CN103745197A - Detection method of license plate and device thereof - Google Patents
Detection method of license plate and device thereof Download PDFInfo
- Publication number
- CN103745197A CN103745197A CN201310739338.4A CN201310739338A CN103745197A CN 103745197 A CN103745197 A CN 103745197A CN 201310739338 A CN201310739338 A CN 201310739338A CN 103745197 A CN103745197 A CN 103745197A
- Authority
- CN
- China
- Prior art keywords
- image
- model
- square frame
- edge
- hog
- 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.)
- Granted
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 41
- 238000000034 method Methods 0.000 claims abstract description 33
- 238000004364 calculation method Methods 0.000 claims description 31
- 239000000284 extract Substances 0.000 claims description 29
- 238000000605 extraction Methods 0.000 claims description 12
- 238000012549 training Methods 0.000 claims description 12
- 238000006243 chemical reaction Methods 0.000 claims description 11
- 238000012545 processing Methods 0.000 abstract description 2
- 230000015572 biosynthetic process Effects 0.000 description 6
- 238000012217 deletion Methods 0.000 description 6
- 230000037430 deletion Effects 0.000 description 6
- 238000005286 illumination Methods 0.000 description 5
- 230000001143 conditioned effect Effects 0.000 description 3
- 238000005259 measurement Methods 0.000 description 3
- 238000012795 verification Methods 0.000 description 3
- HUTDUHSNJYTCAR-UHFFFAOYSA-N ancymidol Chemical compound C1=CC(OC)=CC=C1C(O)(C=1C=NC=NC=1)C1CC1 HUTDUHSNJYTCAR-UHFFFAOYSA-N 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000013519 translation Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
Images
Abstract
The present invention is applicable to the field of image processing, and provides a detection method of license plate and a device thereof, the method includes the following steps of: performing multi-scale detection for the images by using a HOG model and extracting a pane area via the HOG model detection; calculating an edge of the image and an integrogram image of the image after edge computation to gain the edge integrogram iamge of the image; gaining the pane area via the HOG model detection on the edge integrogram iamge, calculating an edge density of the pane area via the HOG model detection, and extracting the edge density greater than or equal to the pane of the threshold value; via a Haar model, detecting the edge density greater than or equal to the pane of the threshold value to gain the exact license plate area. The present invention by the HOG features detects the images, and then performs the detection the edge and the edge density, finally the license plate area is detected by the haar model.
Description
Technical field
The invention belongs to image processing field, relate in particular to a kind of car plate detection method and device.
Background technology
Car plate identification has been widely used in the places such as gateway, parking lot, mall gateway.Car plate identification is generally divided into car plate and detects, segmentation of the characters and their identification and car plate this three step of voting.Car plate detects, and car plate from video, detected, and ticket detection is a very important step in car plate identification, is also a step the most consuming time.Car plate detects and in vehicle image, detects exactly the position of car plate.Because car plate identification equipment is all generally that the image-forming condition of car plate and background is uncontrollable in outdoor application, illumination, complex background, the setting angle of video camera and distance etc. detects and has brought difficulty to car plate.License plate area shared ratio in entire image is less, and from entire image, detect license plate area must search in a large amount of background informations.In actual life, the identification application scenarios of car plate there is is the restriction of pair time, therefore more need fast detecting to arrive car plate.Car plate detection method based on video has a lot, comprises the angle detection algorithm in the binary image based on line template, utilizes genetic algorithm to detect car plate, also has Adaboost based on Haar feature to detect etc.Wherein the Adaboost training pattern based on Haar feature is carried out full figure search.The method verification and measurement ratio is high, but speed is slow, and full figure carries out Haar detection can accurately detect car plate, but operand is large, if be applied in embedded device, the arithmetic speed of embedded device is difficult to reach requirement.To detect angle requirement high for Haar in addition, if car plate angle has surpassed 30 degree, just there will be certain undetected.
Summary of the invention
The object of the embodiment of the present invention is to provide a kind of car plate detection method and device, aim to provide a kind of car plate detection method, by HOG feature, image is detected, then carry out rim detection and marginal density and detect, finally by Haar model, detect license plate area.
The embodiment of the present invention is achieved in that a kind of car plate detection method, and described method comprises the steps:
Use HOG model to carry out multiple scale detecting to image, and extract the boxed area detecting by HOG model;
Image is carried out to edge calculations, and calculate the integrogram of the image after edge calculations, acquire the edge integrogram of image;
On described edge integrogram, obtain the boxed area detecting by HOG model, calculate the marginal density of the boxed area detecting by HOG model, extract the square frame that marginal density is more than or equal to threshold value;
The square frame that is more than or equal to threshold value by Haar model edge density detects, and detects and obtains accurate license plate area.
Further, described method comprises use HOG model carries out the step of multiple scale detecting to image before:
Image is changed, image is converted to gray-scale map;
Image is carried out to convergent-divergent.
Further, described method is being changed image, also comprises before image being converted to the step of gray-scale map:
According to the positive negative sample gathering, set up HOG model.
Further, described method carries out also comprising before detecting step at the square frame that is more than or equal to threshold value by Haar model edge density:
Edge density is more than or equal to the square frame of threshold value divides equally, and calculates the average marginal density of dividing rear each square frame equally, according to average marginal density, obtains average marginal density deviation, and the square frame of calibration offset in threshold range is similar car plate square frame.
Further, the described square frame that is more than or equal to threshold value by Haar model edge density detects, and detects the step that obtains accurate license plate area and comprises:
Adaboost algorithm by cascade is trained and is obtained Haar model Haar feature;
The square frame that is more than or equal to threshold value by Haar model edge density detects, and detects and obtains accurate license plate area.
Another object of the embodiment of the present invention is to provide a kind of car plate pick-up unit, and described device comprises:
Boxed area extraction unit, for using HOG model to carry out multiple scale detecting to image, and extracts the boxed area detecting by HOG model;
Integrogram computing unit, for image is carried out to edge calculations, and calculates the integrogram of the image after edge calculations, acquires the edge integrogram of image;
Square frame extraction unit, for obtain the boxed area detecting by HOG model on described edge integrogram, calculates the marginal density of the boxed area detecting by HOG model, extracts the square frame that marginal density is more than or equal to threshold value;
License plate area detecting unit, detects for be more than or equal to the square frame of threshold value by Haar model edge density, detects and obtains accurate license plate area.
Further, described device also comprises:
Image conversion unit for scaling, for image is changed, is converted to gray-scale map by image; Image is carried out to convergent-divergent.Further, described device also comprises:
Image conversion unit, for image is changed, is converted to gray-scale map by image.
Further, described device also comprises:
Demarcate unit, the square frame that is more than or equal to threshold value for edge density is divided equally, and calculate the average marginal density of dividing rear each square frame equally, and according to average marginal density, obtaining average marginal density deviation, the square frame of calibration offset in threshold range is similar car plate square frame.
Further, described license plate area detecting unit also comprises:
Model training unit, trains and obtains Haar model Haar feature for the Adaboost algorithm by cascade;
Accurately detecting unit, detects for be more than or equal to the square frame of threshold value by Haar model edge density, detects and obtains accurate license plate area.
The embodiment of the present invention is by a kind of car plate detection method and device, first by HOG feature, acquire some boxed area, image is carried out to the edge integrogram that edge calculations calculates image simultaneously, by the boxed area that obtains and edge integrogram, accurately extract again and meet pre-conditioned square frame, finally by Haar model, detect license plate area, owing to first having adopted HOG feature to detect, can realize wide-angle detection, simultaneously the model of HOG feature is to light robust, and the probability that makes finally to detect by Haar model car plate is higher.
Accompanying drawing explanation
Fig. 1 is the realization flow figure of a kind of car plate detection method of providing of first embodiment of the invention;
Fig. 2 is the realization flow figure of a kind of car plate detection method of providing of second embodiment of the invention;
Fig. 3 is the structural drawing of a kind of car plate pick-up unit of providing of second embodiment of the invention; And
Fig. 4 is the structural drawing of a kind of car plate pick-up unit of providing of second embodiment of the invention.
Embodiment
In order to make object of the present invention, technical scheme and advantage clearer, below in conjunction with drawings and Examples, the present invention is further elaborated.Should be appreciated that specific embodiment described herein, only in order to explain the present invention, is not intended to limit the present invention.
Below in conjunction with specific embodiment, specific implementation of the present invention is described in detail:
embodiment mono-:
Fig. 1 shows the realization flow of a kind of car plate detection method that first embodiment of the invention provides, and details are as follows:
S101, is used HOG model to carry out multiple scale detecting to image, and extracts the boxed area detecting by HOG model.
Use HOG(Histogram of oriented gradients) gradient orientation histogram feature characterizes each width positive and negative samples, forms HOG proper vector.Partial gradient amplitude and the direction character of HOG picture engraving, be normalized the proper vector of piece based on Gradient Features.HOG allows between piece overlapped, therefore, to illumination variation and skew in a small amount insensitive, can effectively depict edge feature.The car plate large for angle detects effective.By the HOG proper vector forming, set up HOG model afterwards, by HOG model, carry out multiple scale detecting, to each region that can verify by HOG in image, all use a boxed area to show, extract the boxed area detecting by HOG model.
S102, carries out edge calculations to image, and calculates the integrogram of the image after edge calculations, acquires the edge integrogram of image.
Rule of thumb, car plate is generally all in edge than comparatively dense and the region that is evenly distributed, the edge of computed image full figure, for the convenience of searching for, calculate again the integrogram of the full figure after edge calculations, referred to herein as edge integrogram, acquire the edge integrogram of image.
S103 obtains the boxed area detecting by HOG model on described edge integrogram, calculates the marginal density of the boxed area detecting by HOG model, extracts the square frame that marginal density is more than or equal to threshold value.
Use the boxed area obtaining to go out doubtful license plate area at edge integrogram upper ledge, then by integrogram, calculate the marginal density in these regions, in deletion gray-scale map, marginal density is less than the square frame of threshold value, extracts the square frame that marginal density is more than or equal to threshold value.
S104, the square frame that is more than or equal to threshold value by Haar model edge density detects, and detects and obtains accurate license plate area.
Detection finishes rear remaining square frame to be carried out to cluster operation, and all square frame area occupied of each class are linked up to formation piece, by Haar model, remaining piece is detected and obtains accurate car plate afterwards, detects accurate license plate area.
The embodiment of the present invention by the way, a kind of car plate detection method, first by HOG feature, acquire some boxed area, image is carried out to the edge integrogram that edge calculations calculates image simultaneously, by the boxed area that obtains and edge integrogram, accurately extract again and meet pre-conditioned square frame, finally by Haar model, detect license plate area, owing to first having adopted HOG feature to detect, can realize wide-angle detection, simultaneously the model of HOG feature is to light robust, and the probability that makes finally to detect by Haar model car plate is higher.
embodiment bis-:
Fig. 2 shows the realization flow of a kind of car plate detection method that second embodiment of the invention provides, and details are as follows:
S201, according to the positive negative sample gathering, sets up HOG model.
Select a large amount of car plate pictures and non-car plate picture as the positive and negative samples of training set, the process of choosing can be chosen by artificial mode, uses afterwards HOG feature to characterize each width positive and negative samples, forms proper vector, sets up HOG model.
。
S202, changes image, and image is converted to gray-scale map; Image is carried out to convergent-divergent.
First camera is carried out to image conversion process at the image capturing, image is converted to gray level image, by image is converted to gray level image, can reduce the calculated amount that image is processed; Image after conversion is carried out to convergent-divergent, in concrete implementation process, can and highly respectively reduce half the width of picture.Image is carried out can further reducing the calculated amount to image after convergent-divergent, improve the speed that detects license plate area.
S203, is used HOG model to carry out multiple scale detecting to image, and extracts the boxed area detecting by HOG model.
Use HOG model to carry out multiple scale detecting to image, wherein the video of gateway, a large amount of parking lot can be taken and store by camera to obtaining of HOG model under each period and weather condition.In these videos, manually intercept out the license plate image of each car, license plate image is as positive sample, and manually intercepting out is not wherein the part of car plate, and the non-car plate picture intercepting out is as negative sample.Use HOG feature to characterize each width positive and negative samples, form HOG proper vector.Partial gradient amplitude and the direction character of HOG picture engraving, be normalized the proper vector of piece based on Gradient Features.HOG allows between piece overlapped, therefore, to illumination variation and skew in a small amount insensitive, can effectively depict edge feature.The car plate large for angle detects effective.HOG feature is the gradient statistical information of gray-scale map, and gradient is mainly present in the place at edge.The First-order Gradient of computed image, gradient magnitude:
Gradient direction: Ang (x, y)=arccos (I (x+1, y)-I (x-1, y)/R).In gradient calculation specifically can, use one-dimensional discrete differential masterplate [1 0 1] in the horizontal direction sample image to be processed, obtain horizontal gradient image; Use one-dimensional discrete differential masterplate [1 0 1]
tin vertical direction sample image is processed, obtained VG (vertical gradient) image.What histogram direction was concrete can be 9, and the one dimension histogram of gradients of all pixels in each piecemeal is added to wherein, has just formed final feature.Finally by final feature, set up HOG model, utilize the Adaboost algorithm of cascade to train HOG feature, finally obtain the cascade car plate strong classifier model of 5 layers, re-use HOG model image is carried out to multiple scale detecting, extract the boxed area detecting by HOG model.
S204, carries out edge calculations to image, and calculates the integrogram of the image after edge calculations, acquires the edge integrogram of image.
Image is carried out to edge calculations, calculate the integrogram of the image after edge calculations, calculate the integrogram of the image after edge calculations, acquire the edge integrogram of image.
S205 obtains the boxed area detecting by HOG model on described edge integrogram, calculates the marginal density of the boxed area detecting by HOG model, extracts the square frame that marginal density is more than or equal to threshold value.
The coordinate of the boxed area detecting by the HOG model obtaining afterwards goes out doubtful license plate area at edge integrogram upper ledge, by integrogram, calculate the marginal density in these regions, in deletion gray-scale map, marginal density is less than the square frame of threshold value, extracts the square frame that marginal density is more than or equal to threshold value.Wherein use integrogram can accelerate the calculating of marginal density.For fear of a square frame the double counting that is added of rim value a little, in algorithm, used integrogram.Each point (x, y) on integrogram has comprised the rim value from point (0,0) to all pixels of point (x, y).If represent the gray-scale map after edge calculations with I (x, y), II (x, y) represents the integrogram calculating, so:
a rectangle can be used following mode to calculate arbitrarily: if the upper left corner of rectangle is (x
lt, y
lt), the coordinate in the lower right corner is (x
rb, y
rb), the integrogram of this rectangle can calculate with following formula so: SUM
d=II (x
rb, y
rb)-II (x
lt, y
rb)-II (x
rb, y
lt)+II (x
lt, y
lt), wherein different and variant for the sample of the concrete application scenarios of the set basis of threshold value and training, the threshold value of setting is in the present embodiment 7.Extract the square frame that marginal density is more than or equal to threshold value.
S206, edge density is more than or equal to the square frame of threshold value divides equally, and calculates the average marginal density of dividing rear each square frame equally, according to average marginal density, obtains average marginal density deviation, and the square frame of calibration offset in threshold range is similar car plate square frame.
In background and vehicle body, have some rectangle object, feature and car plate are similar, but marginal distribution is inhomogeneous.For example some car light can be that car plate and marginal density are high by flase drop together with the region on side, but its Kernel density estimate is inhomogeneous.The square frame that we need edge density to be more than or equal to threshold value is divided equally, and concrete can be divided into 5 the square frame detecting, and calculates respectively this average marginal density of 5.Then calculate the deviation of this average density of 5, if deviation is in threshold value, represent the marginal distribution formula equilibrium in this region, it is the region of similar car plate, to the local of similar car plate, can also adopt the cluster of square frame, concrete for adopting the method for arest neighbors, as long as reaching 60%, coincidence between any two just thinks that these two frames belong to a class.After cluster, according to default deletion condition, delete with part square frame: rule of thumb, the square frame that license plate area is searched for out can be a lot, if so the square frame that has classification to comprise after cluster be less than 10, just all square frames of whole class are deleted.If the position that a plurality of scale detect overlaps, the scale in the middle of only retaining, deletes the square frame that other scale detects.Same scale carries out translation in testing process, if adjacent two large area appear in a plurality of square frames that detect, overlaps, and only retains central square frame, and this square frame up and down, left and right direction expands respectively 10%.Finally all square frame area occupied of each class are linked up to formation piece, finally demarcating and linking up formation piece is similar car plate square frame.
S207, the square frame that is more than or equal to threshold value by Haar model edge density detects, and detects and obtains accurate license plate area.
Finally by Haar model, remaining piece is detected, obtain accurate car plate position, wherein Haar model is to utilize the Adaboost algorithm of cascade to train Haar feature, by training, obtain Haar model, concrete can will manually select a large amount of car plate pictures and the non-car plate picture positive and negative samples as training set.Use Haar feature to characterize each width positive and negative samples, form proper vector, obtain the cascade car plate strong classifier of 20 layers.Form Haar model, the square frame that is more than or equal to threshold value by Haar model edge density detects, and detects and obtains accurate license plate area.
By the way, a kind of car plate detection method, first utilizes the HOG model of 5 layers to carry out full figure detection to the embodiment of the present invention, then doubtful license plate area is carried out to marginal density detection and the balanced detection of marginal density.Then use the method for cluster that close region clustering is become to doubtful license plate area.Finally remaining several license plate areas are carried out the method for the Haar model detection of 20 layers.It is fast that the method is carried out Haar detection speed than traditional full figure, and utilize the large car plate of characteristic energy detection angles of HOG, the results show, and the verification and measurement ratio of the method is high, and to light robust.
embodiment tri-:
Fig. 3 shows the structural drawing of a kind of car plate pick-up unit that third embodiment of the invention provides, and for convenience of explanation, only shows the part relevant to the embodiment of the present invention.
Boxed area extraction unit 31, for using HOG model to carry out multiple scale detecting to image, and extracts the boxed area detecting by HOG model.
Boxed area extraction unit is used for using HOG feature to characterize each width positive and negative samples, forms HOG proper vector.Partial gradient amplitude and the direction character of HOG picture engraving, be normalized the proper vector of piece based on Gradient Features.HOG allows between piece overlapped, therefore, to illumination variation and skew in a small amount insensitive, can effectively depict edge feature.The car plate large for angle detects effective.By the HOG proper vector forming, set up HOG model afterwards, by HOG model, carry out multiple scale detecting, to each region that can verify by HOG in image, all use a boxed area to show, extract the boxed area detecting by HOG model.
Rule of thumb, car plate is generally all in edge than comparatively dense and the region that is evenly distributed, by the edge of integrogram computing unit computed image full figure, for the convenience of searching for, calculate again the integrogram of the full figure after edge calculations, referred to herein as edge integrogram, acquire the edge integrogram of image.
Square frame extraction unit 33, for obtain the boxed area detecting by HOG model on described edge integrogram, calculates the marginal density of the boxed area detecting by HOG model, extracts the square frame that marginal density is more than or equal to threshold value.
Square frame extraction unit is for being used the boxed area of acquisition to go out doubtful license plate area at edge integrogram upper ledge, by integrogram, calculate again the marginal density in these regions, in deletion gray-scale map, marginal density is less than the square frame of threshold value, extracts the square frame that marginal density is more than or equal to threshold value.
License plate area detecting unit 34, detects for be more than or equal to the square frame of threshold value by Haar model edge density, detects and obtains accurate license plate area.
License plate area detecting unit carries out cluster operation to remaining square frame after finishing, the all square frame area occupied of each class are linked up to formation piece, by Haar model, remaining piece is detected and obtains accurate car plate afterwards, detect accurate license plate area.
The embodiment of the present invention by the way, a kind of car plate pick-up unit, first by HOG feature, acquire some boxed area, image is carried out to the edge integrogram that edge calculations calculates image simultaneously, by the boxed area that obtains and edge integrogram, accurately extract again and meet pre-conditioned square frame, finally by Haar model, detect license plate area, owing to first having adopted HOG feature to detect, can realize wide-angle detection, simultaneously the model of HOG feature is to light robust, and the probability that makes finally to detect by Haar model car plate is higher.
embodiment tetra-:
Fig. 4 shows the structural drawing of a kind of car plate pick-up unit that fourth embodiment of the invention provides, and for convenience of explanation, only shows the part relevant to the embodiment of the present invention.
Model is set up unit 41, according to the positive negative sample gathering, sets up HOG model.
Model is set up unit according to selecting a large amount of car plate pictures and the non-car plate picture positive and negative samples as training set, the process of choosing can be chosen by artificial mode, use afterwards HOG feature to characterize each width positive and negative samples, form proper vector, set up HOG model.
Image conversion unit for scaling 42, for image is changed, is converted to gray-scale map by image; Image is carried out to convergent-divergent.
Image conversion unit for scaling specifically can comprise: image conversion unit and image scaling unit, wherein first image conversion unit for carrying out image conversion process by camera at the image capturing, image is converted to gray level image, by image is converted to gray level image, can reduce the calculated amount that image is processed; Wherein image scaling unit, for the image after conversion is carried out to convergent-divergent, can and highly respectively reduce half the width of picture in concrete implementation process.Image is carried out can further reducing the calculated amount to image after convergent-divergent, improve the speed that detects license plate area.
Boxed area extraction unit 43, for using HOG model to carry out multiple scale detecting to image, and extracts the boxed area detecting by HOG model.
Boxed area extraction unit is used for using HOG model to carry out multiple scale detecting to image, and wherein the video of gateway, a large amount of parking lot can be taken and store by camera to obtaining of HOG model under each period and weather condition.In these videos, manually intercept out the license plate image of each car, license plate image is as positive sample, and manually intercepting out is not wherein the part of car plate, and the non-car plate picture intercepting out is as negative sample.Use HOG feature to characterize each width positive and negative samples, form HOG proper vector.Partial gradient amplitude and the direction character of HOG picture engraving, be normalized the proper vector of piece based on Gradient Features.HOG allows between piece overlapped, therefore, to illumination variation and skew in a small amount insensitive, can effectively depict edge feature.The car plate large for angle detects effective.HOG feature is the gradient statistical information of gray-scale map, and gradient is mainly present in the place at edge.The First-order Gradient of computed image, gradient magnitude:
Gradient direction: Ang (x, y)=arccos (I (x+1, y)-I (x-1, y)/R).In gradient calculation specifically can, use one-dimensional discrete differential masterplate [1 0 1] in the horizontal direction sample image to be processed, obtain horizontal gradient image; Use one-dimensional discrete differential masterplate [1 0 1]
tin vertical direction sample image is processed, obtained VG (vertical gradient) image.What histogram direction was concrete can be 9, and the one dimension histogram of gradients of all pixels in each piecemeal is added to wherein, has just formed final feature.Finally by final feature, set up HOG model, utilize the Adaboost algorithm of cascade to train HOG feature, finally obtain the cascade car plate strong classifier model of 5 layers, re-use HOG model image is carried out to multiple scale detecting, extract the boxed area detecting by HOG model.
Integrogram computing unit, for image is carried out to edge calculations, calculates the integrogram of the image after edge calculations, calculates the integrogram of the image after edge calculations, acquires the edge integrogram of image.
Square frame extraction unit 45, for obtain the boxed area detecting by HOG model on described edge integrogram, calculates the marginal density of the boxed area detecting by HOG model, extracts the square frame that marginal density is more than or equal to threshold value.
The coordinate of the boxed area that square frame extraction unit detects for the HOG model by obtaining goes out doubtful license plate area at edge integrogram upper ledge, by integrogram, calculate the marginal density in these regions, in deletion gray-scale map, marginal density is less than the square frame of threshold value, extracts the square frame that marginal density is more than or equal to threshold value.Wherein use integrogram can accelerate the calculating of marginal density.For fear of a square frame the double counting that is added of rim value a little, in algorithm, used integrogram.Each point (x, y) on integrogram has comprised the rim value from point (0,0) to all pixels of point (x, y).If represent the gray-scale map after edge calculations with I (x, y), II (x, y) represents the integrogram calculating, so:
a rectangle can be used following mode to calculate arbitrarily: if the upper left corner of rectangle is (x
lt, y
lt), the coordinate in the lower right corner is (x
rb, y
rb), the integrogram of this rectangle can calculate with following formula so: SUM
d=II (x
rb, y
rb)-II (x
lt, y
rb)-II (x
rb, y
lt)+II (x
lt, y
lt), wherein different and variant for the sample of the concrete application scenarios of the set basis of threshold value and training, the threshold value of setting is in the present embodiment 7.Extract the square frame that marginal density is more than or equal to threshold value.
Demarcate unit for there are some rectangle object in background and vehicle body, feature and car plate are similar, but marginal distribution is inhomogeneous.For example some car light can be that car plate and marginal density are high by flase drop together with the region on side, but its Kernel density estimate is inhomogeneous.The square frame that we need edge density to be more than or equal to threshold value is divided equally, and concrete can be divided into 5 the square frame detecting, and calculates respectively this average marginal density of 5.Then calculate the deviation of this average density of 5, if deviation is in threshold value, represent the marginal distribution formula equilibrium in this region, it is the region of similar car plate, to the local of similar car plate, can also adopt the cluster of square frame, concrete for adopting the method for arest neighbors, as long as reaching 60%, coincidence between any two just thinks that these two frames belong to a class.After cluster, according to default deletion condition, delete with part square frame: rule of thumb, the square frame that license plate area is searched for out can be a lot, if so the square frame that has classification to comprise after cluster be less than 10, just all square frames of whole class are deleted.If the position that a plurality of scale detect overlaps, the scale in the middle of only retaining, deletes the square frame that other scale detects.Same scale carries out translation in testing process, if adjacent two large area appear in a plurality of square frames that detect, overlaps, and only retains central square frame, and this square frame up and down, left and right direction expands respectively 10%.Finally all square frame area occupied of each class are linked up to formation piece, finally demarcating and linking up formation piece is similar car plate square frame.
License plate area detecting unit 47, detects for be more than or equal to the square frame of threshold value by Haar model edge density, detects and obtains accurate license plate area.
License plate area detecting unit comprises: model training unit and accurately detecting unit, and model training unit is for being trained and obtain Haar model Haar feature by the Adaboost algorithm of cascade; Accurately detecting unit detects for be more than or equal to the square frame of threshold value by Haar model edge density, detects and obtains accurate license plate area.Concrete be used for finally by Haar model, remaining piece being detected for license plate area detecting unit, obtain accurate car plate position, wherein Haar model is the model that utilizes the Adaboost algorithm of cascade to train Haar feature, will manually select a large amount of car plate pictures and the non-car plate picture positive and negative samples as training set.Use Haar feature to characterize each width positive and negative samples, form proper vector, obtain the cascade car plate strong classifier of 20 layers.Form Haar model, the square frame that is more than or equal to threshold value by Haar model edge density detects, and detects and obtains accurate license plate area.
By the way, a kind of car plate pick-up unit, first utilizes the HOG model of 5 layers to carry out full figure detection to the embodiment of the present invention, then doubtful license plate area is carried out to marginal density detection and the balanced detection of marginal density.Then use the method for cluster that close region clustering is become to doubtful license plate area.Finally remaining several license plate areas are carried out the method for the Haar model detection of 20 layers.It is fast that the method is carried out Haar detection speed than traditional full figure, and utilize the large car plate of characteristic energy detection angles of HOG, the results show, and the verification and measurement ratio of the method is high, and to light robust.
One of ordinary skill in the art will appreciate that all or part of step realizing in above-described embodiment method is to come the hardware that instruction is relevant to complete by program, described program can be stored in computer read/write memory medium, described storage medium, as ROM/RAM, disk, CD etc.
The foregoing is only preferred embodiment of the present invention, not in order to limit the present invention, all any modifications of doing within the spirit and principles in the present invention, be equal to and replace and improvement etc., within all should being included in protection scope of the present invention.
Claims (10)
1. a car plate detection method, is characterized in that, described method comprises the steps:
Use HOG model to carry out multiple scale detecting to image, and extract the boxed area detecting by HOG model;
Image is carried out to edge calculations, and calculate the integrogram of the image after edge calculations, acquire the edge integrogram of image;
On described edge integrogram, obtain the boxed area detecting by HOG model, calculate the marginal density of the boxed area detecting by HOG model, extract the square frame that marginal density is more than or equal to threshold value;
The square frame that is more than or equal to threshold value by Haar model edge density detects, and detects and obtains accurate license plate area.
2. the method for claim 1, is characterized in that, described method comprises use HOG model carries out the step of multiple scale detecting to image before:
Image is changed, image is converted to gray-scale map;
Image is carried out to convergent-divergent.
3. method as claimed in claim 2, is characterized in that, described method is being changed image, also comprises before image being converted to the step of gray-scale map:
According to the positive negative sample gathering, set up HOG model.
4. the method for claim 1, is characterized in that, described method carries out also comprising before detecting step at the square frame that is more than or equal to threshold value by Haar model edge density:
Edge density is more than or equal to the square frame of threshold value divides equally, and calculates the average marginal density of dividing rear each square frame equally, according to average marginal density, obtains average marginal density deviation, and the square frame of calibration offset in threshold range is similar car plate square frame.
5. the method for claim 1, is characterized in that, the described square frame that is more than or equal to threshold value by Haar model edge density detects, and detects the step that obtains accurate license plate area and comprises:
Adaboost algorithm by cascade is trained and is obtained Haar model Haar feature;
The square frame that is more than or equal to threshold value by Haar model edge density detects, and detects and obtains accurate license plate area.
6. a car plate pick-up unit, is characterized in that, described device comprises:
Boxed area extraction unit, for using HOG model to carry out multiple scale detecting to image, and extracts the boxed area detecting by HOG model;
Integrogram computing unit, for image is carried out to edge calculations, and calculates the integrogram of the image after edge calculations, acquires the edge integrogram of image;
Square frame extraction unit, for obtain the boxed area detecting by HOG model on described edge integrogram, calculates the marginal density of the boxed area detecting by HOG model, extracts the square frame that marginal density is more than or equal to threshold value;
License plate area detecting unit, detects for be more than or equal to the square frame of threshold value by Haar model edge density, detects and obtains accurate license plate area.
7. device as claimed in claim 5, is characterized in that, described device also comprises:
Image conversion unit for scaling, for image is changed, is converted to gray-scale map by image; Image is carried out to convergent-divergent.
8. device as claimed in claim 6, is characterized in that, described device also comprises:
Model is set up unit, according to the positive negative sample gathering, sets up HOG model.
9. device as claimed in claim 6, is characterized in that, described device also comprises:
Demarcate unit, the square frame that is more than or equal to threshold value for edge density is divided equally, and calculate the average marginal density of dividing rear each square frame equally, and according to average marginal density, obtaining average marginal density deviation, the square frame of calibration offset in threshold range is similar car plate square frame.
10. device as claimed in claim 6, is characterized in that, described license plate area detecting unit also comprises:
Model training unit, trains and obtains Haar model Haar feature for the Adaboost algorithm by cascade;
Accurately detecting unit, detects for be more than or equal to the square frame of threshold value by Haar model edge density, detects and obtains accurate license plate area.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310739338.4A CN103745197B (en) | 2013-12-27 | 2013-12-27 | A kind of detection method of license plate and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310739338.4A CN103745197B (en) | 2013-12-27 | 2013-12-27 | A kind of detection method of license plate and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103745197A true CN103745197A (en) | 2014-04-23 |
| CN103745197B CN103745197B (en) | 2017-08-29 |
Family
ID=50502214
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310739338.4A Expired - Fee Related CN103745197B (en) | 2013-12-27 | 2013-12-27 | A kind of detection method of license plate and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103745197B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104182769A (en) * | 2014-09-17 | 2014-12-03 | 深圳市捷顺科技实业股份有限公司 | Number plate detection method and system |
| CN104268509A (en) * | 2014-09-17 | 2015-01-07 | 深圳市捷顺科技实业股份有限公司 | Method and system for detecting license plate of dump truck |
| 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 |
| CN104463220A (en) * | 2014-12-19 | 2015-03-25 | 深圳市捷顺科技实业股份有限公司 | License plate detection method and system |
| CN105184291A (en) * | 2015-08-26 | 2015-12-23 | 深圳市捷顺科技实业股份有限公司 | Method and system for detecting multiple types of license plates |
| CN106651774A (en) * | 2016-12-27 | 2017-05-10 | 深圳市捷顺科技实业股份有限公司 | License plate super-resolution model reconstruction method and device |
| CN106846353A (en) * | 2017-01-04 | 2017-06-13 | 美的集团股份有限公司 | Image processing method, image processing apparatus and equipment based on density |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040264776A1 (en) * | 2003-06-27 | 2004-12-30 | Quen-Zong Wu | String extraction method for image based on multiple stroke width patterns matching |
| US20060171564A1 (en) * | 2003-07-10 | 2006-08-03 | James Simon | Autonomous wide-angle license plate recognition |
| CN101246551A (en) * | 2008-03-07 | 2008-08-20 | 北京航空航天大学 | Fast license plate locating method |
| CN102129562A (en) * | 2010-01-15 | 2011-07-20 | 富士通株式会社 | Method and device for identifying icons |
-
2013
- 2013-12-27 CN CN201310739338.4A patent/CN103745197B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040264776A1 (en) * | 2003-06-27 | 2004-12-30 | Quen-Zong Wu | String extraction method for image based on multiple stroke width patterns matching |
| US20060171564A1 (en) * | 2003-07-10 | 2006-08-03 | James Simon | Autonomous wide-angle license plate recognition |
| CN101246551A (en) * | 2008-03-07 | 2008-08-20 | 北京航空航天大学 | Fast license plate locating method |
| CN102129562A (en) * | 2010-01-15 | 2011-07-20 | 富士通株式会社 | Method and device for identifying icons |
Non-Patent Citations (4)
| Title |
|---|
| 张浩鹏 等: "基于灰度方差和边缘密度的车牌定位算法", 《仪器仪表学报》 * |
| 曾丽华 等: "基于边缘与颜色信息的车牌精确定位算法", 《北京航空航天大学学报》 * |
| 李驰: "智能交通中的车牌识别算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 赵元兴: "用于车牌定位的分类器设计与相关算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104182769A (en) * | 2014-09-17 | 2014-12-03 | 深圳市捷顺科技实业股份有限公司 | Number plate detection method and system |
| CN104268509A (en) * | 2014-09-17 | 2015-01-07 | 深圳市捷顺科技实业股份有限公司 | Method and system for detecting license plate of dump truck |
| CN104182769B (en) * | 2014-09-17 | 2018-04-27 | 深圳市捷顺科技实业股份有限公司 | A kind of detection method of license plate and system |
| CN104268509B (en) * | 2014-09-17 | 2018-05-29 | 深圳市捷顺科技实业股份有限公司 | The method and system of dump truck car plate detection |
| 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 |
| CN104318225B (en) * | 2014-11-19 | 2019-02-22 | 深圳市捷顺科技实业股份有限公司 | Detection method of license plate and device |
| CN104463220A (en) * | 2014-12-19 | 2015-03-25 | 深圳市捷顺科技实业股份有限公司 | License plate detection method and system |
| CN105184291A (en) * | 2015-08-26 | 2015-12-23 | 深圳市捷顺科技实业股份有限公司 | Method and system for detecting multiple types of license plates |
| CN105184291B (en) * | 2015-08-26 | 2019-07-30 | 深圳市捷顺科技实业股份有限公司 | A kind of polymorphic type detection method of license plate and system |
| CN106651774A (en) * | 2016-12-27 | 2017-05-10 | 深圳市捷顺科技实业股份有限公司 | License plate super-resolution model reconstruction method and device |
| CN106846353A (en) * | 2017-01-04 | 2017-06-13 | 美的集团股份有限公司 | Image processing method, image processing apparatus and equipment based on density |
| CN106846353B (en) * | 2017-01-04 | 2020-02-21 | 美的集团股份有限公司 | Density-based image processing method, image processing apparatus and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103745197B (en) | 2017-08-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Mnih et al. | Learning to label aerial images from noisy data | |
| CN103699905B (en) | Method and device for positioning license plate | |
| CN103745197A (en) | Detection method of license plate and device thereof | |
| CN109977782B (en) | Cross-store operation behavior detection method based on target position information reasoning | |
| US9754192B2 (en) | Object detection utilizing geometric information fused with image data | |
| WO2018072233A1 (en) | Method and system for vehicle tag detection and recognition based on selective search algorithm | |
| CN111553397B (en) | Cross-domain target detection method based on regional full convolution network and self-adaption | |
| Zhang et al. | Semi-automatic road tracking by template matching and distance transformation in urban areas | |
| CN105260749B (en) | Real-time target detection method based on direction gradient binary pattern and soft cascade SVM | |
| Jung | Detecting building changes from multitemporal aerial stereopairs | |
| US20120027252A1 (en) | Hand gesture detection | |
| CN104517095B (en) | A kind of number of people dividing method based on depth image | |
| CN103310194A (en) | Method for detecting head and shoulders of pedestrian in video based on overhead pixel gradient direction | |
| CN104239867A (en) | License plate locating method and system | |
| CN104978567A (en) | Vehicle detection method based on scenario classification | |
| CN102176208A (en) | Robust video fingerprint method based on three-dimensional space-time characteristics | |
| CN103065163B (en) | A kind of fast target based on static images detects recognition system and method | |
| CN104463220A (en) | License plate detection method and system | |
| Chen et al. | A coarse-to-fine approach for vehicles detection from aerial images | |
| CN104268595A (en) | General object detecting method and system | |
| CN108537223A (en) | A kind of detection method of license plate, system and equipment and storage medium | |
| CN104063682A (en) | Pedestrian detection method based on edge grading and CENTRIST characteristic | |
| Li et al. | Cascade Classifier Using Combination of Histograms of Oriented Gradients for Rapid Pedestrian Detection. | |
| CN104331708B (en) | A kind of zebra crossing automatic detection analysis method and system | |
| CN106295556A (en) | A kind of Approach for road detection based on SUAV Aerial Images |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170829 |