CN107545265A - A kind of intelligent vehicle license plate recognition system - Google Patents
A kind of intelligent vehicle license plate recognition system Download PDFInfo
- Publication number
- CN107545265A CN107545265A CN201710761984.9A CN201710761984A CN107545265A CN 107545265 A CN107545265 A CN 107545265A CN 201710761984 A CN201710761984 A CN 201710761984A CN 107545265 A CN107545265 A CN 107545265A
- Authority
- CN
- China
- Prior art keywords
- license plate
- image
- neighborhood
- vector
- cyclic
- 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
- Medicines Containing Antibodies Or Antigens For Use As Internal Diagnostic Agents (AREA)
Abstract
The invention belongs to technical field of image processing, discloses a kind of intelligent vehicle license plate recognition system, and the intelligent vehicle license plate recognition system includes：Car license recognition unit is used for the Car license recognition unit for gathering license board information and being handled license board information；Parking lot Charging Detail Record unit is electrically connected with Car license recognition unit and carries out parking charge for pair vehicle corresponding with car plate；Passage gate inhibition access unit is electrically connected with Car license recognition unit, for handling run-in information；Car license recognition unit includes：Image capture module, man-machine recording module, License Plate module, Character segmentation module, character recognition module etc..The technology of the present invention has a clear superiority, and solves the automatic charging function of vehicle, and intelligence degree is high, and the function with intelligent entrance guard, based on RFID technique, be combined by wireless radio-frequency and smart card techniques, have the characteristics that using it is simple, easy to maintenance, be easy to control.
Description
Technical Field
The invention belongs to the technical field of image processing, and particularly relates to an intelligent license plate recognition system.
Background
License plate number recognition, also called license plate number recognition or vehicle license plate recognition, is an application of computer video image recognition technology in vehicle license plate recognition, namely extracting and recognizing the license plate number from the image information. Vehicle number identification is an important technology in an intelligent traffic system, and can be widely applied to various vehicles with special numbers, such as: automobiles, trains, non-motor vehicles, etc.
The vehicle number identification has been widely applied, such as electronic police, public security gate, highway speed measurement and charging, parking lot management, sky net monitoring, special car special parking management, etc., and especially in the fields of electronic police, public security gate, highway, etc., the vehicle number identification has basically been provided with the license plate identification function. The parking lot also enters a wide application stage in China from 2012, the coverage rate is estimated to reach 10% by 2013, and the application of sky net monitoring and special car special parking is just started. However, the existing intelligent license plate recognition system has no automatic vehicle charging system, and has single function and low intelligent degree.
The interactive matting technology is widely applied to the fields of image and video editing, three-dimensional reconstruction and the like, and has extremely high application value under the limited user interaction. In the recent matting technology, a Laplace matrix gives a linear relation among pixels on an alpha image, and plays an important role in estimating the alpha image. Interactive matting is the computation of an alpha map of the foreground under limited user interaction, separating the foreground from the background. The input of the matting problem is an original image I and a trimap image provided by a user, and the output is an alpha image, a foreground F and a background B, so that the matting problem is a typical ill-conditioned problem and needs to introduce an assumed condition to solve the alpha image. Matting algorithms can be divided into three categories: a sampling-based method, a propagation-based method, a combined sampling and propagation method.
The linear relation among alpha values of neighborhood pixels is given by a Laplace matting matrix deduced by the prior art, and the method is widely applied to a matting algorithm; the Laplace matting matrix has the limitation, the Laplace matting matrix represents the relationship among pixels in the spatial neighborhood, but cannot reflect the relationship among pixels in non-neighborhoods; the Laplace matting matrix is established on the basis of the assumption of space continuity, and in some regions with abrupt change of foreground and background components, the Laplace matting matrix is difficult to obtain ideal effects.
In recent years, although researchers have conducted some studies on carrier frequency estimation of a single carrier frequency signal under an Alpha stationary distributed noise model, the results of the studies are few. Sun et al propose a new spectral analysis method suitable for Alpha stable distribution based on fractional low order statistics. The method utilizes the fractional low-order covariance spectrum to analyze the frequency characteristics of the noisy signals in all value ranges (alpha is more than 0 and less than or equal to 2), and provides a weighted overlapping average method to estimate the fractional low-order covariance spectrum. The method is applicable to any alpha value, and the variance of the spectral estimation is small. However, no specific algorithm step is provided for carrier frequency estimation in the document, and the carrier frequency can be estimated only by deeply researching the covariance spectrum of the carrier frequency; zhao et al have proposed a fraction low order cyclic spectrum-based MPSK signal carrier frequency estimation method for the problem that the parameter estimation method based on the second order cyclic statistic is seriously degraded in Alpha stable distributed noise, have analyzed the relation of its carrier frequency and corresponding fraction low order cyclic spectrum parameter for PSK signals under different M values, have given the carrier frequency estimation method suitable for all PSK signals on this basis. When the mixed signal-to-noise ratio is-10 dB and alpha is 1.5, the normalized mean square error of the carrier frequency estimation of the BPSK signal is 0.043, and the normalized mean square error of the carrier frequency estimation of the QPSK signal is 0.041, so the carrier frequency estimation performance under the low signal-to-noise ratio still needs to be improved.
In summary, the problems of the prior art are as follows: the existing intelligent license plate recognition system has no automatic vehicle charging system, and has single function and low intelligent degree; the signal quality is poor and the image acquisition authenticity is poor.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides an intelligent license plate recognition system.
