CN103559507A - Method for detecting traffic signs based on combination of color feature and shape feature - Google Patents

Abstract

The invention provides a method for detecting traffic signs based on the combination of a color feature and a shape feature. The method includes the steps that firstly, the HSV color feature is adopted, suspicious areas of the traffic signs in a whole image are extracted, then, the shape feature is adopted, the suspicious areas are divided again, and the areas of which both the color and the shape meet conditions are extracted while the other areas are filtered. The method is small in calculation, good in robustness and not sensitive to the change of an external environment, and the traffic signs in a complex background on the two sides of a road can be accurately detected in real time.

Description

The method for traffic sign detection combining with shape facility based on color

Technical field

The present invention relates to a kind of detection method of road signs, particularly relate to a kind of method for traffic sign detection based under complex background.

Background technology

Traffic Sign Recognition (Traffic Sign Recognition, TSR) system refer to people by advanced infotech, data transmission technology, control technology and computing machine treatment technology etc. effectively integrated use in whole transportation management system, make Ren,Che, road and environment close fit, thus set up a kind of on a large scale in, comprehensive play a role in real time, traffic information prompting system accurately and efficiently.It is an important component part of intelligent transportation system, owing to containing many important transport information in traffic sign, as to the variation of the speed prompt of current driving, road ahead situation, driving behavior restriction etc., therefore in this backup system, how fast, detect accurately and efficiently the traffic sign in road and it is fed back to human pilot or control system, for guaranteeing driving safety, what avoid traffic accident has a very important Research Significance, therefore, be subject to more and more experts and scholars' attention.

In more famous having aspect intelligent transportation system research: European Prometheus (Program for European Traffic with Highest Efficiency and Unprecedented Safety, PROMETHEUS) plan, intelligent vehicle roadnet (the Intelligent Vehicle and Highway Systems that the U.S. proposes, IVHS), and the advanced safety vehicle (Advanced Safety Vehicle, ASV) of Japan etc.At present, in the research of intelligent vehicle field of machine vision in the German UBM(Universitat der Bundeswehr Munchen of mainly containing of forward position) university, Italian professor Broggi leader's seminar, U.S. International Technology application company etc.Wherein, German UBM professor Dick-manns of university leader's intelligent vehicle research group is devoted to the research of dynamic field of machine vision always, and the EMS-Vision vision of development can be simulated human eye functions preferably.In more than 10 year of past, some country has successfully developed some road Identification and trackers based on vision.Wherein, representative system has: LOIS system, GOLD system, RALPH system, SCARF system and ALVINN system.

Before Google's two-and-a-half years, just released pilotless automobile project, wherein, also be unable to do without Traffic Sign Recognition System, as far back as 2011 August Google's pilotless automobile project leader Chris E Musen just announce the said firm more than ten pilotless automobiles under computer is controlled safety traffic 480,000 kilometers.Subsequently the automobile vendor such as BMW, Volvo, Audi also in succession by being equipped with intelligent software, the visual field is auxiliary and environmental monitoring system has been released unmanned concept car.Mainly be designed for city traffic driving, by the judgement of roadside traffic sign and other correlation technique, automobile can independently accelerate and brake.Domestic unmanned vehicle research and development aspect, since the mid-90 in last century, " autonomous driving technology " innovation team that He Hangen teaches and professor Dai Bin leads of the National University of Defense technology starts from scratch, rely on autonomous innovation to realize series of key techniques breakthrough, in calendar year 2001, succeed in developing first pilotless automobile of China, broken the external blockade to the unmanned technology of China.In addition, professor Deng Zhidong of the Xu Youchun of military traffic institute professor, Tsing-Hua University etc., also fruitful aspect unmanned vehicle research and development.More domestic other colleges and universities (as Beijing Institute of Technology, Tsing-Hua University, Xi'an Communications University, Wuhan University etc.), research institution and some Automobile Enterprises have also launched exploratory development and the preliminary experiment of some correlation techniques in succession.

TSR backup system is the multi-disciplinary cross-application such as computer vision, artificial intelligence, image processing, pattern-recognition, along with the development of video technique and computer technology, adopt the video detection identification of image processing method intelligent extraction traffic sign to become possibility.In TSR system, mainly comprise two basic links: be first the detection (cutting apart) of traffic sign, comprise the location of cutting apart of the pre-service of the image to collecting and traffic sign; Next is the identification (classification) of traffic sign, comprises feature extraction, signature analysis and the final identification etc. of traffic sign.And detect, be the key link of TSR system, the quality of its result directly affects the quality of recognition result.

Summary of the invention

Goal of the invention

In order better traffic sign to be identified, the present invention proposes a kind of method for traffic sign detection combining with shape facility based on color.

Technical scheme

The present invention implements by the following technical programs:

The method for traffic sign detection combining with shape facility based on color, is characterized in that: the method step is as follows:

(1), extract color characteristic

Adopt HSV colorimetry model, by H, S, V are composed to corresponding value, come entire image to extract suspicious object color.

(2), extract shape facility

After color is extracted, entire image is only left the region identical with target traffic sign color, now, by regional is carried out to the shape extracting based on region, by asking circularity, rectangular degree, range of extension to judge the shape of regional, and only leave and take circle, rectangle and delta-shaped region.

(3) morphology filters

Adopt two-dimensional Gaussian function as smoothing filter, to remove the noise at the boundary of image, make target image more level and smooth; By dilation and corrosion, reduce burr in image, fuzzy noise and remove the dark-coloured details less than structural element, keeping integral image gray-scale value and large dark areas substantially constant.

(4) suspicious object is extracted

The shape facility extracting, after morphology filters, is carried out and computing with original image, extract colored target image, i.e. Traffic Sign Images.

Advantage and effect

Because the single features of image is too unilateral unavoidably to the content description of image, only can be expressed the part attribute of image, for traffic sign real-time under state of nature, extract, the retrieval of single features is poor effect usually, and the retrieval combining based on CF feature, to the expression of picture material more comprehensively, thereby can maximize favourable factors and minimize unfavourable ones, obtain better retrieval effectiveness.Below with regard to independent application color, application of shape, other colors adopt with shape acting in conjunction method and this patent separately color shape, combine to extract traffic sign icon and contrast.

Accompanying drawing explanation

Fig. 1 is color-shape pairing model and the schematic diagram of just classifying that traffic sign detects;

Fig. 2 is for extracting 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 for adopting separately RGB to carry out blueness sign;

Fig. 6 extracts schematic diagram for adopting separately RGB to carry out blueness sign;

Fig. 7 is that color and shape acting in conjunction detect;

Fig. 8 extracts schematic diagram for adopting this patent method to carry out circle marker;

Fig. 9 is embodiment 1 original graph;

Figure 10 is embodiment 1HSV color model;

Figure 11 is the design sketch that 1 pair of entire image of embodiment extracts red area;

Figure 12 is embodiment 1 use morphologic filtering the design sketch that carries out area filling;

Figure 13 is that embodiment 1 extracts target shape figure;

Figure 14 is that embodiment 1 target is extracted net result figure.

Embodiment

Below in conjunction with accompanying drawing, the present invention is specifically described:

The invention provides a kind of method for traffic sign detection combining with shape facility based on color, the method step is as follows:

(1), extract color characteristic

Adopt HSV colorimetry model, by H, S, V are composed to corresponding value, come entire image to extract suspicious object color; For example, blue, red, yellow etc.

(2), extract shape facility

After color is extracted, entire image is only left the region identical with target traffic sign color, now, by regional is carried out to the shape extracting based on region, by asking circularity, rectangular degree, range of extension to judge the shape of regional, and only leave and take circle, rectangle and delta-shaped region.

(3) morphology filters

Adopt two-dimensional Gaussian function as smoothing filter, to remove the noise at the boundary of image, thereby make target image more level and smooth; By dilation and corrosion, reduce burr in image, the noise such as fuzzy and remove the dark-coloured details less than structural element, keeping integral image gray-scale value and large dark areas substantially constant.

(4) suspicious object is extracted

The shape facility extracting, after morphology filters, is carried out and computing with original image, thereby extracts colored target image, i.e. Traffic Sign Images.

Traffic sign is exactly a kind of very lively language concerning driver, and driver's driving procedure almost relies on vision system to process this traffic language information completely.Want be familiar with or unfamiliar environment in to accomplish to observe traffic rules and regulations, identifying accurately, timely traffic sign is exactly crucial prerequisite.

But often people's notice is not in complete set in driving procedure, is easily subject to the impact of external interference object (such as phone, large small advertisement, passerby, vehicle etc.), and has ignored traffic mark board.Therefore from middle 1980s, various traffic sign recognition methods are just constantly suggested, for reminding driver to reach the object of safe driving.But diversity and complicacy because scene is brought arbitrarily, also have the high request to aspects such as accuracy, real-time and robustnesss, caused up to the present going back the reasonable system of the real effect of neither one.Because research object is the traffic sign under unforeseen complex background, therefore the present invention illumination variation, color incompleteness, shooting angle, blurring or distortion and indicate block aspect all done comprehensive consideration, to reach to get rid of, disturb, automatically, find out timely traffic mark board, detect flag information, and promptly feed back to driver, farthest improve driving safety, the possibility that minimizing accident occurs, makes driving variable obtain safety and light more.

China directly traffic sign relevant with traffic safety is divided into 3 large classes, adds up to 131 kinds (not comprising the sign that can derive from).Wherein, 48 of prohibitory signs, 36 of Warning Marks, 47 of warning notices.Research object of the present invention is these 131 traffic signs.

1) Warning Mark: be used to refer to advancing of vehicle and pedestrian, take blueness as feature, be shaped as circle or rectangle.

2) warning notice: be used for warning vehicle and pedestrian danger place, take yellow as feature, except " emergency lane " is rectangle, shape is the equilateral triangle on angle.

3) prohibitory sign: be used for forbidding or limiting advancing of vehicle and pedestrian, take redness as feature except " ban releasing ", except " Stop and give way " and " giving way " shape is circle.

From analyzing above, traffic sign frame has indigo plant, black, red three kinds of colors, and shape has circle, triangle and rectangle, therefore, can detect according to the color of traffic sign frame and shape localization the regional location at traffic sign place in traffic image.So by entire image first being carried out to the extraction of yellow, redness, blueness and black part, the shape facility that extracts region by scanning is afterwards as triangle, circle and rectangle.Just can probably extract the suspicious region of traffic sign.Fig. 1 is color-shape pairing model and the just classification that traffic sign detects.

Because the single features of image is too unilateral unavoidably to the content description of image, only can be expressed the part attribute of image, different demands for different user, the retrieval of single features is poor effect usually, and the retrieval combining based on CF feature, to the expression of picture material more comprehensively, thereby can maximize favourable factors and minimize unfavourable ones, obtain better retrieval effectiveness.

(1) extract color characteristic

At the color space mainly adopting aspect color extraction, there are RGB, HSI, LAB, HSV etc.Wherein, RGB color space is most basic, most widely used, and other nearly all space can be changed by rgb space.But the information of its response diagram picture is directly perceived not, do not meet the perception of human vision, and in RGB color space, three variablees have certain correlativity, make image ratio be easier to be subject to the impact of illumination.In HSI model, although H, S, tri-component relevances of I are very little, comparatively independent, but when saturation degree and brightness can cause the unstable of tone when lower, cannot be for the detection of traffic sign, only possess certain saturation degree and brightness value and just can obtain stable tone, therefore, inapplicable real-time color detection.LAB color space is by special formulation, to be measured a kind of color mode of International Commission on Illumination's formulation of color standard.Can directly with the geometric distance of color space, carry out the comparison of different colours, therefore can be effectively for measuring little aberration, but because it is nonlinear transformation, so calculated amount is larger, and there is singular point in color space.

Hsv color space is the reduced form in Munsell colour space, is to take the tone (H), saturation degree (S), brightness (V) of color as three elements represent, and be that non-linear color table shows system.Wherein, tone is the attribute of describing pure color, and saturation degree is to describe the tolerance of the degree that pure color diluted by white light, and brightness is a subjective descriptor, has embodied colourless brightness concept, is the key parameter of describing color perception.Hsv color space is consistent to the perception of color with people, and in HSV space, people is more even to the perception of aberration, is the color space that is applicable to human vision property.Below we from vision consistance, integrality, compactedness, several aspects such as naturality, compare above-mentioned color space, as shown in table 1.

The comparison of the several chrominance spaces of table 1

In hsv color space, be conducive to the processing of image as can be seen from the above table, such as rim detection, image, cut apart and target identification etc.Therefore the color that, this patent adopts hsv color space to carry out the first step to entire image is extracted.

(2) extract shape facility

Shape facility is one of core feature of image, and the shape information of image does not change with the variation of color of image, is the invariant feature of object.Picture shape feature can be distinguished the object of identical category, and color and texture usually do not have this feature.And very directly perceived by shape facility differentiation of objects.Therefore, utilize shape facility retrieving images can improve the accuracy and efficiency of retrieval.

Type based on shape image retrieval is divided into two kinds, and a kind of is characteristic key based on image-region, and another kind is the retrieval based on edge feature.Shape description based on edge is that it does not consider the information of shape inside to surrounding the description of the profile of target area.This class description generally has continuous type and two kinds of forms of discrete type.The describing method of continuous type is also referred to as overall type, and it is the integral body extraction eigenvector from profile; The describing method of discrete type is also referred to as structural type, and it is to extract corresponding feature after profile is divided into a lot of fragments.In the image retrieval based on edge feature, the method for the Boundary Extraction operator that we often use mainly contains: Canny operator, Laplacian operator, Sobel operator and Roberts operator etc.Describing method based on region is to regard shape area as an integral body, utilizes all pixels in region, is subject to the impact of noise and change of shape relatively little, and this method for expressing is divided into two types, overall type and local type.The provincial characteristics of shape mainly contain region area, Euler's numbers, dispersion, excentricity, area invariant moment, region skeleton, gather the methods such as not bending moment, Zernike square, angular radius conversion.

Retive boundary is extracted, and extracted region accuracy is higher, and because the color of previous step is extracted, has made entire image binaryzation, there is no complex background, therefore based on extracted region algorithm, can realize the extraction in suspicious object region completely.In sum, this patent adopts overall type region description to extract suspicious object.

(3) morphology filters

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.In the middle of the various noises of image, Gaussian noise is maximum, for filter Normal Distribution in image edge noise we generally by gaussian filtering method, undertaken.Therefore, here we to process image be Bian with having two-dimentional Gaussian function, as smoothing filter, carries out.The expression formula of the Gaussian function of two dimension is as follows:

$G(x,y)=\frac{1}{{2\mathrm{\Π\σ}}^2}\exp (-\frac{x^2+y^2}{{2\mathrm{\σ}}^2})$

This patent utilizes gaussian filtering method the noise at the boundary of image can be eliminated, thereby makes target image more level and smooth, and these pretreatment operation play a good role for the accurate extraction of guaranteeing picture shape feature.

2) dilation and erosion

Mathematical morphology can extract for expressing and the useful picture content of description region shape from image, and in bianry image, the set of all black picture elements is that the morphology that image is complete is described.Morphologic fundamental operation has dilation and erosion.First corroding expands is afterwards called out operation, and the post-etching that first expands is called closed operation.Open operation and can make target area profile become smooth, disconnect narrow interruption and eliminate thin protrusion, meanwhile, can obviously not change again the area of target area.Be defined as:

AοB＝(AΘB)⊕B

Closed operation can diminish narrow interruption and long thin wide gap, eliminate little hole, and fill up the fracture in outline line, the effect on level and smooth its border in the situation that of not obvious change area.Be defined as:

A·B＝(A⊕B)ΘB

Through opening operation and closed operation, point isolated in energy removal of images disturbs with some the fine grained chippings forming, and also can eliminate object boundary point simultaneously, also can make the objective area in image area after corroding compensate to some extent.Then use 3 * 3 rectangular window to carry out medium filtering to image, can play further removal of images noise, spot, further the effect of segmentation object and background.

(4) suspicious object is extracted

After morphology filters, the contour of object in image is just more clear, first needs to extract x, the y direction coordinate starting point of suspicious region, as shown in Figure 2, obtains the zone length of direction, and the region that area is less than certain setting threshold is removed.

Difform traffic sign has different attributes, as barycenter Edge Distance figure, circularity, rectangular degree, elongation etc.As shown in Figure 3, take scheme in dotted arrow indication place be starting point, be rotated counterclockwise one week, ask successively the distance between barycenter and each marginal point, in figure, the second row is the barycenter Edge Distance figure of 3 kinds of difformity traffic signs, wherein horizontal ordinate represents the pixel number that traffic sign edge image comprises, and ordinate represents that barycenter is to the distance value of edge pixel point.

Comprehensive circle, triangle and rectangle feature separately, this patent is chosen circularity, rectangular degree and elongation as the major parameter of Shape Feature Extraction.Wherein, circularity mainly refers to that object approaches the degree of theoretical circle; Rectangular degree has reflected the object degree similar to rectangle; That refers to material because the effect of external force has extended degree after producing distortion to elongation.By obtaining barycenter, to minimum, maximum and the mean value of Edge Distance, obtain circularity, rectangular degree and the elongation in region.

Concrete 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-region, S-region area, W-x direction zone length, H-y direction zone length.

Scholar Miguel A.Garc ' 1a-Garrido is in the paper detecting about traffic sign of delivering for 2012, the same mode that adopts color to combine with shape facility, but it is in shape extracting part, what adopt is that discrete candy operator carries out Boundary Extraction, can be because breakpoint causes recognition failures in leaching process.Example is as shown below.Originally be two adjacent circular traffic signs, due to the breakpoint of Discrete Operator, caused two circle markers by flase drop, to be become the sign of a connection.

From experimental result, can find out, apply merely the effect that color or shape extract all undesirable, while utilizing color, especially in extracting blue sign, because most car plates be also blueness, thus be easy to car plate to miss extraction, thus experimental result affected.And when application of shape feature is extracted, due to the complex background under physical environment, make its extraction effect poorer.To sum up, the traffic sign based on CF acting in conjunction combination detects, and to the expression of picture material more comprehensively, thereby can maximize favourable factors and minimize unfavourable ones, and obtains better retrieval effectiveness.Table 2 by reference recall ratio and precision ratio contrasts three kinds of methods.

Wherein, the sample image of definition datum inquiry is set as N, and the image set of retrieving images is set as M, can obtain:

The number of the associated picture retrieving is:

The number of flase drop is: $B_k=\underset{m=0}{\overset{k-1}{\mathrm{\Σ}}}(1-V_m),$

Undetected number is: $C_k=\underset{m=0}{\overset{M-1}{\mathrm{\Σ}}}{V}_m-A_k,$

The number of nd irrelevant image is:

Here, V _mfor the correlation of benchmark retrieving images and the ordinal number image that is m, V _m∈ (0, l); K is cutoff.The criterion of these two retrieval performance assessments of precision ratio and recall ratio, we can be worth by several above.

(1) precision ratio: precision ratio is commonly used to the ability that evaluate image searching system is rejected irrelevant image, what it obtained is the shared ratio of image of correctly searching in the middle of retrieval set.

Precision ratio:

(2) recall ratio: recall ratio is used for measuring the ability that image indexing system retrieves associated picture, what it obtained is the ratio of the quantity of all similar images in the shared image data base of effective image.

Recall ratio:

The contrast of three kinds of algorithm recall ratios of table 2 and precision ratio

In sum, no matter the traffic sign detection algorithm combining with shape facility based on color that this patent proposes, be at recall ratio or aspect precision ratio, is all better than adopting separately the detection algorithm of colour or shape.With with comparing with the method for shape cooperation by color, the inventive method more can highlight the traffic sign that will extract, and it is little to extract error.And owing to first adopting color to filter for the first time entire image, this Shape Feature Extraction that is next step has been got rid of a large amount of complex backgrounds and interference, greatly save the time of Shape Feature Extraction and improved its Detection accuracy, so the inventive method has good robustness, real-time, is applicable to real world applications.

Below in conjunction with accompanying drawing and concrete embodiment, the present invention is described further,

Embodiment 1:

First from vehicle-mounted vidicon, obtain original image, as shown in Figure 9.

(1) color is extracted

In HSV six water chestnuts cones colour models, form and aspect (H) are in being parallel in the look plane of six water chestnut vertex of a cone faces, and they are around central shaft V rotation and change, red, yellow, and green, green grass or young crops, indigo plant, pinkish red six standard colorss 60 degree of being separated by respectively.Along six water chestnut cone central shaft V, change from top to bottom, central shaft top is white in color (V=1), and bottom is black (V=0), and they represent the greyscale color of netrual colour system.Color saturation (S) along continuous straight runs changes, the color that more approaches six water chestnut cone central shafts, its saturation degree is lower, and the RC color saturation of hexagon is zero (S=0), coincide with the V=1 of highest lightness, the color of high saturation is (S=1) on the edge line in hexagon housing.

The present invention adopts yellow, blueness and the red part in hsv color model extraction entire image at color Extraction parts.Wherein,

Every assignment that blue region detects is:

H(i,j)>=240&&H(i,j)<=255))||((H(i,j)>=0&&H(i,j)<=10)；

Every assignment that red area detects is:

(H(i,j)>=240&&H(i,j)<=255))||((H(i,j)>=0&&H(i,j)<=10)；

Every assignment that yellow area detects is:

H(i,j)>=18&&H(i,j)<=45。

Take redness as example, and the design sketch that adopts this patent method to extract is illustrated in fig. 11 shown below.

(2) morphologic filtering

Complicacy due to street background, color has noise after extracting unavoidably, and the zonule identical with target sign color of mistake extraction, for this situation, first, can compensate inhomogeneous background luminance by modified opening operator, choose suitable structural element image is carried out to opening operation, produce the estimation to whole image background.Modified opening operator is designated as I ° of b, and structural element b (i', j') is defined as I ° of b=(I Θ b) ⊕ b to the opening operation of image I (i, j);

In formula: I is the image after color is extracted; Symbol " Θ " and " ⊕ " are respectively the dilation and erosion computing of structural element b (i ', j ') to image I (i, j), and formula is as follows:

(I⊕b)(i,j)＝max{I(i-i',j-j')+b(i',j')/(i',j')∈D _b}

(IΘb)(i,j)＝min{I(i+i',j+j')-b(i',j')/(i',j')∈D _b}

In formula: D _bit is the field of definition of structural element b (i ', j ').

Opening operation can be removed the bright details less than structural element, keep integral image gray-scale value and large bright area substantially constant, by opening operation, can obtain uniform background estimating, the background that deducts estimation from original image so can generate the image that a width has homogeneous background, eliminate the impact of noise and interference, this process is top cap (top-hat) conversion in morphology, and formula is as follows:

g＝I-(Iοb)

In formula: g is output image.

After converting by top-hat, the dash area of image can be substantially capped, and combine together with former having powerful connections, thereby target area can be extracted at partitioning portion, there will not be shade is used as to Target Segmentation phenomenon out.On the other hand, the phenomenon of jolting, shaking due to vehicle-mounted camera, cause the general contrast of image not high, noise pollution is serious, edge fog, therefore the present invention is after carrying out top cap computing removal shade, and sonar image enhancing is carried out in the computing that execution cap-end, top cap (bottom-hat) combines, and end cap operation definition is:

In formula, Ib is closed operation, structural element b (i ', j ') as follows to the closed operation formula of image I (i, j):

I·b＝(I⊕b)Θb

Closed operation can be removed the dark-coloured details less than structural element, keeps integral image gray-scale value and large dark areas substantially constant.Completed thus morphology preprocessing part, as shown in figure 12.

(3) shape extracting

After morphologic filtering, removed Noise and Interference unnecessary in image, the target that now just can carry out based on shape is extracted, pass through test of many times, author finds when circularity, rectangular degree and elongation are chosen following scope, the best results that extract target area, as shown in figure 13.

Circle marker: C>0.85, R>0.70, E>0.85;

Triangle sign: 0.35<C<0.70,0.4<R<0.65, E>0.8;

Rectangle sign: 0.60<C<0.85, R>0.70, E>0.85.

(4) target is extracted

By the target shape extracting and original image is done and computing, can obtain final target recognition result.Target is extracted net result as shown in figure 14.

(5) method accuracy rate

In this experiment, adopted altogether each 40 of four kinds of representative type of signs, adopt the detection case of the inventive method traffic sign as shown in table 3 below:

Table 3 all kinds traffic sign detection case

。

Claims (1)

1. the method for traffic sign detection combining with shape facility based on color, is characterized in that: the method step is as follows:

(1), extract color characteristic

Adopt HSV colorimetry model, by H, S, V are composed to corresponding value, come entire image to extract suspicious object color;

(2), extract shape facility

After color is extracted, entire image is only left the region identical with target traffic sign color, now, by regional is carried out to the shape extracting based on region, by asking circularity, rectangular degree, range of extension to judge the shape of regional, and only leave and take circle, rectangle and delta-shaped region;

(3), morphology filters

Adopt two-dimensional Gaussian function as smoothing filter, to remove the noise at the boundary of image, make target image more level and smooth; By dilation and corrosion, reduce burr in image, fuzzy noise and remove the dark-coloured details less than structural element, keeping integral image gray-scale value and large dark areas substantially constant;

(4), suspicious object is extracted

The shape facility extracting, after morphology filters, is carried out and computing with original image, extract colored target image, i.e. Traffic Sign Images.

Images