CN103514427B - Vehicle checking method and system - Google Patents
Vehicle checking method and system Download PDFInfo
- Publication number
- CN103514427B CN103514427B CN201210202064.0A CN201210202064A CN103514427B CN 103514427 B CN103514427 B CN 103514427B CN 201210202064 A CN201210202064 A CN 201210202064A CN 103514427 B CN103514427 B CN 103514427B
- Authority
- CN
- China
- Prior art keywords
- distance
- image
- subclass
- histogram
- sample image
- 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.)
- Expired - Fee Related
Links
Landscapes
- Image Analysis (AREA)
Abstract
The invention provides a kind of vehicle checking method and system, the method includes: the scene image obtaining vehicle front as input picture and reads the sample image in sample image storehouse；Extract the range direction edge histogram feature of the two dimension of image and the range direction edge histogram characteristic expansion of described two dimension is become one-dimensional characteristic vector；Cluster and utilize the positive negative sample of sample image to train subclass grader the one-dimensional characteristic vector of sample image；And the one-dimensional characteristic vector of corresponding input picture is included into the subclass corresponding to sample image closest with it；And utilize whether corresponding subclass grader comprises vehicle in judging input picture.The problem that the method for the present invention changes greatly in effectively solving class and between class, similarity is relatively large, improves the systematic function of vehicle detection.
Description
Technical field
The invention belongs to image procossing and field of object detection, relate to a kind of vehicle checking method, particularly relate to a kind of logical
Cross and captured frontal scene is carried out image procossing to detect whether vehicle front exists the method for vehicle and use the method
System.
Background technology
Along with popularizing of automobile, daily transportation is more and more busier, and traffic safety is increasingly becoming Jiao that people are paid close attention to
Point.Accordingly it is desirable to the vehicle participating in traffic can possess more Intelligent Characteristics, thus substitute human pilot and judge one
Divide safety issues.One aspect of traffic safety relates to the composed component of vehicle whether safety own, and another aspect relates to vehicle row
The environment entered whether safety.For the latter, mainly understand the situation on road surface by driver and judge.But, driving
Member be in fatigue or without noticing of in the case of, often cause danger.Accordingly, it is desirable to a kind of method and system can be
Driver replaces driver to carry out the judgement of some road safety situations, to evade danger in the case of failing to judge.And
On road surface, maximum danger is often the collision between vehicle, therefore, occurs even if understanding vehicle periphery, especially front
Vehicle be just particularly important.Accordingly it is desirable to one can be provided on vehicle can to detect other vehicles on road surface
The method and system of position, in order to vehicle control syetem can give the operation that institute's testing result controls vehicle.In prior art
Although proposing this vehicle checking method, but owing in a practical situation, road surface there will be and there is similar shape with vehicle
Between non-vehicle and the vehicle of shape there is bigger difference in the shape of itself, and therefore, prior art is judging obtained image
In vehicle and non-vehicle on often there is the problem that precision is the highest, therefore cause control system vehicle to be sent and actual scene
The instruction that middle vehicle existence is not inconsistent.
Summary of the invention
The present invention is made in view of the above-mentioned problems in the prior art.The present invention is directed to cars different in vehicle detection
The problem that change of shape is bigger, proposes " distance-direction " edge histogram feature, and straight based on this " distance-direction " edge
Side's figure feature automatically generates a series of subclass, and again with the grader of the positive and negative sample training subclass in each subclass, in order to
During detection by with the distance relation of subclass automatically by there may be the image division of vehicle to corresponding subclass, based on through instruction
The sub-classifier practiced discriminates whether to there is vehicle.
To this end, the invention provides a kind of vehicle checking method, including: obtain the scene image of vehicle front as input
Sample image in image and reading sample image storehouse；Extract the distance-limit, direction of the two dimension of each read sample image
Edge histogram feature and the distance-direction edge histogram feature of extraction input picture two dimension；By the distance-side of described two dimension
One-dimensional characteristic vector is become to edge histogram characteristic expansion；The one-dimensional characteristic vector of sample image is clustered thus obtain right
In the subclass of sample image and subclass should be numbered；Utilize the positive negative sample of sample image to train subclass grader；With
And the one-dimensional characteristic vector of corresponding input picture is included into the subclass corresponding to sample image closest with it；And utilize
Whether corresponding subclass grader comprises vehicle in judging input picture.
Vehicle according to the invention detection method, the step of the distance-direction edge histogram feature of described extraction two dimension
Including: in sample image and input picture, for each pixel, each pixel to sample image and input picture
Carry out rim detection thus obtain edge image；In sample image and input picture, for each edge in edge image
Point, calculates its gradient vector, and obtains the gradient direction angle of this marginal point based on the gradient vector calculated；Calculate corresponding diagram
Inconocenter point is to the vertical dimension of the straight line of gradient direction angle by vertically being calculated through marginal point and direction；And to correspondence
In image, the distance calculated and the direction of all marginal points are added up, and set up two-dimensional distance-direction edge histogram, thus
Obtain the two-dimensional distance-direction edge histogram of correspondence image.
Vehicle according to the invention detection method, the described distance calculated to marginal points all in correspondence image and side
To adding up, set up two-dimensional distance-direction edge histogram, thus the two-dimensional distance-edge, direction obtaining correspondence image is straight
Side's figure includes: according to a certain distance interval, distance farthest for image distance central point is divided into multiple distance distribution histogram group
Away from, and set of histograms of adjusting the distance in order is away from carrying out serial number；Angle of circumference is divided into multiple according to certain angle interval
Direction histogram group away from, and in order to direction set of histograms away from carrying out serial number；
By by the orientation angle of each marginal point divided by direction histogram group away from and round downwards, it is thus achieved that this marginal point institute
The direction histogram group belonged to is away from numbering；By by the vertical dimension calculated of each marginal point divided by distance distribution histogram group away from also
Round downwards, it is thus achieved that the distance distribution histogram group belonging to this marginal point is away from numbering；
Gradient vector based on marginal point, calculates the gradient magnitude of marginal point；And between marginal point and image center
Distance inversely proportionally calculates the normalized distance weighting factor of each marginal point；And by the gradient magnitude of marginal point with
Product between the distance weighting factor calculated is to calculate the ballot weights of this marginal point；Exist according to the ballot weights of marginal point
Voted in the grid constituted away from distance distribution histogram group by direction histogram group, thus obtain the Two-dimensional Distance of each image
From-direction edge histogram.
Vehicle according to the invention detection method, becomes one by the distance of described two dimension-direction edge histogram characteristic expansion
The step of dimensional feature vector includes: to each by direction histogram group away from and distance distribution histogram group carry out continuously away from the grid constituted
After numbering, the ballot amount according to each grid forms rectangular histogram.
Vehicle according to the invention detection method, the described one-dimensional characteristic vector by corresponding input picture is included into and its distance
The nearest subclass step corresponding to sample image includes: calculate the one-dimensional characteristic of all sample images that each subclass is comprised
The average vector of vector；The distance of vectorial the obtained average vector with each subclass of one-dimensional characteristic of calculating input image, and
Compare calculated distance；And the one-dimensional characteristic vector of input picture is included into the subclass corresponding with minimum range, and should
The numbering of subclass gives the one-dimensional characteristic vector of this input picture.
According to another embodiment of the invention, additionally provide a kind of vehicle detecting system, including: image input units,
The scene image obtaining vehicle front as input picture and reads the sample image in sample image storehouse；Characteristic vector pickup
Unit, extracts the distance-direction edge histogram feature of the two dimension of each read sample image and extracts input picture two
Distance-direction edge histogram the feature of dimension, and the distance of described two dimension-direction edge histogram characteristic expansion is become one-dimensional spy
Levy vector；Cluster cell, clusters the one-dimensional characteristic vector of sample image thus obtains the subclass corresponding to sample image
And subclass is numbered；
Subclass training unit, utilizes the positive negative sample of sample image to train subclass grader；Subclass division unit, by right
The one-dimensional characteristic vector answering input picture is included into the subclass corresponding to sample image closest with it；And authentication unit,
Utilize whether corresponding subclass grader comprises vehicle in judging input picture.
The detailed description of preferred embodiment of the invention below being considered in conjunction with the accompanying by reading, is better understood with this
Above and other target, feature, advantage and the technology of invention and industrial significance.
Accompanying drawing explanation
Shown in Fig. 1 is the flow chart training subclass grader according to an embodiment of the invention.
Shown in Fig. 2 be according to embodiments of the present invention carry out vehicle in candidate image based on the subclass grader being trained to
The flow chart of detection.
Shown in Fig. 3 is the flow chart of the distance-direction histogram feature extracting image according to embodiments of the present invention.
Shown in Fig. 4 A-4E is the signal of the distance-direction histogram feature extracting image according to embodiments of the present invention
Figure.
Shown in Fig. 5 is the structural representation of classifier training unit according to an embodiment of the invention.
Shown in Fig. 6 is the structural representation of vehicle detection unit according to embodiments of the present invention.
Shown in Fig. 7 is the structural representation of feature extraction unit according to embodiments of the present invention.
Shown in Fig. 8 is the use environment schematic of the vehicle checking method using the present invention.
Detailed description of the invention
In general, method and apparatus of the present invention can be used for in-vehicle camera, the method for the application of the invention and
Device, it is provided that prevent the vehicle using the present invention and other automobile crash.Certainly present invention may also apply to road monitoring or
Detection to vehicle in intelligent transportation.For this reason, it may be necessary to vehicle or monitoring device road pavement situation carry out vehicle detection.The present invention is led to
Cross and gather the positive negative sample of vehicle at various illumination conditions, they are extracted feature, and clusters, for be polymerized to each
Subclass is respectively trained grader, then uses trained grader to carry out vehicle detection.When carrying out vehicle detection, first
First obtain the image of current scene, then generate the candidate region that there may be vehicle, each candidate region is extracted feature, and
According to these features, candidate region is divided into the subclass of correspondence, finally utilizes the grader of corresponding subclass to differentiate this candidate region
Whether there is vehicle.In order to improve the precision of vehicle detection, the present invention propose " distance-direction " histogram feature this without exception
Read.
The embodiment of the present invention is described below in conjunction with the accompanying drawings.
Shown in Fig. 1 is the flow chart training subclass grader according to an embodiment of the invention.As it is shown in figure 1, in step
At rapid S11, classifier training unit reads sample image from the Sample Storehouse of the memorizer 121 of system.Include vehicle pattern
Sample image be referred to as positive sample, and the spurious edition image not comprising vehicle pattern is referred to as negative sample.At step S12, feature
Extraction unit extracts characteristic vector from every width sample image.About specifically how extracting characteristic vector, later herein with reference to attached
Fig. 3 and 7 is described.Subsequently, at step S13, the characteristic vector extracted based on every width figure, carry out clustering processing.Cluster
Common k-nearest neighbor, K means Method can be used.Specifically exactly the characteristic vector of each image is carried out similarity
Relatively, two width sample images corresponding to the characteristic vector within certain threshold range of the similarity between any two are classified as one
Class.It is of course also possible to use other modes existing to cluster.It practice, cluster is a kind of routine techniques, contain various
Cluster mode, is not specifically described at this.Afterwards, at step S14, by sample corresponding for institute's directed quantity of being classified as a class
This image is classified as a subclass, and is numbered each subclass.These subclasses average in feature space and variance will be by
When dividing candidate image when storing for detection.Last at step S15, utilize the methods such as existing machine learning to each
Subclass is trained separate for positive negative sample grader.Training subclass grader can use support vector machine, AdaBoost etc..
After trained subclass grader for each subclass, it is possible to use subclass detects the on-the-spot image obtained
In whether there is vehicle.Shown in Fig. 2 be according to embodiments of the present invention carry out candidate's figure based on the subclass grader being trained to
Flow chart as interior vehicle detection.As in figure 2 it is shown, generally scene image is owing to being continuously shot, the short time is interior or produces a lot of field
Scape image.In order to reduce amount of calculation, improve detection speed, it is necessary first at step S21, from current scene, detect candidate's figure
As region is as input picture.This step can use existing method, utilizes some simple features, as relied on symmetry,
Or Haar+AdaBoost framework etc..Scene image generally uses scene image acquiring unit to obtain.Scene image acquiring unit
Usually onboard camera, it speed based on vehicle can shoot photo, constantly be inputed to by the scene image of vehicle front
Car-mounted computer.Subsequently, at step S22, feature extraction unit proposes characteristic vector from the scene image received.Herein
The feature extraction unit used can be to be same with the feature extraction unit in classifier training unit, it is also possible to individually
The feature extraction unit arranged.The extraction process of the characteristic vector herein performed is retouched later in reference to accompanying drawing 3 and 7 too
State.Then, at step S23, in the characteristic vector of the corresponding scene image of calculating and sample image set between each subclass
Distance.Specifically, it is simply that for each subclass, the average of all sample image characteristics of correspondence vector in this subclass is calculated
Value, thus obtain the center of the averaged feature vector of this subclass, i.e. this subclass, the characteristic vector of the most corresponding scene image
Averaged feature vector with this subclass, it is thus achieved that distance between the two, this distance be then corresponding scene image characteristic vector with
The distance of this subclass.Then, at step S24, using with the closest subclass of the characteristic vector of corresponding scene image as
Corresponding subclass belonging to scene image, and the numbering of this subclass is given the characteristic vector of this correspondence scene image.Then, in step
At rapid S25, the numbering of the subclass of characteristic vector based on corresponding scene image calls the grader of the subclass of reference numeral.?
After, at step S26, use the subclass grader called detects in corresponding scene image whether there is vehicle.The most such as
What uses training aids to detect whether to there is vehicle in candidate image and belong to prior art, do not repeat at this.
The present invention is to improve the precision of vehicle detection, it is proposed that " distance-direction " characteristic vector.Shown in Fig. 3 is basis
The flow chart of the distance-direction histogram feature extracting image of the embodiment of the present invention.Shown in Fig. 4 A-4E is according to the present invention
The schematic diagram of the distance-direction histogram feature extracting image of embodiment.In the present invention, " distance-direction " characteristic vector
In " direction " refer to the gradient direction of each pixel in image, and " distance " is image center arrive through this marginal point and
It is perpendicular to the vertical dimension of the straight line of this gradient direction.Shown in Fig. 4 A is an example of acquired scene image.Wherein
White blockage shown in be a pixel carrying out processing.Due to the object to be detected in image edge in the picture
It is presented as that grey scale pixel value exists acute variation, therefore, in order to reduce amount of calculation, first image is carried out edge extracting process.
Edge extracting is the technology of comparative maturity, and a kind of simple mode is exactly when the left and right in the horizontal direction of certain pixel in image
Difference between the gray value of two neighbors and/or between the gray value of two neighbors up and down in vertical direction
When difference is more than a certain threshold value, then this pixel is marginal point pixel.Much method is available carries out edge extracting, such as utilizes
Canny operator, this operator extraction edge has higher precision.By edge extracting, image is performed binary conversion treatment, will symbol
The marginal point of conjunction condition is set to white, and other pixels are set to black, thus obtains the binary picture comprising white edge point
Picture, thus only edge point is processed in calculating subsequently, such that it is able to reduce amount of calculation.Fig. 4 B is to carry through edge
Take the binaryzation edge image after process.This step is performed by step S1202 in Fig. 3.It was to obtain before carrying out step S1202
Take step S1201 of candidate image.When characteristic vector pickup flow process is directed to scene image, candidate image is scene image
A part.
Afterwards, at step S1203, obtain the pixel of each marginal point of candidate image.Then in step S1204
Place, the gradient vector (Gx, Gy) of the pixel of this marginal point in candidate image.A kind of selectable simple gradient vector meter
Calculation method is Sobel operator, difference calculated level and gradient G x of vertical direction and Gy.Shown in Fig. 4 C is the white in Fig. 4 A
Picture frame carries out the schematic diagram of the process of gradient calculation after amplifying.Prior art has the calculation of a lot of gradient vector, at this
Do not enumerate.Subsequently, at step S1205, formula (1) is utilized to obtain ladder based on the gradient vector (Gx, Gy) at marginal point
Degree direction
Then, at step S1206, utilize formula (2) to obtain this gradient based on the gradient vector (Gx, Gy) at marginal point
Amplitude M of vector:
Certainly, the calculating of the amplitude of gradient vector is not that to realize necessity of the purpose of the present invention indispensable, but is
The precision of further raising characteristic vector pickup.This will be described later when the ballot of distance-direction histogram processes and retouches
State.
Afterwards, in order to extract " distance-direction " histogram feature, need angle of circumference according to certain angle interval segmentation
For multiple directions set of histograms away from, and in order to direction set of histograms away from carrying out serial number, then by each marginal point ladder
Degree direction be included into concrete angle group away from.To this end, at step S1207, utilize formula (3) by by each marginal point
Orientation angle divided by direction histogram group away from and round downwards, it is thus achieved that the direction histogram group belonging to this marginal point away from numbering count
Calculate the gradient direction correspondence group numbering away from (bin)
WhereinFor the direction group sum away from (bin),Represent and round downwards.
Subsequently, at step S1208, its gradient direction of coordinate position based on marginal pointObtain formula (4) representative
The straight line side through gradient direction (hereinafter referred to as " edge direction ") that current edge point and direction are this marginal point vertical
Journey:
Ax+By+C=0 (4)
Wherein A, B, C are the coefficient of equation, and coordinate based on pixel and gradient direction obtain.Being by shown in Fig. 4 D
The schematic diagram of the process that edge direction calculates.
Afterwards, in order to extract " distance-direction " histogram feature, need by distance farthest for image distance central point according to
A certain distance interval be divided into multiple distance distribution histogram group away from, and set of histograms of adjusting the distance in order is away from carrying out serial number,
Then utilize formula (5) by by the vertical dimension calculated of each marginal point divided by distance distribution histogram group away from and take downwards
Whole, it is thus achieved that the distance distribution histogram group belonging to this marginal point calculates training image central point (or scene image central point) away from numbering
Distance to this straight line
Afterwards, formula (6) is utilized to obtainAffiliated distance group is away from (bin) BD:
Wherein NDFor the distance group sum away from (bin), DMaxFor the maximum distance of the point on image to picture centre, this is
The remote pixel being typically on each angle of image is to the distance of image center.
According in actual surface conditions, vehicle position in the picture be generally proximal to image center some.Therefore, for
The data of marginal point are more beneficial for accurately vote foundation " distance-direction " histogram feature so that away from central point
Marginal point has less ballot weights, and the highest the closer to the ballot weights of the marginal point of central point.Similarly, since marginal point
Gradient magnitude the highest, this marginal point be more probably reality marginal point, therefore give the weights that the marginal point of amplitude is higher.
To this end, at step S1209, calculate the distance weight factor W of pixel according to formula (7) for marginal pointL,
Wherein XPAnd YPIt is respectively the current edge point horizontally and vertically distance to picture centre, and Width and Height divides
Not Wei photo width and height.
Then, at step S1210, amplitude and distance weight factor WL according to formula (8) jointing edge point gradient are counted
The ballot weights W of calculation edge pixel:
W=M*WL (8)
Afterwards, at step S1211, to current edge pixel, to the two-dimensional feature vector rectangular histogram grid of its correspondenceVote with weights W.For ease of ballot and verifying below, to each by direction histogram group away from away from
From set of histograms ballot amount formation rectangular histogram according to each grid after the grid constituted carries out serial number, thus by two dimension
Characteristic vector rectangular histogram is launched into one-dimensional characteristic histogram vector, thus decreases and follow-up compare amount of calculation.
At step S1212, it may be judged whether all of marginal point pixel is performed above-mentioned steps S1201-1211 step.
If all marginal point pixels are all disposed, at step S1213, thus obtain " distance-direction " rectangular histogram of present image
Feature.
If not carrying out weight computing of voting, the ballot weights phase of the most each edge pixel point in actual application
Deng.
Shown in Fig. 5 is the structural representation of classifier training unit according to an embodiment of the invention.Classifier training
Unit includes: sample database 11, and its storage is in memory；Feature extraction unit 12；Cluster cell 13, subclass training unit
14 and subclass grader memorizer 15.
Shown in Fig. 6 is the structural representation of vehicle detection unit according to embodiments of the present invention.Vehicle detection unit bag
Including: candidate image extraction unit 21, it extracts candidate image as input picture from captured scene image；Feature extraction
Unit 22, it extracts " distance-direction " histogram feature vector from input picture；Subclass division unit 23, it will be extracted
" distance-direction " histogram feature vector be included into the subclass of concrete sample image, and be input picture extraction " away from
From-direction " histogram feature vector is numbered with the numbering of this subclass；Authentication unit 24, verifies whether described input picture exists car
?；And display unit 25, it is used for showing the result.
Shown in Fig. 7 is the structural representation of feature extraction unit according to embodiments of the present invention.Feature extraction unit bag
Include memorizer 121, the related data of its storage input image data；Edge detection unit 122, it obtains the limit in input picture
Edge information, this marginal information includes: whether as marginal point, the position etc. of marginal point；Gradient calculation unit 123, calculates marginal point
Gradient vector；Metrics calculation unit 124, the vertical dimension of calculating input image central point to edge line, and by its this distance
Be included into corresponding distance group away from；Gradient direction computing unit 125, gradient vector based on marginal point calculates the gradient side of marginal point
To, and its gradient direction is included into corresponding direction group away from；Gradient magnitude computing unit 126, calculates edge based on gradient vector
The gradient magnitude of point；Position weight computing unit 127, based on edge pixel point to the distance of input picture central point, adopts
The position weight of marginal point pixel is calculated with inverse proportion function；And rectangular histogram ballot unit 128, based on marginal point
The ballot weights that gradient magnitude and position weight produce to affiliated direction group away from distance group away from carry out ballot obtain " away from
From-direction " rectangular histogram.
Shown in Fig. 8 is the use environment schematic of the vehicle checking method using the present invention.
The sequence of operations illustrated in the description can be held by the combination of hardware, software or hardware with software
OK.When being performed this series of operation by software, computer program therein can be installed to be built in the meter of specialized hardware
In memorizer in calculation machine so that computer performs this computer program.Or, computer program can be installed to hold
In the general purpose computer of the various types of process of row so that computer performs this computer program.
For example, it is possible to computer program is prestored to the hard disk as record medium or ROM (read only memory)
In.Or, can temporarily or permanently store (record) computer program in removable record medium, such as floppy disk, CD-
ROM (compact disc read-only memory), MO (magneto-optic) dish, DVD (digital versatile disc), disk or semiconductor memory.Can be this
The removable record medium of sample provides as canned software.
The present invention has been described in detail by reference to specific embodiment.It may be evident, however, that in the essence without departing substantially from the present invention
In the case of god, those skilled in the art can perform change and replace embodiment.In other words, the shape of present invention explanation
Formula is open rather than explains with being limited.Idea of the invention to be judged, it is contemplated that appended claim.
Claims (5)
1. a vehicle checking method, including:
The scene image obtaining vehicle front as input picture and reads the sample image in sample image storehouse；
Extract the distance-direction edge histogram feature of the two dimension of each read sample image and extract input picture two dimension
Distance-direction edge histogram feature；
The distance of described two dimension-direction edge histogram characteristic expansion is become one-dimensional characteristic vector；
The one-dimensional characteristic vector of sample image is clustered thus obtains the subclass corresponding to sample image and subclass is carried out
Numbering；
Utilize the positive negative sample of sample image to train subclass grader；And
The one-dimensional characteristic vector of corresponding input picture is included into the subclass corresponding to sample image closest with it；And
Utilize whether corresponding subclass grader comprises vehicle in judging input picture；
Wherein, the step of the described distance-direction edge histogram feature extracting two dimension includes:
In sample image and input picture, for each pixel, each pixel to sample image and input picture
Carry out rim detection thus obtain edge image；
In sample image and input picture, for each marginal point in edge image, calculate its gradient vector, and based on institute
The gradient vector calculated obtains the gradient direction angle of this marginal point；
Calculate correspondence image central point to vertical by the straight line of gradient direction angle that vertically calculated through marginal point and direction
Distance；And
The distance calculated and direction to marginal points all in correspondence image are added up, and set up two-dimensional distance-edge, direction
Rectangular histogram, thus obtain the two-dimensional distance-direction edge histogram of correspondence image.
2. vehicle checking method as claimed in claim 1, the described distance calculated to marginal points all in correspondence image
Add up with direction, set up two-dimensional distance-direction edge histogram, thus obtain the two-dimensional distance-limit, direction of correspondence image
Edge rectangular histogram includes:
By distance farthest for image distance central point according to a certain distance interval be divided into multiple distance distribution histogram group away from, and press
Order adjusts the distance set of histograms away from carrying out serial number；
By angle of circumference according to certain angle interval be divided into multiple directions set of histograms away from, and in order to direction set of histograms
Away from carrying out serial number；
By by the orientation angle of each marginal point divided by direction histogram group away from and round downwards, it is thus achieved that belonging to this marginal point
Direction histogram group is away from numbering；
By by the vertical dimension calculated of each marginal point divided by distance distribution histogram group away from and round downwards, it is thus achieved that this edge
Distance distribution histogram group belonging to Dian is away from numbering；
Gradient vector based on marginal point, calculates the gradient magnitude of marginal point；
And the distance between marginal point and image center inversely proportionally calculates the normalized distance weighting of each marginal point
The factor；And
The ballot of this marginal point is calculated by the product between gradient magnitude and the distance weighting factor calculated of marginal point
Weights；
Throwing in the grid constituted away from distance distribution histogram group by direction histogram group according to the ballot weights of marginal point
Ticket, thus obtain the two-dimensional distance-direction edge histogram of each image.
3. vehicle checking method as claimed in claim 2, becomes the distance of described two dimension-direction edge histogram characteristic expansion
The step of one-dimensional characteristic vector includes:
To each by direction histogram group away from and distance distribution histogram group after the grid constituted carries out serial number according to each side
The ballot amount of lattice forms rectangular histogram.
4. vehicle checking method as claimed in claim 3, the described one-dimensional characteristic vector by corresponding input picture is included into and it
The closest subclass step corresponding to sample image includes:
Calculate the average vector of the one-dimensional characteristic vector of all sample images that each subclass is comprised；
The distance of vectorial the obtained average vector with each subclass of the one-dimensional characteristic of calculating input image, and compare and calculated
Distance；And
The one-dimensional characteristic vector of input picture is included into the subclass corresponding with minimum range, and it is defeated to give this by the numbering of this subclass
Enter the one-dimensional characteristic vector of image.
5. a vehicle detecting system, including:
Image input units, the scene image obtaining vehicle front as input picture and reads the sample in sample image storehouse
Image；
Characteristic vector pickup unit, extract each read sample image two dimension distance-direction edge histogram feature with
And extract distance-direction edge histogram feature that input picture is two-dimentional, and by the distance of described two dimension-direction edge histogram
Characteristic expansion becomes one-dimensional characteristic vector；
Cluster cell, clusters the one-dimensional characteristic vector of sample image thus obtains the subclass corresponding to sample image right
Subclass is numbered；
Subclass training unit, utilizes the positive negative sample of sample image to train subclass grader；
Subclass division unit, is included into closest with it corresponding to sample image by the one-dimensional characteristic vector of corresponding input picture
Subclass；And
Authentication unit, utilizes whether corresponding subclass grader comprises vehicle in judging input picture；
Wherein, described characteristic vector pickup unit includes:
Edge image obtains unit, in sample image and input picture, for each pixel, to sample image and input figure
Each pixel of picture carries out rim detection thus obtains edge image；
Gradient direction angle computing unit, in sample image and input picture, for each marginal point in edge image, meter
Calculate its gradient vector, and obtain the gradient direction angle of this marginal point based on the gradient vector calculated；
Vertical dimension computing unit, calculates correspondence image central point to being the gradient side vertically calculated through marginal point and direction
Vertical dimension to the straight line at angle；And
Edge histogram obtains unit, and the distance calculated and direction to marginal points all in correspondence image are added up, and build
Vertical two-dimensional distance-direction edge histogram, thus obtain the two-dimensional distance-direction edge histogram of correspondence image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210202064.0A CN103514427B (en) | 2012-06-15 | 2012-06-15 | Vehicle checking method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210202064.0A CN103514427B (en) | 2012-06-15 | 2012-06-15 | Vehicle checking method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103514427A CN103514427A (en) | 2014-01-15 |
| CN103514427B true CN103514427B (en) | 2016-12-21 |
Family
ID=49897128
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210202064.0A Expired - Fee Related CN103514427B (en) | 2012-06-15 | 2012-06-15 | Vehicle checking method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103514427B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111768409B (en) * | 2020-09-03 | 2020-12-04 | 腾讯科技（深圳）有限公司 | Box-type structure detection method and device based on artificial intelligence |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101469991B (en) * | 2007-12-26 | 2011-08-10 | 南京理工大学 | All-day structured road multi-lane line detection method |
| CN101881615B (en) * | 2010-05-28 | 2012-07-11 | 清华大学 | Method for detecting visual barrier for driving safety |
| CN101975951B (en) * | 2010-06-09 | 2013-03-20 | 北京理工大学 | Field environment barrier detection method fusing distance and image information |
| JP5867807B2 (en) * | 2010-12-08 | 2016-02-24 | 株式会社リコー | Vehicle identification device |
-
2012
- 2012-06-15 CN CN201210202064.0A patent/CN103514427B/en not_active Expired - Fee Related
Also Published As
| Publication number | Publication date |
|---|---|
| CN103514427A (en) | 2014-01-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US8099213B2 (en) | Road-edge detection | |
| CN103699905B (en) | Method and device for positioning license plate | |
| CN109522784A (en) | For distinguishing the device and method that can be crossed object and object can not be crossed | |
| CN111626277B (en) | Vehicle tracking method and device based on over-station inter-modulation index analysis | |
| CN106169244A (en) | The guidance information utilizing crossing recognition result provides device and method | |
| CN105574552A (en) | Vehicle ranging and collision early warning method based on monocular vision | |
| CN102194102B (en) | Method and device for classifying a traffic sign | |
| CN103456003A (en) | Device and method for tracking object by using characteristic point descriptor, device and method for removing erroneous characteristic | |
| CN105488501A (en) | Method for correcting license plate slant based on rotating projection | |
| JP5388291B2 (en) | Discriminator generation method, computer program, discriminator generation device, and predetermined object detection device | |
| CN104933398A (en) | vehicle identification system and method | |
| CN103049733A (en) | Human face detection method and human face detection equipment | |
| JP6818626B2 (en) | Vehicle type discrimination device, vehicle type discrimination method, and vehicle type discrimination system | |
| CN103116757A (en) | Three-dimension information restoration and extraction method for identifying spilled articles on roads | |
| CN105160292A (en) | Vehicle identification recognition method and system | |
| CN105976392A (en) | Maximum-output-probability-based vehicle tyre detection method and apparatus | |
| CN104463220A (en) | License plate detection method and system | |
| Mammeri et al. | North-American speed limit sign detection and recognition for smart cars | |
| CN104700068A (en) | SVM based detection method of safety belt of driver | |
| Blanc et al. | LaRASideCam: A fast and robust vision-based blindspot detection system | |
| Premachandra et al. | Image based automatic road surface crack detection for achieving smooth driving on deformed roads | |
| CN103514427B (en) | Vehicle checking method and system | |
| JP2013122739A (en) | Analysis device, analysis program and analysis method | |
| CN106778473A (en) | A kind of model recognizing method | |
| CN112329886A (en) | Double-license plate recognition method, model training method, device, equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20161221 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |