CN106503638B - Image procossing, vehicle color identification method and system for color identification - Google Patents
Image procossing, vehicle color identification method and system for color identification Download PDFInfo
- Publication number
- CN106503638B CN106503638B CN201610893880.9A CN201610893880A CN106503638B CN 106503638 B CN106503638 B CN 106503638B CN 201610893880 A CN201610893880 A CN 201610893880A CN 106503638 B CN106503638 B CN 106503638B
- Authority
- CN
- China
- Prior art keywords
- target
- pixel
- color
- vehicle
- target object
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
- G06V20/584—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads of vehicle lights or traffic lights
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/08—Detecting or categorising vehicles
Abstract
The present invention relates to a kind of image procossing, vehicle color identification method and systems for color identification, the above-mentioned image processing method for color identification, include the following steps: to choose multiple videos from video acquisition system, several frame images are chosen from each video respectively, training set is constructed, the target context code book of video acquisition system is identified according to the training set；According to the corresponding foreground target pixel of target object in the target context code book recognition target image；The corresponding pixel of target object reflector segment in the foreground target pixel is rejected, the corresponding subject pixel point of main part is obtained；The color identification that target object is carried out according to the subject pixel point, determines the color of target object；Its accuracy for effectively increasing identified color.
Description
Technical field
The present invention relates to technical field of image processing, more particularly to a kind of image procossing for color identification, vehicle
Color identification method and system.
Background technique
Currently, video recording system etc. has been answered extensively in the video acquisition system that a certain fixed point obtains corresponding place video image
In people's lives and work.Acquisition and respective handling are carried out to the video in above-mentioned video acquisition system, identify target
The color of object is of great significance to security or related management work；For example, can be in urban road or community gate inhibition
Vehicle monitoring video is obtained in the bayonet system in equal places, carries out the color identification of wherein some vehicle, and then determines above-mentioned vehicle
Identity, for the work such as vehicle retrieval, traffic administration.
Tradition is in the color identifying schemes of target image in video, there is method to utilize HSI color space and reference colour
Poor formula calculates color value, recycles the BP neural network to carry out the color classification of the target images such as vehicle；There is method to use
The matched method of color template realizes the classification of target image color；There is method to complete target using H-S two-dimensional color histogram
The classification of color of image；There are also method by hsv color space acutance and saturation degree component separate, then constructed with this
Classifier completes color classification, accordingly to be identified to target image color.However some targets in video acquisition system
Object, the main part (body portion of such as vehicle) including needing to carry out color identification, it is also possible to including interference color identification
Reflector segment (the vehicle window part on such as vehicle)；Above-mentioned color identifying schemes are easy by reflector segment in target object
Interference, keeps identified color accuracy low.
Summary of the invention
Based on this, it is necessary to for the technical problem that traditional scheme keeps identified color accuracy low, provide a kind of use
In image procossing, vehicle color identification method and the system of color identification.
A kind of image processing method for color identification, includes the following steps:
Multiple videos are chosen from video acquisition system, choose several frame images, construction training from each video respectively
Collection identifies the target context code book of video acquisition system according to the training set；
According to the corresponding foreground target pixel of target object in the target context code book recognition target image；
The corresponding pixel of target object reflector segment in the foreground target pixel is rejected, it is corresponding to obtain main part
Subject pixel point；
The color identification that target object is carried out according to the subject pixel point, determines the color of target object.
A kind of image processing system for color identification, comprising:
It chooses module and chooses several frames from each video respectively for choosing multiple videos from video acquisition system
Image constructs training set, and the target context code book of video acquisition system is identified according to the training set；
Identification module, for according to the corresponding foreground target of target object in the target context code book recognition target image
Pixel；
Module is rejected to obtain for rejecting the corresponding pixel of target object reflector segment in the foreground target pixel
To subject pixel point；
Determining module, the color for carrying out target object according to the subject pixel point identifies, determines target object
Color.
Above-mentioned image processing method and system for color identification, can be according to multiple videos in video acquisition system
It determines target context code book, utilizes the corresponding foreground target pixel of target object in upper target context code book recognition target image
Point, then the corresponding pixel of reflector segment of color identification interference will be easy to cause to reject in foreground target pixel, then
Corresponding color identification is carried out to the subject pixel point of energy accurate characterization target object color, to determine the color of target object,
Effectively increase the accuracy of identified color.
A kind of vehicle color identification method, includes the following steps:
Vehicle monitoring video is obtained from bayonet system；In vehicle monitoring video identification target vehicle, with target vehicle
Vehicle body is main body portion, using the vehicle window of target vehicle as reflector segment；
Body color is identified according to the above-mentioned image processing method for the identification of target object color, according to the vehicle identified
Body color determines the color of target vehicle.
A kind of vehicle color identifying system, comprising:
Module is obtained, for obtaining vehicle monitoring video from bayonet system；Target vehicle is chosen in vehicle monitoring video,
With the vehicle body of target vehicle for main body portion, using the vehicle window of target vehicle as reflector segment；
Vehicle color determining module, for identifying vehicle according to the above-mentioned image processing system for the identification of target object color
Body color determines the color of target vehicle according to the body color identified.
Above-mentioned vehicle color identification method and system can choose multiple automobile video frequencies from bayonet system and determine background mesh
Coding sheet, using the corresponding foreground target pixel of upper target context code book identification target vehicle, then by foreground target pixel
In be easy to cause color identification interference the corresponding pixel of vehicle window rejected, then to can accurate characterization vehicle color vehicle body
Pixel carries out corresponding color identification, makes identified vehicle color accuracy with higher.
Detailed description of the invention
Fig. 1 is the image processing method flow chart for color identification of one embodiment；
Fig. 2 is the positional diagram between the main shaft and vehicle of one embodiment；
Fig. 3 is the positional diagram between the main shaft and vehicle of one embodiment；
Fig. 4 is the A-B plane subregion schematic diagram of one embodiment；
Fig. 5 is the vehicle color identification process schematic diagram of one embodiment；
Fig. 6 is the image processing system schematic diagram for color identification of one embodiment.
Specific embodiment
With reference to the accompanying drawing to the image procossing identified for color of the invention, vehicle color identification method and system
Specific embodiment is described in detail.
The image processing method flow chart identified for color of one embodiment is shown with reference to Fig. 1, Fig. 1, including such as
Lower step:
S10 chooses multiple videos from video acquisition system, chooses several frame images from each video respectively, construction
Training set identifies the target context code book of video acquisition system according to the training set；
Above-mentioned video acquisition system, which may include bayonet system at vehicle outpost etc., to be acquired the place in fixed-site and regards
The system of frequency image.Multiple videos can be randomly selected from video acquisition system, then are chosen from above-mentioned each video several
Frame image, makes selected image and the training set according to above-mentioned image configuration includes the spy of corresponding fixed-site as far as possible
Sign.Above-mentioned target context code book includes removing in target object to be identified (such as the vehicle by corresponding outpost) and fixed-site
Place background outside above-mentioned target object.Above-mentioned target context code book has its corresponding code word CBComposition, according to respectively from
Several frame images in each video are determined, integrality and accuracy with higher.
S20, according to the corresponding foreground target pixel of target object in the target context code book recognition target image；
Target image includes target object and place background, and foreground target pixel is the corresponding picture of above-mentioned target object
Vegetarian refreshments.
In one embodiment, it is described according to target object in the target context code book recognition target image it is corresponding before
The step of scape target pixel points may include:
It identifies all code words in the target context code book, non-background picture is searched in the corresponding pixel of target image
Vegetarian refreshments obtains foreground target pixel.
The corresponding pixel of above-mentioned target image includes background pixel point and non-background pixel point, and background pixel point refers to place
The corresponding pixel of background, non-background pixel point refer to the corresponding pixel of target object, as foreground target pixel.
As one embodiment, above-mentioned the step of searching non-background pixel point in the corresponding pixel of target image, can be with
Include:
The rgb value for obtaining any one pixel p in target image calculates the lightness I of pixel p according to the rgb value,
Wherein,R is the R component value of pixel p, and G is the G component value of pixel p, and B is B points of pixel p
Magnitude；
The pixel that following condition A or condition B are unsatisfactory in target image is determined as non-background pixel point:
Condition A:Color_distance (p, CB)≤e, wherein p indicates the pixel in target image, CBTarget context
Color distance is sought in code word in code book, Color_distance () expression, and e indicates preset color threshold；Above-mentioned e is usual
It is set as a lesser threshold value, such as 0.3 or 0.5 is equivalent；
Condition B: Indicate the corresponding minimum lightness of pixel in target image,
Indicate the corresponding maximum lightness of pixel in target image.Lightness function brightness's is defined as:Wherein,min
{ } expression is minimized, m1For preset first lightness parameter, m2For preset second lightness parameter；Above-mentioned first lightness parameter
m1It can be set to 0.5, the second lightness parameter m2It can be set to 1.5.
All code words have been checked in traversal, i.e., target context code book has been substituted into condition A and condition B and carries out non-background pixel
After the identification of point, if some pixel is mismatched with all code words, it can determine whether that the pixel is foreground target pixel.
After successively all pixels point detection of image to be processed is finished, foreground target pixel can be determined.
S30 rejects the corresponding pixel of target object reflector segment in the foreground target pixel, obtains main part
Corresponding subject pixel point；
Above-mentioned foreground target pixel may include the corresponding subject pixel point of main part, that is, need to carry out color identification
Pixel, further include the corresponding pixel of reflector segment due to reflective interference corresponding color identification, such as vehicle window in vehicle
Corresponding pixel.