The invention is realized in this way, an intelligent license plate recognition system, the intelligent license plate recognition system includes: the system comprises a license plate recognition unit, a parking lot charging unit and a passage entrance guard access unit;
the license plate recognition unit is used for collecting license plate information and processing the license plate information;
the parking lot charging unit is electrically connected with the license plate recognition unit and is used for charging the parking of the vehicle corresponding to the license plate;
the access control access unit is electrically connected with the license plate recognition unit and is used for processing vehicle access channel information;
the license plate recognition unit includes: the system comprises an image acquisition module, a man-machine input module, a license plate positioning module, a character segmentation module and a character recognition module;
the image acquisition module is used for acquiring license plate image information; the method specifically comprises the following steps: constructing a Laplace matrix by using a mobile least square method instead of a least square method, and using a KNN neighborhood to replace a spatial neighborhood to obtain a linear relation of non-neighborhood pixels on an alpha image, thereby calculating a mobile Laplace matrix and obtaining the alpha image;
solving a cyclic covariant function of a received PSK signal containing Alpha stable distributed noise; fourier transform is carried out on the cyclic covariant function to obtain a cyclic covariant spectrum of the cyclic covariant function; extracting a section with the cycle frequency epsilon =0Hz according to the obtained cycle covariate spectrum; respectively searching peak values of positive and negative half shafts of the obtained section, finding positive and negative frequency values corresponding to the peak values, taking absolute values, and then calculating an average value to be used as carrier frequency estimation of PSK signals under Alpha stable distribution noise of carrier frequency estimation values and outputting the carrier frequency estimation; extracting the output image characteristic vector and outputting a finally acquired image;
the method for moving least square matting is as follows:
in a grayscale image, the window w i The alpha value in the neighborhood satisfies the local linear condition, and the local linear relation is solved by using a moving least square method, which is expressed as follows:
weight ω, ω in equation (1) i Is the neighborhood w k The weight value in (1); formula (1) is represented in the form of the following matrix:
for each neighborhood w k ,G k Is defined as | w k A | × 2 matrix; g k Each row comprising a vector (I) i ,1)，W k Is the weight value omega corresponding to each row vector i Vector of composition, G k ' is G k W of (2) k Weighting, the corresponding per-row vector is represented as (W) k .I i ,W k )，The vector is formed by alpha values corresponding to all pixels in the neighborhood;
coefficient a k ,b k The solution is as follows:
order toJ (α) is represented by the following formula:
δ i,j is the Kronecker delta function, mu k And σ 2 Respectively a small window w k Inner based on W k Weighted mean and variance, | w k II is the number of pixels in the window, L is the moving Las matting matrix;
the cyclic covariate function of the received signal comprises:
the signal contains an MPSK signal that obeys the S distributed noise, expressed as:
where E is the average power of the signal,M＝2 k m =1,2.. M, q (T) denotes a rectangular pulse waveform, T denotes a symbol period, f denotes a symbol period c Represents the carrier frequency, phi 0 Representing the initial phase, if w (t) is non-gaussian noise following a S α S distribution, its self-covariant function is defined as:
wherein (x (t- τ)) <p-1> ＝|x(t-τ) p-2 x*(t-τ)，γ x(t-τ) Is the dispersion coefficient of x (t), the cyclic covariance of x (t) is defined as:
where ε is the cycle frequency and T is one symbol period;
the image feature vector extraction method comprises the following specific steps:
step one, collecting N samples to be used as a training set X, and solving an average value m of the samples by adopting the following formula:
wherein x is i E sample training set X = (X) 1 ，x 2 ，…，x N )；
Step two, obtaining a dispersion matrix S:
obtaining an eigenvalue lambdai of the dispersion matrix and a corresponding eigenvector ei, wherein ei is a principal component, and arranging the eigenvalues of lambdai 1, lambdai 2 and … in sequence from large to small;
taking p values, λ 1, λ 2, …, λ p, determines license plate space E = (E1, E2, …, eP), where the point in training sample X where each element projects into this space is given by:
x'i＝Etxi，t＝1，2，…，N；
the p-dimensional vector obtained by the formula is obtained by carrying out PCA dimensionality reduction on the original vector;
the feature extraction is based on sparse representation, and multi-image recognition is carried out by adopting an image recognition algorithm;
the specific method for carrying out multi-image recognition by the image recognition algorithm comprises the following steps:
detecting the license plates of the current frame and sequencing according to coordinates to obtain the recognition results of a plurality of license plates of the current frame; calculating the recognition results of the adjacent n frames of the corresponding license plate according to the recognition results of the license plates of the current frame; counting the identities of all license plates, and determining the final identity of a target by more than half n/2 of uniform identities;
wherein, the reconstruction error { r) between the picture to be identified and each category of the preset image library is calculated 1 ，r 2 ……r n }，r 1 <r 2 <……<r n The obtained similarity value is according toDetermining a final recognition result; wherein T is 1 Is a ratio value, T1=0.65;
the man-machine input module is electrically connected with the image acquisition module, can perform manual input when the image acquisition module is insensitive, and is favorable for preventing suspicious license plates from escaping punishment;
the license plate positioning module is electrically connected with the image acquisition module and can effectively position a license plate according to image acquisition information; the license plate positioning method of the license plate positioning module comprises the following steps: leaving a circumscribed matrix of each subspace according to the size and the proportion of the license plate, namely a suspected license plate area; setting a jump function f (i, j), accurately positioning the suspected license plate area, and determining the upper and lower boundaries of the license plate area:
wherein c (i, j) is
c(i,j)＝LBP 8,1 (i,j)-LBP 8,1 (i,j-1)
In the above two formulas, i =1,2,3,4, … N, j =2,3,4, … M, so the sum S (i) of the transition times of any row i is:
if the sum S (i is more than or equal to 12) of the jumping times of any line, the line can belong to the license plate area; scanning the whole image from top to bottom, finding out all rows meeting S (i is more than or equal to 12), and recording the row number i of the row; if continuous h rows meet S (i is more than or equal to 12), a rectangular area with the width of M and the height of h can be obtained, and the area can be a license plate area, so that the area without the characteristics in the vehicle image is excluded;
the character segmentation module is electrically connected with the license plate positioning module and is used for segmenting and reading characters on a license plate;
the character recognition module is electrically connected with the character segmentation module and is used for respectively recognizing the segmented characters and finally outputting a result;
the parking lot charging unit includes: card reader, controller, charging device;
the card reader is used for reading the vehicle owner identification card;
the controller deducts the fee of the card of the vehicle owner according to the fee displayed on the fee counter;
the access control unit comprises an RFID control unit and an electromagnetic door lock.
