CN107305634A - A kind of license plate locating method returned based on integrated random fern and shape - Google Patents

A kind of license plate locating method returned based on integrated random fern and shape Download PDF

Info

Publication number
CN107305634A
CN107305634A CN201610554203.4A CN201610554203A CN107305634A CN 107305634 A CN107305634 A CN 107305634A CN 201610554203 A CN201610554203 A CN 201610554203A CN 107305634 A CN107305634 A CN 107305634A
Authority
CN
China
Prior art keywords
mrow
msub
shape
license plate
random
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
Application number
CN201610554203.4A
Other languages
Chinese (zh)
Inventor
不公告发明人
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hangzhou Ling Technology Co Ltd
Original Assignee
Hangzhou Ling Technology Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Hangzhou Ling Technology Co Ltd filed Critical Hangzhou Ling Technology Co Ltd
Publication of CN107305634A publication Critical patent/CN107305634A/en
Pending legal-status Critical Current

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/20Image preprocessing
    • G06V10/25Determination of region of interest [ROI] or a volume of interest [VOI]
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/21Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/24Classification techniques
    • G06F18/241Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411Classification 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
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/40Extraction of image or video features
    • G06V10/50Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00Scenes; Scene-specific elements
    • G06V20/60Type of objects
    • G06V20/62Text, e.g. of license plates, overlay texts or captions on TV images
    • G06V20/625License plates

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Multimedia (AREA)
  • Other Investigation Or Analysis Of Materials By Electrical Means (AREA)

Abstract

The present invention provides a kind of license plate locating method returned based on integrated random fern and shape, this method 1) extract candidate license plate region using integrated random fern algorithm;2) zoom to normal size, extract histograms of oriented gradients feature, non-license plate area is excluded using SVMs;3) method returned using shape obtains the coordinate on four summits of car plate.By the processing of this method, normalized car plate relatively easily can be obtained using four obtained summits.The present invention improves the car plate detection rate in the case of complex background using integrated random fern, while the method returned using shape significantly simplify License Plate Character Segmentation and the identification in later stage, lifts Car license recognition effect.

Description

A kind of license plate locating method returned based on integrated random fern and shape
Technical field
The present invention relates to a kind of method of License Plate, and in particular to a kind of car returned based on integrated random fern and shape Board localization method.
Background technology
The identification of the number-plate number is a comprehensive crossover for being related to Digital Image Processing, computer vision and pattern-recognition Science.By identifying Car license recognition the identity of vehicle, there is incomparable superiority in vehicle management, stop large-scale The places such as parking lot, cell and school district vehicle registration, gather around and have broad application prospects.
License plate recognition technology includes License Plate and Car license recognition two large divisions.License Plate is the basis of Car license recognition, The performance quality of License Plate has important influence to whole Vehicle License Plate Recognition System.
Existing license plate locating method mainly has:Based on statistics edge saltus step and morphologic location algorithm, based on color The location algorithm of segmentation and location algorithm based on textural characteristics etc..However, the popular feature of these algorithms is environment-adapting ability Difference, runs into the situation that illumination variation, car plate are smudgy and background level is complicated, or be transplanted in addition by a kind of working environment During a kind of working environment, the fluctuation that can not be estimated occurs in License Plate accuracy rate.It follows that the Shandong of existing License Plate Rod can not meet the requirement to vehicle license location technique under complex background.
The content of the invention
The present invention combines the present Research of domestic License Plate at present, considers the excellent of existing algorithm of locating license plate of vehicle Shortcoming, proposes a kind of new targeting scheme based on machine learning, that is, utilizes integrated random fern (Boosted Random Ferns, BRFs) and SVMs (SVM) Cascade algorithms positioning licence plate position and utilization shape regression algorithm acquisition car plate Four summits.The problem of present invention can effectively solve traditional algorithm of locating license plate of vehicle poor robustness, and to a certain extent Simplify the subsequent step of Car license recognition.
Implementer case of the present invention in current problem is solved is as follows:
A kind of face sheltering detection system based on deep learning, the detection algorithm comprises the following steps:
1) integrated random fern algorithm extracts candidate license plate region:1. gray proces are carried out to the license plate image got, returned One change is handled.Because histograms of oriented gradients (HOG) feature rotates and to be stained sensitiveness weak to illumination condition, car plate, therefore use HOG feature construction Weak Classifiers.Integrated random fern algorithm is, by the built-up strong classifier of multiple Weak Classifiers, can to integrate The local feature of picture structure is embodied, there is good classification ability.2. many points are carried out using the framework of sliding window to license plate image Resolution is detected, each window is classified using integrated random fern algorithm, retains the window for being divided into positive class, and window is carried out Multiresolution merges, and obtains the license plate area of candidate;
2) candidate region got to previous step, uniformly zooms to normal size, extracts HOG features, uses svm classifier Device excludes non-license plate area:
3) license plate area that previous step is got is returned using shape regression algorithm, is accurately obtained four tops of car plate Point coordinates;
In order that the object, technical solutions and advantages of the present invention are clearer, illustrate below in conjunction with the accompanying drawings to present invention work It is described in further detail.
Illustrate that with reference to accompanying drawing 1 and accompanying drawing 2 the car plate coarse positioning algorithm of integrated random fern algorithm is implemented:
This algorithm carries out coarse positioning to car plate using integrated random fern algorithm (BRFs), and BRFs is as strong classifier comprising many Individual Weak Classifier, is expressed as:
Wherein ht(x) it is Weak Classifier, βcFor strong classifier threshold value.Weak Classifier ht(x) random fern grader is used, is gathered around There is F characteristic point pair, each characteristic point two random characters to being made up of.Each characteristic point mapping relations following to setting up:
WhereinWithIt is characterized two random characters in space.
Therefore, feature space is divided into 2 by random fern graderFIndividual space, each space output 0 or 1.So constitute F length binary string, try to achieve the decimal value.Therefore random fern grader is set up following binary system and closed with the decimal system System:
z:F(2)→z(10) (3)
And random fern grader can use Bayes's relation to represent:
Wherein C represents car plate classification, fiFeature pair is represented, B represents background classification
The foundation that pick-up board classification and the other log ratio of background classes are classified as random fern, while assuming uniform priori Probability P (C)=P (B), and remove P (f1, f2..., fF), above-mentioned relation is changed into:
And be expressed as by F observed value:
Wherein k is F observed value, giRepresent the locus of image
In summary, random fern grader is expressed as:
The structure and application BRFs that this algorithm includes BRFs carry out two parts of detection:
1st, the structure of BRFs graders
1) N number of car plate sample image is prepared as positive sample, M non-car plate backgrounds are used as negative sample, composing training sample Set;
2) by all sample gray processings, and 14 × 42 sizes are zoomed to;
3) to all sample extraction HOG features;
4) the Weak Classifier pond being made up of random fern Weak Classifier is built.The training iteration of integrated random fern grader is set Number of times is T, and is iterated.
A) first iteration when, the probability for initializing each sample is:
Dt(i)=1/ (N+M) (9)
B) the z Distribution values of the sample under present weight are calculated:
Selection makes the minimum Weak Classifier of lower face amount as the optimal classification device of this iteration from Weak Classifier pond, is put into In strong classifier.
C) current sample weights are updated, and continue iteration:
5) final strong classifier is obtained:
2nd, detected using BRFs
1) license plate image got is set as I, by I gray processings, obtains IgGray level image;
2) first order is positioned:In order to improve accuracy of identification, I is described with image pyramidg, and it is divided into five groups;It is basic herein On each group of image be placed in metric space portrayed, be further divided into five layers;On each layer, sliding window model is set up, is performed Below step:
A) I is scanned with the window of 14 × 42 sizesg, obtain great amount of images window area Igw
B) using the BRFs graders obtained when training to IgwClassified;Candidate license plate is classified as be categorized as positive class Window area.
3) multiresolution merging is carried out to candidate region obtained above, obtains the car plate window area of candidate.With reference to Fig. 1 Illustrate that non-license plate area excludes implementing for algorithm with Fig. 2:
Non- license plate area may be contained for the candidate license plate region that integrated random fern algorithm is obtained, it is therefore desirable to exclude non- License plate area.The algorithm includes the training and prediction of SVM classifier.
The step of training, is as follows:
1) car plate for reusing coarse positioning collection is positive sample;The image without car plate is collected, using car plate coarse positioning Algorithm, obtained candidate license plate window is as negative sample;
2) all samples are zoomed into 24 × 84 sizes;
3) to all sample extraction HOG features;
4) using the SVM training graders with RBF kernel functions.
Prediction steps are as follows:
1) the candidate license plate region for obtaining coarse positioning, uniformly zooms to 28 × 84 sizes:
2) HOG features are extracted;
3) SVM classifier that application training is obtained is classified, and grader is classified as into license plate area for positive class.With reference to figure 1 and Fig. 2 illustrates implementing for car plate shape regression algorithm.
Four summits for obtaining car plate, this algorithm uses shape regression algorithm.Four vertex representations of car plate are shape S =[x1, y1, x2, y2, x3, y3, x4, y4], license plate image is given, the target of recurrence is estimation car plate shapeSo thatIt is minimum.
This regression algorithm returns device R using integrated T strong random ferntAlgorithm, each strong random fern return device by K it is weak with Machine fern returns deviceComposition.For giving license plate image I and initial car plate shape S0(can be the average shape of training set, Can also be the shape of training sample at random), each weak random fern returns device iteration and updates car plate shape:
Wherein,I-th of weak random fern in device is returned for the strong random ferns of t and returns device, updates former shape St-1To new shape Shape St
Weak random fern returns device and feature space is divided into 2FIndividual space, each space exports the renewal δ S of shapeb.For Each weak random fern returns the output valve δ S of each division spatially on devicebNeed to calculate in training and obtain.According to following Formula calculate obtain:
Wherein SiFormer shape is represented, this non trivial solution is:
It is above-mentioned to be changed in order to prevent over-fitting:
Wherein, β is free parameter, and this algorithm is 1000, automatically controls over-fitting degree
Return device algorithm and include training and two steps of regression forecasting:
1st, the step of training is as follows:
1) a large amount of images containing car plate are collected, four summits that car plate is calibrated by hand constitute car plate shape S, and use Algorithm of locating license plate of vehicle obtains the rectangle frame of car plate, then rectangle frame and car plate shape of each training sample by image, comprising car plate Constitute;
2) histogram equalization processing is carried out to all sample images;
3) for original shape of the random selected shape from other samples of each sample as this sample.All shapes The normalization of reference axis is carried out according to car plate rectangle frame;
4) repetitive exercise T strong random fern returns device, and each strong random fern returns device and returns device comprising K weak random ferns.
A it is) random that 400 points are selected from license plate area, so as to constitute 4002Individual point pair.Calculate regression residuals and be used as mesh Mark, the accumulation regression forecasting value for initializing each sample is 0 shape, and K weak random ferns of repetitive exercise return device.
A) by the regression residuals target projection of all samples to random direction, a series of scalars are obtained;
B) choose F point pair successively from all-pair using the feature selecting algorithm based on correlation, each put to Pixel difference and a) in scalar variance it is maximum;
C) to F point of acquisition to choosing random threshold value;
D) recurrence output of all division spaces in current sample set that this weak random fern returns device is calculated;
E) obtain weak random fern and return device, and update accumulation regression forecasting value.
B) obtain strong random fern and return device, and update regression residuals target.
5) final recurrence device is obtained.
2nd, the step of regression forecasting is as follows:
1) each car plate candidate rectangle region obtained for location algorithm, carries out car plate regression forecasting car plate shape;
2) random that 5 initial car plate shapes are selected from training sample, car plate shape carries out the reference axis of rectangular area Normalization;
3) it is predicted using recurrence device obtained above, obtains 5 regression results, try to achieve average value, it is as final Car plate shape.
Brief description of the drawings
Fig. 1 is the flow chart of whole algorithm
Fig. 2 is algorithm steps design sketch.

Claims (7)

1. a kind of license plate locating method returned based on integrated random fern and shape, it is characterised in that comprise the following steps:
1) integrated random fern algorithm extracts candidate license plate region;
2) the non-license plate candidate areas of SVM based on HOG features are excluded:
3) four apex coordinates of car plate are obtained using shape regression algorithm.
2. a kind of license plate locating method returned based on integrated random fern and shape according to claim 1, its feature is existed In described integrated random fern algorithm extracts candidate license plate region and comprised the following steps:
1) gray proces, normalized are carried out to the license plate image got;
2) multiresolution detection is carried out using the framework of sliding window to license plate image.Set up license plate image pyramid;
3) for each tomographic image of image pyramid, HOG characteristic response figures are extracted, HOG is used and arrived without symbol gradient, i.e., 0 degree 180 degree, cell sizes are 6*6, and direction number during statistic histogram is 4;
4) it is scanned using size for 14*42 sliding window on characteristic response figure, obtains a large amount of windows;
5) each window is classified using integrated random fern algorithm, retains the window for being divided into positive class;
6) multiresolution merging is carried out to window obtained above, obtains the license plate area of candidate.
3. integrated random fern algorithm according to claim 2 extracts candidate license plate region, it is characterised in that described use Integrated random fern algorithm carries out classification to each window includes structure and the prediction of integrated random fern.
Integrated random fern includes multiple Weak Classifiers as strong classifier, is expressed as:
Wherein ht(x) it is Weak Classifier, βcFor strong classifier threshold value.Weak Classifier ht(x) random fern grader is used, possesses F Characteristic point pair, each characteristic point two random characters to being made up of.Each characteristic point mapping relations following to setting up:
<mrow> <mi>f</mi> <mrow> <mo>(</mo> <mi>x</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfenced open = "{" close = ""> <mtable> <mtr> <mtd> <mrow> <mn>1.</mn> <msub> <mi>x</mi> <msub> <mi>&amp;Omega;</mi> <mi>i</mi> </msub> </msub> <mo>&gt;</mo> <msub> <mi>x</mi> <msub> <mi>&amp;Omega;</mi> <mi>j</mi> </msub> </msub> </mrow> </mtd> </mtr> <mtr> <mtd> <mrow> <mn>0</mn> <mo>,</mo> <mi>o</mi> <mi>t</mi> <mi>h</mi> <mi>e</mi> <mi>r</mi> </mrow> </mtd> </mtr> </mtable> </mfenced> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>2</mn> <mo>)</mo> </mrow> </mrow>
WhereinWithIt is characterized two random characters in space.
Therefore, feature space is divided into 2 by random fern graderFIndividual space, each space output 0 or 1.The F length so constituted The binary string of degree, tries to achieve the decimal value.Therefore random fern grader sets up following binary system and decimal system relation:
z:F(2)→z(10) (3)
And random fern grader can use Bayes's relation to represent:
<mrow> <mi>P</mi> <mrow> <mo>(</mo> <mi>C</mi> <mo>|</mo> <msub> <mi>f</mi> <mn>1</mn> </msub> <mo>,</mo> <msub> <mi>f</mi> <mn>2</mn> </msub> <mo>,</mo> <mo>...</mo> <mo>,</mo> <msub> <mi>f</mi> <mi>F</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mrow> <mi>P</mi> <mrow> <mo>(</mo> <msub> <mi>f</mi> <mn>1</mn> </msub> <mo>,</mo> <msub> <mi>f</mi> <mn>2</mn> </msub> <mo>,</mo> <mo>...</mo> <mo>,</mo> <msub> <mi>f</mi> <mi>F</mi> </msub> <mo>|</mo> <mi>C</mi> <mo>)</mo> </mrow> <mi>P</mi> <mrow> <mo>(</mo> <mi>C</mi> <mo>)</mo> </mrow> </mrow> <mrow> <mi>P</mi> <mrow> <mo>(</mo> <msub> <mi>f</mi> <mn>1</mn> </msub> <mo>,</mo> <msub> <mi>f</mi> <mn>2</mn> </msub> <mo>,</mo> <mo>...</mo> <mo>,</mo> <msub> <mi>f</mi> <mi>F</mi> </msub> <mo>)</mo> </mrow> </mrow> </mfrac> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>4</mn> <mo>)</mo> </mrow> </mrow>
<mrow> <mi>P</mi> <mrow> <mo>(</mo> <mi>B</mi> <mo>|</mo> <msub> <mi>f</mi> <mn>1</mn> </msub> <mo>,</mo> <msub> <mi>f</mi> <mn>2</mn> </msub> <mo>,</mo> <mo>...</mo> <mo>,</mo> <msub> <mi>f</mi> <mi>F</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mrow> <mi>P</mi> <mrow> <mo>(</mo> <msub> <mi>f</mi> <mn>1</mn> </msub> <mo>,</mo> <msub> <mi>f</mi> <mn>2</mn> </msub> <mo>,</mo> <mo>...</mo> <mo>,</mo> <msub> <mi>f</mi> <mi>F</mi> </msub> <mo>|</mo> <mi>B</mi> <mo>)</mo> </mrow> <mi>P</mi> <mrow> <mo>(</mo> <mi>B</mi> <mo>)</mo> </mrow> </mrow> <mrow> <mi>P</mi> <mrow> <mo>(</mo> <msub> <mi>f</mi> <mn>1</mn> </msub> <mo>,</mo> <msub> <mi>f</mi> <mn>2</mn> </msub> <mo>,</mo> <mo>...</mo> <mo>,</mo> <msub> <mi>f</mi> <mi>F</mi> </msub> <mo>)</mo> </mrow> </mrow> </mfrac> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>5</mn> <mo>)</mo> </mrow> </mrow>
Wherein C represents car plate classification, fiFeature pair is represented, B represents background classification.
The foundation that pick-up board classification and the other log ratio of background classes are classified as random fern, while assuming uniform prior probability P (C)=P (B), and remove P(f1, f2..., fF), above-mentioned relation is changed into:
<mrow> <mi>log</mi> <mfrac> <mrow> <mi>P</mi> <mrow> <mo>(</mo> <mi>C</mi> <mo>|</mo> <msub> <mi>f</mi> <mn>1</mn> </msub> <mo>,</mo> <msub> <mi>f</mi> <mn>2</mn> </msub> <mo>,</mo> <mo>&amp;CenterDot;</mo> <mo>&amp;CenterDot;</mo> <mo>&amp;CenterDot;</mo> <mo>,</mo> <msub> <mi>f</mi> <mi>F</mi> </msub> <mo>)</mo> </mrow> </mrow> <mrow> <mi>P</mi> <mrow> <mo>(</mo> <mi>B</mi> <mo>|</mo> <msub> <mi>f</mi> <mn>1</mn> </msub> <mo>,</mo> <msub> <mi>f</mi> <mn>2</mn> </msub> <mo>,</mo> <mo>...</mo> <mo>,</mo> <msub> <mi>f</mi> <mi>F</mi> </msub> <mo>)</mo> </mrow> </mrow> </mfrac> <mo>=</mo> <mi>log</mi> <mfrac> <mrow> <mi>P</mi> <mrow> <mo>(</mo> <msub> <mi>f</mi> <mn>1</mn> </msub> <mo>,</mo> <msub> <mi>f</mi> <mn>2</mn> </msub> <mo>,</mo> <mo>...</mo> <mo>,</mo> <msub> <mi>f</mi> <mi>F</mi> </msub> <mo>|</mo> <mi>C</mi> <mo>)</mo> </mrow> </mrow> <mrow> <mi>P</mi> <mrow> <mo>(</mo> <msub> <mi>f</mi> <mn>1</mn> </msub> <mo>,</mo> <msub> <mi>f</mi> <mn>2</mn> </msub> <mo>,</mo> <mo>...</mo> <mo>,</mo> <msub> <mi>f</mi> <mi>F</mi> </msub> <mo>|</mo> <mi>B</mi> <mo>)</mo> </mrow> </mrow> </mfrac> <mo>=</mo> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>F</mi> </munderover> <mi>log</mi> <mfrac> <mrow> <mi>P</mi> <mrow> <mo>(</mo> <msub> <mi>f</mi> <mi>i</mi> </msub> <mo>|</mo> <mi>C</mi> <mo>)</mo> </mrow> </mrow> <mrow> <mi>P</mi> <mrow> <mo>(</mo> <msub> <mi>f</mi> <mi>i</mi> </msub> <mo>|</mo> <mi>B</mi> <mo>)</mo> </mrow> </mrow> </mfrac> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>6</mn> <mo>)</mo> </mrow> </mrow>
And be expressed as by F observed value:
<mrow> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>F</mi> </munderover> <mi>log</mi> <mfrac> <mrow> <mi>P</mi> <mrow> <mo>(</mo> <msub> <mi>z</mi> <mi>i</mi> </msub> <mo>=</mo> <mi>k</mi> <mo>|</mo> <mi>C</mi> <mo>,</mo> <msub> <mi>g</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> </mrow> <mrow> <mi>P</mi> <mrow> <mo>(</mo> <msub> <mi>z</mi> <mi>i</mi> </msub> <mo>=</mo> <mi>k</mi> <mo>|</mo> <mi>B</mi> <mo>,</mo> <msub> <mi>g</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> </mrow> </mfrac> <mo>,</mo> <mi>k</mi> <mo>=</mo> <mn>1</mn> <mo>,</mo> <mn>2</mn> <mo>,</mo> <mo>...</mo> <mo>,</mo> <mi>F</mi> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>7</mn> <mo>)</mo> </mrow> </mrow>
Wherein k is F observed value, giRepresent the locus of image.
In summary, random fern grader is expressed as:
<mrow> <msub> <mi>h</mi> <mi>t</mi> </msub> <mrow> <mo>(</mo> <mi>x</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mn>1</mn> <mn>2</mn> </mfrac> <mi>log</mi> <mfrac> <mrow> <mi>P</mi> <mrow> <mo>(</mo> <msub> <mi>z</mi> <mi>t</mi> </msub> <mo>=</mo> <mi>k</mi> <mo>|</mo> <mi>C</mi> <mo>,</mo> <msub> <mi>g</mi> <mi>t</mi> </msub> <mo>)</mo> </mrow> </mrow> <mrow> <mi>P</mi> <mrow> <mo>(</mo> <msub> <mi>z</mi> <mi>t</mi> </msub> <mo>=</mo> <mi>k</mi> <mo>|</mo> <mi>B</mi> <mo>,</mo> <msub> <mi>g</mi> <mi>t</mi> </msub> <mo>)</mo> </mrow> </mrow> </mfrac> <mo>,</mo> <mi>k</mi> <mo>=</mo> <mn>1</mn> <mo>,</mo> <mn>2</mn> <mo>,</mo> <mo>...</mo> <mo>,</mo> <mi>F</mi> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>8</mn> <mo>)</mo> </mrow> </mrow>
1) structure of integrated random fern grader
(1) N number of car plate sample image is prepared as positive sample, M non-car plate backgrounds are used as negative sample, composing training sample set Close;
(2) by all sample gray processings, and 14 × 42 sizes are zoomed to;
(3) to all sample extraction HOG features;
(4) the Weak Classifier pond being made up of random fern Weak Classifier is built.The training iteration time of integrated random fern grader is set Number is T, and is iterated;
A) first iteration when, the probability for initializing each sample is:
Dt(i)=1/ (N+M) (9) b) the z Distribution values of the sample under present weight are calculated,
<mrow> <mi>P</mi> <mrow> <mo>(</mo> <msub> <mi>z</mi> <mi>t</mi> </msub> <mo>=</mo> <mi>k</mi> <mo>|</mo> <mi>C</mi> <mo>,</mo> <msub> <mi>g</mi> <mi>t</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <munder> <mi>&amp;Sigma;</mi> <mrow> <mi>i</mi> <mo>:</mo> <msub> <mi>z</mi> <mi>t</mi> </msub> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mi>k</mi> </mrow> </munder> <msub> <mi>D</mi> <mi>t</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>)</mo> </mrow> <mo>,</mo> <mi>k</mi> <mo>=</mo> <mn>1</mn> <mo>,</mo> <mo>...</mo> <mo>,</mo> <mi>K</mi> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>10</mn> <mo>)</mo> </mrow> </mrow>
C) selection makes following Q from Weak Classifier pondtIt is worth minimum Weak Classifier as the optimal classification device of this iteration, is put into In strong classifier.
<mrow> <msub> <mi>Q</mi> <mi>t</mi> </msub> <mo>=</mo> <mn>2</mn> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>k</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>K</mi> </munderover> <msqrt> <mrow> <mi>P</mi> <mrow> <mo>(</mo> <msub> <mi>z</mi> <mi>i</mi> </msub> <mo>=</mo> <mi>k</mi> <mo>|</mo> <mi>C</mi> <mo>,</mo> <msub> <mi>g</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mi>P</mi> <mrow> <mo>(</mo> <msub> <mi>z</mi> <mi>i</mi> </msub> <mo>=</mo> <mi>k</mi> <mo>|</mo> <mi>C</mi> <mo>,</mo> <msub> <mi>g</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> </mrow> </msqrt> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>11</mn> <mo>)</mo> </mrow> </mrow>
D) current sample weights are updated, and continue iteration:
<mrow> <msub> <mi>D</mi> <mrow> <mi>t</mi> <mo>+</mo> <mn>1</mn> </mrow> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mrow> <msub> <mi>D</mi> <mi>t</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>)</mo> </mrow> <mi>exp</mi> <mo>&amp;lsqb;</mo> <mo>-</mo> <msub> <mi>y</mi> <mi>i</mi> </msub> <msub> <mi>h</mi> <mi>t</mi> </msub> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mo>&amp;rsqb;</mo> </mrow> <mrow> <msubsup> <mi>&amp;Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mrow> <mi>N</mi> <mo>+</mo> <mi>M</mi> </mrow> </msubsup> <msub> <mi>D</mi> <mi>t</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>)</mo> </mrow> <mi>exp</mi> <mo>&amp;lsqb;</mo> <mo>-</mo> <msub> <mi>y</mi> <mi>i</mi> </msub> <msub> <mi>h</mi> <mi>t</mi> </msub> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mo>&amp;rsqb;</mo> </mrow> </mfrac> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>12</mn> <mo>)</mo> </mrow> </mrow>
(5) final strong classifier is obtained:
<mrow> <mi>E</mi> <mrow> <mo>(</mo> <mi>x</mi> <mo>)</mo> </mrow> <mo>=</mo> <mi>s</mi> <mi>i</mi> <mi>g</mi> <mi>n</mi> <mrow> <mo>(</mo> <msubsup> <mi>&amp;Sigma;</mi> <mrow> <mi>t</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>T</mi> </msubsup> <msub> <mi>h</mi> <mi>t</mi> </msub> <mo>(</mo> <mi>x</mi> <mo>)</mo> <mo>-</mo> <msub> <mi>&amp;beta;</mi> <mi>e</mi> </msub> <mo>)</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>13</mn> <mo>)</mo> </mrow> </mrow>
2) each window that sliding window detection framework is produced is predicted using random fern grader.
4. a kind of license plate locating method returned based on integrated random fern and shape according to claim 1, its feature is existed In the non-license plate candidate areas of the SVM based on HOG features, which are excluded, to be comprised the following steps:
1) for the step 1 described in claim 1) the candidate license plate region that gets, uniformly zoom to 28*84 sizes;
2) HOG features are extracted;
3) classified using SVM classifier, exclude the candidate region for being divided into negative class, obtain license plate area.
5. the non-license plate candidate areas of the SVM according to claim 4 based on HOG features are excluded, it is characterised in that described Include the training and prediction of SVM classifier using SVM classifier classification.
1) it is positive sample to reuse the car plate collected by the structure of integrated random fern grader;The image without car plate is collected, Application integration random fern algorithm extracts candidate license plate zone algorithm, and obtained candidate license plate window is as negative sample;
2) all samples are zoomed into 28 × 84 sizes;
3) to all sample extraction HOG features;
4) using the SVM training graders with RBF kernel functions.
SVM classifier prediction steps are as follows:
1) the candidate license plate region for obtaining coarse positioning, uniformly zooms to 28 × 84 sizes;
2) HOG features are extracted;
3) SVM classifier that application training is obtained is classified, and grader is classified as into license plate area for positive class.
6. a kind of license plate locating method returned based on integrated random fern and shape according to claim 1, its feature is existed In the step of described use shape regression algorithm obtains four apex coordinates of car plate is as follows:
1) for the step 2 described in claim 1) obtained each license plate area, carry out car plate regression forecasting;
2) random that 5 initial car plate shapes are selected from training sample, car plate shape carries out the reference axis normalizing of rectangular area Change;
3) shape regression algorithm is used, 5 regression results is obtained, tries to achieve average value, as final car plate shape.
7. use shape regression algorithm according to claim 6 obtains four apex coordinates of car plate, it is characterised in that described Shape regression algorithm include return device structure and prediction.
Four vertex representations of car plate are shape S=[x1, y1, x2, y2, x3, y3, x4, y4], license plate image is given, the target of recurrence is Estimate car plate shapeSo thatIt is minimum.
This shape regression algorithm returns device R using integrated T strong random ferntAlgorithm, each strong random fern return device by K it is weak with Machine fern returns deviceComposition.For give license plate image I and initial car plate shape (can be the average shape of training set, Can be the shape of training sample at random), each weak random fern returns device iteration and updates car plate shape:
<mrow> <msup> <mi>S</mi> <mi>t</mi> </msup> <mo>=</mo> <msup> <mi>S</mi> <mrow> <mi>t</mi> <mo>-</mo> <mn>1</mn> </mrow> </msup> <mo>+</mo> <msubsup> <mi>r</mi> <mi>i</mi> <mi>t</mi> </msubsup> <mrow> <mo>(</mo> <mi>I</mi> <mo>,</mo> <msup> <mi>S</mi> <mi>t</mi> </msup> <mo>)</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>14</mn> <mo>)</mo> </mrow> </mrow>
Wherein,I-th of weak random fern in device is returned for the strong random ferns of t and returns device, updates former shape St-1To new shape St
Weak random fern returns device and feature space is divided into 2FIndividual space, each space exports the renewal δ S of shapeb.For each Weak random fern returns the output valve δ S of each division spatially on devicebNeed to calculate in training and obtain.According to following public affairs Formula is calculated and obtained:
<mrow> <msub> <mi>&amp;delta;S</mi> <mi>b</mi> </msub> <mo>=</mo> <mi>arg</mi> <munder> <mi>min</mi> <mrow> <mi>i</mi> <mo>&amp;Element;</mo> <msub> <mi>&amp;Omega;</mi> <mi>b</mi> </msub> </mrow> </munder> <munder> <mo>&amp;Sigma;</mo> <mrow> <mi>i</mi> <mo>&amp;Element;</mo> <msub> <mi>&amp;Omega;</mi> <mi>b</mi> </msub> </mrow> </munder> <mo>|</mo> <mo>|</mo> <mover> <msub> <mi>S</mi> <mi>i</mi> </msub> <mo>^</mo> </mover> <mo>-</mo> <mrow> <mo>(</mo> <msub> <mi>S</mi> <mi>i</mi> </msub> <mo>+</mo> <mi>&amp;delta;</mi> <mi>S</mi> <mo>)</mo> </mrow> <mo>|</mo> <mo>|</mo> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>15</mn> <mo>)</mo> </mrow> </mrow> 3
Wherein SiFormer shape is represented, this non trivial solution is:
<mrow> <msub> <mi>&amp;delta;S</mi> <mi>b</mi> </msub> <mo>=</mo> <mfrac> <mrow> <msub> <mi>&amp;Sigma;</mi> <mrow> <mi>i</mi> <mo>&amp;Element;</mo> <msub> <mi>&amp;Omega;</mi> <mi>b</mi> </msub> </mrow> </msub> <mrow> <mo>(</mo> <mover> <msub> <mi>S</mi> <mi>i</mi> </msub> <mo>^</mo> </mover> <mo>-</mo> <msub> <mi>S</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> </mrow> <mrow> <mo>|</mo> <msub> <mi>&amp;Omega;</mi> <mi>b</mi> </msub> <mo>|</mo> </mrow> </mfrac> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>16</mn> <mo>)</mo> </mrow> </mrow>
It is above-mentioned to be changed in order to prevent over-fitting:
<mrow> <msub> <mi>&amp;delta;S</mi> <mi>b</mi> </msub> <mo>=</mo> <mfrac> <mn>1</mn> <mrow> <mn>1</mn> <mo>+</mo> <mi>&amp;beta;</mi> <mo>/</mo> <mo>|</mo> <msub> <mi>&amp;Omega;</mi> <mi>b</mi> </msub> <mo>|</mo> </mrow> </mfrac> <mfrac> <mrow> <msub> <mi>&amp;Sigma;</mi> <mrow> <mi>i</mi> <mo>&amp;Element;</mo> <msub> <mi>&amp;Omega;</mi> <mi>b</mi> </msub> </mrow> </msub> <mrow> <mo>(</mo> <mover> <msub> <mi>S</mi> <mi>i</mi> </msub> <mo>^</mo> </mover> <mo>-</mo> <msub> <mi>S</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> </mrow> <mrow> <mo>|</mo> <msub> <mi>&amp;Omega;</mi> <mi>b</mi> </msub> <mo>|</mo> </mrow> </mfrac> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>17</mn> <mo>)</mo> </mrow> </mrow>
Wherein, β is free parameter, and this algorithm is 1000, automatically controls over-fitting degree.
Return device algorithm and include training and two steps of regression forecasting:
1) the step of training is as follows:
(1) a large amount of images containing car plate are collected, four summits that car plate is calibrated by hand constitute car plate shape, and use car plate Location algorithm obtains the rectangle frame of car plate, then rectangle frame and car plate shape of each training sample by image, comprising car plate are constituted;
(2) histogram equalization processing is carried out to all sample images;
(3) for original shape of the random selected shape from other samples of each sample as this sample.All shapes according to The normalization of reference axis is carried out according to car plate rectangle frame;
(4) repetitive exercise T strong random fern returns device, and each strong random fern returns device and returns device comprising K weak random ferns.
A it is) random that 400 points are selected from license plate area, so as to constitute 4002Individual point pair.Regression residuals are calculated as target, initially The accumulation regression forecasting value for changing each sample is 0 shape, and K weak random ferns of repetitive exercise return device.
A) by the regression residuals target projection of all samples to random direction, a series of scalars are obtained;
B) choose F point pair successively from all-pair using the feature selecting algorithm based on correlation, each put to pixel Difference and a) in scalar variance it is maximum;
C) to F point of acquisition to choosing random threshold value;
D) recurrence output of all division spaces in current sample set that this weak random fern returns device is calculated;
E) obtain weak random fern and return device, and update accumulation regression forecasting value.
B) obtain strong random fern and return device, and update regression residuals target.
(5) final recurrence device is obtained.
2) it is predicted using shape recurrence device and obtains target shape.
CN201610554203.4A 2016-04-21 2016-07-12 A kind of license plate locating method returned based on integrated random fern and shape Pending CN107305634A (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN2016102500580 2016-04-21
CN201610250058 2016-04-21

Publications (1)

Publication Number Publication Date
CN107305634A true CN107305634A (en) 2017-10-31

Family

ID=60151041

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610554203.4A Pending CN107305634A (en) 2016-04-21 2016-07-12 A kind of license plate locating method returned based on integrated random fern and shape

Country Status (1)

Country Link
CN (1) CN107305634A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108629335A (en) * 2018-06-05 2018-10-09 华东理工大学 Adaptive face key feature points selection method
CN110400313A (en) * 2019-08-01 2019-11-01 北京灵医灵科技有限公司 A kind of the soft tissue separation method and separation system of nuclear magnetic resonance image

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102968646A (en) * 2012-10-25 2013-03-13 华中科技大学 Plate number detecting method based on machine learning
CN104318225A (en) * 2014-11-19 2015-01-28 深圳市捷顺科技实业股份有限公司 License plate detection method and device

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102968646A (en) * 2012-10-25 2013-03-13 华中科技大学 Plate number detecting method based on machine learning
CN104318225A (en) * 2014-11-19 2015-01-28 深圳市捷顺科技实业股份有限公司 License plate detection method and device

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
MICHAEL VILLAMIZAR ET AL.: "Efficient Rotation Invariant Object Detection using Boosted Random Ferns", 《 2010 IEEE COMPUTER SOCIETY CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 *
XUDONG CAO ET AL.: "Face Alignment by Explicit Shape Regression", 《2012 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 *
杨晨晖 等: "基于随机蕨的实时车辆匹配", 《厦门大学学报(自然科学版)》 *
罗艳 等: "基于多示例学习和随机蕨丛检测的在线目标跟踪", 《电子与信息学报》 *
黄叶珏 等: "基于在线多示例提升随机蕨丛的目标跟踪", 《计算机应用》 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108629335A (en) * 2018-06-05 2018-10-09 华东理工大学 Adaptive face key feature points selection method
CN110400313A (en) * 2019-08-01 2019-11-01 北京灵医灵科技有限公司 A kind of the soft tissue separation method and separation system of nuclear magnetic resonance image
CN110400313B (en) * 2019-08-01 2021-01-01 北京灵医灵科技有限公司 Soft tissue separation method and separation system for nuclear magnetic resonance image

Similar Documents

Publication Publication Date Title
Sun et al. Research on the hand gesture recognition based on deep learning
CN108304873A (en) Object detection method based on high-resolution optical satellite remote-sensing image and its system
Jia et al. Visual tracking via adaptive structural local sparse appearance model
Du et al. Leaf shape based plant species recognition
CN102254303B (en) Methods for segmenting and searching remote sensing image
CN103049763B (en) Context-constraint-based target identification method
Wang et al. Recognition of leaf images based on shape features using a hypersphere classifier
CN106909902B (en) Remote sensing target detection method based on improved hierarchical significant model
CN107563413B (en) Unmanned aerial vehicle aerial image farmland block object accurate extraction method
CN108346159A (en) A kind of visual target tracking method based on tracking-study-detection
CN106570874B (en) Image marking method combining image local constraint and object global constraint
CN107633226B (en) Human body motion tracking feature processing method
CN107103317A (en) Fuzzy license plate image recognition algorithm based on image co-registration and blind deconvolution
CN103886619B (en) A kind of method for tracking target merging multiple dimensioned super-pixel
CN108280397A (en) Human body image hair detection method based on depth convolutional neural networks
CN105740945A (en) People counting method based on video analysis
CN107766890A (en) The improved method that identification segment learns in a kind of fine granularity identification
CN111753828A (en) Natural scene horizontal character detection method based on deep convolutional neural network
CN108038435A (en) A kind of feature extraction and method for tracking target based on convolutional neural networks
CN108647682A (en) A kind of brand Logo detections and recognition methods based on region convolutional neural networks model
CN107480585A (en) Object detection method based on DPM algorithms
CN114758288A (en) Power distribution network engineering safety control detection method and device
CN105574545B (en) The semantic cutting method of street environment image various visual angles and device
CN109685045A (en) A kind of Moving Targets Based on Video Streams tracking and system
CN103985143A (en) Discriminative online target tracking method based on videos in dictionary learning

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
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20171031

WD01 Invention patent application deemed withdrawn after publication