CN104050450A - Vehicle license plate recognition method based on video - Google Patents
Vehicle license plate recognition method based on video Download PDFInfo
- Publication number
- CN104050450A CN104050450A CN201410267812.2A CN201410267812A CN104050450A CN 104050450 A CN104050450 A CN 104050450A CN 201410267812 A CN201410267812 A CN 201410267812A CN 104050450 A CN104050450 A CN 104050450A
- Authority
- CN
- China
- Prior art keywords
- image
- license plate
- character
- car plate
- region
- 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.)
- Pending
Links
Landscapes
- Traffic Control Systems (AREA)
Abstract
The invention provides a vehicle license plate recognition method based on a video. According to the vehicle license plate recognition method based on the video, moving vehicles are detected and separated out with the vehicle video which is obtained through actual photographing by means of a camera serving as input, the accurate position of a vehicle license plate area is determined by conducting vertical edge extraction on a target vehicle image obtained after pre-processing, a vehicle license plate image is separated out, color correction, binaryzation and inclination correction are conducted on a vehicle license plate image, each character in the positioned vehicle license plate area is separated to serve as an independent character, feature extraction is conducted one each character, obtained feature vectors are classified through a classifier which is well trained in advance, a classification result serves as a preliminary recognition result, secondary recognition is conducted on the stained vehicle license plate characters according to a template matching algorithm imitating the visual characteristics of human eyes, and then a final vehicle license plate recognition result is obtained. The vehicle license plate recognition method based on the video has the advantages that hardware cost is reduced, the management efficiency of an intelligent transportation system is improved, the anti-jamming performance and the robustness are high, the recognition efficiency is high, and the recognition speed is high.
Description
Technical field
The invention belongs to Digital Image Processing and mode identification technology, be specifically related to a kind of licence plate recognition method.
Background technology
High speed development along with Chinese national economy, domestic motor vehicles, highway, urban road, parking lot etc. are more and more, requirement to traffic control, safety management is also increased day by day, intelligent transportation system ITS (Intelligent Transportation System) has become the development trend of 21 century control of traffic and road, is more and more subject to people's attention.The intelligent traffic administration system that wherein Vehicle License Plate Recognition System is robotization as the core of intelligent traffic administration system provides efficient, practical means, can be widely used in the aspects such as traffic monitoring, site survey of traffic accident, break in traffic rules and regulations record automatically, criminal tracking, highway toll collection system, the management of parking lot automatic safe, the management of intelligent garden, significant to improving management level and the automaticity of field of traffic.
Although had the car plate identification product of a lot of maturations seemingly at present, but really can apply to actual car plate identification product appears at abroad mostly, and, inside various for China's car plate standard comprises baroque Chinese character, car plate by the problem such as stained or deliberately the block fine solution of neither one still, make at present domestic go back neither one can be under various outdoor conditions efficient, stable, the car plate identification product that discrimination is high of all weather operations.Therefore, this vacancy is attracting numerous people to go constantly to study and explore.The car plate identification product occurring for current China mainly contains following some problem:
1. to identify the mode of car plate in the majority to rely on hardware trigger system to carry out video capture for current existing Vehicle License Plate Recognition System, and as magnetic induction loop, infrared, radio frequency etc., its toggle rate is higher, but the installation and maintenance cost of hardware device is higher.And along with the fast development of computer technology, the operational speed of a computer significantly improves, Vehicle License Plate Recognition System based on video more and more receives publicity, thereby its advantage is to have avoided the dependence of hardware device to reduce cost, its shortcoming is that operand is large, the image of identifying is the frame in video, and this image is subject to the impact of vehicle traveling direction, speed and other external interference, and the picture quality of obtaining is not as still image.
2. the car plate identification product that scarcely depends at present hardware trigger is not accomplished real-time detection truly, and its essence is still the car plate identification under static condition, and this speciality fails to play revolutionary improvement for current traffic.For example, this type of car plate identification product can be used for expressway tol lcollection bayonet socket, parking fee collective system bayonet socket etc., thereby it needs vehicle stop, photographic images is identified, although reduce like this human intervention, but still fail to accomplish non-parking charge, transport solution blockage problem and the intelligent transportation field of the automatic Vehicle License Plate Recognition System Jiang Dui of not stopping truly China plays revolutionary effect fundamentally.
3. current car plate identification product, for car plate standard more single.China's car plate standard is of a great variety and differ greatly, and car plate has four kinds of different background colors, and same car plate also may occur the character of different colours; Character kind comprises Chinese character, numeral and alphabetical, and wherein Chinese character part complex structure, identifies and get up to compare numeral and alphabetical difficult many; Part font and the common car plate of military vehicle characters on license plate are also different; Car plate rivet position is also along with car plate kind changes to some extent; The different types of characters on license plate arrangement architecture of part is also different, for example WJ car plate.These Dou Dui China apply to reality by Vehicle License Plate Recognition System and have produced very large obstruction.Therefore setting up the Vehicle License Plate Recognition System that can include the various car plate standards of China is the another focus of current this research field.
Most car plate identification product to as if target is clear, background is simple, the uniform image of illumination, especially for car plate, have the still good solution of neither one of car plate stained or that malice is blocked.This is again the major issue that Vehicle License Plate Recognition System is difficult to use in China's reality, should give attention.
Summary of the invention
In order to overcome the deficiencies in the prior art, the invention provides a kind of car plate based on video and know method for distinguishing.The method is usingd video that high-definition camera takes in real time on highway as input, has reduced hardware cost; Can realize the real-time detection of vehicle under normal transport condition, improved the efficiency of management of intelligent transportation system simultaneously; Susceptibility for illumination, background, picture quality is lower, has stronger anti-interference and robustness, can meet the round-the-clock real time execution of system; Especially, the identification that realizes the various car plate standards of Liao Dui China and have stained car plate, and discrimination is high, speed is fast, makes Vehicle License Plate Recognition System really apply to the actual possibility that becomes of China.
The technical solution adopted for the present invention to solve the technical problems comprises moving vehicle detection, car plate location, license plate image pre-service, Character segmentation and character recognition.
It is using the automobile video frequency normally travelling on the highway with high-definition camera real scene shooting as input that described moving vehicle detects, and therefrom detects moving vehicle and is split, and as the input picture of next step car plate identification, specific implementation step is:
The first step: the image even partition that is m * n by each frame sign in described video is M * N equal-sized sub-block, and the size of each block is 24 * 24 pixels;
Second step: successively by present frame I (i, j, k) and former frame I (i, j, k-1) poor, D (i, j, k)=| I (i, j, k)-I (i, j, k-1) |, accumulative total D (i, j, k) each sub-block in pixel value be greater than the number of pixels of threshold value Th1 (Th1=100), and to deposit its correspondence in size be the matrix R of M * N;
The 3rd step: judge successively each value in matrix R, when this value is more than or equal to threshold value Th2 (Th2=58), judge that in original image, corresponding sub-block has moving target to occur, this piece in original image is composed white, R (i even, j) >=Th2, R (i, j)=1, if R is (i, j) < Th2, R (i, j)=0;
The 4th step: by there will be some white boxed area in image after above-mentioned processing, these regions are the moving target of preliminary judgement, when the size of the white portion being communicated with is greater than threshold value Th3 (Th3=m * n * 53%), judges and occur moving vehicle herein, otherwise determine that it is the nontarget area that noise causes, by its eliminating;
The 5th step: the definite white portion of previous step is target vehicle region, by rank scanning, determine vehicle level and vertical position, again accordingly position from original image by vehicles segmentation out, and using the image that contains moving vehicle as input picture, so that subsequent operation;
Described car plate location is by pretreated target vehicle image is carried out to Vertical edge detection, and in conjunction with license plate area gray scale textural characteristics, geometric characteristic, color characteristic etc., determine the accurate location of license plate area, and by License Plate Segmentation out, specific implementation step is:
The first step: input picture is done respectively to the top cap conversion of horizontal direction and vertical direction,, with original image and the image subtraction after opening operation is processed, the image obtaining is pretreated image;
Second step: use the vertical rim detection mask of Sobel operator to carry out rim detection to pretreated image, obtain only containing the bianry image of vertical marginal information, thereby retain to greatest extent the marginal information of license plate area;
The 3rd step: carry out closed operation to only containing the bianry image of vertical marginal information, image neighboring edge is coupled together, then image is carried out to opening operation, make to disturb to connect to disconnect, obtain some regular connected regions as candidate's license plate area;
The 4th step: calculate the length breadth ratio of each candidate region and the hop value of horizontal direction projection, if all meet the corresponding condition of setting, think that this region is license plate area, and by target license plate zone location and split.Geometric characteristic by car plate knows, car plate is rectangular, although its big or small position in different images, can not determine, its Aspect Ratio remains in certain interval substantially.The high H of the wide W of car plate is than being 3:1, i.e. W=3 * H, the area S of car plate and perimeter L square meet S=L
