CN106709530A - License plate recognition method based on video - Google Patents
License plate recognition method based on video Download PDFInfo
- Publication number
- CN106709530A CN106709530A CN201710040890.2A CN201710040890A CN106709530A CN 106709530 A CN106709530 A CN 106709530A CN 201710040890 A CN201710040890 A CN 201710040890A CN 106709530 A CN106709530 A CN 106709530A
- Authority
- CN
- China
- Prior art keywords
- character
- license plate
- image
- svm classifier
- features
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/22—Image preprocessing by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/14—Image acquisition
- G06V30/148—Segmentation of character regions
- G06V30/153—Segmentation of character regions using recognition of characters or words
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/469—Contour-based spatial representations, e.g. vector-coding
- G06V10/473—Contour-based spatial representations, e.g. vector-coding using gradient analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/62—Text, e.g. of license plates, overlay texts or captions on TV images
- G06V20/625—License plates
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention provides a license plate recognition method based on a video, and the method comprises the steps: 1), setting a detection region and a virtual triggering line; 2), firstly carrying out the coarse positioning of a corresponding region during the triggering of the triggering line, and then carrying out the discrimination of a license plate based on the features of HOG (Histogram of Oriented Gradients) and an SVM classifier; 3), taking a plurality of continuous license plate images, carrying out the character segmentation through a projection method, extracting the HOG features after the normalization of each character, and obtaining the L1-BRD distance between different character features corresponding to different license plate images; 4), calculating the weight value of each character, obtaining the features of weights of a plurality of frames, and enabling the features to serve as the recognition features of the character; 5), recognizing the character of the license plate through combining with a multi-class SVM classifier. The method combines the multi-frame weighing histogram features and the multi-class SVM classifier, and then carries out the recognition of the characters of the license plate. The method can effectively eliminate impact of the characters of a single-frame license plate during segmentation or the impact caused by noise, and improves the license plate character recognition accuracy.
Description
Technical field
The present invention relates to intelligent transportation field, and in particular to a kind of licence plate recognition method based on video.
Background technology
Intelligent transportation is increasingly widely applied, and has become the indispensable part of traffic control system, it
It is that the traffic surveillance videos obtained by video monitoring system are automatically analyzed using computer vision technique to condition of road surface.
License plate recognition technology is a very important part in intelligent transportation, is applied to traffic control system, vehicle and comes in and goes out
The fields such as management system.
The algorithm of current Car license recognition has a lot, can substantially be divided into based on template matches, neutral net and supporting vector
The class method of machine (SVM) three：Method calculating speed based on template matches is fast, but to low-resolution image and inclined car plate
Discrimination is relatively low；Method based on neutral net easily causes the situation of local vacation saturation under the conditions of sample size is insufficient,
Character identification rate is extremely difficult to degree very high；SVM is based on statistical learning, and in the case of finite sample, training obtains minimum
Error, set up an optimal separating hyper plane in higher dimensional space, make the distance of the characteristic vector of positive negative sample and hyperplane most
Bigization, SVM has good adaptability under the limitation of small sample, process when high dimensional pattern is recognized also have it is very big excellent
Gesture.
However, not being based on the licence plate recognition method of the suitable treatment multi-frame joint image of video at present, car is limited
The development of board identification.
Based on the above, there is provided the car plate of a kind of histogram feature that can combine multi-frame joint and SVM classifier is known
Other method is necessary.
The content of the invention
The shortcoming of prior art in view of the above, it is an object of the invention to provide a kind of Car license recognition based on video
A kind of method, to realize licence plate recognition method of the histogram feature and SVM classifier that can combine multi-frame joint.
In order to achieve the above objects and other related objects, the present invention provides a kind of licence plate recognition method based on video, bag
Include step：1) detection zone is set on road, virtual firing line together is set within a detection region；2) when virtual triggering
When line is triggered, coarse positioning first is carried out to respective regions, being then based on histograms of oriented gradients HOG features and SVM classifier is carried out
Car plate identifies the true and false；3) the car plate picture for taking continuous multiple frames carries out Character segmentation using sciagraphy, is carried after each character normalization
HOG features are taken, the L1-BRD distances between different car plate picture correspondence character features are obtained；4) weight of each character is calculated, is obtained
Go out multiframe weighting feature, using this as character recognition feature；5) characters on license plate is carried out with reference to polytypic SVM classifier
Identification.
As a kind of preferred scheme of the licence plate recognition method based on video of the invention, step 2) in, judge firing line
Triggering includes step：Camera collection video is installed in detection zone front upper place, gray processing treatment is carried out to every two field picture, take tactile
The gray value of adjacent two field pictures on hair line, makes the difference the sum for seeking absolute value, if being more than predetermined threshold value, is determined with object by inspection
Survey region.
As a kind of preferred scheme of the licence plate recognition method based on video of the invention, step 2) license plate area is carried out
Coarse positioning includes step：The first step, extracts candidate license plate region in detection zone, and gaussian filtering is carried out to image, reduces
Influence of noise, then carries out gray processing and obtains gray level image；Second step, the gray level image to obtaining carries out binaryzation and obtains two-value
Image；3rd step, morphologic closed operation is carried out to bianry image；4th step, connected component labeling is carried out to bianry image；5th
Step, each connected domain to marking takes minimum enclosed rectangle, calculates rectangle deflection angle, and filter out angular deflection default
Rectangular area in angle；6th step, calculates the ratio of width to height of rectangular area that the 5th step is filtered out, and filters out depth-width ratio and exist
Rectangular area in preset range；7th step, level is adjusted to by rotating by the rectangular area that the 6th step is filtered out, the rectangle
Region correspondence original image is license plate area.
As a kind of preferred scheme of the licence plate recognition method based on video of the invention, step 2) in, based on HOG features
The identification true and false, including step are carried out to the car plate of coarse positioning with SVM classifier：The first step, size normalizing is carried out to license plate area
Change is processed；Second step, the method training SVM classifier based on Machine self-learning, the HOG features that can be based on license plate area are known
The true and false of other car plate；3rd step, must be to the car plate identification identification true and false based on the SVM classifier for training.
As a kind of preferred scheme of the licence plate recognition method based on video of the invention, step 3) to the car plate that navigates to
Character segmentation, including step are carried out using sciagraphy：The first step, gray processing is carried out to license plate area, obtains gray level image；Second
Step, binaryzation is carried out to gray level image, obtains bianry image；3rd step, counts white pixel points N um1 in the bianry image
And black pixel number Num2, if Num1>Num2, color is negated to bianry image；4th step, in vertical direction projection downwards, system
The number of the white pixel point of each column is counted, gray-scale statistical histogram is obtained, then scanned from left to right, record white pixel points
More than the abscissa value of threshold value, the continuous line segment of m is obtained, then image cut and obtain the m image containing character
Block；5th step, to the image block of each character, is projected in the horizontal direction, the often capable white pixel points of statistics, is obtained
Gray-scale statistical histogram per a line, scans from top to bottom, and record pixel number cuts the row less than threshold value more than the row of threshold value
Slice off, obtain the m character split, wherein, m is natural number, and m=3~10.
Further, step 3) in the L1-BRD distances that obtain between different car plate pictures correspondence character features include：Record
Two frames obtain the centre coordinate of car plate, if distance is in certain threshold value, are recorded as unifying car plate bin-to-bin distance descriptions
The distance between two corresponding bin of histogram, ifRepresent the histogram for counting at present, common n bin, hiTable
Show i-th value of bin, after obtaining k frame car plates, after Character segmentation, respectively obtain k group character sets, every group contains m character,
K groups character L1-BRD between any two is calculated apart from di,j,i,j∈[1,k]；TakeDistance matrix D is can obtain, L1-BRD is calculated
Such as formula (1)：
Further, step 4) include：
Calculate Distance matrix D such as formula (2)：
By every group correspondence character feature and all character features between distance quadratic sum and the quadratic sum of all distances
Ratio as the character weights, k character feature h of correspondence positioni, i ∈ [1, k] can obtain k frames weighted histogram special
H is levied, such as formula (3), (4), (5)：
As a kind of preferred scheme of the licence plate recognition method based on video of the invention, step 5) in, polytypic SVM
Grader builds hyperplane using SVM classifier to N classes and other N-1 classes samples, by man-to-man method to determine
There is the positive and negative of sample.
As described above, the licence plate recognition method based on video of the invention, has the advantages that：
The present invention will be based on after the histogram feature of multiframe weighting is combined with many classification SVM classifiers to characters on license plate
It is identified, the method for the present invention can effectively eliminate the influence that single frames characters on license plate is caused in segmentation or noise, improves car plate
The accuracy rate of character recognition.Step of the present invention is simple, and effect is significant is with a wide range of applications in intelligent transportation field.
Brief description of the drawings
Fig. 1 is shown as the licence plate recognition method step 1 based on video of the invention) structural representation that is presented.
The step of Fig. 2 is shown as licence plate recognition method based on video of the invention schematic flow sheet.
Component label instructions
S11~S15 steps 1)~step 5)
Specific embodiment
Embodiments of the present invention are illustrated below by way of specific instantiation, those skilled in the art can be by this specification
Disclosed content understands other advantages of the invention and effect easily.The present invention can also be by specific realities different in addition
The mode of applying is embodied or practiced, the various details in this specification can also based on different viewpoints with application, without departing from
Various modifications or alterations are carried out under spirit of the invention.