In one embodiment, above-mentioned target object (such as vehicle) is zhou duicheng tuxing；The light-reflecting portion of the target object
Divide and is symmetrically distributed in above main part；Under normal conditions, it is reflector segment above main shaft, is main body portion below main shaft；
It is described to reject the corresponding pixel of target object reflector segment in the foreground target pixel, obtain subject pixel
Point the step of include:
Identify the main shaft of the corresponding target object of foreground target pixel；
Main shaft foreground target pixel below is determined as subject pixel point.
The vehicle window reflector segment of such as vehicle usually will appear the high optical phenomenon of mirror-reflection, even by color correction
Also these high optical phenomenons can not be restored and give acquisition image bring interference, if carried out according to such noisy image corresponding
Color identification, can the identified result of high degree influence.The corresponding pixel of target object reflector segment is rejected, it can
To guarantee the accuracy of identified color.
It is above-mentioned after the distribution for determining the foreground target pixel is to be horizontal distribution as one embodiment,
The step of main shaft of the corresponding target object of identification foreground target pixel may include:
Central moment is set are as follows:Wherein,
TBIndicate foreground target pixel, (x, y) ∈ TBIndicate that pixel x and y belong to foreground target pixel, p and q are respectively indicated down
Variable is marked, round numbers can be distinguished, p and q take different integers, are respectively provided with corresponding λp,q, for example, p=2, q=0, then this
When λp,qFor λ2,0；
According to towards angle calculation formulaIt calculates towards angleWhereinIt is above-mentionedIt is a temporary variable symbol, in the parametric procedure minimized
The intermediate variable used；Above-mentioned steps can be by rightFirst derivative is sought to calculate towards angle
If describedValue be located at sectionOn, then determine that the distribution of the foreground target pixel is
It is horizontal distribution.
In the present embodiment, for the locating region of the reflector segments such as accurate detection vehicle window, the prospect mesh is determined first
The distribution of pixel is marked to be horizontal distribution, and then accurately detects the direction of the main shaft of target object.By taking vehicle as an example,
Above-mentioned main shaft is with the vehicle window of vehicle above and below along parallel, main shaft being denoted as in whole image towards angleIf target object is
Vehicle,Positioned at sectionOn, then the positional diagram between above-mentioned main shaft and vehicle can with as shown in Fig. 2,
Otherwise it can be determined that vehicle target is vertical distribution in the target image, the positional diagram between main shaft and vehicle can
With as shown in Figure 3.In Fig. 2, vehicle is horizontal distribution, and vehicle is divided into two parts by cut-off rule (main shaft), is located at and divides
Main part after the part removal vehicle window of secant lower end；In Fig. 3, vehicle is vertical distribution, and vehicle is divided by cut-off rule
Two parts, positioned at the part of cut-off rule lower end remove vehicle window after vehicle target (main part).
S40 identifies according to the color that the subject pixel point carries out target object, determines the color of target object.
In one embodiment, the process packet of the above-mentioned color identification that target object is carried out according to the subject pixel point
It includes:
The rgb value of the subject pixel point is subjected to standardization calculating, obtains the R component value I after standardization calculatesR, G point
Magnitude IGWith B component value IB, wherein IR=r/ (r+g+b), IG=g/ (r+g+b), IBPixel based on=b/ (r+g+b), r
R component value, the G component value of pixel based on g, the B component value of pixel based on b；
According to the IR、IGAnd IBJudge whether subject pixel point belongs to secondary colour；
If the subject pixel point belongs to secondary colour, subject pixel point input secondary colour classifier is subjected to color
Identification；
If the subject pixel point is not belonging to secondary colour, subject pixel point input monochromatic color classifier is subjected to face
Color identification.
Such as vehicle target object generally includes secondary colour and monochromatic color, first can carry out pixel tone mirror to target object
Not, to judge whether subject pixel point belongs to secondary colour, then subject pixel point is inputted into corresponding color sorter (such as secondary colour
Classifier or monochromatic color classifier) identification of corresponding color is carried out, to carry out target object color identification.
As one embodiment, subject pixel point can be converted into LAB color space, using A-B plane, using pole
Coordinate pair A-B plane is sampled, and is quantization unit with 10 in polar diameter direction；In polar angle direction, with 90 degree for quantization unit；
A-B plane is divided into 28 sub-regions in total, as shown in figure 4, each subregion corresponds to a Color Statistical table, uses mLAB
(k)=# q | q ∈ binkIndicate k-th of Color Statistical table in pixel quantity, can construct one 28 dimension vector mLAB=
[mLAB(1),mLAB(2),...mLAB(28)] ', which is the part LAB of blend color feature vector.
For RGB component, 6 dimensional vectors of a record component pixel quantity relativeness will be constructed, the 6 of the vector
What a component recorded respectively is the quantity of B > G, G > R, G > B, G > R, R > B, the corresponding pixel of R > G, with symbol mRGB=[mRGB
(1),mRGB(2),...mRGB(6)] ' indicate.The two vectors are merged to the feature vector m for constituting one 34 dimensionDG, it is input to SVM
Classify in classifier, completes the building of the classifier of monochromatic color.
As one embodiment, 3 color components in RGB color are quantified as 8 grades, thus a total of 24 respectively
The quantity of this corresponding pixel of 24 grades is respectively constituted the feature vector of one 24 dimension, is denoted as by a color grade
mquan, in addition, recording the quantity of B > G, G > R, G > B, G > R, R > B, the corresponding pixel of R > G in 3 channels, it is denoted as: mRGB.Two
Person merges, and obtains the feature vector m of one 30 dimensionDC.This feature vector is input in SVM classifier and is classified, is completed
The building of the classifier of secondary colour.
It is above-mentioned according to the I as one embodimentR、IGAnd IBJudge whether subject pixel point belongs to the process of secondary colour
May include:
Calculate the Euclidean distance d between subject pixel point and secondary colour central axis vectorp, wherein dp=(IR-0.333)2
+(IG-0.333)2+(IB-0.333)2；
Calculate the Euclidean distance d between subject pixel point and hybrid mean value vectormean, wherein dmean=(r-ur)2+(g-
ug)2+(b-ub)2, urIndicate the mean value of the R component value of all subject pixel points, ugIndicate the G component value of all subject pixel points
Mean value, ubIndicate the mean value of the B component value of all subject pixel points；
Calculate separately dmeanMean μmean, dpMean μd, dmeanVariances sigmameanAnd dpVariances sigmad；
According toIt calculates subject pixel point and belongs to mixing
The probability P (gray | p) of color；
P (gray | p) the subject pixel point for being greater than probability threshold value is judged to belonging to the subject pixel point of secondary colour.
P (gray | p) is greater than probability threshold value, i.e. and P (gray | p) > ThpIn, above-mentioned probability threshold value ThpIt can be set to 80%
Equivalence passes through what is averagely obtained to all mixing color pixels when constructing two training sample databases.Due to vehicle
Etc. the complexity of target object in structure and the interference of ambient noise, target object to be detected usually both containing mixing colour
Element also contains single-frequency color pixel.Can thus be judged by following rule: mixing color pixel accounting is more than 80%, then recognizes
It is to belong to secondary colour for the vehicle target color；Otherwise, then it is assumed that the vehicle target color is to belong to monochromatic color.
It,, can be by vehicle color in conjunction with the vehicle color acquired in bayonet system by taking vehicle as an example as one embodiment
Classification is totally divided into two major class: secondary colour, monochromatic color, which is known as the identification of pixel tone.Then big according to two respectively
Class constructs training sample database, it may be assumed that secondary colour training sample database, monochromatic color training sample.Sample in two training sample databases is not
There are any intersections.First setting vehicle color classification include: black, it is white, silver color, navy blue, bottle green, yellow, red, green
Color, blue.
The above-mentioned process for carrying out color identification to vehicle can be as shown in figure 5, carry out picture to vehicle target pixel first
Plain color tune identifies,, can be by above-mentioned vehicle if belonging to secondary colour to judge whether the subject pixel point of vehicle belongs to secondary colour
Subject pixel point carry out secondary colour feature extraction, then input secondary colour classifier (SVM) carry out color identification, above-mentioned mixing
Color includes black, white, silver color, navy blue, bottle green；It, can be by the subject pixel point of above-mentioned vehicle if belonging to monochromatic color
Secondary colour feature extraction is carried out, then inputs monochromatic color classifier (SVM) and carries out color identification, above-mentioned monochromatic color includes yellow, red
Color, green, blue.
It, can be according to multiple views in video acquisition system provided by the present invention for the image processing method of color identification
Frequency determines target context code book, utilizes the corresponding foreground target pixel of target object in upper target context code book recognition target image
Point, then the corresponding pixel of reflector segment of color identification interference will be easy to cause to reject in foreground target pixel, then
Corresponding color identification is carried out to the subject pixel point of energy accurate characterization target object color, to determine the color of target object,
Effectively increase the accuracy of identified color.
In one embodiment, the Cheng Qian that crosses of the above-mentioned color identification that target object is carried out according to the subject pixel point goes back
May include:
The subject pixel point is divided into M pixel subregion, single region is denoted as T 'i, selected in M pixel region
A frame image R is selected as construction frame, obtaining the corresponding image region of construction frame R is Ri；
H is seti(p)=αiRi(p)+βi, wherein p is T 'iIn any one pixel, WithIt is T ' respectivelyiPixel mean value and RiPixel mean value,WithIt is T ' respectivelyiPicture
Plain variance and RiPixel variance, hiIt (p) is the corresponding color mapping function of i-th of subregion；
According toCalculate αiAnd βiValue:|T′i| expression is picture
The pixel capacity in sub-prime region；
H is seti(p) for the weight w of pixel pi(p),Operator | | a-b | | table
Show the Euclidean distance asked between a and b；
According toCalculate normalized weight
According toAnd hi(p) weighting global color correction function H (p) is determined, wherein
The subject pixel point is substituted into weighting global color correction function H (p) and is corrected.
It, can be in order to overcome color acquisition difference caused by the factors such as different illumination conditions, shooting angle in the present embodiment
Color mapping and correction are carried out to subject pixel point.To reject the subject pixel point of reflector segment as input image data.Due to
Real scene image construction is extremely complex, therefore, it is necessary to subregion implement the mapping of color.The first step can use
The Open-Source Tools that Felzenszwalb is provided, are divided into M region for input picture T ' (subject pixel point), single region is denoted as
T′i；In addition, selecting the preferable frame image R of image quality as construction frame at random, its foreground target pixel is removed, constructs R pairs of frame
The image region answered is Ri.Meanwhile in mapping function construction process, it is necessary to keep the invariable of respective pixel coordinate.
Second step completes mapping in the following way in LAB color space:
hi(p)=αiRi(p)+βi, wherein p is subregion T 'iIn any one pixel, WithIt is input image data T ' respectivelyiMean value and construction frame image data Ri's
Mean value,WithIt is input image data T respectivelyi' variance and construction frame image data RiVariance, hi(p) it is i-th
The corresponding color mapping function of subregion；Come the α that gets parms by way of minimizing energy functioniAnd βiValue,
To above formula right end expression respectively to parameter alphaiAnd βiIt seeks first derivative and solves, can obtain:
Wherein, | T 'i| expression be subregion pixel capacity.
Third step constructs global color correction function H (p) by the mapping function of subregion；Define sub-district domain mapping letter
Number hiThe weight w of () for pixel pi(p):
Wherein, operator | | a-b | | it indicates to seek the Euclidean distance of a and b.For M
Just there is M local mapping function h in a regioni()；In order to guarantee the standardization of different colours weight computing result, it is necessary to power
Weight function standardized (normalization) handle:
It can obtain weighting global color correction function:
To which the subject pixel point is substituted into weighting global color correction function H (p)
In be corrected.
Refering to what is shown in Fig. 6, Fig. 6 is the image processing system schematic diagram for color identification of one embodiment, packet
It includes:
Module 10 is chosen to choose from each video respectively several for choosing multiple videos from video acquisition system
Frame image constructs training set, and the target context code book of video acquisition system is identified according to the training set；
Identification module 20, for according to the corresponding prospect mesh of target object in the target context code book recognition target image
Mark pixel；
Module 30 is rejected, for rejecting the corresponding pixel of target object reflector segment in the foreground target pixel,
Obtain subject pixel point；
Determining module 40, the color for carrying out target object according to the subject pixel point identifies, determines target object
Color.
Image processing system provided by the present invention for color identification and the figure provided by the present invention for color identification
As processing method correspond, it is described for color identification image processing method embodiment illustrate technical characteristic and its
Beneficial effect suitable for the embodiment of the image processing system identified for color, hereby give notice that.
A kind of vehicle color identification method, includes the following steps:
Vehicle monitoring video is obtained from bayonet system；In vehicle monitoring video identification target vehicle, with target vehicle
Vehicle body is main body portion, using the vehicle window of target vehicle as reflector segment；
Body color is identified according to the above-mentioned image processing method for the identification of target object color, according to the vehicle identified
Body color determines the color of target vehicle.
Above-mentioned bayonet system includes the bayonet system in the places such as urban road or community gate inhibition.Above-mentioned target vehicle is upper
The target object in the image processing method for the identification of target object color is stated, is identified using above-mentioned for target object color
Image processing method carry out the color identification of automobile body part (main part), and then determine corresponding vehicle color, can
To improve the accuracy of identified vehicle color.
Vehicle color identification method provided by the invention can choose multiple automobile video frequencies from bayonet system and determine background
Target code book, using the corresponding foreground target pixel of upper target context code book identification target vehicle, then by foreground target pixel
It is easy to cause the corresponding pixel of vehicle window of color identification interference to be rejected in point, then to the vehicle of energy accurate characterization vehicle color
Body pixel carries out corresponding color identification, makes identified vehicle color accuracy with higher.
A kind of vehicle color identifying system, comprising:
Module is obtained, for obtaining vehicle monitoring video from bayonet system；Target vehicle is chosen in vehicle monitoring video,
With the vehicle body of target vehicle for main body portion, using the vehicle window of target vehicle as reflector segment；
Vehicle color determining module, for identifying vehicle according to the above-mentioned image processing system for the identification of target object color
Body color determines the color of target vehicle according to the body color identified.
Vehicle color identifying system provided by the invention and vehicle color identification method provided by the invention correspond,
The technical characteristic and its advantages that the embodiment of the vehicle color identification method illustrates are suitable for vehicle color identification system
In the embodiment of system, hereby give notice that.
Each technical characteristic of embodiment described above can be combined arbitrarily, for simplicity of description, not to above-mentioned reality
It applies all possible combination of each technical characteristic in example to be all described, as long as however, the combination of these technical characteristics is not deposited
In contradiction, all should be considered as described in this specification.
The embodiments described above only express several embodiments of the present invention, and the description thereof is more specific and detailed, but simultaneously
It cannot therefore be construed as limiting the scope of the patent.It should be pointed out that coming for those of ordinary skill in the art
It says, without departing from the inventive concept of the premise, various modifications and improvements can be made, these belong to protection of the invention
Range.Therefore, the scope of protection of the patent of the invention shall be subject to the appended claims.
Claims (13)
1. a kind of image processing method for color identification, which comprises the steps of:
Multiple videos are chosen from video acquisition system, choose several frame images from each video respectively, construct training set, root
According to the target context code book of training set identification video acquisition system；
According to the corresponding foreground target pixel of target object in the target context code book recognition target image；The target pair
As for zhou duicheng tuxing；The reflector segment of the target object is symmetrically distributed in above main part；
The corresponding pixel of target object reflector segment in the foreground target pixel is rejected, the corresponding master of main part is obtained
Body image vegetarian refreshments, comprising: determine the distribution of the foreground target pixel to be horizontal distribution, identify foreground target pixel
Main shaft foreground target pixel below is determined as subject pixel point by the main shaft of corresponding target object；
The color identification that target object is carried out according to the subject pixel point, determines the color of target object.
2. the image processing method according to claim 1 for color identification, which is characterized in that described according to the back
The step of target object corresponding foreground target pixel, includes: in scape target code book recognition target image
It identifies all code words in the target context code book, searches non-background pixel in the corresponding pixel of target image
Point obtains foreground target pixel.
3. the image processing method according to claim 2 for color identification, which is characterized in that described in target image
The step of searching non-background pixel point in corresponding pixel include:
The rgb value for obtaining any one pixel p in target image calculates the lightness I of pixel p according to the rgb value, whereinR is the R component value of pixel p, and G is the G component value of pixel p, and B is the B component value of pixel p；
The pixel that following condition A or condition B are unsatisfactory in target image is determined as non-background pixel point:
Condition A:Color_distance (p, CB)≤e, wherein p indicates the pixel in target image, CBTarget context code book
In code word, Color_distance () expression seeks color distance, and e indicates preset color threshold；
Condition B: Indicate the corresponding minimum lightness of pixel in target image,It indicates
The corresponding maximum lightness of pixel in target image；Lightness function brightness's is defined as:Wherein,min{}
Expression is minimized, m1For preset first lightness parameter, m2For preset second lightness parameter.
4. the image processing method according to claim 1 for color identification, which is characterized in that before the judgement is described
The distribution of scape target pixel points be horizontal distribution, comprising:
Central moment is set are as follows:Wherein,TBIt indicates
Foreground target pixel, (x, y) ∈ TBIndicate that pixel x and y belong to foreground target pixel, p and q respectively indicate subscript change
Amount；
According to towards angle calculation formulaIt calculates towards angle
Wherein For temporary variable symbol；
If describedValue be located at sectionOn, then the distribution of the foreground target pixel is determined to be water
Divide cloth equally.
5. the image processing method according to claim 1 for color identification, which is characterized in that described according to the master
Volumetric pixel point carry out target object color identification process include:
The rgb value of the subject pixel point is subjected to standardization calculating, obtains the R component value I after standardization calculatesR, G component value
IGWith B component value IB, wherein IR=r/ (r+g+b), IG=g/ (r+g+b), IBThe R of pixel points based on=b/ (r+g+b), r
Magnitude, the G component value of pixel based on g, the B component value of pixel based on b；
According to the IR、IGAnd IBJudge whether subject pixel point belongs to secondary colour；
If the subject pixel point belongs to secondary colour, subject pixel point input secondary colour classifier is subjected to color knowledge
Not；
If the subject pixel point is not belonging to secondary colour, subject pixel point input monochromatic color classifier is subjected to color knowledge
Not.
6. the image processing method according to claim 5 for color identification, which is characterized in that described according to the IR、
IGAnd IBJudge whether subject pixel point belongs to the process of secondary colour and include:
Calculate the Euclidean distance d between subject pixel point and secondary colour central axis vectorp, wherein dp=(IR-0.333)2+(IG-
0.333)2+(IB-0.333)2；
Calculate the Euclidean distance d between subject pixel point and hybrid mean value vectormean, wherein dmean=(r-ur)2+(g-ug)2+
(b-ub)2, urIndicate the mean value of the R component value of all subject pixel points, ugIndicate the equal of the G component value of all subject pixel points
Value, ubIndicate the mean value of the B component value of all subject pixel points；
Calculate separately dmeanMean μmean, dpMean μd, dmeanVariances sigmameanAnd dpVariances sigmad；
According toIt calculates subject pixel point and belongs to secondary colour
Probability P (gray | p)；Wherein, the p in P (gray | p) represents subject pixel point, and gray represents the subject pixel for belonging to secondary colour
Point；
P (gray | p) the subject pixel point for being greater than probability threshold value is judged to belonging to the subject pixel point of secondary colour.
7. a kind of image processing system for color identification characterized by comprising
It chooses module and chooses several frame images from each video respectively for choosing multiple videos from video acquisition system,
Training set is constructed, the target context code book of video acquisition system is identified according to the training set；
Identification module, for according to the corresponding foreground target pixel of target object in the target context code book recognition target image
Point；The target object is zhou duicheng tuxing；The reflector segment of the target object is symmetrically distributed in above main part；
Module is rejected to be led for rejecting the corresponding pixel of target object reflector segment in the foreground target pixel
Body image vegetarian refreshments, comprising: determine the distribution of the foreground target pixel to be horizontal distribution, identify foreground target pixel
Main shaft foreground target pixel below is determined as subject pixel point by the main shaft of corresponding target object；
Determining module, the color for carrying out target object according to the subject pixel point identifies, determines the color of target object.
8. a kind of computer equipment, including memory and processor, the memory are stored with computer program, feature exists
In the processor realizes the figure described in any one of claims 1 to 6 for color identification when executing the computer program
As the step of processing method.
9. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the computer program quilt
Processor realizes the step of image processing method described in any one of claims 1 to 6 for color identification when executing.
10. a kind of vehicle color identification method, which comprises the steps of:
Vehicle monitoring video is obtained from bayonet system；In vehicle monitoring video identification target vehicle, with the vehicle body of target vehicle
For main body portion, using the vehicle window of target vehicle as reflector segment；
Image processing method according to any one of claims 1 to 6 for color identification identifies body color, according to institute
The body color of identification determines the color of target vehicle.
11. a kind of vehicle color identifying system characterized by comprising
Module is obtained, for obtaining vehicle monitoring video from bayonet system；Target vehicle is chosen in vehicle monitoring video, with mesh
The vehicle body for marking vehicle is main body portion, using the vehicle window of target vehicle as reflector segment；
Vehicle color determining module, for according to claim 1 to the image procossing for being used for color identification described in 6 any one
Method identifies body color, and the color of target vehicle is determined according to the body color identified.
12. a kind of computer equipment, including memory and processor, the memory are stored with computer program, feature exists
In when the processor executes the computer program the step of vehicle color identification method described in realization claim 10.
13. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the computer program
The step of vehicle color identification method described in any one of claim 10 is realized when being executed by processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610893880.9A CN106503638B (en) | 2016-10-13 | 2016-10-13 | Image procossing, vehicle color identification method and system for color identification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610893880.9A CN106503638B (en) | 2016-10-13 | 2016-10-13 | Image procossing, vehicle color identification method and system for color identification |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106503638A CN106503638A (en) | 2017-03-15 |
| CN106503638B true CN106503638B (en) | 2019-09-13 |
Family
ID=58293946
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610893880.9A Active CN106503638B (en) | 2016-10-13 | 2016-10-13 | Image procossing, vehicle color identification method and system for color identification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106503638B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107154032B (en) * | 2017-04-20 | 2019-01-29 | 腾讯科技（深圳）有限公司 | A kind of image processing method and device |
| CN110519489B (en) * | 2019-06-20 | 2021-04-06 | 杭州海康威视数字技术股份有限公司 | Image acquisition method and device |
| CN110751053B (en) * | 2019-09-26 | 2022-02-22 | 高新兴科技集团股份有限公司 | Vehicle color identification method, device, equipment and storage medium |
| CN111340896B (en) * | 2020-02-21 | 2023-10-27 | 北京迈格威科技有限公司 | Object color recognition method, device, computer equipment and storage medium |
| CN111626188B (en) * | 2020-05-26 | 2022-05-06 | 西南大学 | Indoor uncontrollable open fire monitoring method and system |
| CN112016433A (en) * | 2020-08-24 | 2020-12-01 | 高新兴科技集团股份有限公司 | Vehicle color identification method based on deep neural network |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6088482A (en) * | 1998-10-22 | 2000-07-11 | Symbol Technologies, Inc. | Techniques for reading two dimensional code, including maxicode |
| CN101436252A (en) * | 2008-12-22 | 2009-05-20 | 北京中星微电子有限公司 | Method and system for recognizing vehicle body color in vehicle video image |
| CN102034080A (en) * | 2009-09-24 | 2011-04-27 | 北京汉王智通科技有限公司 | Vehicle color identification method and device |
| CN103530366A (en) * | 2013-10-12 | 2014-01-22 | 湖北微模式科技发展有限公司 | Vehicle searching method and system based on user-defined features |
| CN105005766A (en) * | 2015-07-01 | 2015-10-28 | 深圳市迈科龙电子有限公司 | Vehicle body color identification method |
-
2016
- 2016-10-13 CN CN201610893880.9A patent/CN106503638B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6088482A (en) * | 1998-10-22 | 2000-07-11 | Symbol Technologies, Inc. | Techniques for reading two dimensional code, including maxicode |
| CN101436252A (en) * | 2008-12-22 | 2009-05-20 | 北京中星微电子有限公司 | Method and system for recognizing vehicle body color in vehicle video image |
| CN102034080A (en) * | 2009-09-24 | 2011-04-27 | 北京汉王智通科技有限公司 | Vehicle color identification method and device |
| CN103530366A (en) * | 2013-10-12 | 2014-01-22 | 湖北微模式科技发展有限公司 | Vehicle searching method and system based on user-defined features |
| CN105005766A (en) * | 2015-07-01 | 2015-10-28 | 深圳市迈科龙电子有限公司 | Vehicle body color identification method |
Non-Patent Citations (2)
| Title |
|---|
| Digital Color Classification for Colorful Cross Stitch Threads Using RGB+Euclidean Distance and LAB+CIE94;David Setiabudi et al.;《2016 International Conference on Information,Communication Technology and System》;20161012;第150-156页 * |
| 一种基于码本背景模型的运动目标检测方法;赵占杰 等;《北京石油化工学院学报》;20100331;第18卷(第1期);摘要，第1节，第2节3段，第3节 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106503638A (en) | 2017-03-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106503638B (en) | Image procossing, vehicle color identification method and system for color identification | |
| CN108596277B (en) | Vehicle identity recognition method and device and storage medium | |
| CN109271991B (en) | License plate detection method based on deep learning | |
| CN112686812B (en) | Bank card inclination correction detection method and device, readable storage medium and terminal | |
| CN106650731B (en) | Robust license plate and vehicle logo recognition method | |
| CN103035013B (en) | A kind of precise motion shadow detection method based on multi-feature fusion | |
| CN105809651B (en) | Image significance detection method based on the comparison of edge non-similarity | |
| CN108875600A (en) | A kind of information of vehicles detection and tracking method, apparatus and computer storage medium based on YOLO | |
| CN103034838B (en) | A kind of special vehicle instrument type identification based on characteristics of image and scaling method | |
| CN109214420A (en) | The high texture image classification method and system of view-based access control model conspicuousness detection | |
| CN103824091B (en) | A kind of licence plate recognition method for intelligent transportation system | |
| CN107392968B (en) | The image significance detection method of Fusion of Color comparison diagram and Color-spatial distribution figure | |
| CN106610969A (en) | Multimodal information-based video content auditing system and method | |
| CN107273832B (en) | License plate recognition method and system based on integral channel characteristics and convolutional neural network | |
| Mo et al. | Deshadowing of urban airborne imagery based on object-oriented automatic shadow detection and regional matching compensation | |
| CN102842032A (en) | Method for recognizing pornography images on mobile Internet based on multi-mode combinational strategy | |
| CN109918971B (en) | Method and device for detecting number of people in monitoring video | |
| CN110309781B (en) | House damage remote sensing identification method based on multi-scale spectrum texture self-adaptive fusion | |
| CN102024144A (en) | Container number identification method | |
| Christlein et al. | A study on features for the detection of copy-move forgeries | |
| CN104182763A (en) | Plant type identification system based on flower characteristics | |
| CN109740572A (en) | A kind of human face in-vivo detection method based on partial color textural characteristics | |
| CN108491498A (en) | A kind of bayonet image object searching method based on multiple features detection | |
| CN110598560A (en) | Night monitoring and identifying method and system based on neural network enhancement | |
| WO2021000829A1 (en) | Multi-dimensional identity information identification method and apparatus, computer device and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |