CN107301421A - The recognition methods of vehicle color and device - Google Patents
The recognition methods of vehicle color and device Download PDFInfo
- Publication number
- CN107301421A CN107301421A CN201610235077.6A CN201610235077A CN107301421A CN 107301421 A CN107301421 A CN 107301421A CN 201610235077 A CN201610235077 A CN 201610235077A CN 107301421 A CN107301421 A CN 107301421A
- Authority
- CN
- China
- Prior art keywords
- color
- connected region
- image
- pixel
- quantized interval
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/56—Extraction of image or video features relating to colour
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a kind of recognition methods of vehicle color and device, this method includes：It is the second color space by the first color space conversion of the vehicle image got, wherein, the first color space and the second color space are made up of three Color Channels respectively；Color Channel in second color space is carried out to quantify to obtain the connected region in the center color value vector sum bianry image of Color Channel, wherein, connected region includes cluster connected region and non-cluster connected region；The multidimensional color feature vector for representing vehicle image color is obtained according to the connected region in the color value vector sum bianry image of center；Classification based training is carried out to multidimensional color characteristic.Pass through the present invention, spatial positional information where losing certain pixel when solving in correlation technique for vehicle color identification using color histogram, and there is the problem of disturbing to different images using subjective fixed threshold during color convergence vector, reach the effect for improving the degree of accuracy that vehicle color is recognized.
Description
Technical field
Field, recognition methods and device in particular to a kind of vehicle color are recognized the present invention relates to color of image.
Background technology
Field is recognized in color of image, generally using color histogram, the technology such as adaptive color threshold value and color template matching,
Color analysis, and then judgment object color are carried out to coloured image, can quickly detect that with identification there is different colours and shape
Target.In terms of traffic, the management system recognized based on car plate, logo, body color can be to parking lot, high speed crossing
Carry out automatic management.
Vehicle color identification refers to, in traffic monitoring scene, it is known that if predefined Ganlei's vehicle color, is monitored when passing through
When video tracking is to the vehicle moved, by the analysis to capturing image or video, its body color can be categorized as to a certain class
Color.For existing adaptive color threshold method, conventional coloured image Target detection and identification method is mostly based on fixed
Color threshold, this method is simple, and processing speed is fast, but Shandong during due to the influence of space light change, target identification and detection
Rod is poor.
Fig. 1 a~1c is different scenes and the schematic diagram of illumination condition vehicle color identification classification in correlation technique, such as Fig. 1 a~1c institutes
Show, illustrate the effect that video monitoring vehicle color is recognized under different scenes and illumination condition, wherein, Fig. 1 a and Fig. 1 b are same
Experiment effect under one scene different illumination conditions, Fig. 1 a and Fig. 1 c is the experiment effect under different scenes.But due to vehicle face
Color is related to several factors, the reflectance spectrum distribution of such as material, the spatial distribution of sunshine, response of the video camera to Color Channel
Deng, therefore excessively distorted for crossing the vehicle color under sunburst irradiation, the challenge is not solved also well.With this
The immediate prior art of patent includes：
1) color histogram；
Histogram is obtained by statistic and describes quantative attribute in image on color, the statistical of color of image can be reflected
Cloth and key colour.Color histogram is divided into color histogram and cumulative color histogram again.
What color histogram reflected is the composition distribution of color in image, that is, occurs in which color and a variety of colors occur general
Rate.When the feature in image can not take toilet to be possible to value, some null values occur in statistic histogram.These null values
Occur that influence can be brought on the calculating of similarity measurement, so that the color that similarity measurement can not be between correct response image
Difference.To solve this problem, cumulative color histogram is further provided on the basis of color histogram.In cumulative Nogata
In figure, adjacent color is related in frequency, and null value common in general histogram is eliminated compared to general histogram, also gram
General histogram has been taken to quantify meticulous or cross the defect that coarse search effect declines.
Color histogram is relative to image to observe the geometric transformations such as the little Pan and Zoom of rotation and amplitude of the axle as axle center
It is insensitive, the change (such as fuzzy, noise) for picture quality is also not very sensitive.This characteristic of color histogram
So that it is relatively more suitable for retrieving the occasion of the global color similitude of image, i.e., weighed by comparing the difference of color histogram
Difference of the two images in color overall situation distribution.
The method for setting up color histogram is fairly simple, but has the following disadvantages：First, color histogram does not account for figure
As the characteristics of itself, only from pixel level analysis characteristics of image；Secondly, the characteristics of not accounting for color space, is not enough to reaction
The 3D distribution situations in different colours space；Finally, its color level and representativeness are poor.