Refer to Fig. 1~Fig. 2.It should be noted that the diagram provided in the present embodiment only illustrates this in a schematic way
The basic conception of invention, package count when only display is with relevant component in the present invention rather than according to actual implementation in illustrating then
Mesh, shape and size are drawn, and the kenel of each component, quantity and ratio can be a kind of random change during its actual implementation, and its
Assembly layout kenel is likely to increasingly complex.
As shown in Fig. 1~Fig. 2, the present embodiment provides a kind of licence plate recognition method based on video, including step：
As shown in Fig. 1~Fig. 2, step 1 is carried out first) S11, a detection zone A is set on road, in detection zone
It is middle that virtual firing line L together is set.
As shown in figure 1, then carrying out step 2) S12, when virtual firing line is triggered, it is slightly fixed that first respective regions are carried out
Position, being then based on histograms of oriented gradients HOG features and SVM classifier carries out the car plate identification true and false.
As an example, judging that firing line triggering includes step：Camera collection video is installed in detection zone front upper place, it is right
Gray processing treatment is carried out per two field picture, the gray value of adjacent two field pictures in firing line is taken, the sum for seeking absolute value is made the difference, if being more than
Predetermined threshold value, then be determined with object by detection zone, i.e., virtual firing line triggering.
As an example, coarse positioning is carried out to license plate area includes step：The first step, extracts candidate's car in detection zone
Board region, gaussian filtering is carried out to image, reduces influence of noise, is then carried out gray processing and is obtained gray level image；Second step, to
To gray level image carry out binaryzation and obtain bianry image；3rd step, morphologic closed operation is carried out to bianry image；4th
Step, connected component labeling is carried out to bianry image；5th step, each connected domain to marking takes minimum enclosed rectangle, calculates square
Shape deflection angle, and filter out rectangular area of the angular deflection in predetermined angle；6th step, calculates what the 5th step was filtered out
The ratio of width to height of rectangular area, and filter out depth-width ratio rectangular area within a preset range；7th step, by rotating the 6th step
The rectangular area for filtering out is adjusted to level, and rectangular area correspondence original image is license plate area.
Specifically, including：
Step 2-1), it is known that car plate has a large amount of marginal informations, and vehicle has many horizontal edges, to extract car
Board region, it is to avoid other regions interference of car body, using sobel edge detection operators, to the derivation of image level direction, can hang down
Nogata is to edge.
Step 2-2), to step 2-1) obtain gray level image and carry out binaryzation with OSTU methods.
Step 2-3), after obtaining bianry image, in order to subsequently obtain license plate area, it is necessary to allow fringe region to connect with connected domain
Lead to, so carrying out morphologic closed operation to bianry image.
Step 2-4), to step 2-3) bianry image that arrives, connected component labeling is carried out, recognized by the priori to car plate size
Know, threshold value is set and excludes area more than 500*150 and the region less than 50*15.
Step 2-5), minimum enclosed rectangle is taken to each connected domain, the apex coordinate of boundary rectangle can be obtained, by top
Point coordinates can calculate rectangle deflection angle θ, filter out region of the angular deflection in positive and negative 15 degree.
Step 2-6), the general size of Chinese car plate is 440mm*140mm, and area is 440*140, and the ratio of width to height is 3.14,
To the rectangular area for 7) obtaining, the ratio of width to height bi is calculated, filter out 3<bi<4 region.
Step 2-7), to step 2-6) rectangular area that obtains, level is adjusted to by the region rotated deflection.The square
Shape region correspondence original image is the candidate region of car plate.
As an example, carrying out the identification true and false, including step to the car plate of coarse positioning based on HOG features and SVM classifier：The
One step, size normalized is carried out to license plate area；Second step, the method training SVM classifier based on Machine self-learning, makes
The true and false of its HOG feature recognition car plate that can be based on license plate area；3rd step, must be to car plate based on the SVM classifier for training
The identification identification true and false.
In the present embodiment, substantial amounts of license plate candidate area, as coarse positioning can be extracted by substantial amounts of test video
The car plate for going out, manually the car plate to coarse positioning demarcate, true car plate be labeled as positive sample, be negative non-car plate sample labeling
Sample, puts into SVM classifier after extraction HOG features and is trained, then using the SVM classifier for training to slightly arriving surely
Car plate identified.
Specifically, comprise the following steps：
Step 2-a), in practical application, longitudinal grid of vehicle, car light, the pattern of vehicle and complex background can all influence
The result of positioning so that coarse positioning to license plate candidate area contain some non-license plate areas.To the car plate candidate for having obtained
Region, is normalized to 128*32 sizes.