The RFID control unit adopts an RFID radio frequency chip which transmits a reading signal to the vehicle owner identification card,
the RFID control unit is connected with the electromagnetic door lock and used for controlling the opening and closing of the electromagnetic door lock.
Further, the cyclic covariate spectrum of the received signal is performed as follows:
the cyclic covariate spectrum is the fourier transform of the cyclic covariate function, expressed as:
the derivation of the cyclic covariate spectrum is as follows:
when M is greater than or equal to 4, inAt the position of the air compressor, the air compressor is started,
when the M =2, the signal strength of the signal is high,
wherein Q (f) is the Fourier transform of Q (t), and
further, the carrier frequency estimation is realized by extracting a section with the cyclic frequency epsilon =0Hz in the cyclic covariate spectrum, and the following steps are carried out:
the envelope of the cyclic covariate spectrum on the n =0, i.e. epsilon =0Hz section is:
when f = +/-f c The envelope then takes a maximum value.
Further, a weight omega is introduced i The method is applied to a color model, and the moving least square matting method under the color model is as follows:
the linear relationship between the channels of the color image is represented by:
c is the number of channels of the color image, and after considering the information of each channel, equation (1) is converted into the following equation:
after the formula (2) is simplified, the mobile Laplace matrix under the color model is obtained by solving the following formula:
J(α)＝αLα T ；
in the formula (3), I is a matrix formed by 3*1 color vectors corresponding to all pixels in a small neighborhood, and is mu k W of I k Weighted average, Σ k Is I at W k Covariance matrix under weighting.
Further, the KNN neighborhood of the moving least square matting method expands the space neighborhood in the Laplace matrix to the KNN neighborhood, and the point of the KNN space is jointly determined by five dimensions (R, G, B, X and Y); and realizing efficient searching of the KNN neighborhood by using KD-TREE.
Further, the method for solving the large kernel in the moving least square sectional graph comprises the following steps: solving an alpha value by using a conjugate gradient method;
for equation Lx = b, the key of the conjugate gradient method is to construct a conjugate vector p and solve the corresponding residual error; the conjugate gradient method is solved by an iterative method, and in each iterative process, a new conjugate vector is solved by the following formula:
further, the coefficient of the conjugation direction is solved by:
the new x value and residual are solved using the following equation:
solving element q corresponding to point i in Lp vector by using the formula i ：
W k Is the neighborhood corresponding to pixel k, | w k II is the size of the neighborhood, i is the neighborhood W surrounding pixel k k One pixel of q i Is the I-th element of the q-vector, I i Representing three channels R, G, B, p for the 3-dimensional vector corresponding to pixel i i Is the element corresponding to pixel i in the conjugate vector, μ k Is a 3-dimensional vector, being the neighborhood W k In (II) i The mean value of the vector is calculated,is a neighborhood W k Conjugate vector p corresponding to middle element i i The average value of (a) of (b),is the corresponding 3-dimensional vector of pixel k,is a scalar quantity corresponding to the pixel k.
The invention has the advantages and positive effects that: the intelligent charging system has the advantages of solving the automatic charging function of the vehicle, having high intelligent degree, having the function of intelligent access control, being based on the RFID technology, combining the radio frequency technology and the intelligent card technology, and having the characteristics of simple use, convenient maintenance, convenient control and the like.
The Laplace matting matrix method using the moving least square method has the advantages that the complex foreground and foreground regions and the complex mixed foreground and background regions can obtain better effects. Deriving a moving Laplace matrix by using minimum moving quadratic multiplication instead of a least square method; compared with the least square method, the linear condition solved by the moving least square method is more accurate; the KNN neighborhood is used to replace the spatial neighborhood so that the Laplace matrix can reflect the relationship of the alpha values of the non-inter-neighborhood pixels. The Laplace matting matrix method using the moving least square method solves the alpha image according to the matrix, so that the foreground matting processing can be carried out on the image under the complex background, compared with the prior method, the method is more effective, the more accurate alpha image can be solved, and good effects can be obtained in the areas with complex foreground and background in the image, particularly in the mixed areas of the foreground and the background color and the areas with large local cavities and large changes.
The invention can estimate the carrier frequency of the PSK signal under Alpha stable distributed noise; the method has better estimation performance in the low signal-to-noise ratio environment; compared with the existing method, the method has better estimation performance under the same simulation experiment environment and the same signal parameter setting conditions of code element rate, carrier frequency, sampling point number, signal-to-noise ratio and the like. Thereby obtaining good image data.
Compared with 91.27% in the prior art, the data obtained by the image acquisition method is improved to 96.83%; this is a key point of the present invention.
Drawings
FIG. 1 is a block diagram of an intelligent license plate recognition system according to an embodiment of the present invention;
FIG. 2 is a block diagram of a license plate recognition unit provided in the practice of the present invention;
FIG. 3 is a block diagram of a parking lot billing unit provided by the implementation of the present invention;
fig. 4 is a block diagram of a gateway access unit provided in the present invention.
In the figure: 1. a license plate recognition unit; 2. a parking lot charging unit; 3. a passage entrance guard entrance and exit unit; 4. an image acquisition module; 5. a man-machine input module; 6. a license plate positioning module; 7. a character segmentation module; 8. a character recognition module; 9. a card reader; 10. a controller; 11. a charging unit; 12. an RFID control unit; 13. provided is an electromagnetic door lock.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The application of the principles of the present invention will be further described with reference to the accompanying drawings and specific embodiments.
As shown in fig. 1 to 4, an intelligent license plate recognition system provided in an embodiment of the present invention includes: the system comprises a license plate recognition unit 1, a parking lot charging unit 2 and a passage entrance guard access unit 3.
The license plate recognition unit 1 is used for collecting license plate information and processing the license plate information;
the parking lot charging unit 2 is electrically connected with the license plate recognition unit 1 and is used for charging parking of vehicles corresponding to the license plate;
and the passage entrance guard access control unit 3 is electrically connected with the license plate recognition unit 1 and is used for processing the vehicle access passage information.
The license plate recognition unit 1 comprises an image acquisition module 4, a man-machine input module 5, a license plate positioning module 6, a character segmentation module 7 and a character recognition module 8.
The image acquisition module 4 is used for acquiring license plate image information;
the man-machine input module 5 is electrically connected with the image acquisition module 4, can perform manual input when the image acquisition module 4 is insensitive, and is favorable for preventing suspicious license plates from escaping punishment;
the license plate positioning module 6 is electrically connected with the image acquisition module 4 and can effectively position a license plate according to image acquisition information;
the character segmentation module 7 is electrically connected with the license plate positioning module 6 and is used for segmenting and reading characters on a license plate;
the character recognition module 8 is electrically connected with the character segmentation module 7 and is used for respectively recognizing the segmented characters and finally outputting results;
the parking lot charging 2 unit comprises a card reader 9, a controller 10 and a charger 11.
The card reader 9 is used for reading a vehicle owner identification card;
the controller 10 deducts the card of the vehicle owner according to the fee displayed on the fee counter 11.
The passage entrance guard access unit 3 comprises an RFID control unit 12 and an electromagnetic door lock 13.
The RFID control unit 12 adopts an RFID radio frequency chip, the RFID radio frequency chip transmits a reading signal to the vehicle owner identification card,
the RFID control unit 12 is connected with an electromagnetic door lock 13 and used for controlling the opening and closing of the electromagnetic door lock.
The invention is further described with reference to specific examples.
The image acquisition module is used for acquiring license plate image information; the method specifically comprises the following steps: constructing a Laplace matrix by using a mobile least square method instead of a least square method, and using a KNN neighborhood to replace a spatial neighborhood to obtain a linear relation of non-neighborhood pixels on an alpha image, thereby calculating a mobile Laplace matrix and obtaining the alpha image;
solving a cyclic covariant function of a received PSK signal containing Alpha stable distributed noise; fourier transform is carried out on the cyclic covariant function to obtain a cyclic covariant spectrum of the cyclic covariant function; extracting a section with the cycle frequency epsilon =0Hz according to the obtained cycle covariate spectrum; respectively searching peak values of positive and negative half shafts of the obtained section, finding positive and negative frequency values corresponding to the peak values, taking absolute values, and then calculating an average value to be used as carrier frequency estimation of PSK signals under Alpha stable distribution noise of carrier frequency estimation values and outputting the carrier frequency estimation; extracting the output image characteristic vector and outputting a finally acquired image;
the method of moving least squares matting is as follows:
in a grayscale image, the window w i The alpha value in the neighborhood satisfies the local linear condition, and the local linear relation is solved by using a moving least square method, which is expressed as follows:
the weight value omega, omega in the formula (1) i Is the neighborhood w k The weight value in (1); formula (1) is represented in the form of the following matrix:
for each neighborhood w k ,G k Is defined as | w k A | × 2 matrix; g k Each row comprising a vector (I) i ,1)，W k Is the weight value omega corresponding to each row vector i Vector of composition, G k ' is G k W of (2) k Weighting, the corresponding per-row vector is represented as (W) k .I i ,W k )，The vector is composed of alpha values corresponding to all pixels in the neighborhood;
coefficient a k ,b k The solution is as follows:
order toJ (α) is represented by the following formula:
δ i,j is the Kronecker delta function, mu k And σ 2 Respectively a small window w k Inner based on W k Weighted mean and variance, | w k II is the number of pixels in the window, L is the moving Las matting matrix;
the cyclic co-varying function of the received signal comprises:
the signal contains an MPSK signal that obeys the S distributed noise, expressed as:
where E is the average power of the signal,M＝2 k m =1,2.. M, q (T) denotes a rectangular pulse waveform, T denotes a symbol period, f denotes a symbol period c Represents the carrier frequency, phi 0 Representing the initial phase, if w (t) is non-gaussian noise following a S α S distribution, its self-covariant function is defined as:
wherein (x (t- τ)) ,p-1> ＝|x(t-τ) p-2 x*(t-τ)，γ x(t-τ) Is the dispersion coefficient of x (t), the cyclic covariances of x (t) are defined as:
where ε is the cycle frequency and T is one symbol period;
the image feature vector extraction method comprises the following specific steps:
step one, collecting N samples to be used as a training set X, and calculating a sample average value m by adopting the following formula:
wherein x is i E to the sample training set X = (X) 1 ，x 2 ，…，x N )；
Step two, obtaining a dispersion matrix S:
obtaining an eigenvalue lambdai of the dispersion matrix and a corresponding eigenvector ei, wherein ei is a principal component, and arranging the eigenvalues of lambdai 1, lambdai 2 and … in sequence from large to small;
taking p values, λ 1, λ 2, …, λ p, determines license plate space E = (E1, E2, …, eP), where the point in training sample X where each element projects into this space is given by:
x'i＝Etxi，t＝1，2，…，N；
the p-dimensional vector obtained by the formula is obtained by carrying out PCA dimensionality reduction on the original vector;
the feature extraction is based on sparse representation, and multi-image recognition is carried out by adopting an image recognition algorithm;
the specific method for carrying out multi-image recognition by the image recognition algorithm comprises the following steps:
detecting the license plates of the current frame and sequencing according to coordinates to obtain recognition results of a plurality of license plates of the current frame; calculating the recognition results of the adjacent n frames of the corresponding license plate according to the recognition results of the license plates of the current frame; counting the identities of all license plates, and determining the final identity of a target by more than half n/2 of uniform identities;
wherein, the reconstruction between the picture to be identified and each category of the preset image library is calculatedError { r } 1 ，r 2 ……r n }，r 1 <r 2 <……<r n The obtained similarity value is according toDetermining a final recognition result; wherein T is 1 Is a ratio value, T1=0.65;
the license plate positioning module is electrically connected with the image acquisition module and can effectively position a license plate according to image acquisition information; the license plate positioning method of the license plate positioning module comprises the following steps: leaving a circumscribed matrix of each subspace according to the size and the proportion of the license plate, namely a suspected license plate area; setting a jump function f (i, j), accurately positioning the suspected license plate area, and determining the upper and lower boundaries of the license plate area:
wherein c (i, j) is
c(i,j)＝LBP 8,1 (i,j)-LBP 8,1 (i,j-1)
In the above two formulas, i =1,2,3,4, … N, j =2,3,4, … M, so the number of transitions in any row i and S (i) are:
if the sum S (i is more than or equal to 12) of the jumping times of any line, the line can belong to the license plate area; scanning the whole image from top to bottom, finding out all rows meeting S (i is more than or equal to 12), and recording the row number i of the row; if continuous h rows meet S (i is more than or equal to 12), a rectangular area with the width of M and the height of h can be obtained, and the area can be a license plate area, so that the area without the characteristics in the vehicle image is excluded;
the cyclic covariate spectrum of the received signal is performed as follows:
the cyclic covariate spectrum is the fourier transform of the cyclic covariate function, expressed as:
the derivation of the cyclic covariate spectrum is as follows:
when M is greater than or equal to 4, atAt the position of the air compressor, the air compressor is started,
when the M =2, the signal processing unit is in a state of M =2,
wherein Q (f) is the Fourier transform of Q (t), and
the carrier frequency estimation is realized by extracting a section with the cyclic frequency epsilon =0Hz in the cyclic covariate spectrum, and the method is carried out as follows:
the envelope of the cyclic covariate spectrum on the section n =0, namely epsilon =0Hz, is as follows:
when f = +/-f c The envelope then takes a maximum value.
Introducing a weight omega i The method is applied to a color model, and the moving least square matting method under the color model is as follows:
the linear relationship between the channels of the color image is represented by:
c is the number of channels of the color image, and equation (1) is converted into the following equation after considering the information of each channel:
after the formula (2) is simplified, the mobile Laplace matrix under the color model is obtained by solving the following formula:
J(α)＝αLα T ；
in the formula (3), I is a matrix formed by 3*1 color vectors corresponding to all pixels in a small neighborhood, and is mu k W of I k Weighted average, Σ k Is I at W k Covariance matrix under weighting.
The KNN neighborhood of the mobile least square matting method expands the space neighborhood in the Laplace matrix to the KNN neighborhood, and the point of the KNN space is jointly determined by five dimensions (R, G, B, X and Y); and realizing efficient searching of the KNN neighborhood by using KD-TREE.
The method for solving the large kernel in the moving least square sectional graph comprises the following steps: solving an alpha value by using a conjugate gradient method;
for equation Lx = b, the key of the conjugate gradient method is to construct a conjugate vector p and solve the corresponding residual error; the conjugate gradient method is solved by an iterative method, and in each iterative process, a new conjugate vector is solved by the following formula:
the coefficient of the conjugation direction is solved by:
the new x value and residual are solved using the following equation:
solving the element q corresponding to the point i in the Lp vector by the following formula i ：
W k Is the neighborhood corresponding to pixel k, | w k II is the size of the neighborhood, i is the neighborhood W surrounding pixel k k One pixel of q i Is the I-th element of the q-vector, I i Representing three channels R, G, B, p for the 3-dimensional vector corresponding to pixel i i Is the element corresponding to pixel i in the conjugate vector, μ k Is a 3-dimensional vector, which is the neighborhood W k In (II) i The mean value of the vector is calculated,is a neighborhood W k Conjugate vector p corresponding to element i in i The average value of (a) of (b),is the corresponding 3-dimensional vector for pixel k,is a scalar quantity corresponding to the pixel k.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.
Claims (7)
1. The utility model provides an intelligence license plate identification system which characterized in that, intelligence license plate identification system includes: the system comprises a license plate recognition unit, a parking lot charging unit and a passage entrance guard access unit;
the license plate recognition unit is used for acquiring license plate information and processing the license plate information;
the parking lot charging unit is electrically connected with the license plate recognition unit and is used for charging the parking of the vehicle corresponding to the license plate;
the access control access unit is electrically connected with the license plate recognition unit and is used for processing vehicle access channel information;
the license plate recognition unit includes: the system comprises an image acquisition module, a man-machine input module, a license plate positioning module, a character segmentation module and a character recognition module;
the image acquisition module is used for acquiring license plate image information; the method specifically comprises the following steps: constructing a Laplace matrix by using a mobile least square method instead of a least square method, and using a KNN neighborhood to replace a spatial neighborhood to obtain a linear relation of non-neighborhood pixels on an alpha image, thereby calculating a mobile Laplace matrix and obtaining the alpha image;
solving a cyclic covariant function of a received PSK signal containing Alpha stable distributed noise; fourier transform is carried out on the cyclic covariant function to obtain a cyclic covariant spectrum of the cyclic covariant function; extracting a section with the cycle frequency epsilon =0Hz according to the obtained cycle covariate spectrum; respectively searching peak values of positive and negative half shafts of the obtained section, finding positive and negative frequency values corresponding to the peak values, taking absolute values, and then calculating an average value to be used as carrier frequency estimation of PSK signals under Alpha stable distribution noise of carrier frequency estimation values and outputting the carrier frequency estimation; extracting the output image characteristic vector and outputting a finally acquired image;
the method for moving least square matting is as follows:
in a grayscale image, the window w i The alpha value in the neighborhood satisfies the local linear condition, and the local linear relation is solved by using a moving least square method, which is expressed as follows:
weight ω, ω in equation (1) i Is the neighborhood w k The weight value in (1); formula (1) is represented in the form of the following matrix:
for each neighborhood w k ,G k Is defined as | w k A | × 2 matrix; g k Each row comprising a vector (I) i ,1)，W k Is the weight value omega corresponding to each row vector i Vector of composition, G k ' is G k W of (2) k Weighting, the corresponding per-row vector is represented as (W) k .I i ,W k )，The vector is formed by alpha values corresponding to all pixels in the neighborhood;
coefficient a k ,b k The solution is as follows:
order toJ (α) is represented by the following formula:
δ i,j is the Kronecker delta function, mu k And σ 2 Respectively a small window w k Inner based on W k Weighted mean and variance, | w k II is the number of pixels in the window, L is the moving Las matting matrix;
the cyclic covariate function of the received signal comprises:
the signal contains an MPSK signal that obeys the S distributed noise, expressed as:
where E is the average power of the signal,M＝2 k m =1,2.. M, q (T) denotes a rectangular pulse waveform, T denotes a symbol period, f denotes a symbol period c Represents the carrier frequency, phi 0 Representing the initial phase, if w (t) is non-gaussian noise following a S α S distribution, its self-covariant function is defined as:
wherein (x (t- τ)) <p-1> ＝|x(t-τ) p-2 x*(t-τ)，γ x(t-τ) Is the dispersion coefficient of x (t), the cyclic covariance of x (t) is defined as:
where ε is the cycle frequency and T is one symbol period;
the image feature vector extraction method comprises the following specific steps:
step one, collecting N samples to be used as a training set X, and calculating a sample average value m by adopting the following formula:
wherein x is i E sample training set X = (X) 1 ，x 2 ，…，x N )；
Step two, obtaining a dispersion matrix S:
solving an eigenvalue lambada i of the dispersion matrix and a corresponding eigenvector ei, wherein the ei is a principal component, and sequentially arranging the eigenvalues of lambada 1, lambada 2 and … from large to small;
taking p values, λ 1, λ 2, …, λ p, determines the license plate space E = (E1, E2, …, eP), where the point in the training sample X where each element is projected into this space is given by:
x'i＝Etxi，t＝1，2，…，N；
the p-dimensional vector obtained by the formula is obtained by carrying out PCA dimensionality reduction on the original vector;
the feature extraction is based on sparse representation, and multi-image recognition is carried out by adopting an image recognition algorithm;
the specific method for carrying out multi-image recognition by the image recognition algorithm comprises the following steps:
detecting the license plates of the current frame and sequencing according to coordinates to obtain the recognition results of a plurality of license plates of the current frame; calculating the recognition results of the adjacent n frames of the corresponding license plate according to the recognition results of the license plates of the current frame; counting the identities of all license plates, and determining the final identity of a target by more than half n/2 of uniform identities;
wherein, the reconstruction error { r) between the picture to be identified and each category of the preset image library is calculated 1 ，r 2 ……r n }，r 1 <r 2 <……<r n The obtained similarity value is according toDetermining a final recognition result; wherein T is 1 Is a ratio value, T1=0.65;
the man-machine input module is electrically connected with the image acquisition module, can perform manual input when the image acquisition module is insensitive, and is favorable for preventing suspicious license plates from escaping punishment;
the license plate positioning module is electrically connected with the image acquisition module and can effectively position a license plate according to image acquisition information; the license plate positioning method of the license plate positioning module comprises the following steps: leaving a circumscribed matrix of each subspace as a suspected license plate area according to the size and the proportion of the license plate; setting a hopping function f (i, j), accurately positioning a suspected license plate region, and determining the upper and lower boundaries of the license plate region:
wherein c (i, j) is
c(i,j)＝LBP 8,1 (i,j)-LBP 8,1 (i,j-1)
In the above two formulas, i =1,2,3,4, … N, j =2,3,4, … M, so the number of transitions in any row i and S (i) are:
if the sum S (i is more than or equal to 12) of the jumping times of any line, the line can belong to the license plate area; scanning the whole image from top to bottom, finding out all rows meeting S (i is more than or equal to 12), and recording the row number i of the row; if continuous h rows meet S (i is more than or equal to 12), a rectangular area with the width of M and the height of h can be obtained, and the area can be a license plate area, so that areas without the characteristics in the vehicle image are excluded;
the character segmentation module is electrically connected with the license plate positioning module and is used for segmenting and reading characters on a license plate;
the character recognition module is electrically connected with the character segmentation module and is used for respectively recognizing the segmented characters and finally outputting a result;
the parking lot charging unit includes: the system comprises a card reader, a controller and a charging device;
the card reader is used for reading the vehicle owner identification card;
the controller deducts the fee of the card of the vehicle owner according to the fee displayed on the fee counter;
the access control unit comprises an RFID control unit and an electromagnetic door lock.
The RFID control unit adopts an RFID radio frequency chip which transmits a reading signal to the vehicle owner identification card,
the RFID control unit is connected with the electromagnetic door lock and used for controlling the opening and closing of the electromagnetic door lock.
2. The intelligent license plate recognition system of claim 1, wherein the cyclic covariate spectrum of the received signal is performed by:
the cyclic covariate spectrum is the fourier transform of the cyclic covariate function, expressed as:
the derivation of the cyclic covariate spectrum is as follows:
when M is greater than or equal to 4, inAt the position of the air compressor, the air compressor is started,
when the M =2, the signal strength of the signal is high,
wherein Q (f) is the Fourier transform of Q (t), and
3. the intelligent license plate recognition system of claim 1, wherein the carrier frequency estimation is achieved by extracting a cross section of a cyclic frequency e =0Hz in a cyclic covariate spectrum by:
the envelope of the cyclic covariate spectrum on the n =0, i.e. epsilon =0Hz section is:
when f = ± f c The envelope then takes a maximum value.
4. The intelligent license plate recognition system of claim 1, wherein a weight ω is introduced i The method is applied to a color model, and the moving least square matting method under the color model is as follows:
the linear relationship between the channels of the color image is represented by:
c is the number of channels of the color image, and after considering the information of each channel, equation (1) is converted into the following equation:
after the formula (2) is simplified, the mobile Laplace matrix under the color model is obtained by solving the following formula:
J(α)＝αLα T ；
in the formula (3), I is a matrix formed by 3*1 color vectors corresponding to all pixels in a small neighborhood, and is mu k W of I k Weighted average, Σ k Is I at W k Covariance matrix under weighting.
5. The intelligent license plate recognition system of claim 1, wherein the KNN neighborhood of the moving least squares matting method extends the spatial neighborhood in the Laplace matrix to a KNN neighborhood, the points of the KNN space being jointly determined by five dimensions (R, G, B, X, Y); and realizing efficient searching of the KNN neighborhood by using KD-TREE.
6. The intelligent license plate recognition system of claim 1, wherein the method for solving for the large kernel in the moving least squares cutout comprises: solving an alpha value by using a conjugate gradient method;
for equation Lx = b, the key of the conjugate gradient method is to construct a conjugate vector p and solve the corresponding residual error; the conjugate gradient method is solved by an iterative method, and in each iterative process, the new conjugate vector is solved by the following formula:
7. the intelligent license plate recognition system of claim 6, wherein the coefficients of the conjugate direction are solved by:
the new x value and residual are solved using the following equation:
solving the element q corresponding to the point i in the Lp vector by the following formula i ：
W k Is the neighborhood corresponding to pixel k, | w k II is the size of the neighborhood, i is the neighborhood W surrounding pixel k k One pixel of q i Is the I-th element of the q-vector, I i Representing three channels R, G, B, p for the 3-dimensional vector corresponding to pixel i i Is the element corresponding to pixel i in the conjugate vector, μ k Is a 3-dimensional vector, being the neighborhood W k In (II) i The mean value of the vector is calculated,is a neighborhood W k Conjugate vector p corresponding to element i in i The average value of (a) of (b),is the corresponding 3-dimensional vector of pixel k,is a scalar quantity corresponding to the pixel k.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710580152 | 2017-07-17 | ||
| CN2017105801527 | 2017-07-17 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107545265A true CN107545265A (en) | 2018-01-05 |
Family
ID=60957961
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710761984.9A Pending CN107545265A (en) | 2017-07-17 | 2017-08-30 | A kind of intelligent vehicle license plate recognition system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107545265A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108399946A (en) * | 2018-03-05 | 2018-08-14 | 湖北省第三人民医院 | A kind of nursing work load distribution assistance system |
| CN108460808A (en) * | 2018-03-20 | 2018-08-28 | 河南省洛阳正骨医院（河南省骨科医院） | Dynamic acquisition system for human body bone joint kinematics |
| CN108491706A (en) * | 2018-03-14 | 2018-09-04 | 黄冈职业技术学院 | A kind of multi-application computer and office equipment cabinet |
| CN108578138A (en) * | 2018-05-15 | 2018-09-28 | 山东中医药高等专科学校 | A kind of new-type surgical nursing dressing change device, control system and control method |
| CN108635153A (en) * | 2018-04-08 | 2018-10-12 | 鞠红梅 | A kind of occupational health diagnostic couch |
| CN108734148A (en) * | 2018-05-29 | 2018-11-02 | 河南牧业经济学院 | A kind of public arena image information collecting unmanned aerial vehicle control system based on cloud computing |
| CN111027363A (en) * | 2019-04-28 | 2020-04-17 | 刘晓英 | Auxiliary signal parameter analysis system |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103761504A (en) * | 2013-12-31 | 2014-04-30 | 江苏图云智能科技发展有限公司 | Face recognition system |
| CN103810474A (en) * | 2014-02-14 | 2014-05-21 | 西安电子科技大学 | Car plate detection method based on multiple feature and low rank matrix representation |
| CN104038454A (en) * | 2014-06-20 | 2014-09-10 | 西安电子科技大学 | Method for estimating carrier frequency of PSK (phase shift keying) signal in Alpha-stable distribution noise |
| CN104239862A (en) * | 2014-09-11 | 2014-12-24 | 中国电子科技集团公司第二十九研究所 | Face recognition method |
| CN104616011A (en) * | 2015-02-13 | 2015-05-13 | 中国人民解放军国防科学技术大学 | MRF (Multi-Reference Frame) license plate denoising algorithm based on combined apriorism of gradient information and block area |
| CN105096326A (en) * | 2015-08-13 | 2015-11-25 | 丽水学院 | Laplace cutout matrix method by using moving least square method |
| CN105631951A (en) * | 2016-02-25 | 2016-06-01 | 江苏中兴新泰物联网科技园有限公司 | Digital city vehicle management system |
| WO2016097890A1 (en) * | 2014-12-15 | 2016-06-23 | Airbus Group Singapore Pte. Ltd. | Automated method for selecting training areas of sea clutter and detecting ship targets in polarimetric synthetic aperture radar imagery |
| CN106650731A (en) * | 2016-12-23 | 2017-05-10 | 中山大学 | Robust license plate and logo recognition method |
-
2017
- 2017-08-30 CN CN201710761984.9A patent/CN107545265A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103761504A (en) * | 2013-12-31 | 2014-04-30 | 江苏图云智能科技发展有限公司 | Face recognition system |
| CN103810474A (en) * | 2014-02-14 | 2014-05-21 | 西安电子科技大学 | Car plate detection method based on multiple feature and low rank matrix representation |
| CN104038454A (en) * | 2014-06-20 | 2014-09-10 | 西安电子科技大学 | Method for estimating carrier frequency of PSK (phase shift keying) signal in Alpha-stable distribution noise |
| CN104239862A (en) * | 2014-09-11 | 2014-12-24 | 中国电子科技集团公司第二十九研究所 | Face recognition method |
| WO2016097890A1 (en) * | 2014-12-15 | 2016-06-23 | Airbus Group Singapore Pte. Ltd. | Automated method for selecting training areas of sea clutter and detecting ship targets in polarimetric synthetic aperture radar imagery |
| CN104616011A (en) * | 2015-02-13 | 2015-05-13 | 中国人民解放军国防科学技术大学 | MRF (Multi-Reference Frame) license plate denoising algorithm based on combined apriorism of gradient information and block area |
| CN105096326A (en) * | 2015-08-13 | 2015-11-25 | 丽水学院 | Laplace cutout matrix method by using moving least square method |
| CN105631951A (en) * | 2016-02-25 | 2016-06-01 | 江苏中兴新泰物联网科技园有限公司 | Digital city vehicle management system |
| CN106650731A (en) * | 2016-12-23 | 2017-05-10 | 中山大学 | Robust license plate and logo recognition method |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108399946A (en) * | 2018-03-05 | 2018-08-14 | 湖北省第三人民医院 | A kind of nursing work load distribution assistance system |
| CN108491706A (en) * | 2018-03-14 | 2018-09-04 | 黄冈职业技术学院 | A kind of multi-application computer and office equipment cabinet |
| CN108460808A (en) * | 2018-03-20 | 2018-08-28 | 河南省洛阳正骨医院（河南省骨科医院） | Dynamic acquisition system for human body bone joint kinematics |
| CN108635153A (en) * | 2018-04-08 | 2018-10-12 | 鞠红梅 | A kind of occupational health diagnostic couch |
| CN108578138A (en) * | 2018-05-15 | 2018-09-28 | 山东中医药高等专科学校 | A kind of new-type surgical nursing dressing change device, control system and control method |
| CN108734148A (en) * | 2018-05-29 | 2018-11-02 | 河南牧业经济学院 | A kind of public arena image information collecting unmanned aerial vehicle control system based on cloud computing |
| CN111027363A (en) * | 2019-04-28 | 2020-04-17 | 刘晓英 | Auxiliary signal parameter analysis system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107545265A (en) | A kind of intelligent vehicle license plate recognition system | |
| CN101246547B (en) | Method for detecting moving objects in video according to scene variation characteristic | |
| CN103310194B (en) | Pedestrian based on crown pixel gradient direction in a video shoulder detection method | |
| CN103455797B (en) | Detection and tracking method of moving small target in aerial shot video | |
| US7881531B2 (en) | Error propogation and variable-bandwidth mean shift for feature space analysis | |
| CN105528794A (en) | Moving object detection method based on Gaussian mixture model and superpixel segmentation | |
| CN106127812B (en) | A kind of passenger flow statistical method of the non-gate area in passenger station based on video monitoring | |
| Johnson et al. | AutoGAD: An improved ICA-based hyperspectral anomaly detection algorithm | |
| CN106557740B (en) | The recognition methods of oil depot target in a kind of remote sensing images | |
| CN102156702A (en) | Fast positioning method for video events from rough state to fine state | |
| CN111915583B (en) | Vehicle and pedestrian detection method based on vehicle-mounted thermal infrared imager in complex scene | |
| CN115620212B (en) | Behavior identification method and system based on monitoring video | |
| CN114627447A (en) | Road vehicle tracking method and system based on attention mechanism and multi-target tracking | |
| Mu et al. | Hyperspectral image classification based on active learning and spectral-spatial feature fusion using spatial coordinates | |
| CN105138987A (en) | Vehicle detection method based on aggregation channel characteristic and motion estimation | |
| Mo et al. | Vehicles detection in traffic flow | |
| CN106529472B (en) | Object detection method and device based on large scale high-resolution high spectrum image | |
| Song et al. | Feature extraction and target recognition of moving image sequences | |
| CN108133218A (en) | Infrared target detection method, equipment and medium | |
| CN105512622A (en) | Visible remote-sensing image sea-land segmentation method based on image segmentation and supervised learning | |
| Wang et al. | Precise localization of concealed objects in millimeter-wave images via semantic segmentation | |
| CN110458064B (en) | Low-altitude target detection and identification method combining data driving type and knowledge driving type | |
| Lee et al. | Real-time automatic vehicle management system using vehicle tracking and car plate number identification | |
| CN107895162A (en) | Saliency algorithm of target detection based on object priori | |
| CN110516666A (en) | The license plate locating method combined based on MSER and ISODATA |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB02 | Change of applicant information |
Address after: Room 506, Huafeng Building, 225 Zhongxing Road, Yuecheng District, Shaoxing City, Zhejiang Province Applicant after: Zhejiang Zhishen Science and Technology Co.,Ltd. Address before: Room 402-1, No. 225, Zhongxing Zhongxing Road, Shaoxing, Zhejiang Province, 312000 Applicant before: ZHEJIANG ZHISHEN DIGITAL TECHNOLOGY Co.,Ltd. |
|
| CB02 | Change of applicant information | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180105 |
|
| RJ01 | Rejection of invention patent application after publication |