2/ 22, introduce matching degree parameter α, β, establish α=(3 * H)/W, β=(22 * S)/L
2, work as matching degree | α-1|≤0.05, | β-1|≤0.05 o'clock illustrates that corresponding connected region is license plate area; From the Gray Projection of license plate area, the frequent and rule of the Gray Level Jump of car plate, its saltus step numerical value is 14, so the gray-scale value number of transitions J in calculated candidate region
1if, | J
1-Th
jump|>=3, get rid of this region, wherein Th
jump=14;
Described license plate image pre-service is license plate image to be carried out to color correction, binaryzation and slant correction process, and specific implementation step is:
The first step: will make low-pass filtering treatment after described license plate area gray processing, get maximum gradation value f corresponding to image after low-pass filtering treatment
maxaverage gray value f with former license plate image
avemake comparisons, if f
max-f
ave> 0, judges that car plate background color is for light color (comprising white and yellow), if f
max-f
ave< 0, is judged as dark color (comprising blueness and black);
Second step: ask license plate area B passage (blue component) histogram and do low-pass filtering treatment, find out the maximum gradation value g in B passage histogram, g=max (B (i, j));
The 3rd step: if judgement car plate background color is light color, the license plate image after processing is negated and (if picture data type is unit8, deducted each pixel value with 255; If picture data type is double, with 1, deduct each pixel value); If dark car plate is without correction;
The 4th step: proofread and correct red pixel point.When judgement car plate background color is light color, if g > is Th
color(g=max (B (i, j)), Th
color=120) judge that car plate background color is as white, the rgb value of the red pixel point in license plate image is proofreaied and correct for [40,40, g-60], if g < is Th
colorwithout correction; When car plate background color is dark, if g < is Th
color(g=max (B (i, j)), Th
color=120) judge that car plate background color, as black, is multiplied by 2 by the rgb value of the red pixel point in license plate image, strengthen brightness, if g > is Th
colorwithout correction;
The 5th step: the above-mentioned license plate grey level image through color correction is done to top cap conversion, eliminate the impact of uneven illumination;
The 6th step: the grey level histogram of asking the license plate image after the cap conversion of top, find two crests maximum in histogram, using the average of these two crest manipulative indexings as threshold value, gray level image is carried out to binaryzation, if there are not above-mentioned two crests, use Otsu method instead gray level image is carried out to binaryzation;
The 7th step: to described bianry image doing mathematics morphology operations, first use 2 * 30 1 matrix to carry out closed operation, use again 5 * 41 matrix to carry out opening operation, then the image after using canny operator to morphology processing is made rim detection, by nose section of Hough change detection, ask the inclination alpha of this line segment again, α is the horizontal tilt angle of image, by this angle [alpha] reverse rotation image, obtain the license plate image after horizontal tilt is proofreaied and correct;
The 8th step: the license plate image after described horizontal tilt is proofreaied and correct, detects gray-scale value number of transitions J by line scanning
2if, | J
2-Th
jump|≤2 (Th
jump=14), think that behavior character is expert at, otherwise excised as upper and lower side frame;
The 9th step: to the car plate bianry image of excision upper and lower side frame with progressive angle θ (20 ° of 0 ° of < θ <, step-length is 0.3) do shear transformation, the variance of projection in image level direction after computational transformation, the angle beta that maximum variance is corresponding is vertical bank angle, by angle beta, image is done to shear transformation, the license plate image after finally being proofreaied and correct.
Described Character segmentation, is exactly by each Character segmentation in oriented license plate area out, becomes single character.Due to China People's Armed Police car plate and common characters on license plate array structure difference very large, the corresponding employing of the present invention is common to be cut apart pattern and People's Armed Police and cuts apart pattern and process respectively, with People's Armed Police's car plate and the character arrangement mode of common car plate and different People's Armed Police's car plate and the common car plates distinguished of structure, when being People's Armed Police's car plate, judgement car plate adopt People's Armed Police to cut apart pattern, otherwise adopt the common pattern of cutting apart, specific implementation step is:
The first step: the image after proofreading and correct is made to medium filtering, and carry out zone marker, thereby delete interference region.Then extract each regional center, find two central points of row-coordinate maximum, be designated as A (x
1, y
1), B (x
2, y
2).If meeting, the row-coordinate that A, B are ordered and row-coordinate difference impose a condition, i.e. x
1< H * 0.6, x
2< H * 0.6, | x
1-x
2| < 3, wherein x
1and x
2represent respectively the row-coordinate that A, B are ordered, H represents license plate image height, thinks that this car plate is People's Armed Police's car plate; Otherwise, judge that this car plate is common car plate;
Second step: this car plate is People's Armed Police's car plate if judge, using affiliated area that A, B the order decimal fractions region as People's Armed Police's car plate, this Wu Ge region, right side, region is as character zone, (People's Armed Police's car plate is totally 9 characters for seven character zones altogether, the first two character " WJ " is changeless, does not need to cut apart and identify);
The 3rd step: this car plate is common car plate, cuts apart character by Gray Projection method if judge.The license plate image after described correction is made to medium filtering, and carry out horizontal direction projection,, to each row summation, obtain one-dimension array S, detect the reference position of this array non-zero region;
The 4th step: calculate the width w in the non-vanishing region of each section of projection, find out the width satisfying condition, i.e. 0.08w
p< w < 0.125w
p, w wherein
pfor car plate width;
The 5th step: the region that above-mentioned character duration is satisfied condition is as character zone;
The 6th step: calculating character number of regions n, if n < 7 thinks and has character " 1 " to have (width of " 1 " character does not meet above-mentioned condition);
The 7th step: carry out postsearch screening to not meeting the region of above-mentioned condition, if its width w meets 0.01w
p< w < 0.05w
p(w wherein
pfor car plate width), determine that it is " 1 ", otherwise cast out this region;
The 8th step: recalculate character zone number m, if m=6 think and lack Chinese character, using first region in first character left side as Chinese character region; If m=7, termination character is cut apart; If other, think that the impaired serious or image of car plate is pseudo-license plate image.
Described character recognition is that each character is carried out to feature extraction, resulting proper vector is classified by the good sorter of training in advance and using classification results as preliminary recognition result, characters on license plate by stained is carried out to secondary identification by imitating the template matching algorithm of human-eye visual characteristic, finally obtain license plate recognition result, specific implementation step is:
The first step: use the method for bilinear interpolation to carry out size conversion to split 7 groups of characters on license plate images, obtain the normalized image that 7 groups of pixel sizes are 32 * 16;
Second step: normalized image is carried out respectively to feature extraction: first described normalized image is divided into 4 * 4 piece, the gray-scale value summation in each piece, can obtain 8 * 4 array; Then described normalized image border is expanded with 0, it is done to rim detection with Canny operator, only retain maximal margin;
The 3rd step: the piece that is 4 * 4 to the above-mentioned image even partition that only retains maximal margin, to the gray-scale value summation in each piece, obtains the array of another group 8 * 4;
The 4th step: by above-mentioned two stretching merging of array, form the vector of one 64 dimension, as the proper vector of this character picture;
The 5th step: with sample training BP network, the BP neural network classifier that obtains training.Described BP neural network classifier trains in advance, and its training method is: first according to the samples pictures of actual photographed, obtain a large amount of pictures of each character, extract its feature as training sample; Then create BP neural network, determine that (the network number of plies is made as 2 layers to its each parameter, and transfer function is made as logsig, and training function is made as traingdx, and validity function is mse, and frequency of training is made as 10000, and target mse minimum value is made as 10
-5, target minimal gradient is made as 10
-17);
The 6th step: extract the proper vector of character to be identified, send into the BP neural network classifier training and carry out Classification and Identification, obtain preliminary recognition result;
The 7th step: when car plate has the situation such as stained, incomplete, may cause above-mentioned preliminary classification correctly to identify, now utilize the visual characteristic of human eye that the character picture going wrong is carried out to secondary identification by template matching method, thereby finally obtain correct recognition result;
The visual characteristic of described human eye refers to when target information is imperfect, and the mankind as the case may be lower global knowledge relevant to target infer, can infer according to local knowledge overall.This characteristic can be applied in and in template matching method, identify stained car plate.The steps include: first dirty character image to be extracted; Then be divided into upper left, lower-left, upper right, the equal-sized sub-block in four of bottom rights, the sub-block of character to be identified is mated with the corresponding blocks of template in template base; Finally using characters on license plate corresponding to the most approaching template of matching result as final recognition result.
The invention has the beneficial effects as follows:
The motion detection algorithm used of vehicle is improved frame-to-frame differences method.Each frame in video is carried out to piecemeal, and take each piece carries out moving region and detects and to have reduced calculated amount and memory space as unit, and the interference that many threshold determinations mechanism causes noise and environment in addition has good inhibition; Finally only be extracted in the movement destination image in certain area coverage, in the vehicle image that can guarantee like this to obtain, car plate size is more stable, for follow-up car plate location provides good image source.Thereby the method only depends on video input and has avoided the cost and the high problem of maintenance cost that rely on hardware device to bring, thereby improved algorithm can improve the real-time that arithmetic speed meets system greatly in addition, well having solved current car plate identification product on the market mostly can only be for the problem of static state or quasistatic image.
Car plate location adopts the method for local edge screening and mathematical morphology.The method is only utilized the half-tone information of image, does not utilize chromatic information, can prevent because the car plate that camera color distortion causes is located inaccurate problem, and have higher location rate and stronger robustness.The first step is the pre-service to image, can weaken non-license plate area marginal information, outstanding license plate area marginal information; Second step carries out local edge screening to image, because license plate area edge gradient changes obvious, frequent and has certain rule, so only retain the vertical marginal information of image according to this feature, can maximizedly filter out car plate edge, gets rid of and disturbs; The 3rd step is used the opening and closing operation in mathematical morphology can be connected fast or cut off the white portion in bianry image with region-filling algorithm, like this, can obtain rapidly some license plate candidate areas by opening and closing operation; Thereby the 4th step is set up decision mechanism by priori and is filtered out rapidly and accurately license plate area.The method takes full advantage of edge gradient feature, geometric characteristic, gray scale textural characteristics of car plate etc. and carrys out positioning licence plate as priori, for the car plate identification of next stage is laid a good foundation, the method has made up existing algorithm of locating license plate of vehicle and for the image of background complexity, has easily formed the problem of flase drop in addition, and locating speed comparatively fast can meet system real time, less to condition restriction such as the color of car plate, shooting angle, illumination, result shows that different shooting environmental is all had to good locating effect.
License plate image pre-service mainly comprises binaryzation and slant correction, and algorithm used herein is all applicable to China's Mainland standard car plate, and binaryzation is respond well, and slant correction is accurate, and has stronger robustness.The first step to the four steps, detected car plate background color, if directly license plate image is carried out to gray processing and binaryzation, may there are two kinds of binary images of black matrix wrongly written or mispronounced character and white gravoply, with black engraved characters, cannot carry out subsequent treatment, after background color detects and guaranteed binaryzation, license plate image is black matrix wrongly written or mispronounced character, has proofreaied and correct the red glyphs color that white background and black matrix car plate may occur simultaneously, and after assurance binaryzation, red glyphs information can not lost; The 5th step and the 6th step, first cap conversion in top can be eliminated the impact of uneven illumination on car plate, prevent car plate Characters Stuck and unsharp effect after binaryzation, the binaryzation method of use based on histogram analysis can be extracted license plate image binary-state threshold better, compare Otsu method, the method can obtain characters on license plate more clearly; The 7th step and the 8th step are image to be done to horizontal tilt proofread and correct, algorithm used is the slant correction method based on character edge information and Hough conversion, this algorithm is the improvement of the slant correction algorithm based on Hough conversion to tradition, it is no longer the straight line that detects license plate image upper and lower side frame, but detect place, the upper and lower edge of character straight line, effectively prevented that car plate positioning image from not containing the situation of frame, and algorithm stability is good, detection angles is accurate, and horizontal tilt calibration result is good; The 9th step has been excised the upper and lower side frame of the rear image of horizontal tilt correction, for next step provides good basis; The tenth step effect is vertical skew correction, and algorithm used is angle corresponding to horizontal projection maximum variance after detection shear transformation, and this algorithm accuracy of detection is high, good stability.
Character segmentation step has taken into full account China's Mainland car plate standard, be divided into common car plate and People's Armed Police's car plate, common License Plate Character Segmentation is used to the characters on license plate cutting algorithm based on projection, People's Armed Police's License Plate Character Segmentation is used to the characters on license plate cutting algorithm based on connected domain, thereby prevented that single algorithm is for the inapplicable situation that can not carry out car plate identification of heteroid car plate.To two class car plates, use corresponding algorithm all can accurately cut out the character in car plate, there is good stability.
Character recognition algorithm used is by BP neural network and template matches combines and utilize the visual characteristic of human eye to overcome the defect that existing licence plate recognition method None-identified car plate has stained situation.Compare single stencil matching or neural network algorithm, this algorithm identified rate is high, and robustness is good.The first step is transformed to unified size by character picture; Second step extracts feature to character picture, and this feature had both comprised image overall information, comprises again marginal information, can distinguish preferably other characters; The 3rd step is carried out BP neural network classification, and BP neural network can be carried out Nonlinear Classification, with it, character feature is classified, and can obtain higher accuracy; The 4th step is for part, to have the undesirable characters of classification such as stained or uneven illumination, uses stencil matching to carry out secondary identification recognizable set is increased greatly, thereby improved the applicability of system.
Accompanying drawing explanation
Fig. 1 is the licence plate recognition method general steps process flow diagram based on video;
Fig. 2 is the step block diagram that motion detection arrives car plate location;
Fig. 3 is for detecting the step block diagram of car plate background color and slant correction;
Fig. 4 is for differentiating car plate standard step block diagram;
Fig. 5 is red glyphs color correction effect schematic diagram;
Fig. 6 is slant correction treatment effect schematic diagram;
Fig. 7 is People's Armed Police's License Plate Character Segmentation effect schematic diagram;
Fig. 8 is all kinds License Plate Character Segmentation design sketch;
Fig. 9 is dirty character recognition effect figure.
Embodiment
Below in conjunction with drawings and Examples, the present invention is further described.
The invention provides a kind of licence plate recognition method based on video, key step flow process as shown in Figure 1, comprise: from video, detect moving vehicle and extract vehicle image, vehicle image is carried out to car plate location and obtain license plate image, license plate image is carried out the detection of car plate background color and proofreaies and correct and process, the license plate image of proofreading and correct after processing is carried out to Character segmentation and character recognition.
For above-mentioned steps, in conjunction with FB(flow block), describe in detail below:
Detect vehicle go forward side by side driving board location steps flow chart as shown in Figure 2, its detailed step is:
The image even partition that is m * n by each frame sign in described video is M * N equal-sized sub-block, and the size of each block is 24 * 24 pixels;
Successively by present frame I (i, j, k) and former frame I (i, j, k-1) poor, D (i, j, k)=| I (i, j, k)-I (i, j, k-1) |, accumulative total D (i, j, k) each sub-block in pixel value be greater than the number of pixels of threshold value Th1 (Th1=100), and to deposit its correspondence in size be the matrix R of M * N;
Judge successively each value in matrix R, when this value is greater than threshold value Th2 (Th2=58), judge that in original image, corresponding sub-block has moving target to occur, this piece in original image is composed white, R (i even, j) >=Th2, R (i, j)=1, if R is (i, j) < Th2, R (i, j)=0;
By there will be some white boxed area in image after above-mentioned processing, these regions are the moving target of preliminary judgement, when the size of the white portion being communicated with is greater than threshold value Th3 (Th3=m * n * 53%), judges and occur moving vehicle herein, otherwise determine that it is the nontarget area that noise causes, by its eliminating;
The definite white portion of previous step is target vehicle region, by rank scanning, determines vehicle level and vertical position, more accordingly position from original image by vehicles segmentation out, and using the image that contains moving vehicle as input picture, so that subsequent operation;
Input picture is done respectively to the top cap conversion of horizontal direction and vertical direction,, with original image and the image subtraction after opening operation is processed, the image obtaining is pretreated image;
Use the vertical rim detection mask of Sobel operator to carry out rim detection to pretreated image, obtain only containing the bianry image of vertical marginal information, thereby retain to greatest extent the marginal information of license plate area;
To only containing the bianry image of vertical marginal information, carry out closed operation, image neighboring edge is coupled together, then image is carried out to opening operation, make to disturb to connect to disconnect, obtain some regular connected regions as candidate's license plate area;
Calculate the length breadth ratio of each candidate region and the hop value of horizontal direction projection, if all meet the corresponding condition of setting, think that this region is license plate area, and by target license plate zone location and split.Geometric characteristic by car plate knows, car plate is rectangular, although its big or small position in different images, can not determine, its Aspect Ratio remains in certain interval substantially.The high H of the wide W of car plate is than being 3:1, i.e. W=3 * H, the area S of car plate and perimeter L square meet S=L
2/ 22, introduce matching degree parameter α, β, establish α=(3 * H)/W, β=(22 * S)/L
2, work as matching degree | α-1|≤0.05, | β-1|≤0.05 o'clock illustrates that corresponding connected region is license plate area; From the Gray Projection of license plate area, the frequent and rule of the Gray Level Jump of car plate, its saltus step numerical value is 14, so the gray-scale value number of transitions J in calculated candidate region
1if, | J
1-Th
jump|>=3, get rid of this region, wherein Th
jump=14;
To after described license plate area gray processing, make low-pass filtering treatment, get maximum gradation value f corresponding to image after low-pass filtering treatment
maxaverage gray value f with former license plate image
avemake comparisons, if f
max-f
ave> 0, judges that car plate background color is for light color (comprising white and yellow), if f
max-f
ave< 0, is judged as dark color (comprising blueness and black);
Ask license plate area B passage (blue component) histogram and do low-pass filtering treatment, find out the maximum gradation value g in B passage histogram, g=max (B (i, j));
If judgement car plate background color is light color, the license plate image after processing is negated and (if picture data type is unit8, with 255, deducted each pixel value; If picture data type is double, with 1, deduct each pixel value); If dark car plate is without correction;
Proofread and correct red pixel point.When judgement car plate background color is light color, if g > is Th
color(g=max (B (i, j)), Th
color=120) judge that car plate background color is as white, the rgb value of the red pixel point in license plate image is proofreaied and correct as [40,40, g-60], if
without correction; When car plate background color is dark, if g < is Th
color(g=max (B (i, j)), Th
color=120) judge that car plate background color, as black, is multiplied by 2 by the rgb value of the red pixel point in license plate image, strengthen brightness, if g > is Th
colorwithout correction;
The above-mentioned license plate grey level image through color correction is done to top cap conversion, eliminate the impact of uneven illumination;
Ask the grey level histogram of the license plate image after the cap conversion of top, find two crests maximum in histogram, using the average of these two crest manipulative indexings as threshold value, gray level image is carried out to binaryzation, if there are not above-mentioned two crests, use Otsu method instead gray level image is carried out to binaryzation;
To described bianry image doing mathematics morphology operations, first use 2 * 30 1 matrix to carry out closed operation, use again 5 * 41 matrix to carry out opening operation, then the image after using canny operator to morphology processing is made rim detection, by nose section of Hough change detection, ask the inclination alpha of this line segment again, α is the horizontal tilt angle of image, by this angle [alpha] reverse rotation image, obtain the license plate image after horizontal tilt is proofreaied and correct;
License plate image after described horizontal tilt is proofreaied and correct, detects gray-scale value number of transitions J by line scanning
2if, | J
2-Th
jump|≤2 (Th
jump=14), think that behavior character is expert at, otherwise excised as upper and lower side frame;
To the car plate bianry image of excision upper and lower side frame with progressive angle θ (20 ° of 0 ° of < θ <, step-length is 0.3 °) do shear transformation, the variance of projection in image level direction after computational transformation, the angle beta that maximum variance is corresponding is vertical bank angle, by angle beta, image is done to shear transformation, the license plate image after finally being proofreaied and correct.
Image after proofreading and correct is made to medium filtering, and carry out zone marker, thereby delete interference region.Then extract each regional center, find two central points of row-coordinate maximum, be designated as A (x
1, y
1), B (x
2, y
2).If meeting, the row-coordinate that A, B are ordered and row-coordinate difference impose a condition, i.e. x
1< H * 0.6, x
2< H * 0.6, | x
1-x
2| < 3, (x wherein
1and x
2represent respectively the row-coordinate that A, B are ordered, H represents license plate image height), think that this car plate is People's Armed Police's car plate; Otherwise, judge that this car plate is common car plate;
If judge, this car plate is People's Armed Police's car plate, using affiliated area that A, B the order decimal fractions region as People's Armed Police's car plate, this Wu Ge region, right side, region is as character zone, (People's Armed Police's car plate is totally 9 characters for seven character zones altogether, the first two character " WJ " is changeless, does not need to cut apart and identify);
If judge, this car plate is common car plate, cuts apart character by Gray Projection method.The license plate image after described correction is made to medium filtering, and carry out horizontal direction projection,, to each row summation, obtain one-dimension array S, detect the reference position of this array non-zero region;
The width w that calculates the non-vanishing region of each section of projection, finds out the width satisfying condition, i.e. 0.08w
p< w < 0.125w
p(w wherein
pfor car plate width);
The region that above-mentioned character duration is satisfied condition is as character zone;
Calculating character number of regions n, if n < 7 thinks and has character " 1 " to have (width of " 1 " character does not meet above-mentioned condition);
To not meeting the region of above-mentioned condition, carry out postsearch screening, if its width w meets 0.01w
p< w < 0.05w
p(w wherein
pfor car plate width), determine that it is " 1 ", otherwise cast out this region;
Recalculate character zone number m, if m=6 think and lack Chinese character, using first region in first character left side as Chinese character region; If m=7, termination character is cut apart; If other, think that the impaired serious or image of car plate is pseudo-license plate image.
To split 7 groups of characters on license plate images, use the method for bilinear interpolation to carry out size conversion, obtain the normalized image that 7 groups of pixel sizes are 32 * 16;
Normalized image is carried out respectively to feature extraction: first described normalized image is divided into 4 * 4 piece, the gray-scale value summation in each piece, can obtain 8 * 4 array; Then described normalized image border is expanded with 0, it is done to rim detection with Canny operator, only retain maximal margin;
The piece that is 4 * 4 to the above-mentioned image even partition that only retains maximal margin, to the gray-scale value summation in each piece, obtains the array of another group 8 * 4;
By above-mentioned two stretching merging of array, form the vector of one 64 dimension, as the proper vector of this character picture;
With sample training BP network, the BP neural network classifier that obtains training.Described BP neural network classifier trains in advance, and its training method is: first according to the samples pictures of actual photographed, obtain a large amount of pictures of each character, extract its feature as training sample; Then create BP neural network, determine that (the network number of plies is made as 2 layers to its each parameter, and transfer function is made as logsig, and training function is made as traingdx, and validity function is mse, and frequency of training is made as 10000, and target mse minimum value is made as 10
-5, target minimal gradient is made as 10
-17);
Extract the proper vector of character to be identified, send into the BP neural network training and carry out Classification and Identification, obtain preliminary recognition result;
When car plate has the situation such as stained, incomplete, may cause above-mentioned preliminary classification correctly to identify, now utilize the visual characteristic of human eye that the character picture going wrong is carried out to secondary identification by template matching method, thereby finally obtain correct recognition result;
The visual characteristic of described human eye refers to when target information is imperfect, and the mankind as the case may be lower global knowledge relevant to target infer, can infer according to local knowledge overall.This characteristic can be applied in and in template matching method, identify stained car plate.The steps include: first dirty character image to be extracted; Then be divided into upper left, lower-left, upper right, the equal-sized sub-block in four of bottom rights, the sub-block of character to be identified is mated with the corresponding blocks of template in template base; Finally using characters on license plate corresponding to the most approaching template of matching result as final recognition result.
For outstanding advantage of the present invention, below in conjunction with design sketch, be illustrated:
Fig. 5 is red glyphs color correction and license plate binary effect schematic diagram, and as shown in the figure, the picture left above is original car plate gray-scale map, this car plate background color is black, and " neck " word is red, after gray processing, brightness is starkly lower than other characters, and overall binaryzation can cause character to lose, as shown in top right plot.Through red-correction, improve red area Character Intensity (if white background car plate, should lower red glyphs brightness), as shown in the figure of lower-left, after overall binaryzation, more clearly reserved character, as shown in bottom-right graph.Because Chinese car plate structure difference is large, have in many kind car plates and there will be red glyphs, such as " police " word in police car car plate, in military vehicle car plate, Chinese character is all red, in part military vehicle car plate, second and the 3rd character is also redness, in People's Armed Police's car plate, " W " " J " and decimal fractions are all red, and these are all processed and caused inconvenience license plate binary.By red glyphs color correction, can address this problem more satisfactoryly.
Fig. 6 is that car plate is proofreaied and correct treatment effect schematic diagram, as shown in the figure, first license plate image is original binary image, can find out, car plate in the horizontal direction car plate has the inclination of certain angle, and this can proofread and correct processing by horizontal tilt, and character shape has certain deformation, become parallelogram pattern, this can process by vertical skew correction.Second license plate image is the design sketch after horizontal tilt is proofreaied and correct, and can find out, characters on license plate has distributed in the same horizontal line.The 3rd license plate image, for excising the design sketch after upper and lower side frame, removed the interference of upper and lower side frame to Character segmentation, but character also has deformation.The 4th design sketch that license plate image is vertical skew correction, can find out, character shape has returned to regular shape.
Fig. 7 is People's Armed Police's car plate cutting effect figure, and as shown in the figure, upper figure is the license plate image after overcorrect is processed, and after the cutting of People's Armed Police's pattern, the character obtaining is as figure below.
Fig. 8 is various different types of car plate cutting effect schematic diagram, and the car plate of the first row and the second row is military vehicle car plate, and two car plates are white background, all contain red glyphs, have all obtained the effect of good cutting from Character segmentation figure; The car plate of the third line is taken from bus, is car plate of the yellow end, and cutting effect is good; Fourth line is car plate at the bottom of normal blue, can find out that this car plate has uneven illumination impact, after overcorrect is processed, has eliminated this and has disturbed, and cutting effect is good; Fifth line is driving school's car plate, and background color is yellow, and over-exposed, through overcorrect, processes, and cutting effect is good.
Fig. 9 is the recognition effect figure that has dirty character image.First by mapping software, the character picture of real scene shooting car plate being added to stained thing at random blocks character picture, with the stained car plate of this emulation, then by the visual characteristic template matching method of freeing of this method, it is identified, can find out that final recognition effect is good.Identification error after only having " E " and " B " character picture to be blocked, remaining all numeral and letter all can be correctly validated, and this has the certain significance for the identification problem that solves stained car plate.
Claims (1)
1. the licence plate recognition method based on video, is characterized in that comprising the steps:
(1) moving vehicle detects, and specific implementation step is:
(1a) the image even partition that is m * n by each frame sign in the automobile video frequency normally travelling on the highway of high-definition camera real scene shooting is M * N equal-sized sub-block, and the size of each block is 24 * 24 pixels;
(1b) successively by present frame I (i, j, k) and former frame I (i, j, k-1) make poor D (i, j, k), accumulative total D (i, j, k) each sub-block in pixel value be greater than the number of pixels of threshold value Th1=100, and to deposit its correspondence in size be the matrix R of M * N;
(1c) judge successively each value in matrix R, when this value is more than or equal to threshold value Th2=58, judge that in original image, corresponding sub-block has moving target to occur, this piece in original image is composed white, R (i even, j) >=Th2, R (i, j)=1, if R is (i, j) < Th2, R (i, j)=0;
(1d) when the size of the white portion being communicated with is greater than threshold value Th3=m * n * 53%, judges and occur moving vehicle herein, otherwise determine that it is the nontarget area that noise causes, by its eliminating;
(1e) the definite white portion of previous step is target vehicle region, by rank scanning, determines vehicle level and vertical position, more accordingly position from original image by vehicles segmentation out, and using the image that contains moving vehicle as input picture;
(2) car plate location, specific implementation step is:
(2a) input picture is done respectively to the top cap conversion of horizontal direction and vertical direction, obtained pretreated image;
(2b) use the vertical rim detection mask of Sobel operator to carry out rim detection to pretreated image, obtain only containing the bianry image of vertical marginal information;
(2c) to only containing the bianry image of vertical marginal information, successively carry out closed operation and opening operation, obtain candidate's license plate area;
(2d) calculate the length breadth ratio of each candidate region and the hop value of horizontal direction projection, if the high H of the wide W of car plate than being 3:1, the area S of car plate and perimeter L square meet S=L
2/ 22, introduce matching degree parameter α, β, establish α=(3 * H)/W, β=(22 * S)/L
2, work as matching degree | α-1|≤0.05, | β-1|≤0.05 o'clock illustrates that corresponding connected region is license plate area; The gray-scale value number of transitions J in calculated candidate region
1if, | J
1-Th
jump|>=3, get rid of this region, wherein Th
jump=14;
(3) license plate image pre-service, specific implementation step is:
(3a) will after described license plate area gray processing, make low-pass filtering treatment, get maximum gradation value f corresponding to image after low-pass filtering treatment
maxaverage gray value f with former license plate image
avemake comparisons, if f
max-f
ave> 0, judges that car plate background color is for light color, if f
max-f
ave< 0, is judged as dark color;
(3b) ask license plate area blue component histogram and do low-pass filtering treatment, find out the maximum gradation value g in histogram;
If (3c) judgement car plate background color is light color, the license plate image after processing is negated; If dark car plate is without correction;
(3d) when judgement car plate background color is light color, if g > is Th
color, Th
colorjudge that car plate background color is as white for=120, the rgb value of the red pixel point in license plate image is proofreaied and correct for [40,40, g-60], if g < is Th
colorwithout correction; When car plate background color is dark, if g < is Th
colorjudge that car plate background color, as black, is multiplied by 2 by the rgb value of the red pixel point in license plate image, if g > is Th
colorwithout correction;
(3e) the above-mentioned license plate grey level image through color correction is done to top cap conversion;
(3f) ask the grey level histogram of the license plate image after the cap conversion of top, find two crests maximum in histogram, using the average of these two crest manipulative indexings as threshold value, gray level image is carried out to binaryzation, if there are not above-mentioned two crests, use Otsu method instead gray level image is carried out to binaryzation;
(3g) to bianry image doing mathematics morphology operations, first use 2 * 30 1 matrix to carry out closed operation, use again 5 * 41 matrix to carry out opening operation, then the image after using canny operator to morphology processing is made rim detection, by nose section of Hough change detection, ask the inclination alpha of this line segment again, α is the horizontal tilt angle of image, by this angle [alpha] reverse rotation image, obtain the license plate image after horizontal tilt is proofreaied and correct;
(3h) license plate image after described horizontal tilt correction is detected to gray-scale value number of transitions J by line scanning
2if, | J
2-Th
jump|≤2, think that behavior character is expert at, otherwise excised as upper and lower side frame;
(3i) the car plate bianry image of excision upper and lower side frame is done to shear transformation with progressive angle θ, 20 ° of 0 ° of < θ <, step-length is 0.3 °, the variance of projection in image level direction after computational transformation, the angle beta that maximum variance is corresponding is vertical bank angle, by angle beta, image is done to shear transformation, the license plate image after finally being proofreaied and correct;
(4) Character segmentation, specific implementation step is:
(4a) license plate image after proofreading and correct is made to medium filtering, and carry out zone marker, thereby delete interference region; Then extract each regional center, find two central point A and the B of row-coordinate maximum, if x
1< H * 0.6, x
2< H * 0.6, | x
1-x
2| < 3, wherein x
1and x
2represent respectively the row-coordinate that A, B are ordered, H represents license plate image height, thinks that this car plate is People's Armed Police's car plate; Otherwise, judge that this car plate is common car plate;
If (4b) judge that this car plate is People's Armed Police's car plate, using affiliated area that A, B order as the decimal fractions region of People's Armed Police's car plate, this Wu Ge region, right side, region is as character zone, altogether seven character zones;
If (4c) judge that this car plate is common car plate, cuts apart character by Gray Projection method;
(4d) calculate the width w in the non-vanishing region of each section of projection, find out the width satisfying condition, i.e. 0.08w
p< w < 0.125w
p, w wherein
pfor car plate width;
(4e) region above-mentioned character duration being satisfied condition is as character zone;
(4f) calculating character number of regions n, if n < 7 thinks and has character 1 to exist;
(4g) to not meeting the region of step (4d) condition, carry out postsearch screening, if its width w meets 0.01w
p< w < 0.05w
p, determine that it is character 1, otherwise cast out this region;
(4h) recalculate character zone number m, if m=6 think and lack Chinese character, using first region in first character left side as Chinese character region; If m=7, termination character is cut apart; If other, think that the impaired serious or image of car plate is pseudo-license plate image;
(5) character recognition, specific implementation step is:
(5a) to split 7 groups of characters on license plate images, use the method for bilinear interpolation to carry out size conversion, obtain the normalized image that 7 groups of pixel sizes are 32 * 16;
(5b) normalized image is carried out respectively to feature extraction: first described normalized image is divided into 4 * 4 piece, the gray-scale value summation in each piece, can obtain 8 * 4 array; Then described normalized image border is expanded with 0, it is done to rim detection with Canny operator, only retain maximal margin;
(5c) piece that is 4 * 4 to the above-mentioned image even partition that only retains maximal margin, to the gray-scale value summation in each piece, obtains the array of another group 8 * 4;
(5d), by above-mentioned two stretching merging of array, form the vector of one 64 dimension, as the proper vector of this character picture;
(5e) use the samples pictures BP network of actual photographed, the BP neural network classifier that obtains training;
(5f) extract the proper vector of character to be identified, send into the BP neural network classifier training and carry out Classification and Identification, obtain preliminary recognition result;
(5g) when above-mentioned preliminary classification cannot correctly be identified, dirty character image is extracted; Then be divided into upper left, lower-left, upper right, the equal-sized sub-block in four of bottom rights, the sub-block of character to be identified is mated with the corresponding blocks of template in template base; Finally using characters on license plate corresponding to the most approaching template of matching result as final recognition result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410267812.2A CN104050450A (en) | 2014-06-16 | 2014-06-16 | Vehicle license plate recognition method based on video |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410267812.2A CN104050450A (en) | 2014-06-16 | 2014-06-16 | Vehicle license plate recognition method based on video |
Publications (1)
Publication Number | Publication Date |
---|---|
CN104050450A true CN104050450A (en) | 2014-09-17 |
Family
ID=51503263
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410267812.2A Pending CN104050450A (en) | 2014-06-16 | 2014-06-16 | Vehicle license plate recognition method based on video |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104050450A (en) |
Cited By (73)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104392232A (en) * | 2014-11-07 | 2015-03-04 | 北京邮电大学 | Method for correcting video metadata in traffic scenes |
CN104408931A (en) * | 2014-10-29 | 2015-03-11 | 合肥指南针电子科技有限责任公司 | Incomplete sign license plate identification system and method |
CN104573680A (en) * | 2015-02-10 | 2015-04-29 | 深圳市哈工大交通电子技术有限公司 | Image detection method, image detection device and traffic violation detection system |
CN104616025A (en) * | 2015-01-19 | 2015-05-13 | 南京邮电大学 | Medical thermometer recognizing system |
CN104680163A (en) * | 2015-02-10 | 2015-06-03 | 柳州市金旭节能科技有限公司 | Licence plate recognition system |
CN104794698A (en) * | 2015-05-06 | 2015-07-22 | 中国人民解放军国防科学技术大学 | Low-light license plate image restoration algorithm based on license plate tracking and frame accumulation optimizing |
CN104809790A (en) * | 2015-04-29 | 2015-07-29 | 无锡市崇安区科技创业服务中心 | Parking management method based on image analysis |
CN104836986A (en) * | 2015-04-15 | 2015-08-12 | 华东师范大学 | Video information intelligent display method and apparatus |
CN105069456A (en) * | 2015-07-30 | 2015-11-18 | 北京邮电大学 | License plate character segmentation method and apparatus |
CN105205489A (en) * | 2015-08-27 | 2015-12-30 | 华南理工大学 | License plate detection method based on color texture analyzer and machine learning |
CN105488797A (en) * | 2015-11-25 | 2016-04-13 | 安徽创世科技有限公司 | License plate location method for HSV space |
CN105512662A (en) * | 2015-06-12 | 2016-04-20 | 北京卓视智通科技有限责任公司 | Detection method and apparatus for unlicensed vehicle |
CN105825212A (en) * | 2016-02-18 | 2016-08-03 | 江西洪都航空工业集团有限责任公司 | Distributed license plate recognition method based on Hadoop |
CN105975970A (en) * | 2016-05-30 | 2016-09-28 | 北京智芯原动科技有限公司 | SVF-based license plate positioning method and apparatus |
CN106326900A (en) * | 2016-08-30 | 2017-01-11 | 北海和思科技有限公司 | Time-saving garage railing control system |
CN106407984A (en) * | 2015-07-31 | 2017-02-15 | 腾讯科技(深圳)有限公司 | Target object recognition method and device |
CN106503712A (en) * | 2016-11-16 | 2017-03-15 | 广西大学 | One kind is based on stroke density feature character recognition method |
CN106503711A (en) * | 2016-11-16 | 2017-03-15 | 广西大学 | A kind of character recognition method |
CN106778752A (en) * | 2016-11-16 | 2017-05-31 | 广西大学 | A kind of character recognition method |
CN107067008A (en) * | 2017-01-17 | 2017-08-18 | 深圳怡化电脑股份有限公司 | A kind of recognition methods of similar character and system |
CN107067002A (en) * | 2017-03-09 | 2017-08-18 | 华东师范大学 | Road licence plate recognition method in a kind of dynamic video |
CN107145487A (en) * | 2016-03-01 | 2017-09-08 | 深圳中兴力维技术有限公司 | Image search method and device |
CN107180230A (en) * | 2017-05-08 | 2017-09-19 | 上海理工大学 | General licence plate recognition method |
CN107273896A (en) * | 2017-06-15 | 2017-10-20 | 浙江南自智能科技股份有限公司 | A kind of car plate detection recognition methods based on image recognition |
CN107330878A (en) * | 2017-06-21 | 2017-11-07 | 苏州卡睿知光电科技有限公司 | A kind of method and device for setting up eyeglass identification model |
CN107403130A (en) * | 2017-04-19 | 2017-11-28 | 北京粉笔未来科技有限公司 | A kind of character identifying method and character recognition device |
CN107451589A (en) * | 2017-07-07 | 2017-12-08 | 南京邮电大学 | A kind of method of the night license plate binary based on improved OSTU algorithms |
CN107507334A (en) * | 2017-08-31 | 2017-12-22 | 深圳怡化电脑股份有限公司 | A kind of banknote denomination recognition methods, device, equipment and storage medium |
CN107563380A (en) * | 2017-09-08 | 2018-01-09 | 上海理工大学 | A kind of vehicle license plate detection recognition method being combined based on MSER and SWT |
CN107895492A (en) * | 2017-10-24 | 2018-04-10 | 河海大学 | A kind of express highway intelligent analysis method based on conventional video |
CN108090484A (en) * | 2016-11-23 | 2018-05-29 | 杭州海康威视数字技术股份有限公司 | A kind of licence plate recognition method and device |
CN108229466A (en) * | 2016-12-15 | 2018-06-29 | 杭州海康威视数字技术股份有限公司 | A kind of licence plate recognition method and device |
CN108335396A (en) * | 2018-04-15 | 2018-07-27 | 珠海市华清创新科技有限公司 | A kind of artificial intelligence visitor management system with Car license recognition |
CN108364010A (en) * | 2018-03-08 | 2018-08-03 | 广东工业大学 | A kind of licence plate recognition method, device, equipment and computer readable storage medium |
CN108537099A (en) * | 2017-05-26 | 2018-09-14 | 华南理工大学 | A kind of licence plate recognition method of complex background |
CN108595730A (en) * | 2018-01-19 | 2018-09-28 | 南京工程学院 | A kind of gate system passage logic design method |
CN108717545A (en) * | 2018-05-18 | 2018-10-30 | 北京大账房网络科技股份有限公司 | A kind of bank slip recognition method and system based on mobile phone photograph |
CN108777021A (en) * | 2018-05-18 | 2018-11-09 | 北京大账房网络科技股份有限公司 | It is a kind of to mix the bank slip recognition method and system swept based on scanner |
CN108830827A (en) * | 2017-05-02 | 2018-11-16 | 通用电气公司 | Neural metwork training image generation system |
CN108960100A (en) * | 2018-06-22 | 2018-12-07 | 广西大学 | A kind of recognition methods of the sugarcane sugarcane section based on image procossing |
CN109100370A (en) * | 2018-06-26 | 2018-12-28 | 武汉科技大学 | A kind of pcb board defect inspection method based on sciagraphy and connected domain analysis |
CN109145915A (en) * | 2018-07-27 | 2019-01-04 | 武汉科技大学 | License plate rapid distortion antidote under a kind of complex scene |
CN109550712A (en) * | 2018-12-29 | 2019-04-02 | 杭州慧知连科技有限公司 | A kind of chemical fiber wire tailfiber open defect detection system and method |
CN109657682A (en) * | 2018-11-29 | 2019-04-19 | 国网河北省电力有限公司电力科学研究院 | A kind of electric energy meter registration recognition methods based on deep neural network and the soft cutting of multi-threshold |
CN109697857A (en) * | 2019-01-18 | 2019-04-30 | 合肥米佑信息技术有限公司 | Intelligent traffic control system based on image recognition and neural network algorithm |
CN109740547A (en) * | 2019-01-04 | 2019-05-10 | 平安科技(深圳)有限公司 | A kind of image processing method, equipment and computer readable storage medium |
CN109753967A (en) * | 2018-12-29 | 2019-05-14 | 北京师范大学 | A kind of picture character recognition methods |
CN109829460A (en) * | 2019-04-02 | 2019-05-31 | 安徽省安瑞机电科技有限公司 | A kind of License Plate Character Segmentation recognition methods and its system applied to parking management system |
CN109902680A (en) * | 2019-03-04 | 2019-06-18 | 四川长虹电器股份有限公司 | The detection of picture rotation angle and bearing calibration based on convolutional neural networks |
CN110414507A (en) * | 2019-07-11 | 2019-11-05 | 和昌未来科技(深圳)有限公司 | Licence plate recognition method, device, computer equipment and storage medium |
CN110443220A (en) * | 2019-08-13 | 2019-11-12 | 树根互联技术有限公司 | Digital table image-recognizing method, device, electronic equipment and storage medium |
CN110458161A (en) * | 2019-07-15 | 2019-11-15 | 天津大学 | A kind of mobile robot doorplate location method of combination deep learning |
CN110807415A (en) * | 2019-10-31 | 2020-02-18 | 南通大学 | Traffic checkpoint vehicle intelligent retrieval system and method based on annual inspection marks |
CN110956169A (en) * | 2018-09-27 | 2020-04-03 | 杭州海康威视数字技术股份有限公司 | License plate recognition method and device and electronic equipment |
CN111160366A (en) * | 2019-12-18 | 2020-05-15 | 山东鹰格信息工程有限公司 | Color image identification method |
CN111178359A (en) * | 2018-11-09 | 2020-05-19 | 中移(杭州)信息技术有限公司 | License plate number recognition method, device and equipment and computer storage medium |
CN111291748A (en) * | 2020-01-15 | 2020-06-16 | 广州玖峰信息科技有限公司 | Cascade distributed artificial intelligence case number identification system |
US10706330B2 (en) | 2015-10-01 | 2020-07-07 | Intellivision Technologies Corp | Methods and systems for accurately recognizing vehicle license plates |