Step 2-b), first have to train SVM classifier, the first step of training to obtain sample data.Regarded using a large amount of tests
Frequently, step 1 is carried out to it)~step 2) operation, a large amount of license plate candidate areas are obtained, the inside is divided into two class pictures, real car
Board picture and be not car plate picture, by the use of these pictures as sample data.
Step 2-c), the second step of training, picture is labelled, as learning data, picture is classified by hand, will
Genuine car plate picture is designated as positive sample, and non-car plate picture is designated as negative sample.
Step 2-d), align negative sample and extract HOG features respectively, take 4*4 pixels/cell, 2*2cells/block.
Step 2-e), because sample size is larger, and sample dimension is relatively low, the Selection of kernel function rbf cores of SVM.With before
The sample training SVM classifier for having marked.
Step 2-f), the license plate candidate area for obtaining carries out car plate identification using the SVM classifier that training is obtained.
As shown in figure 1, then carrying out step 3) S13, take the car plate picture of continuous multiple frames and enter line character point using sciagraphy
Cut, HOG features are extracted after each character normalization, obtain the L1-BRD distances between different car plate picture correspondence character features.
As an example, step 3) car plate to navigating to carries out Character segmentation, including step using sciagraphy：
The first step, gray processing is carried out to license plate area, obtains gray level image.
Second step, binaryzation is carried out to gray level image, obtains bianry image；The present embodiment is using OSTU methods to the ash that obtains
Degree image carries out binaryzation.
3rd step, counts white pixel points N um1 and black pixel number Num2 in the bianry image, if Num1>Num2,
Color is negated to bianry image.
4th step, projection downwards, counts the number of the white pixel point of each column in vertical direction, obtains gray-scale statistical straight
Fang Tu, then scans from left to right, and record white pixel points obtain m continuous line segment, so more than the abscissa value of threshold value
Afterwards image cut and obtain the m image block containing character.
5th step, to the image block of each character, is projected in the horizontal direction, the often capable white pixel point of statistics
Number, obtains the gray-scale statistical histogram of every a line, scans from top to bottom, and record pixel number, will be less than threshold more than the row of threshold value
The row of value cuts away, and obtains the m character split, wherein, m is natural number, and m=3~10.Because Chinese license plate number is 7
Position, therefore, in the present embodiment, take m=7.
As an example, the L1-BRD distances obtained between different car plate picture correspondence character features include：Two frames are recorded to obtain
The centre coordinate of car plate, if distance is in certain threshold value, is recorded as unifying car plate bin-to-bin distance two Nogatas of description
Scheme the distance between corresponding bin, ifRepresent the histogram for counting at present, common n bin, hiRepresent i-th
The value of bin, after obtaining k frame car plates, after Character segmentation, respectively obtains k group character sets, and every group contains m character, calculates k groups
Character L1-BRD between any two is apart from di,j,i,j∈[1,k]；To make its difference bigizationner, takeCan obtain Distance matrix D,
Calculate L1-BRD such as formulas (1)：
In this embodiment, it is preferred that, take k=5.
BRD can more accurately measures characteristic vector between distance, but it is more sensitive to noise, and L1 distance receive noise
Influence very little, so the method using L1 distances and BRD distances is combined, L1-BRD is both products.
Specifically, the present invention uses bin-ratio-based Histogram distances, Bin ratio be defined as histogram bin it
Between ratio.We define a rate matrix H=(hi/hj)∈Rn, it contains the ratio between all bin of histogram.Give
2 histograms, it is each element after the normalization of their ratio matrix squares that we define their bin ratio-based distances
Between difference summation.
BRD (Bin Ratio-Based Histogram Distance) is intra-cross-bin distances, and previous
Bin-to-bin distances are relative.BRD has robustness to local matching and histogram normalization, and multiple with linear calculating
Miscellaneous degree.
L2 histograms of the normalization containing n bins is a column vector h ∈ Rn, such as formula (1.1)：
After existing two L2 normalization containing n the histogram feature p and q of bins, seek its squared-distance.Can obtain spy
Between levying apart from BRD (Bin Ratio-Based Histogram Distance), obtain formula (1.2)：
Wherein P and Q are that p and q compares value matrix.
The distance of formula (1.2) expression is simultaneously unstable, and when pi and qi very littles, the change of very little can all cause gained range difference
It is different very big, in order to avoid such case, item 1/q is added in (1.2)i+1/piDivisor is done, formula (1.4) is obtained：
Contrast L1 distances and L2 are directed to the distance of n-dimensional vector apart from these, and BRD distances are the n × n vector of ratio, institute
Contained than L1 distance and L2 apart from more information content with it.dBRDThe specific derivation process of (p, q) has been given above-mentioned, should
Distance only O (n2) time complexity.
BRD can more accurately measures characteristic vector between distance, but it is more sensitive to noise, and L1 distance receive noise
Influence very little, so the method using L1 distances and BRD distances is combined, i.e. both products, as a result such as formula (1.5), can be with
Effectively eliminate influence when noise jamming and numerical value very little.
As shown in figure 1, then carrying out step 4) S14, the weight of each character is calculated, the feature of multiframe weighting is drawn, will
This as character recognition feature；
As an example, step 4) include：
Calculate Distance matrix D such as formula (2)：
By every group correspondence character feature and all character features between distance quadratic sum and the quadratic sum of all distances
Ratio as the character weights, k character feature h of correspondence positioni, i ∈ [1, k] can obtain k frames weighted histogram special
H is levied, such as formula (3), (4), (5)：
As shown in figure 1, finally carrying out step 5) S15, characters on license plate is identified with reference to polytypic SVM classifier.
As a kind of preferred scheme of the licence plate recognition method based on video of the invention, step 5) in, polytypic SVM
Grader builds hyperplane using SVM classifier to N classes and other N-1 classes samples, by man-to-man method to determine
There is the positive and negative of sample.
Recognize that the grader of characters on license plate uses SVM classifier, SVM mainly solves two class classification problems, when many points for the treatment of
Need to construct suitable SVM multi-categorizers during class.It is exactly that a certain classification is treated as one herein using the method for one-to-many combination
Problem is thus converted into two classification problems by class, the sample of remaining other classifications as another kind of.We to N classes and its
Its N-1 classes sample builds hyperplane, and the positive and negative of all samples is determined by man-to-man method.Structure is needed in assorting process
Make N number of grader.
The number-plate number one of China has 7, and first is the abbreviation of each province, and second is letter, and 5 contain 10 afterwards
Individual numeral and 24 capitalizations in addition to " I " " O ".So the present invention needs to construct the SVM classifier of 3 types.
A large amount of characters on license plate samples are divided into 34 classes, are then trained using the SVM classifier of the above method, will trained
The SVM classifier for obtaining is classified to above-mentioned 5 frame weighted histogram feature, identifies characters on license plate.
As described above, the licence plate recognition method based on video of the invention, has the advantages that：
The present invention will be based on after the histogram feature of multiframe weighting is combined with many classification SVM classifiers to characters on license plate
It is identified, the method for the present invention can effectively eliminate the influence that single frames characters on license plate is caused in segmentation or noise, improves car plate
The accuracy rate of character recognition.Step of the present invention is simple, and effect is significant is with a wide range of applications in intelligent transportation field.
So, the present invention effectively overcomes various shortcoming of the prior art and has high industrial utilization.
The above-described embodiments merely illustrate the principles and effects of the present invention, not for the limitation present invention.It is any ripe
The personage for knowing this technology all can carry out modifications and changes under without prejudice to spirit and scope of the invention to above-described embodiment.Cause
This, those of ordinary skill in the art is complete with institute under technological thought without departing from disclosed spirit such as
Into all equivalent modifications or change, should be covered by claim of the invention.
Claims (8)
1. a kind of licence plate recognition method based on video, it is characterised in that including step：
1) detection zone is set on road, virtual firing line together is set within a detection region；
2) when virtual firing line is triggered, coarse positioning first is carried out to respective regions, is then based on histograms of oriented gradients HOG features
The car plate identification true and false is carried out with SVM classifier；
3) the car plate picture for taking continuous multiple frames carries out Character segmentation using sciagraphy, and HOG features are extracted after each character normalization,
Obtain the L1-BRD distances between different car plate picture correspondence character features；
4) calculate the weight of each character, draw multiframe weighting feature, using this as character recognition feature；
5) characters on license plate is identified with reference to polytypic SVM classifier.
2. the licence plate recognition method based on video according to claim 1, it is characterised in that：Step 2) in, judge triggering
Line triggering includes step：Camera collection video is installed in detection zone front upper place, gray processing treatment is carried out to every two field picture, take
The gray value of adjacent two field pictures in firing line, makes the difference the sum for seeking absolute value, if being more than predetermined threshold value, is determined with object process
Detection zone.
3. the licence plate recognition method based on video according to claim 1, it is characterised in that：Step 2) license plate area is entered
Row coarse positioning includes step：
The first step, extracts candidate license plate region in detection zone, and gaussian filtering is carried out to image, reduces influence of noise, so
After carry out gray processing and obtain gray level image；
Second step, the gray level image to obtaining carries out binaryzation and obtains bianry image；
3rd step, morphologic closed operation is carried out to bianry image；
4th step, connected component labeling is carried out to bianry image；
5th step, each connected domain to marking takes minimum enclosed rectangle, calculates rectangle deflection angle, and it is inclined to filter out angle
Turn the rectangular area in predetermined angle；
6th step, calculate the ratio of width to height of rectangular area that the 5th step is filtered out, and filters out depth-width ratio within a preset range
Rectangular area；
7th step, level is adjusted to by rotating by the rectangular area that the 6th step is filtered out, and rectangular area correspondence original image is
It is license plate area.
4. the licence plate recognition method based on video according to claim 1, it is characterised in that：Step 2) in, it is special based on HOG
SVM classifier of seeking peace carries out the identification true and false, including step to the car plate of coarse positioning：
The first step, size normalized is carried out to license plate area；
Second step, the method training SVM classifier based on Machine self-learning, can be based on the HOG feature recognitions of license plate area
The true and false of car plate；
3rd step, must be to the car plate identification identification true and false based on the SVM classifier for training.
5. the licence plate recognition method based on video according to claim 1, it is characterised in that：Step 3) to the car that navigates to
Board carries out Character segmentation, including step using sciagraphy：
The first step, gray processing is carried out to license plate area, obtains gray level image；
Second step, binaryzation is carried out to gray level image, obtains bianry image；
3rd step, counts white pixel points N um1 and black pixel number Num2 in the bianry image, if Num1>Num2, to two
Value image negates color；
4th step, projection downwards, counts the number of the white pixel point of each column in vertical direction, obtains gray-scale statistical Nogata
Figure, then scans from left to right, and record white pixel points obtain m continuous line segment, then more than the abscissa value of threshold value
Image cut and obtains the m image block containing character；
5th step, to the image block of each character, is projected in the horizontal direction, the often capable white pixel points of statistics, is obtained
To the gray-scale statistical histogram of every a line, scan from top to bottom, record pixel number will be less than the row of threshold value more than the row of threshold value
Cut away, obtain the m character split, wherein, m is natural number, and m=3~10.
6. the licence plate recognition method based on video according to claim 5, it is characterised in that：Step 3) in obtain different cars
L1-BRD distances between board picture correspondence character feature include：The centre coordinate that two frames obtain car plate is recorded, if distance is certain
In threshold value, then it is recorded as unifying the distance between two corresponding bin of histogram of car plate bin-to-bin distance descriptions, ifRepresent the histogram for counting at present, common n bin, hiI-th value of bin is represented, after obtaining k frame car plates, word
After symbol segmentation, k group character sets are respectively obtained, every group contains m character, calculates k groups character L1-BRD distances between any two
di,j,i,j∈[1,k]；TakeDistance matrix D is can obtain, L1-BRD such as formulas (1) are calculated：
。
7. the licence plate recognition method based on video according to claim 6, it is characterised in that：Step 4) include：
Calculate Distance matrix D such as formula (2)：
By the ratio of quadratic sum and the quadratic sum of all distances of distance between the feature and all character features of every group of correspondence character
As the weights of the character, k character feature h of correspondence positioni, i ∈ [1, k] can obtain k frame weighted histogram feature h,
Such as formula (3), (4), (5)：
8. the licence plate recognition method based on video according to claim 1, it is characterised in that：Step 5) in, it is polytypic
SVM classifier builds hyperplane using SVM classifier to N classes and other N-1 classes samples, is determined by man-to-man method
All samples it is positive and negative.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710040890.2A CN106709530A (en) | 2017-01-17 | 2017-01-17 | License plate recognition method based on video |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710040890.2A CN106709530A (en) | 2017-01-17 | 2017-01-17 | License plate recognition method based on video |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106709530A true CN106709530A (en) | 2017-05-24 |
Family
ID=58908732
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710040890.2A Pending CN106709530A (en) | 2017-01-17 | 2017-01-17 | License plate recognition method based on video |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106709530A (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107368821A (en) * | 2017-08-04 | 2017-11-21 | 浙江工业大学 | A kind of licence plate recognition method based on SVMs |
| CN107563396A (en) * | 2017-08-10 | 2018-01-09 | 南京大学 | The construction method of protection screen intelligent identifying system in a kind of electric inspection process |
| CN108009545A (en) * | 2017-12-22 | 2018-05-08 | 大连运明自动化技术有限公司 | A kind of automobile engine cylinder-body sequence number vision automatic recognition method |
| CN108108735A (en) * | 2017-12-22 | 2018-06-01 | 大连运明自动化技术有限公司 | A kind of Automobile trade mark automatic identifying method |
| CN108509912A (en) * | 2018-04-03 | 2018-09-07 | 深圳市智绘科技有限公司 | Multipath network video stream licence plate recognition method and system |
| CN109509157A (en) * | 2018-11-09 | 2019-03-22 | 重庆巴奥科技有限公司 | A kind of information of vehicles enhancing recognition methods based on large nuber of images information |
| CN109685070A (en) * | 2019-01-11 | 2019-04-26 | 上海大学(浙江·嘉兴)新兴产业研究院 | A kind of image pre-processing method |
| CN109800752A (en) * | 2018-08-09 | 2019-05-24 | 辽宁工业大学 | A kind of automotive license plate Character segmentation identification algorithm based on machine vision |
| CN109993171A (en) * | 2019-03-12 | 2019-07-09 | 电子科技大学 | A kind of registration number character dividing method based on multi-template and more ratios |
| CN110533019A (en) * | 2018-05-23 | 2019-12-03 | 杭州海康威视数字技术股份有限公司 | License plate locating method, device and storage medium |
| CN111079763A (en) * | 2019-12-05 | 2020-04-28 | 北京嘉楠捷思信息技术有限公司 | Training sample generation, model training, character recognition method and device thereof |
| CN111611995A (en) * | 2020-04-02 | 2020-09-01 | 陕西土豆数据科技有限公司 | Method for positioning license plate of road video image |
| CN113128500A (en) * | 2021-04-08 | 2021-07-16 | 浙江工业大学 | Mask-RCNN-based non-motor vehicle license plate recognition method and system |
| CN114333338A (en) * | 2020-09-30 | 2022-04-12 | 瑞昱新加坡有限公司 | License plate identification system and method |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103093428A (en) * | 2013-01-23 | 2013-05-08 | 中南大学 | Space-time united image sequence multi-scale geometric transformation denoising method |
| CN103871044A (en) * | 2012-12-14 | 2014-06-18 | 阿里巴巴集团控股有限公司 | Image signature generating method and image verifying method and device |
| CN105551264A (en) * | 2015-12-25 | 2016-05-04 | 中国科学院上海高等研究院 | Speed detection method based on license plate characteristic matching |
-
2017
- 2017-01-17 CN CN201710040890.2A patent/CN106709530A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103871044A (en) * | 2012-12-14 | 2014-06-18 | 阿里巴巴集团控股有限公司 | Image signature generating method and image verifying method and device |
| CN103093428A (en) * | 2013-01-23 | 2013-05-08 | 中南大学 | Space-time united image sequence multi-scale geometric transformation denoising method |
| CN105551264A (en) * | 2015-12-25 | 2016-05-04 | 中国科学院上海高等研究院 | Speed detection method based on license plate characteristic matching |
Non-Patent Citations (2)
| Title |
|---|
| WEIMING HU ET.AL: "Bin Ratio-Based Histogram Distances and Their Application to Image Classification", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
| 黄建才 著: "《输电线路绝缘子泄漏电流去噪和特征提取的研究》", 31 October 2015 * |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107368821A (en) * | 2017-08-04 | 2017-11-21 | 浙江工业大学 | A kind of licence plate recognition method based on SVMs |
| CN107563396A (en) * | 2017-08-10 | 2018-01-09 | 南京大学 | The construction method of protection screen intelligent identifying system in a kind of electric inspection process |
| CN107563396B (en) * | 2017-08-10 | 2019-08-30 | 南京大学 | The construction method of protection screen intelligent identifying system in a kind of electric inspection process |
| CN108009545A (en) * | 2017-12-22 | 2018-05-08 | 大连运明自动化技术有限公司 | A kind of automobile engine cylinder-body sequence number vision automatic recognition method |
| CN108108735A (en) * | 2017-12-22 | 2018-06-01 | 大连运明自动化技术有限公司 | A kind of Automobile trade mark automatic identifying method |
| CN108509912B (en) * | 2018-04-03 | 2021-09-28 | 深圳市智绘科技有限公司 | License plate recognition method and system for multi-channel network video stream |
| CN108509912A (en) * | 2018-04-03 | 2018-09-07 | 深圳市智绘科技有限公司 | Multipath network video stream licence plate recognition method and system |
| CN110533019B (en) * | 2018-05-23 | 2022-08-12 | 杭州海康威视数字技术股份有限公司 | License plate positioning method and device and storage medium |
| CN110533019A (en) * | 2018-05-23 | 2019-12-03 | 杭州海康威视数字技术股份有限公司 | License plate locating method, device and storage medium |
| CN109800752A (en) * | 2018-08-09 | 2019-05-24 | 辽宁工业大学 | A kind of automotive license plate Character segmentation identification algorithm based on machine vision |
| CN109800752B (en) * | 2018-08-09 | 2022-01-07 | 辽宁工业大学 | Automobile license plate character segmentation and recognition algorithm based on machine vision |
| CN109509157A (en) * | 2018-11-09 | 2019-03-22 | 重庆巴奥科技有限公司 | A kind of information of vehicles enhancing recognition methods based on large nuber of images information |
| CN109509157B (en) * | 2018-11-09 | 2021-04-30 | 江苏中天安驰科技有限公司 | Vehicle information enhanced identification method based on massive image information |
| CN109685070A (en) * | 2019-01-11 | 2019-04-26 | 上海大学(浙江·嘉兴)新兴产业研究院 | A kind of image pre-processing method |
| CN109993171A (en) * | 2019-03-12 | 2019-07-09 | 电子科技大学 | A kind of registration number character dividing method based on multi-template and more ratios |
| CN109993171B (en) * | 2019-03-12 | 2022-05-03 | 电子科技大学 | License plate character segmentation method based on multiple templates and multiple proportions |
| WO2021109775A1 (en) * | 2019-12-05 | 2021-06-10 | 嘉楠明芯(北京)科技有限公司 | Methods and devices for generating training sample, training model and recognizing character |
| CN111079763A (en) * | 2019-12-05 | 2020-04-28 | 北京嘉楠捷思信息技术有限公司 | Training sample generation, model training, character recognition method and device thereof |
| CN111079763B (en) * | 2019-12-05 | 2023-08-08 | 嘉楠明芯(北京)科技有限公司 | Training sample generation, model training and character recognition method and device |
| CN111611995A (en) * | 2020-04-02 | 2020-09-01 | 陕西土豆数据科技有限公司 | Method for positioning license plate of road video image |
| CN111611995B (en) * | 2020-04-02 | 2023-05-23 | 土豆数据科技集团有限公司 | Method applied to positioning of highway video image license plate |
| CN114333338A (en) * | 2020-09-30 | 2022-04-12 | 瑞昱新加坡有限公司 | License plate identification system and method |
| CN113128500A (en) * | 2021-04-08 | 2021-07-16 | 浙江工业大学 | Mask-RCNN-based non-motor vehicle license plate recognition method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106709530A (en) | License plate recognition method based on video | |
| Panahi et al. | Accurate detection and recognition of dirty vehicle plate numbers for high-speed applications | |
| CN104951784B (en) | A kind of vehicle is unlicensed and license plate shading real-time detection method | |
| Kim et al. | Scene text extraction in natural scene images using hierarchical feature combining and verification | |
| Babu et al. | Vehicle number plate detection and recognition using bounding box method | |
| CN107506763A (en) | A kind of multiple dimensioned car plate precise positioning method based on convolutional neural networks | |
| CN108734189A (en) | Vehicle License Plate Recognition System based on atmospherical scattering model and deep learning under thick fog weather | |
| Liu et al. | Hybrid cascade structure for license plate detection in large visual surveillance scenes | |
| CN106529532A (en) | License plate identification system based on integral feature channels and gray projection | |
| CN105551264A (en) | Speed detection method based on license plate characteristic matching | |
| Azad et al. | A novel and robust method for automatic license plate recognition system based on pattern recognition | |
| CN104050684A (en) | Video moving object classification method and system based on on-line training | |
| CN105184291A (en) | Method and system for detecting multiple types of license plates | |
| CN103699876B (en) | Method and device for identifying vehicle number based on linear array CCD (Charge Coupled Device) images | |
| CN110689003A (en) | Low-illumination imaging license plate recognition method and system, computer equipment and storage medium | |
| Mammeri et al. | North-American speed limit sign detection and recognition for smart cars | |
| CN102129569B (en) | Based on body detection device and the method for multiple dimensioned contrast characteristic | |
| CN110046618B (en) | License plate recognition method based on machine learning and maximum extremum stable region | |
| Moghassemi et al. | Iranian License Plate Recognition using connected component and clustering techniques | |
| Rattanakornphan et al. | Thai License Plate Recognition Based on YOLO Detector and CNN-Classifier | |
| He et al. | Combining global and local features for detection of license plates in video | |
| Hong et al. | License plate extraction and recognition of a Thai vehicle based on MSER and BPNN | |
| Jamtsho et al. | Bhutanese License Plate Recognition Using Hu’s Moments and Centroid Difference | |
| Feng et al. | Effective license plate detection using fast candidate region selection and covariance feature based filtering | |
| Tian et al. | A new algorithm for license plate localization in open environment using color pair and stroke width features of character |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170524 |
|
| RJ01 | Rejection of invention patent application after publication |