kind of method for eliminating vehicle shadow and device | |
| CN101828201A (en) | Image processing device and method, and learning device, method, and program | |
| CN102968782A (en) | Automatic digging method for remarkable objects of color images | |
| CN103914699A (en) | Automatic lip gloss image enhancement method based on color space | |
| Le et al. | Real time traffic sign detection using color and shape-based features | |
| CN102819728A (en) | Traffic sign detection method based on classification template matching | |
| CN109949593A (en) | A kind of traffic lights recognition methods and system based on crossing priori knowledge | |
| CN105446333B (en) | Visual agv navigation system | |
| CN109446919A (en) | A kind of vision lane keeping method based on end-to-end study | |
| CN107679470A (en) | A kind of traffic mark board detection and recognition methods based on HDR technologies | |
| CN105678318A (en) | Traffic label matching method and apparatus | |
| Harbas et al. | Detection of roadside vegetation using features from the visible spectrum |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |