CN104036262B - A kind of method and system of LPR car plates screening identification - Google Patents
A kind of method and system of LPR car plates screening identification Download PDFInfo
- Publication number
- CN104036262B CN104036262B CN201410306452.2A CN201410306452A CN104036262B CN 104036262 B CN104036262 B CN 104036262B CN 201410306452 A CN201410306452 A CN 201410306452A CN 104036262 B CN104036262 B CN 104036262B
- Authority
- CN
- China
- Prior art keywords
- car plate
- car
- region
- plate
- gray
- 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.)
- Expired - Fee Related
Links
- 238000000034 method Methods 0.000 title claims abstract description 27
- 238000012216 screening Methods 0.000 title claims abstract description 18
- 238000000605 extraction Methods 0.000 claims abstract description 8
- 238000012545 processing Methods 0.000 claims description 32
- 238000006243 chemical reaction Methods 0.000 claims description 29
- 238000010586 diagram Methods 0.000 claims description 16
- 238000004422 calculation algorithm Methods 0.000 claims description 11
- 239000006002 Pepper Substances 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000005530 etching Methods 0.000 claims description 3
- 238000001914 filtration Methods 0.000 claims description 3
- 230000015572 biosynthetic process Effects 0.000 claims description 2
- 230000006698 induction Effects 0.000 abstract description 5
- 238000001514 detection method Methods 0.000 description 12
- 230000011218 segmentation Effects 0.000 description 8
- 238000009826 distribution Methods 0.000 description 7
- 229910052704 radon Inorganic materials 0.000 description 7
- SYUHGPGVQRZVTB-UHFFFAOYSA-N radon atom Chemical group [Rn] SYUHGPGVQRZVTB-UHFFFAOYSA-N 0.000 description 7
- 230000000694 effects Effects 0.000 description 6
- 230000004927 fusion Effects 0.000 description 6
- 230000010339 dilation Effects 0.000 description 5
- 230000003628 erosive effect Effects 0.000 description 5
- 238000007726 management method Methods 0.000 description 5
- 230000008859 change Effects 0.000 description 4
- 238000005520 cutting process Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 230000003044 adaptive effect Effects 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 3
- 238000010276 construction Methods 0.000 description 3
- 238000012937 correction Methods 0.000 description 3
- 230000009466 transformation Effects 0.000 description 3
- 241001062009 Indigofera Species 0.000 description 2
- 238000002679 ablation Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 150000003839 salts Chemical class 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 description 1
- 235000007926 Craterellus fallax Nutrition 0.000 description 1
- 240000007175 Datura inoxia Species 0.000 description 1
- 238000009825 accumulation Methods 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 239000003513 alkali Substances 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 238000000205 computational method Methods 0.000 description 1
- 230000007797 corrosion Effects 0.000 description 1
- 238000005260 corrosion Methods 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000001066 destructive effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 239000000945 filler Substances 0.000 description 1
- 238000007667 floating Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 230000014759 maintenance of location Effects 0.000 description 1
- 238000002156 mixing Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 238000013316 zoning Methods 0.000 description 1
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
- Traffic Control Systems (AREA)
Abstract
The step of the step of being gathered the present invention relates to a kind of LPR car plates screening knowledge method for distinguishing, including optimal car plate and License Plate.The step of the step of the step of described optimal car plate is gathered is filtered including car plate and extraction car plate frame number；The step of car plate is filtered：By defining for ROI, car is filtered out from camera farther out or nearer situation；The step of extracting car plate frame number：By candid photograph counter controls, N images are continuously captured to the car plate for entering identification region, stop to capture more than N；The frame class car plate of previous entrance identification region and the frame class car plate of next entrance identification region captured interval more than M seconds, captured counter O reset.May not only be applied to parking management system can also realize the intelligent management of company car, complete the function of automatic work attendance.And the acquisition of car plate does not recycle the mode of embedded ground induction coil, lifetime of system is improved.
Description
Technical field
The present invention relates to traffic surveillance and control system technical field, and in particular to the system that a kind of screening recognizes car plate.
Background technology
With economic growth and the continuous improvement of people's purchasing power, domestic automobile quantity increases on a large scale, especially private
The sharp increase of family's car quantity, the parking stall being equipped with originally considerably beyond city, the problems such as causing domestic parking difficulty increasingly highlights,
How to be realized in the case where parking stall resource is certain turns into of concern heavy to parking lot science, specification, unified management
Point.
Car license recognition is as the important component in modern intelligent traffic monitoring technical field, in terms of parking lot management
Play important role.In addition, the intelligent management of company car can also be realized to the application that Intelligent license-plate of vehicle is recognized, completes
The function of automatic work attendance.
Car license recognition is that based on the technologies such as computer vision, image procossing, pattern-recognition, front-end camera is clapped
The vehicle image or video sequence taken the photograph are analyzed, and obtain each unique number-plate number of automobile, so as to complete car plate knowledge
Other process.It is well known that the identification process of car plate is roughly divided into three steps, i.e.,：The positioning of car plate, Character segmentation and word
The identification of symbol.
At present, the traditional means of vehicle detection are that ground induction coil is buried under road surface, when vehicle passes through ground induction coil, ground sense
The inductance value of coil changes, and causes present coil and the electric signal of the output of the detection circuit of other circuits to change,
Generation vehicle detection signal, generally frequency signal, send processor to be handled, obtain vehicle into and out of the information in parking lot.
In practical application, using ground induction coil detect vehicle method construction maintenance is costly, road pavement is destructive big, freezed, salt
Alkali, heavy traffic influence it is big, service life is short, general only 2 years.
The content of the invention
The defect of the prior art for more than, LPR identifying systems of the invention may not only be applied to parking management system also
The intelligent management of company car can be achieved, the function of automatic work attendance is completed.And the acquisition of car plate does not recycle embedded ground induction coil
Mode, improve lifetime of system.
The identification substantially flow of system car plate is as shown in figure 1, present invention focuses on the screening and identification to car plate, specifically
Technical scheme is as follows：
Method for distinguishing is known in a kind of LPR car plates screening, comprises the following steps：
The step of optimal car plate is gathered：
(1) the step of car plate is filtered：
By defining for identification region (being referred to as region of interest ROI here), can filter out car from camera farther out or
The nearer situation of person.
(2) the step of extracting car plate frame number：Vehicle License Plate Recognition System is provided with counter, and the car plate for entering identification region is connected
It is continuous to capture N (generally 3) images, stop to capture more than N；The frame class car plate of previous entrance identification region enters with next
The frame class car plate of identification region captures interval more than M seconds (being usually 1 second), captures counter O reset.
The step of License Plate：
(1) the step of gray processing processing color diagram：Utilize formula
Gray (i, j)=R (i, j) * 0.299+G (i, j) * 0.587+B (i, j) * 0.114
Gray processing processing is carried out to the color diagram collected.
(2) top-hat conversion is carried out to gray-scale map, can effectively removes highlight area；It is preferred that, in top-hat conversion
Kernel values are 20*1；
(3) continue to carry out gray-scale map binary conversion treatment formation binary map, the selection of its threshold value is preferably to utilize Da-Jin algorithm
Ask for；
(4) continue that arithmetic operation is opened and closed to binary map, can rapidly eliminate the salt-pepper noise in image, preferably just
Method is as follows：
Step 1：Closed operation operation, salt-pepper noise area reduction are carried out to image；The process of execution is over an input image
Carry out first expanding the operation of post-etching with a suitable structural element B.The step can in filler body exiguous space, connection it is adjacent
Nearly object, its smooth border when substantially not changing object area.For structural element B selection, need according to camera acquisition
Image in depending on car plate size, take the effects of 24x 2 best through experiment test B.
Step 2：Opening operation operation is carried out to image, non-salt-pepper noise region can be fused；Opening operation can be eliminated in image
Small objects, at very thin place's separating objects, the border of smooth larger object substantially do not change its area.Through experiment test knot
Constitutive element B takes 2x8 optimal.
Step 3：Etching operation is carried out to image, salt-pepper noise region can reduce again；Structural element takes 24x2.
Step 4：All regions are traveled through, (region that area is less than 40 is defined as the region removal by region area less than 40
Salt-pepper noise region)；Specific practice is all connected domains in search bianry image, is known using priori such as car plate area and length-width ratios
Know area is smaller and the ungratified region of length-width ratio is filtered out, it is appropriate that area threshold is typically set to 40.
(5) it is preferred, in addition to region is the step of merge：
Step 1：All profile boundary rectangles are put into set A in binary map；
Step 2：The color diagram of all binary maps in set A is found, set A is traveled through, if color of object (indigo plant, Huang etc.) ratio
More than setting value, it is optimal that setting value here, which takes 0.02, then is put into candidate rectangle set B；
Step 3：The geometric center of all rectangles in set of computations B, judges the position relationship of rectangle in set B, if two
The distance of individual square boundary is in the range of restriction, here, and setting range takes 10-15 to be optimal, and two rectangular centre point lines
Angle between horizontal line or vertical curve is in given threshold, then it represents that two rectangles in same level or vertical direction, this
When, the setting-out in two rectangles merges it, here given threshold take -10 degree to+10 spend between be optimal.
(6) it is preferred, in addition to the step of search rectangular region：
Step 1：The corresponding rectangle of the binary map merged through region is put into set S；
Step 2：Area is weeded out less than setting value or Aspect Ratio not in the rectangle of setting range, will be grasped through above-mentioned rejecting
Remaining car plate is included into car plate Candidate Set after work.
(7) the step of screening car plate：
Step 5.1：Calculate the ratio of color of object (indigo plant, Huang etc.) in candidate license plate region；
Step 5.2：The maximum region of color of object ratio is found, if the maximum of color of object ratio is more than setting value
0.03 carries out step 5.3 (through statistics, setting value 0.03 is optimal), is otherwise directly abandoned；
Step 5.3：Find the binaryzation artwork master of target license plate；
Step 5.4：1/2 height in region is taken, black and white saltus step scanning is carried out from left to right, if the number of times of black and white saltus step does not surpass
Setting value (through statistics, setting value is optimal when being 14) is crossed then to abandon.
The invention further relates to a kind of system of LPR car plates screening identification, including optimal car plate acquisition module and License Plate
Module；
Described optimal car plate acquisition module is screened and known for obtaining dream car bridge queen feeding License Plate module
Not.
Described optimal car plate acquisition module includes car plate filter element and car plate frame number extraction unit, described car plate mistake
Filtering unit is used for license plate image of the collecting vehicle from camera suitable distance, and described car plate frame number extraction unit is used to capture and mistake
The pseudo- car plate of filter.
Described License Plate module include gray processing processing unit, top-hat converter units, binary conversion treatment unit,
Opening and closing operation unit, car plate screening unit, described gray processing processing unit are used to carry out gray processing to the color diagram collected
Processing forms and top-hat converter units is sent into after gray-scale map, and described top-hat converter units are used to image is carried out to remove bloom
Binary conversion treatment unit is sent into after regional processing, the binary conversion treatment unit is used to send after carrying out binary conversion treatment to gray-scale map
Enter car plate screening unit, described car plate screening unit is used to screen out non-license plate image.
Brief description of the drawings
Fig. 1 is the substantially schematic flow sheet of Vehicle License Plate Recognition System；
Fig. 2 is flow chart of the method for the present invention；
Fig. 3 screens the method flow diagram of car plate for the present invention；
Process and design sketch that Fig. 4 merges for region in the present invention；
Fig. 5 is that car plate rotates schematic diagram；
Fig. 6 is the car plate schematic diagram of more standard；
Fig. 7 is webbing frame car plate schematic diagram；
Fig. 8 is that left frame accurately cuts schematic diagram,
Fig. 9 is the license plate sloped angle schematic diagram of acquisition.
Embodiment
The invention will be further described below in conjunction with the accompanying drawings, a kind of LPR licence plate recognition methods：
1. obtain preferable car plate
Because system is that image is directly directly extracted from the video flowing of camera, vehicle enters state during identification region
With continuity, randomness.Car license recognition based on video flowing must just solve how could to extract from continuous two field picture
Optimal car plate.So-called optimal car plate refers to the factors such as size, position, the inclined degree of car plate to system current algorithm adaptability
It is best.Here the extraction of optimal car plate mainly passes through the number of same car image of the setting to identification region and acquisition.
1.1ROI setting
By defining for identification region (being referred to as region of interest ROI here), can filter out car from camera farther out or
The nearer situation of person.When vehicle is away from camera, the license plate area of extraction is too small, lack of resolution；When vehicle distances shooting
When close, the fracture of license plate area can be caused again, and therefore the selected of ROI is the most important condition for filtering non-ideal car plate.ROI's
It is selected the parameter used to be determined according to the distribution of the position of specific camera, parameter and vehicle optimum position.Here
ROI region is set as：
ROI=RECT (0, height/5, width, height/2)
Wherein width, height are respectively the wide and height of input picture.Setting ROI can also reduce a large amount of unnecessary
Computing, it is known that the requirement of real-time of Vehicle License Plate Recognition System is very high, if our time exceedings for being spent in identification I
Patient limit, then even if accuracy again good system also without practical value.
In practice, the distance that camera is shot is car plate in 10m, the photo shot during vehicle distances camera 3m-8m
Size, definition is relatively good, therefore ROI=(0, height/5, width, height/2).
1.2 extract car plate frame number
Because the number of times that same car is captured when entering identification region is a lot, therefore we need that limitation is captured
Number, it is sufficient that in general same car captures 3 different at the time of.It is assumed here that for N need to be captured, then being grabbed at this
Is identified in the N of bat images as one group, last recognition result is judged that (result of judge can be with by N number of result jointly
It is：Character is defined less than 7 by 7 in same group, all for 7 by being defined for capturing first).Extracting the frame number specified needs
A candid photograph counter is set in a program, complete one per treatment then Jia 1, stop to capture when condition is unsatisfactory for.Need
Noticing is, in the case of ideal.But in actual deployment, especially rainy day, ground rain return light
Spot is more to form pseudo- license plate area, and result is vehicle leakage knowledge caused by being now easily caused counter failure, this situation.(i.e.
If cause counters error to exceed the candid photograph number set by rainwater, if be not cleared now, then treat next
Car will directly be ignored when coming in causes leakage to be known).Here solution is to set a timer, in actual positioning
In, car plate position generally is continuous, and in the case where frame period is not more than 1s, counter will be incremented by from 1 in a short time
To N.Mostly passed by and caused by the pedestrian of the blue clothes of dress due to pseudo- license plate area in the rainy day, thus the interval counted be with
Machine, and the time in counting interval must be longer than time under normal conditions.Therefore way here is：
Step1. capture former frame class car plate and obtain current time pre_time=GetLocalTime ()；Step2. capture
Next frame class car plate obtains current time curr_time=GetLocalTime ()；
If-pre_time. time difference DiffTime=curr_time, DiffTime time difference then captured twice number
Value is more than 3s and so can be assumed that this candid photograph twice is not same vehicle, or now counter works are due to pseudo- car plate
Caused by region, then can will now capture counter O reset processing.
2nd, License Plate
The positioning of car plate mainly includes two steps, one is the coarse positioning of car plate, the second is finely positioning.It is fixed in car plate
Coarse positioning is different with the target that finely positioning is handled among position.In coarse positioning, main target is in whole input picture
The correct approximate location for searching out car plate, can also can only include car plate comprising car plate.Car plate physical location should not herein
Ask very accurate, but car plate must be included in the region of positioning.And the target of finely positioning is the car plate obtained in coarse positioning enters
Drive a vehicle the standardization (including rotation correction, lower edges, the cutting of left and right edges of car plate etc.) of board, last car plate should be wrapped only
Background containing car plate and character.
2.1 coarse positioning (LPRCoarseLocate)
The groundwork of coarse positioning includes carrying out necessary pretreatment to image, and such as gray processing, binaryzation, filter are made an uproar, then
Person performs mathematical morphology operation, carries out dilation erosion to binary map, is that follow-up search rectangular region is prepared.Main stream
Cheng Wei：
2.1.1 gray processing (cvCvtColor)
Camera is when image is gathered, and the data gathered are generally based on the 3- passages of RGB color
(channels) coloured image, it is necessary to which the operation carried out is often that need not locate under color diagram in the coarse positioning of car plate
Reason.Amount of calculation needed for processing coloured image be in very huge (be single pass 3 times), coarse positioning using color diagram not
Surging for amount of calculation is will result only in, but also the precision of positioning can be influenceed.Herein, it would be desirable to which color diagram is subjected to gray processing
Processing.The formula for turning gray-scale map based on RGB color is as follows：
Gray (i, j)=R (i, j) * 0.299+G (i, j) * 0.587+B (i, j) * 0.114
Wherein Gray (i, j) is gray value of the target image on point (i, j), and R, G, B are three components of image.
2.1.2.Top-hat
Top-hat conversion is also referred to as top cap conversion, and Top-hat conversion can effectively remove highlight area.Vehicle is in daylight
Lower or night, due to all can illumination interference, too bright region can not only influence binaryzation effect, but also disturb car plate area
Domain.The influence for the removal bloom that Top-hat conversion can be effectively is carried out to gray level image.It should be noted that Top-hat is converted
Core size selection, with reference to the characteristics of car plate and reduce in general kernel=20x 1 taken to the damage of license plate area
It is more suitable.
Top-hat conversion principle be：
The step of computing, is subtracted to carry out opening operation to original image first using artwork.
2.1.3 binaryzation (cvThreshold)
After the processing for performing previous step, the dimension of gray space is between 0-255, but this becomes alternatively to we are such
Still be not the result that we want afterwards, it is opposite we what is desired is that the only figure (generally black, white dichromatism) of dichromatism, target is with white
Color is filled, and background is the binary map of black.Binaryzation is the important step for carrying out image segmentation, and principle is as follows：
The main difficult point of binaryzation is the selection of threshold value.Under general scenario, there is the selection of two kinds of threshold values, one is based on complete
The fixed threshold of office, two be Adaptive Thresholding.The former computational methods are the average value for obtaining gray-scale map, then in average value
Basis on fluctuate, depending on the effect of specific floating numerical value with reference to actual environment and binaryzation.Adaptive algorithm
Threshold value is asked using Da-Jin algorithm (OTSU).OTSU algorithms can be described as the simple high efficiency method of adaptive polo placement single threshold.Specifically such as
Under：
If gray level image gray level is L, then tonal range is [0, L-1], and the optimal threshold of image is calculated using OTSU algorithms
It is worth and is:
T=Max [w0 (t) * (u0 (t)-u) ^2+w1 (t) * (u1 (t)-u) ^2] variable declaration therein：When the threshold of segmentation
When being worth for t, w0 is background ratio, and u0 is background mean value, and w1 is prospect ratio, and u1 is prospect average, and u is equal for entire image
Value.
Make above transition formula evaluation maximum t, as segmentation figure picture optimal threshold.
The above method has been realized in OpenCv storehouses, only needs incoming CV_THRESH_OTSU parameters.
2.1.4 dilation erosion (cvMorphologyEx)
It is fine using the effect being used in for the use of filter is made an uproar of mathematical morphology in the picture.If there are many spiced salt in image
Noise, then corroded when with suitable structural element, can quickly eliminate this noise.And expansive working is in region
Syncretizing effect is notable.Typically the image containing car plate is handled, the approximate location of car plate is searched out, protrusion is also the most the most
Effective way is exactly the operation using dilation erosion.In senior image procossing, opening operation and closed operation can be promoted to.
Its essence of both computings is also the combination of two kinds of operations of dilation erosion, only in the function that Opencv is provided, two kinds of fortune
After passing through optimization at last.The main step of coarse positioning car plate is：
1) performs 24x 2 closed operation CV_MOP_CLOSE
2) performs 2x 8 opening operation CV_MOP_OPEN
3) performs 24x 2 closed operation
Present invention optimizes dilation erosion operation, closed operation is first carried out before opening operation, so, can during progress opening operation
To reduce the mutual fusion of salt-pepper noise.
2.1.5 region fusion (MergeProbability)
The fusion in region is not essential, and the proposition of the step is the car plate for having particularity for those.In some feelings
Under condition, due to morphology kernel limitation, and license plate area it is larger when, it may appear that the crack conditions between character.
Whether blending algorithm is contemplated that the contact between two regions of situation judgement first, i.e., in same level or same
On vertical direction, next to that to consider the feature of car plate in itself.Distribution of color, the Texture features priori of such as car plate.Together
When, in order to avoid the mistake fusion in region, it should also do the limitation on some region distances.Such as given threshold, distance is in the model
Condition is met within enclosing.Here the algorithm steps of region fusion are provided：
Step 1. obtains all profile boundary rectangles and is put into set A.
Step 2. travels through A, and investigates the distribution of rectangle color, if color of object (blue, yellow) ratio is larger, is put into time
Select in rectangular set B.
Step 3. investigates rectangle position relation in set B, if in same level (same vertical) direction, and meet limit
It is fixed apart from when, the setting-out in two regions merges it.For region fusion, main difficult point is how reasonably to sentence
Determine the correctness in region.Because the region acquired by us might not only include license plate area, it is also possible to viscous with noise
Connect together, can so cause region is excessive to cross broadband situation, regional location becomes less obvious.Here method acquired by algorithm
It is, the geometric center of zoning that the position in region is judged using geometric center.Reliability will more high point.
2.1.6 search rectangular region
In China, the car plate of standard is (440mm x 140mm) with strict size and ratio, therefore rectangle conduct
One important priori of car plate plays an important role in the positioning of car plate.It is morphologic swollen to being carried out comprising license plate image
After swollen corrosion, what is left in image is the different block of many size shapes (block), now needs the institute to whole image
There is block to scan for being put into a set, it is assumed here that to be S.We need further not meeting car to those in S
The block of board feature is weeded out.The condition of screening is：
1. area is too small, and such as less than 100；
2. Aspect Ratio is not met, such as take：
Being included into car plate Candidate Set (herein also for the extension for supporting many Car license recognitions in the future) for above-mentioned condition will be met.
Correctly car plate to be filtered out from Candidate Set, then need more accurate candidate region to screen.
2.1.7 candidate region is screened
Car plate is correctly screened, other features of more careful consideration car plate in itself are now just needed, because in candidate
The class region of concentration and the similarity of car plate are very high.Mentioned in numerous documents, the more significant feature of car plate has：Face
Color, texture and ratio.Screened with reference to these three features in Candidate Set, specific way is：
1) counts ratio and the distribution of car plate background color (blue yellowish-white black, for different vehicles), if what the ratio was accounted for
Proportion is very big, then carries out step 2, is otherwise directly abandoned.
2) takes 1/2 height in region, and black and white saltus step scanning is carried out from left to right.So-called black and white saltus step refers in two-value
In figure, pixel by leucismus it is black or by black to white change frequency.The number of times of the car plate black and white saltus step of standard>=14.
The deformation and correction of 2.2 images
In actual identification scene, due to the inclination of car plate caused by the deployment of camera or the angle of traveling.Car
The inclination of board can not such as be corrected well, then the region segmentation influence on rear part is larger, and such as lower edges cutting causes word
Imperfect, character the mistake of symbol is cut.The premise corrected first to license plate image is to need to obtain inclined angle, next to that car plate
Rotation.
2.2.1 tilt angle calculation (FindAngle)
The calculating at inclination angle is a difficult point for the rotation correction of image, and the method for calculating typically has two kinds, and one is suddenly
Husband's conversion (Hough Transform) detection of straight lines obtains inclination angle；One is Radon transform (Radon Transform).
The general principle of Hough transform：Its general principle is the duality with line using point, and original image space is given
Fixed curve negotiating curve representation form is changed into a point of parameter space.Thus the detection of given curve in original image
Problem is converted into the spike problem found in parameter space.Namely detection overall permanence is converted into detection local characteristicses.Such as
Hough transform can be extended to detection of straight lines, ellipse, circle, camber line etc..
The basic thought of Hough transform：A point under coordinates of original image coordinates system has corresponded to one in parameter coordinate system
Bar straight line, the straight line of same parameter coordinate system has corresponded to a point under original coordinate system, then, is under original coordinate system
A little, their slope and intercept is identical for the institute of existing straight line, so they correspond to same point under parameter coordinate system.
So after by under each spot projection under original coordinate system to parameter coordinate system, see under parameter coordinate system either with or without aggregation
Point, such accumulation point has just corresponded to the straight line under original coordinate system.
The general principle of Radon conversion：Along different straight lines, (distance of straight line and origin is d, deflection in one plane
For alfa) line integral is done to f (x, y), obtained picture F (d, alfa) is exactly function f Radon conversion.That is, plane
The transform value of each point of (d, alfa) has corresponded to some line integral value of original function.
The basic thought of Radon conversion：Radon conversion can be understood as projection of the image in ρ θ spaces, ρ θ spaces it is every
Some correspondence straight line, and Radon conversion is integration of the image slices vegetarian refreshments on every straight line.Therefore, high ash in image
The straight line of angle value can form bright spot in ρ θ spaces, and the line segment of low gray value forms dim spot in ρ θ spaces.Detection to straight line turns
Turn in domain transformation to bright spot, the detection of dim spot.
Note：The calculating of the anglec of rotation herein is only defined in the inclined situation of horizontal direction, the inclination angle of vertical direction
Degree does not consider temporarily.
Specific method is as follows:
Step 1：Gray processing operation is carried out to original license plate area；
In order to the information for the more horn of plenty for retaining car plate, the figure that the present invention is inputted when calculating license plate sloped angle α
As the subgraph to be intercepted on acquired original image, in order to avoid the redundant computation that coloured image is brought, license plate image need to be entered
The processing of row gray processing.
Step 2：Smooth operation is carried out to gray-scale map；
In normal circumstances, because picture noise often exists in the form of Gaussian Profile, noise information is mainly showed
HFS in the picture.Although the traditional Gaussian filter of construction can filter out noise, also can to a certain extent
Suppress marginal information.Dependence based on subsequent step of the present invention to marginal information, while noise can be reduced to the full extent again
Interference, present invention construction is a kind of to be based on median filter.The wave filter can reduce edge letter as far as possible while filter is made an uproar
The loss of breath.
Step 3：Gray scale stretching is carried out to gray-scale map；
The purpose of gray scale stretching is to improve the contrast of image, performs a T conversion, the pixel coverage of original image is existed
[p0,pk] brightness p transform to a new range [q0,qk] in brightness q.Transformation relation is as follows：
Q (i, j)=T (p (i, j))
Step 4：Sobel computings are carried out to gray-scale map；
Sobel operators can detect that the present invention constructs a level using the characteristic to marginal information as needed
Direction Sobel cores, the computing is performed to image, can destroy the marginal information of vertical direction, it is decomposed into isolated point, and
Retention level directional information.
Step 5：Gray-scale map is converted into binary map；
Sobel output images in step 4 are subjected to binary conversion treatment.Followed by a kind of removal specified type noise
Method filters out the point isolated in image, and the line segment of horizontal direction is now only included in image.
Step 6：Use the straight line in hough change detection binary maps；
The polar equation of the Hough transform of a point (x, y) is in plane right-angle coordinate：
λ=x*cos θ+y*sin θ
After the conversion, point (x, y) forms the curve changed with θ in Hough transform space, and such same is straight
All curves that point on line is formed by Hough transform space will have a common intersection point.Algorithm sets a cumulative number
Group A (θ, λ), finally asks for the corresponding θ of maximum A (θ, λ), λ is used as straight line polar equation parameter to be asked.
Step 7：α angles are tried to achieve according to straight slope；
It is license plate sloped angle α that parameter θ is tried to achieve in step 6.
2.2.2 car plate rotation (ImageRotate)
There is the angle of rotation, then can be carried out have rotated.Image is rotated just need to first construct its rotation
Matrix.
Then affine transformation is carried out using spin matrix as the cvWarpAffine that parameter is inputted in OpenCv storehouses.Note：
The upper and lower side frame of extended area that can be appropriate when progress image conversion, prevents postrotational car plate from exceeding region.
2.3 are accurately positioned (LPRPreciseLocate)
It is the preliminary approximate region for finding car plate in coarse positioning, car plate now still includes other noises.
It is so-called be accurately positioned be specific bit car plate in only include car plate background color and character, other frames, rivet etc. are all
It need to filter out.It is the most important thing to be accurately located among the flow entirely recognized, and it not only holds the result from coarse positioning, but also
It is related to the effect of segmentation and the identification of later stage character.
Being accurately positioned mainly has two themes：
1) how precise ablation upper and lower side frame
2) how precise ablation left and right side frame
Certainly, the problem of being also related in actual operation in many details, such as webbing frame car plate.One good
Method should can be competent at above-mentioned or more conditions of compatibility.Here it is accurately positioned and is divided into two scenes：
1) is directed to the Accurate Segmentation of common car plate：
Common car plate refers to that on the premise of image clearly car plate background has good contrast, car plate color with character
The spacing of standard, car plate frame inward flange and character is more than 5mm.As shown in fig. 7, the method for such car plate processing is：
The distribution of color of the analysis background colors of Step 1..Due to the background of car plate be known a certain color (it is assumed here that
For blue bottom car plate), the background color distribution to car plate carries out the bezel locations (not including frame) for the determination car plate that analysis can be substantially,
Because to color detection, the frame of car plate is unlikely to be blueness.Car plate is scanned from both direction, i.e., from upper
Down, from the bottom up；Stop scanning and carrying out the operation for cutting off frame when scanning is to background color.
The upper and lower side frames of Step 2. are accurately cut.Analysis to distribution of color, can remove upper and lower side frame, but upper side frame
The influence of rivet may be still suffered from, it would be desirable to which what is done is the external minimum rectangle of a reserved character.Specific way is to be based on
A kind of method that scan line removes upper and lower side frame.
Arthmetic statement：A) intercepts the width of car plate 0.3 to 0.7 width, from left to right, from the height of car plate 1/2 to coboundary
It is scanned the frequency n hop of simultaneously statistical pixel saltus step.Assuming that scanning has nhop=0, records current position top to the i-th row,
Stop this time to scan.B) is same, from left to right, and time of simultaneously statistical pixel saltus step is scanned from the height of car plate 1/2 to lower boundary
Number.Assuming that scanning has nhop=0, records current position bottom to jth row.Then the up-and-down boundary of car plate for [top,
Bottom] where scope.
The accurate cutting of the left and right side frames of Step 3..According to the priori of car plate, there is one fixed width between each character
Interval, above-mentioned method no longer adapts to the scanning of vertical direction.Frame based on vertical direction is removed, and is generally basede on projection
Mode defines the border of character.Compared with standard car plate and be not required to do too many processing.
2) is directed to the Accurate Segmentation of webbing frame car plate：The characteristics of webbing frame car plate is character and minimum enclosed rectangle and side
The distance between inward flange of frame very little, or even be sticked together.Such car plate brings difficulty to the removal of frame, particularly
It is more difficult to remove in the case of license plate sloped.
For such car plate, Basic practice can according to situation 1, but when upper and lower side frame is handled, preferably on most and
Erase 2pixel width more lowermost edge, prevent the adhesion between character and frame.For webbing frame judgement by from upper
When down and from the bottom up scanning, the line of demarcation that can not find frame and character is then determined as webbing frame.For the feelings of left and right side frame
Shape, because the frame situation on both sides is inconsistent, generally using different methods.
A) left frames：As shown in Figure 8 to car plate carry out vertical direction histogram projection, then from left to right (0~
Between 0.3width) histogram is scanned, the width of each column is tracked, if less than 5pixels width, then can
To think that the column is produced by frame, now left hand edge is retracted.
B) left frames：Because the character of right hand edge is possible to as " 1 ", therefore the method that can not use and remove left frame.
It can be weeded out for left frame when Character segmentation.
Claims (2)
1. method for distinguishing is known in a kind of screening of LPR car plates, it is characterised in that the step of being gathered including optimal car plate and License Plate
Step；
The step of the step of the step of described optimal car plate is gathered is filtered including car plate and extraction car plate frame number；
The step of car plate is filtered：By defining for ROI, car is filtered out from camera farther out or nearer situation；
The step of extracting car plate frame number：By candid photograph counter controls, N images are continuously captured to the car plate for entering identification region,
Stop to capture more than N；Between the frame class car plate of previous entrance identification region and the frame class car plate of next entrance identification region are captured
Every more than M seconds, counter O reset is captured；
The step of described License Plate, includes：
Step one：Utilize formula
Gray (i, j)=R (i, j) * 0.299+G (i, j) * 0.587+B (i, j) * 0.114
Gray processing processing is carried out to the color diagram collected and forms gray-scale map；
Step 2：Top-hat conversion is carried out to gray-scale map；
Step 3：Continue to carry out gray-scale map binary conversion treatment formation binary map；
Step 4：Arithmetic operation is opened and closed to binary map in continuation；
Step 5：The step of screening car plate：
Step 5.1：Calculate the ratio of color of object in candidate license plate region；
Step 5.2：Step 5.3 is carried out if target license plate background color ratio is more than setting value, is otherwise directly abandoned；
Step 5.3：Find the binaryzation artwork master of target license plate；
Step 5.4：1/2 height in region is taken, black and white saltus step scanning is carried out from left to right, if the number of times of black and white saltus step is not less than setting
Definite value is then abandoned；
Kernel values are 20*1 in the top-hat conversion of the step 2；
The step 3 asks for threshold value using Da-Jin algorithm；
Opening and closing operation operation comprises the following steps in the step 4：
Step 4.1：Computing is opened and closed to image, salt-pepper noise region is reduced；
Step 4.2：Opening operation operation is carried out to image, non-salt-pepper noise region is merged；
Step 4.3：Etching operation is carried out to image, salt-pepper noise region is reduced again；
Step 4.4：All regions are traveled through, the region that region area is less than setting value is removed；
Car plate is opened and closed described step four is additionally provided with the step of region is merged and search rectangular region after calculation process
Step：
The step of described region is merged includes：
Step 1：All profile boundary rectangles are put into set A in binary map；
Step 2：The color diagram of all binary maps in set A is found, set A is traveled through, if color of object ratio is more than setting value,
It is put into candidate rectangle set B；
Step 3：The geometric center of all rectangles in set of computations B, judges the position relationship of rectangle in set B, if two squares
Angle of the distance on shape border in the range of restriction and between horizontal line or vertical curve is in given threshold, then in two rectangles
Middle setting-out, merges it；
The step of search rectangular region, includes：
Step 1：Corresponding rectangle is put into set S in the binary map that will be merged through region；
Step 2：Area is weeded out less than setting value or Aspect Ratio not in the rectangle of setting range, after being operated through above-mentioned rejecting
Remaining car plate is included into car plate Candidate Set.
2. a kind of system of LPR car plates screening identification, it is characterised in that including optimal car plate acquisition module and License Plate mould
Block；
Described optimal car plate acquisition module is screened and recognized for obtaining dream car bridge queen feeding License Plate module；
Described optimal car plate acquisition module includes car plate filter element and car plate frame number extraction unit, and described car plate filtering is single
License plate image of the member for collecting vehicle from camera suitable distance, described car plate frame number extraction unit is used to capture and filter puppet
Car plate；
Described License Plate module includes gray processing processing unit, top-hat converter units, binary conversion treatment unit, opening and closing
Arithmetic element, car plate screening unit, described gray processing processing unit are used to carry out gray processing processing to the color diagram collected
Formed and top-hat converter units are sent into after gray-scale map, described top-hat converter units are used to image is carried out to remove highlight area
Binary conversion treatment unit is sent into after processing, the binary conversion treatment unit is used to send into car after carrying out binary conversion treatment to gray-scale map
Board screening unit, described car plate screening unit is used to screen out non-car plate.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410306452.2A CN104036262B (en) | 2014-06-30 | 2014-06-30 | A kind of method and system of LPR car plates screening identification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410306452.2A CN104036262B (en) | 2014-06-30 | 2014-06-30 | A kind of method and system of LPR car plates screening identification |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104036262A CN104036262A (en) | 2014-09-10 |
| CN104036262B true CN104036262B (en) | 2017-09-19 |
Family
ID=51467028
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410306452.2A Expired - Fee Related CN104036262B (en) | 2014-06-30 | 2014-06-30 | A kind of method and system of LPR car plates screening identification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104036262B (en) |
Families Citing this family (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104778450B (en) * | 2015-03-30 | 2019-04-02 | 深圳市道通科技股份有限公司 | A kind of preprocess method and device of vehicle identification code image |
| CN105005759B (en) * | 2015-05-04 | 2018-10-02 | 南京理工大学 | The positioning of the bayonet image front window of multiple features fusion and extracting method |
| CN106529535A (en) * | 2016-11-10 | 2017-03-22 | 南京富士通南大软件技术有限公司 | License plate positioning method in complex lighting environment based on wavelet transform |
| CN106910195B (en) * | 2017-01-22 | 2020-06-16 | 北京奇艺世纪科技有限公司 | Webpage layout monitoring method and device |
| CN107292898B (en) * | 2017-05-04 | 2019-09-10 | 浙江工业大学 | A kind of license plate shadow Detection and minimizing technology based on HSV |
| CN107479061A (en) * | 2017-08-25 | 2017-12-15 | 重庆交通大学 | A kind of safe early warning method and system based on image recognition and radar range finding |
| CN108805007A (en) * | 2018-04-18 | 2018-11-13 | 浙江百世技术有限公司 | Class's line hole capital after selling all securities state judging method based on image recognition |
| CN108665712A (en) * | 2018-07-02 | 2018-10-16 | 大连海事大学 | A kind of vehicle gets over line act of violating regulations monitoring method and system |
| CN109460044A (en) * | 2019-01-10 | 2019-03-12 | 轻客小觅智能科技（北京）有限公司 | A kind of robot method for homing, device and robot based on two dimensional code |
| CN109858484B (en) * | 2019-01-22 | 2022-10-14 | 电子科技大学 | Multi-class transformation license plate correction method based on deflection evaluation |
| CN109753963A (en) * | 2019-02-22 | 2019-05-14 | 山东大学 | A kind of tire identification automatic recognition system and its working method based on line-structured light |
| CN110503094A (en) * | 2019-08-14 | 2019-11-26 | 中国电子科技集团公司第二十八研究所 | Professional certificate photo name board recognition methods, device |
| CN111783790A (en) * | 2020-07-14 | 2020-10-16 | 成都四方伟业软件股份有限公司 | Hand-drawn picture preprocessing method and device |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101183425A (en) * | 2007-12-20 | 2008-05-21 | 四川川大智胜软件股份有限公司 | Guangdong and Hong Kong license plate locating method |
| CN101246551A (en) * | 2008-03-07 | 2008-08-20 | 北京航空航天大学 | Fast license plate locating method |
| CN101398894A (en) * | 2008-06-17 | 2009-04-01 | 浙江师范大学 | Automobile license plate automatic recognition method and implementing device thereof |
| CN101739566A (en) * | 2009-12-04 | 2010-06-16 | 重庆大学 | Self-adapting projection template method-based automobile plate positioning method |
| CN103065138A (en) * | 2012-12-06 | 2013-04-24 | 新疆公众信息产业股份有限公司 | Recognition method of license plate number of motor vehicle |
| CN103824066A (en) * | 2014-03-18 | 2014-05-28 | 厦门翼歌软件科技有限公司 | Video stream-based license plate recognition method |
-
2014
- 2014-06-30 CN CN201410306452.2A patent/CN104036262B/en not_active Expired - Fee Related
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101183425A (en) * | 2007-12-20 | 2008-05-21 | 四川川大智胜软件股份有限公司 | Guangdong and Hong Kong license plate locating method |
| CN101246551A (en) * | 2008-03-07 | 2008-08-20 | 北京航空航天大学 | Fast license plate locating method |
| CN101398894A (en) * | 2008-06-17 | 2009-04-01 | 浙江师范大学 | Automobile license plate automatic recognition method and implementing device thereof |
| CN101739566A (en) * | 2009-12-04 | 2010-06-16 | 重庆大学 | Self-adapting projection template method-based automobile plate positioning method |
| CN103065138A (en) * | 2012-12-06 | 2013-04-24 | 新疆公众信息产业股份有限公司 | Recognition method of license plate number of motor vehicle |
| CN103824066A (en) * | 2014-03-18 | 2014-05-28 | 厦门翼歌软件科技有限公司 | Video stream-based license plate recognition method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104036262A (en) | 2014-09-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104036262B (en) | A kind of method and system of LPR car plates screening identification | |
| CN103077384B (en) | A kind of method and system of vehicle-logo location identification | |
| CN105893949B (en) | A kind of method for detecting lane lines under complex road condition scene | |
| CN103116751B (en) | A kind of Method of Automatic Recognition for Character of Lcecse Plate | |
| CN103235938B (en) | The method and system of car plate detection and indentification | |
| Nienaber et al. | Detecting potholes using simple image processing techniques and real-world footage | |
| CN102708356B (en) | Automatic license plate positioning and recognition method based on complex background | |
| CN109784344A (en) | A kind of non-targeted filtering method of image for ground level mark identification | |
| CN105488492B (en) | A kind of color image preprocess method, roads recognition method and relevant apparatus | |
| CN106650553A (en) | License plate recognition method and system | |
| CN105718870A (en) | Road marking line extracting method based on forward camera head in automatic driving | |
| CN102314599A (en) | Identification and deviation-detection method for lane | |
| CN103324935B (en) | Vehicle is carried out the method and system of location and region segmentation by a kind of image | |
| CN102419820A (en) | Method for rapidly detecting car logo in videos and images | |
| CN101833859A (en) | Self-triggering license plate identification method based on virtual coil | |
| CN104778444A (en) | Method for analyzing apparent characteristic of vehicle image in road scene | |
| CN102842037A (en) | Method for removing vehicle shadow based on multi-feature fusion | |
| CN106887004A (en) | A kind of method for detecting lane lines based on Block- matching | |
| Youjin et al. | A robust lane detection method based on vanishing point estimation | |
| CN109726717A (en) | A kind of vehicle comprehensive information detection system | |
| CN102902957A (en) | Video-stream-based automatic license plate recognition method | |
| CN105117726A (en) | License plate positioning method based on multi-feature area accumulation | |
| CN104112139A (en) | Method and system for accurate positioning of LPR license plate | |
| CN103049788B (en) | Based on space number for the treatment of object detection system and the method for computer vision | |
| CN109886168B (en) | Ground traffic sign identification method based on hierarchy |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170919 |