CN111429516A (en) * | 2020-03-23 | 2020-07-17 | 上海眼控科技股份有限公司 | Corner point positioning method and device of frame number, computer equipment and storage medium |
CN111783758A (en) * | 2020-06-05 | 2020-10-16 | 杭州嘉楠耘智信息科技有限公司 | Character positioning method and device and automatic meter reading system |
CN111832337A (en) * | 2019-04-16 | 2020-10-27 | 高新兴科技集团股份有限公司 | License plate recognition method and device |
CN111860521A (en) * | 2020-07-21 | 2020-10-30 | 西安交通大学 | Method for segmenting distorted code-spraying characters layer by layer |
CN112070081A (en) * | 2020-08-20 | 2020-12-11 | 广州杰赛科技股份有限公司 | Intelligent license plate recognition method based on high-definition video |
CN112182294A (en) * | 2020-09-28 | 2021-01-05 | 天地伟业技术有限公司 | Video structured human-vehicle detection algorithm |
CN112215240A (en) * | 2020-10-13 | 2021-01-12 | 珠海博明视觉科技有限公司 | Optimization method for improving 2D complex edge detection precision |
CN113506279A (en) * | 2021-07-22 | 2021-10-15 | 浙江大华技术股份有限公司 | Method and device for determining inclination angle of object, storage medium and electronic device |
CN114241461A (en) * | 2021-10-29 | 2022-03-25 | 四川天翼网络服务有限公司 | Single-layer/double-layer license plate augmentation method based on machine vision technology |
WO2022121021A1 (en) * | 2020-12-10 | 2022-06-16 | 广州广电运通金融电子股份有限公司 | Identity card number detection method and apparatus, and readable storage medium and terminal |
CN114913705A (en) * | 2022-05-09 | 2022-08-16 | 江苏蓝策电子科技有限公司 | Vehicle positioning method based on Bluetooth AOA |
CN114973754A (en) * | 2022-05-09 | 2022-08-30 | 江苏蓝策电子科技有限公司 | Vehicle positioning system based on Bluetooth AOA |
US11587327B2 (en) | 2015-10-01 | 2023-02-21 | Intellivision Technologies Corp | Methods and systems for accurately recognizing vehicle license plates |
CN115862170A (en) * | 2023-02-27 | 2023-03-28 | 成都智源易行科技有限公司 | License plate recognition auxiliary system and method for ETC charging system |
CN117275102A (en) * | 2023-10-13 | 2023-12-22 | 佛山佳立电气科技有限公司 | Stereo garage offline charging system |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2008234291A (en) * | 2007-03-20 | 2008-10-02 | Toshiba Corp | Character recognition device and character recognition method |
CN101944174A (en) * | 2009-07-08 | 2011-01-12 | 西安电子科技大学 | Identification method of characters of licence plate |
-
2014
- 2014-06-16 CN CN201410267812.2A patent/CN104050450A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2008234291A (en) * | 2007-03-20 | 2008-10-02 | Toshiba Corp | Character recognition device and character recognition method |
CN101944174A (en) * | 2009-07-08 | 2011-01-12 | 西安电子科技大学 | Identification method of characters of licence plate |
Non-Patent Citations (2)
Title |
---|
杨康: ""基于视频的车辆检测理论与方法研究"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
马爽等: ""一种基于多特征提取的实用车牌识别方法"", 《计算机应用研究》 * |
Cited By (93)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104408931A (en) * | 2014-10-29 | 2015-03-11 | 合肥指南针电子科技有限责任公司 | Incomplete sign license plate identification system and method |
CN104392232B (en) * | 2014-11-07 | 2017-09-29 | 北京邮电大学 | The bearing calibration of video metadata under a kind of traffic scene |
CN104392232A (en) * | 2014-11-07 | 2015-03-04 | 北京邮电大学 | Method for correcting video metadata in traffic scenes |
CN104616025A (en) * | 2015-01-19 | 2015-05-13 | 南京邮电大学 | Medical thermometer recognizing system |
CN104573680B (en) * | 2015-02-10 | 2018-09-11 | 深圳市哈工大交通电子技术有限公司 | Image detecting method, image detection device and traffic offence detecting system |
CN104680163A (en) * | 2015-02-10 | 2015-06-03 | 柳州市金旭节能科技有限公司 | Licence plate recognition system |
CN104573680A (en) * | 2015-02-10 | 2015-04-29 | 深圳市哈工大交通电子技术有限公司 | Image detection method, image detection device and traffic violation detection system |
CN104836986A (en) * | 2015-04-15 | 2015-08-12 | 华东师范大学 | Video information intelligent display method and apparatus |
CN104809790A (en) * | 2015-04-29 | 2015-07-29 | 无锡市崇安区科技创业服务中心 | Parking management method based on image analysis |
CN104794698A (en) * | 2015-05-06 | 2015-07-22 | 中国人民解放军国防科学技术大学 | Low-light license plate image restoration algorithm based on license plate tracking and frame accumulation optimizing |
CN105512662A (en) * | 2015-06-12 | 2016-04-20 | 北京卓视智通科技有限责任公司 | Detection method and apparatus for unlicensed vehicle |
CN105069456A (en) * | 2015-07-30 | 2015-11-18 | 北京邮电大学 | License plate character segmentation method and apparatus |
CN105069456B (en) * | 2015-07-30 | 2018-10-09 | 北京邮电大学 | A kind of registration number character dividing method and device |
CN106407984A (en) * | 2015-07-31 | 2017-02-15 | 腾讯科技(深圳)有限公司 | Target object recognition method and device |
CN105205489A (en) * | 2015-08-27 | 2015-12-30 | 华南理工大学 | License plate detection method based on color texture analyzer and machine learning |
CN105205489B (en) * | 2015-08-27 | 2018-07-20 | 华南理工大学 | Detection method of license plate based on color and vein analyzer and machine learning |
US12056589B2 (en) | 2015-10-01 | 2024-08-06 | Intellivision Technologies Corp | Methods and systems for accurately recognizing vehicle license plates |
US11587327B2 (en) | 2015-10-01 | 2023-02-21 | Intellivision Technologies Corp | Methods and systems for accurately recognizing vehicle license plates |
US10706330B2 (en) | 2015-10-01 | 2020-07-07 | Intellivision Technologies Corp | Methods and systems for accurately recognizing vehicle license plates |
CN105488797A (en) * | 2015-11-25 | 2016-04-13 | 安徽创世科技有限公司 | License plate location method for HSV space |
CN105825212A (en) * | 2016-02-18 | 2016-08-03 | 江西洪都航空工业集团有限责任公司 | Distributed license plate recognition method based on Hadoop |
CN107145487A (en) * | 2016-03-01 | 2017-09-08 | 深圳中兴力维技术有限公司 | Image search method and device |
CN105975970A (en) * | 2016-05-30 | 2016-09-28 | 北京智芯原动科技有限公司 | SVF-based license plate positioning method and apparatus |
CN106326900B (en) * | 2016-08-30 | 2019-09-27 | 湖北亘华工科有限公司 | A kind of time saving control system of garage railing |
CN106326900A (en) * | 2016-08-30 | 2017-01-11 | 北海和思科技有限公司 | Time-saving garage railing control system |
CN106503712A (en) * | 2016-11-16 | 2017-03-15 | 广西大学 | One kind is based on stroke density feature character recognition method |
CN106503711A (en) * | 2016-11-16 | 2017-03-15 | 广西大学 | A kind of character recognition method |
CN106778752A (en) * | 2016-11-16 | 2017-05-31 | 广西大学 | A kind of character recognition method |
CN108090484A (en) * | 2016-11-23 | 2018-05-29 | 杭州海康威视数字技术股份有限公司 | A kind of licence plate recognition method and device |
CN108090484B (en) * | 2016-11-23 | 2020-04-03 | 杭州海康威视数字技术股份有限公司 | License plate recognition method and device |
CN108229466A (en) * | 2016-12-15 | 2018-06-29 | 杭州海康威视数字技术股份有限公司 | A kind of licence plate recognition method and device |
CN108229466B (en) * | 2016-12-15 | 2020-12-11 | 杭州海康威视数字技术股份有限公司 | License plate recognition method and device |
CN107067008A (en) * | 2017-01-17 | 2017-08-18 | 深圳怡化电脑股份有限公司 | A kind of recognition methods of similar character and system |
CN107067002A (en) * | 2017-03-09 | 2017-08-18 | 华东师范大学 | Road licence plate recognition method in a kind of dynamic video |
CN107403130A (en) * | 2017-04-19 | 2017-11-28 | 北京粉笔未来科技有限公司 | A kind of character identifying method and character recognition device |
CN108830827A (en) * | 2017-05-02 | 2018-11-16 | 通用电气公司 | Neural metwork training image generation system |
CN107180230A (en) * | 2017-05-08 | 2017-09-19 | 上海理工大学 | General licence plate recognition method |
CN107180230B (en) * | 2017-05-08 | 2020-06-23 | 上海理工大学 | Universal license plate recognition method |
CN108537099A (en) * | 2017-05-26 | 2018-09-14 | 华南理工大学 | A kind of licence plate recognition method of complex background |
CN107273896A (en) * | 2017-06-15 | 2017-10-20 | 浙江南自智能科技股份有限公司 | A kind of car plate detection recognition methods based on image recognition |
CN107330878A (en) * | 2017-06-21 | 2017-11-07 | 苏州卡睿知光电科技有限公司 | A kind of method and device for setting up eyeglass identification model |
CN107451589A (en) * | 2017-07-07 | 2017-12-08 | 南京邮电大学 | A kind of method of the night license plate binary based on improved OSTU algorithms |
CN107507334A (en) * | 2017-08-31 | 2017-12-22 | 深圳怡化电脑股份有限公司 | A kind of banknote denomination recognition methods, device, equipment and storage medium |
CN107563380A (en) * | 2017-09-08 | 2018-01-09 | 上海理工大学 | A kind of vehicle license plate detection recognition method being combined based on MSER and SWT |
CN107895492A (en) * | 2017-10-24 | 2018-04-10 | 河海大学 | A kind of express highway intelligent analysis method based on conventional video |
CN108595730A (en) * | 2018-01-19 | 2018-09-28 | 南京工程学院 | A kind of gate system passage logic design method |
CN108364010A (en) * | 2018-03-08 | 2018-08-03 | 广东工业大学 | A kind of licence plate recognition method, device, equipment and computer readable storage medium |
CN108364010B (en) * | 2018-03-08 | 2022-03-22 | 广东工业大学 | License plate recognition method, device, equipment and computer readable storage medium |
CN108335396A (en) * | 2018-04-15 | 2018-07-27 | 珠海市华清创新科技有限公司 | A kind of artificial intelligence visitor management system with Car license recognition |
CN108717545A (en) * | 2018-05-18 | 2018-10-30 | 北京大账房网络科技股份有限公司 | A kind of bank slip recognition method and system based on mobile phone photograph |
CN108777021A (en) * | 2018-05-18 | 2018-11-09 | 北京大账房网络科技股份有限公司 | It is a kind of to mix the bank slip recognition method and system swept based on scanner |
CN108960100A (en) * | 2018-06-22 | 2018-12-07 | 广西大学 | A kind of recognition methods of the sugarcane sugarcane section based on image procossing |
CN109100370A (en) * | 2018-06-26 | 2018-12-28 | 武汉科技大学 | A kind of pcb board defect inspection method based on sciagraphy and connected domain analysis |
CN109145915B (en) * | 2018-07-27 | 2021-08-06 | 武汉科技大学 | Rapid distortion correction method for license plate under complex scene |
CN109145915A (en) * | 2018-07-27 | 2019-01-04 | 武汉科技大学 | License plate rapid distortion antidote under a kind of complex scene |
CN110956169A (en) * | 2018-09-27 | 2020-04-03 | 杭州海康威视数字技术股份有限公司 | License plate recognition method and device and electronic equipment |
CN111178359A (en) * | 2018-11-09 | 2020-05-19 | 中移(杭州)信息技术有限公司 | License plate number recognition method, device and equipment and computer storage medium |
CN109657682A (en) * | 2018-11-29 | 2019-04-19 | 国网河北省电力有限公司电力科学研究院 | A kind of electric energy meter registration recognition methods based on deep neural network and the soft cutting of multi-threshold |
CN109657682B (en) * | 2018-11-29 | 2022-02-11 | 国网河北省电力有限公司电力科学研究院 | Electric energy representation number identification method based on deep neural network and multi-threshold soft segmentation |
CN109550712A (en) * | 2018-12-29 | 2019-04-02 | 杭州慧知连科技有限公司 | A kind of chemical fiber wire tailfiber open defect detection system and method |
CN109753967A (en) * | 2018-12-29 | 2019-05-14 | 北京师范大学 | A kind of picture character recognition methods |
CN109740547A (en) * | 2019-01-04 | 2019-05-10 | 平安科技(深圳)有限公司 | A kind of image processing method, equipment and computer readable storage medium |
WO2020140610A1 (en) * | 2019-01-04 | 2020-07-09 | 平安科技(深圳)有限公司 | Image processing method and device, and computer-readable storage medium |
CN109697857A (en) * | 2019-01-18 | 2019-04-30 | 合肥米佑信息技术有限公司 | Intelligent traffic control system based on image recognition and neural network algorithm |
CN109902680A (en) * | 2019-03-04 | 2019-06-18 | 四川长虹电器股份有限公司 | The detection of picture rotation angle and bearing calibration based on convolutional neural networks |
CN109829460A (en) * | 2019-04-02 | 2019-05-31 | 安徽省安瑞机电科技有限公司 | A kind of License Plate Character Segmentation recognition methods and its system applied to parking management system |
CN111832337A (en) * | 2019-04-16 | 2020-10-27 | 高新兴科技集团股份有限公司 | License plate recognition method and device |
CN110414507A (en) * | 2019-07-11 | 2019-11-05 | 和昌未来科技(深圳)有限公司 | Licence plate recognition method, device, computer equipment and storage medium |
CN110458161A (en) * | 2019-07-15 | 2019-11-15 | 天津大学 | A kind of mobile robot doorplate location method of combination deep learning |
CN110458161B (en) * | 2019-07-15 | 2023-04-18 | 天津大学 | Mobile robot doorplate positioning method combined with deep learning |
CN110443220A (en) * | 2019-08-13 | 2019-11-12 | 树根互联技术有限公司 | Digital table image-recognizing method, device, electronic equipment and storage medium |
CN110807415A (en) * | 2019-10-31 | 2020-02-18 | 南通大学 | Traffic checkpoint vehicle intelligent retrieval system and method based on annual inspection marks |
CN110807415B (en) * | 2019-10-31 | 2023-04-07 | 南通大学 | Traffic checkpoint vehicle intelligent retrieval system and method based on annual inspection marks |
CN111160366A (en) * | 2019-12-18 | 2020-05-15 | 山东鹰格信息工程有限公司 | Color image identification method |
CN111291748A (en) * | 2020-01-15 | 2020-06-16 | 广州玖峰信息科技有限公司 | Cascade distributed artificial intelligence case number identification system |
CN111429516A (en) * | 2020-03-23 | 2020-07-17 | 上海眼控科技股份有限公司 | Corner point positioning method and device of frame number, computer equipment and storage medium |
CN111783758A (en) * | 2020-06-05 | 2020-10-16 | 杭州嘉楠耘智信息科技有限公司 | Character positioning method and device and automatic meter reading system |
CN111860521B (en) * | 2020-07-21 | 2022-04-22 | 西安交通大学 | Method for segmenting distorted code-spraying characters layer by layer |
CN111860521A (en) * | 2020-07-21 | 2020-10-30 | 西安交通大学 | Method for segmenting distorted code-spraying characters layer by layer |
CN112070081A (en) * | 2020-08-20 | 2020-12-11 | 广州杰赛科技股份有限公司 | Intelligent license plate recognition method based on high-definition video |
CN112070081B (en) * | 2020-08-20 | 2024-01-09 | 广州杰赛科技股份有限公司 | Intelligent license plate recognition method based on high-definition video |
CN112182294A (en) * | 2020-09-28 | 2021-01-05 | 天地伟业技术有限公司 | Video structured human-vehicle detection algorithm |
CN112215240A (en) * | 2020-10-13 | 2021-01-12 | 珠海博明视觉科技有限公司 | Optimization method for improving 2D complex edge detection precision |
CN112215240B (en) * | 2020-10-13 | 2024-02-20 | 珠海博明视觉科技有限公司 | Optimization method for improving 2D complex edge detection precision |
WO2022121021A1 (en) * | 2020-12-10 | 2022-06-16 | 广州广电运通金融电子股份有限公司 | Identity card number detection method and apparatus, and readable storage medium and terminal |
CN113506279A (en) * | 2021-07-22 | 2021-10-15 | 浙江大华技术股份有限公司 | Method and device for determining inclination angle of object, storage medium and electronic device |
CN114241461A (en) * | 2021-10-29 | 2022-03-25 | 四川天翼网络服务有限公司 | Single-layer/double-layer license plate augmentation method based on machine vision technology |
CN114973754A (en) * | 2022-05-09 | 2022-08-30 | 江苏蓝策电子科技有限公司 | Vehicle positioning system based on Bluetooth AOA |
CN114913705A (en) * | 2022-05-09 | 2022-08-16 | 江苏蓝策电子科技有限公司 | Vehicle positioning method based on Bluetooth AOA |
CN114913705B (en) * | 2022-05-09 | 2024-02-02 | 江苏蓝策电子科技有限公司 | Vehicle positioning method based on Bluetooth AOA |
CN114973754B (en) * | 2022-05-09 | 2024-02-02 | 江苏蓝策电子科技有限公司 | Vehicle positioning system based on bluetooth AOA |
CN115862170A (en) * | 2023-02-27 | 2023-03-28 | 成都智源易行科技有限公司 | License plate recognition auxiliary system and method for ETC charging system |
CN117275102A (en) * | 2023-10-13 | 2023-12-22 | 佛山佳立电气科技有限公司 | Stereo garage offline charging system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104050450A (en) | Vehicle license plate recognition method based on video | |
CN105373794B (en) | A kind of licence plate recognition method | |
CN110197589B (en) | Deep learning-based red light violation detection method | |
CN105488454B (en) | Front vehicles detection and ranging based on monocular vision | |
CN103235938B (en) | The method and system of car plate detection and indentification | |
CN105913041B (en) | It is a kind of based on the signal lamp recognition methods demarcated in advance | |
CN109299674B (en) | Tunnel illegal lane change detection method based on car lamp | |
CN102375982B (en) | Multi-character characteristic fused license plate positioning method | |
CN106156768B (en) | The vehicle registration certificate detection method of view-based access control model | |
CN104134079B (en) | A kind of licence plate recognition method based on extremal region and extreme learning machine | |
CN109255350B (en) | New energy license plate detection method based on video monitoring | |
CN107301405A (en) | Method for traffic sign detection under natural scene | |
CN103824081B (en) | Method for detecting rapid robustness traffic signs on outdoor bad illumination condition | |
CN105718870A (en) | Road marking line extracting method based on forward camera head in automatic driving | |
CN105654073B (en) | A kind of speed automatic control method of view-based access control model detection | |
CN102880863B (en) | Method for positioning license number and face of driver on basis of deformable part model | |
CN102915433B (en) | Character combination-based license plate positioning and identifying method | |
CN106128115A (en) | Fusion method for detecting road traffic information based on double cameras | |
CN103077384A (en) | Method and system for positioning and recognizing vehicle logo | |
CN105528609A (en) | Vehicle license plate location method based on character position | |
CN103927548B (en) | Novel vehicle collision avoiding brake behavior detection method | |
CN106887004A (en) | A kind of method for detecting lane lines based on Block- matching | |
CN104978567A (en) | Vehicle detection method based on scenario classification | |
CN112017445B (en) | Pedestrian violation prediction and motion trail tracking system and method | |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20140917 |
|
WD01 | Invention patent application deemed withdrawn after publication |