2) color convergence vector；
Color convergence vector is a complex method in histogram innovatory algorithm, and it draws the color of each in histogram cluster
It is divided into polymerization and non-polymeric two parts.In image similarity comparison procedure, their similitude is respectively compared, is then integrated
A similar value is obtained after balance, so as to obtain a result.Coloured image is after quantization, it is possible to by calculating connected domain, will
Pixel in image is divided into polymerization and non-polymeric two class.
Whether in the search method based on color convergence vector, it is that threshold value is typically set to when polymerizeing to judge a certain connected region
A certain fixed value, more than or equal to this fixed value, is then judged to polymerization, otherwise non-polymeric.This way is often led to when judging
Go to extremes, it is most likely that all regions with certain color are all judged to polymerization or non-polymeric, have so also reformed into and have passed through
Histogram judges the similitude of image, so as to lose the advantage of aggregated vector method.
For the above mentioned problem in correlation technique, not yet there is effective solution at present.
The content of the invention
The invention provides a kind of recognition methods of vehicle color and device, at least to solve for vehicle color to know in correlation technique
Spatial positional information where certain pixel Cai Yong not be lost during color histogram, and using subjective fixed threshold during color convergence vector
There is the problem of disturbing to different images.
According to an aspect of the invention, there is provided a kind of recognition methods of vehicle color, including：By the vehicle image got
The first color space conversion be the second color space, wherein, first color space and second color space respectively by
Three Color Channel compositions；Color Channel in second color space is carried out to quantify to obtain the center color value of Color Channel
Connected region in vector sum bianry image, wherein, the connected region includes cluster connected region and non-cluster connected region；
Obtained according to the connected region in bianry image described in the center color value vector sum for representing many of the vehicle image color
Tie up color feature vector；Classification based training is carried out to the multidimensional color characteristic.
Further, the Color Channel of first color space includes：Red, green, blueness；Second color space
The parameters of three Color Channels include：Hue, saturation, intensity.
Further, the Color Channel in second color space is carried out quantifying the center color value vector for obtaining Color Channel
Include with the connected region in bianry image：By the parameter of three Color Channels of second color space：Tone, saturation degree,
Brightness uniformity equal interval quantizing；The center color value of each quantized interval is obtained according to the quantized interval after quantization；After quantization
Quantized interval obtain cluster connected region and non-cluster connected region in the bianry image between each quantization back zone.
Further, the connected region obtained according to the quantized interval after quantization in the bianry image between each quantization back zone includes：
Obtain the pixel in the quantized interval after quantifying；The connected region area of the pixel in same quantized interval is more than
The connected region of average connected region area is divided into cluster connected region；By the company of the pixel in same quantized interval
The connected region that logical region area is less than average connected region area is divided into non-cluster connected region.
Further, classification based training is carried out to the multidimensional color characteristic in the following manner：Support vector machines.
According to another aspect of the present invention there is provided a kind of identifying device of vehicle color, including：Modular converter, for inciting somebody to action
First color space conversion of the vehicle image got is the second color space, wherein, first color space and described the
Second colors space is made up of three Color Channels respectively；Quantization modules, for entering to the Color Channel in second color space
Row quantifies to obtain the connected region in the center color value vector sum bianry image of Color Channel, wherein, the connected region includes
Cluster connected region and non-cluster connected region；Processing module, for according to bianry image described in the center color value vector sum
In connected region obtain multidimensional color feature vector for representing the vehicle image color；Sort module, for described
Multidimensional color characteristic carries out classification based training.
Further, the Color Channel of first color space includes：Red, green, blueness；Second color space
The parameters of three Color Channels include：Hue, saturation, intensity.
Further, the quantization modules include：Quantifying unit, for by three Color Channels of second color space
Parameter：The uniform equal interval quantizing of hue, saturation, intensity；First processing units, for being obtained according to the quantized interval after quantization
To the center color value of each quantized interval；Second processing unit, for being obtained according to the quantized interval after quantization after each quantization
Cluster connected region and non-cluster connected region in interval bianry image.
Further, the second processing unit includes：Subelement is obtained, for obtaining the pixel in the quantized interval after quantifying
Point；First divides subelement, for the connected region area of the pixel in same quantized interval to be more than into average connection
The connected region of region area is divided into cluster connected region；Second divides subelement, for that described will be in same quantized interval
The connected region area of pixel be less than the connected region of average connected region area and be divided into non-cluster connected region.
Further, classification based training is carried out to the multidimensional color characteristic in the following manner：Support vector machines.
By the present invention, the center color value of each quantized interval is calculated using the quantized interval point using Color Channel, and to every
Classification polymerization connected region and the non-polymeric connected region of bianry image between individual quantization back zone, are solved in correlation technique for car
Colour recognition loses spatial positional information where certain pixel when using color histogram, and using color convergence it is vectorial when it is subjective
There is the problem of disturbing in fixed threshold, reached the effect for the degree of accuracy for improving vehicle color identification to different images.
Brief description of the drawings
Accompanying drawing described herein is used for providing a further understanding of the present invention, constitutes the part of the application, of the invention shows
Meaning property embodiment and its illustrate be used for explain the present invention, do not constitute inappropriate limitation of the present invention.In the accompanying drawings：
Fig. 1 a~1c is different scenes and the schematic diagram of illumination condition vehicle color identification classification in correlation technique；
Fig. 2 is the flow chart of the recognition methods of vehicle color according to embodiments of the present invention；
Fig. 3 is the structured flowchart of the identifying device of vehicle color according to embodiments of the present invention；
Fig. 4 is the alternative construction block diagram one of the identifying device of vehicle color according to embodiments of the present invention；
Fig. 5 is the alternative construction block diagram two of the identifying device of vehicle color according to embodiments of the present invention；
Fig. 6 is the model space schematic diagram one according to alternative embodiment of the present invention；
Fig. 7 is the model space schematic diagram two according to alternative embodiment of the present invention；
Fig. 8 is the model space schematic diagram three according to alternative embodiment of the present invention.
Embodiment
Describe the present invention in detail below with reference to accompanying drawing and in conjunction with the embodiments.It should be noted that in the case where not conflicting,
The feature in embodiment and embodiment in the application can be mutually combined.
It should be noted that term " first ", " second " in description and claims of this specification and above-mentioned accompanying drawing etc. is to use
In distinguishing similar object, without for describing specific order or precedence.
A kind of recognition methods of vehicle color is provided in the present embodiment, and Fig. 2 is vehicle color according to embodiments of the present invention
The flow chart of recognition methods, as shown in Fig. 2 the flow comprises the following steps：
Step S202：It is the second color space by the first color space conversion of the vehicle image got, wherein, the first color
Space and the second color space are made up of three Color Channels respectively；
Step S204：Color Channel in second color space is carried out to quantify to obtain the center color value vector sum two of Color Channel
It is worth the connected region in image, wherein, connected region includes cluster connected region and non-cluster connected region；
Step S206：Obtained according to the connected region in the color value vector sum bianry image of center for representing vehicle image color
Multidimensional color feature vector；
Step S208：Classification based training is carried out to multidimensional color characteristic.
From the above-mentioned steps S202 to step S208 of the embodiment of the present invention, calculate every using the quantized interval point of Color Channel
The center color value of individual quantized interval, and connected region and non-polymeric company are polymerize to the classification of the bianry image between each quantization back zone
Logical region, space bit confidence where losing certain pixel when solving in correlation technique for vehicle color identification using color histogram
Breath, and there is the problem of disturbing to different images using subjective fixed threshold during color convergence vector, reach raising vehicle face
The effect of the degree of accuracy of color identification.
It should be noted that the Color Channel for the first color space being related in embodiment includes：Red, green, blueness；
The parameter of three Color Channels of the second color space includes：Hue, saturation, intensity.Wherein, RGB color：
RGB color is a kind of color standard of industrial quarters, by the way that to red (R), green (G), blue (B) three colors are led to
The change in road and their superpositions each other, obtain miscellaneous color.RGB is to represent red, green, blue three
The color of passage.Hsv color space：HSV is a kind of color space created according to the intuitive nature of color, also referred to as six
Pyramid body Model.The parameter of color is respectively in this model：Tone (H), saturation degree (S), brightness (V).
The Color Channel in the second color space is carried out to quantify to obtain Color Channel in the present embodiment step S204 being related to
Center color value vector sum bianry image in connected region mode, in the optional embodiment of the present embodiment, Ke Yitong
Following manner is crossed to realize：
Step S204-1：By the parameter of three Color Channels of the second color space：Hue, saturation, intensity is uniform at equal intervals
Quantify；
Step S204-2：The center color value of each quantized interval is obtained according to the quantized interval after quantization；
Step S204-3：The cluster connected region in the bianry image between each quantization back zone is obtained according to the quantized interval after quantization
With non-cluster connected region.
And the binary map obtained according to the quantized interval after quantization between each quantization back zone for being related in above-mentioned steps S104-3
The mode of connected region as in, can be realized in the following way：Obtain the pixel in the quantized interval after quantifying；With
Divided in the connected region that the connected region area of the pixel in same quantized interval is more than to average connected region area
For cluster connected region；The connected region area of the pixel in same quantized interval is less than average connected region area
Connected region be divided into non-cluster connected region.
It should be noted that color center value：In referring to that the i-th class quantization is interval, choose suitableGeneration
Color value during the class quantization of table i-th is interval,It is an one-dimensional vector, xhi,xsi,xviRepresent that tri- colors of H, S, V are led to respectively
The value in road.Bianry image is connective：And in bianry image, it is assumed that there are m (m≤8) adjacent pixels around target pixel points,
If the pixel grey scale belongs to a quantized interval with a certain pixel x in this m pixel, then have into the pixel and pixel x
There is connectedness.
It should be noted that carrying out classification based training to multidimensional color characteristic in the following manner：Support vector machines.
Through the above description of the embodiments, those skilled in the art can be understood that the side according to above-described embodiment
Method can add the mode of required general hardware platform to realize by software, naturally it is also possible to by hardware, but in many cases before
Person is more preferably embodiment.Understood based on such, technical scheme substantially makes tribute to prior art in other words
The part offered can be embodied in the form of software product, and the computer software product is stored in a storage medium (such as
ROM/RAM, magnetic disc, CD) in, including some instructions to cause a station terminal equipment (can be mobile phone, computer,
Server, or the network equipment etc.) method that performs each embodiment of the invention.
A kind of identifying device of vehicle color is additionally provided in the present embodiment, and the device is used to realize above-described embodiment and preferred real
Mode is applied, repeating no more for explanation had been carried out.As used below, term " module " can realize predetermined function
The combination of software and/or hardware.Although the device described by following examples is preferably realized with software, hardware, or
The realization of the combination of software and hardware is also that may and be contemplated.
Fig. 3 is the structured flowchart of the identifying device of vehicle color according to embodiments of the present invention, as shown in figure 3, the device includes：
Modular converter 32, for being the second color space by the first color space conversion of the vehicle image got, wherein, the first face
The colour space and the second color space are made up of three Color Channels respectively；Quantization modules 34, are of coupled connections with modular converter 32, use
Carry out quantifying to obtain the company in the center color value vector sum bianry image of Color Channel in the Color Channel in the second color space
Logical region, wherein, connected region includes cluster connected region and non-cluster connected region；Processing module 36, with quantization modules 34
It is of coupled connections, for being obtained according to the connected region in the color value vector sum bianry image of center for representing vehicle image color
Multidimensional color feature vector；Sort module 38, is of coupled connections with processing module 36, for carrying out classification instruction to multidimensional color characteristic
Practice.
It should be noted that the Color Channel of the first color space includes in the present embodiment：Red, green, blueness；Second
The parameter of three Color Channels of color space includes：Hue, saturation, intensity.
Fig. 4 is the alternative construction block diagram one of the identifying device of vehicle color according to embodiments of the present invention, as shown in figure 4, the amount
Changing module 34 includes：Quantifying unit 42, for by the parameter of three Color Channels of the second color space：Tone, saturation degree,
Brightness uniformity equal interval quantizing；First processing units 44, are of coupled connections with quantifying unit 42, for according to the quantization area after quantization
Between obtain the center color value of each quantized interval；Second processing unit 46,46 are of coupled connections with first processing units 44, use
Cluster connected region and non-cluster connected region in the bianry image that the quantized interval after according to quantization obtains between each quantization back zone
Domain.
Fig. 5 is the alternative construction block diagram two of the identifying device of vehicle color according to embodiments of the present invention, as shown in figure 5, this
Two processing units 46 include：Subelement 52 is obtained, for obtaining the pixel in the quantized interval after quantifying；First divides son list
Member 54, is of coupled connections with obtaining subelement 52, for the connected region area of the pixel in same quantized interval is big
Cluster connected region is divided into the connected region of average connected region area；Second divides subelement 56, and it is single to divide son with first
Member 54 is of coupled connections, for the connected region area of the pixel in same quantized interval to be less than into average connected region face
Long-pending connected region is divided into non-cluster connected region.
It should be noted that can be divided in the following manner multidimensional color characteristic in the optional embodiment of the present embodiment
Class is trained：Support vector machines.
It should be noted that above-mentioned modules can be by software or hardware to realize, for the latter, can by with
Under type is realized, but not limited to this：Above-mentioned module is respectively positioned in same processor；Or, above-mentioned module is located at multiple places respectively
Manage in device.
The present invention is illustrated with reference to the alternative embodiment of the present invention；
This alternative embodiment is directed to recognizes vehicle color using color video frequency image, and applied to the core of traffic monitoring video
Problem, compared to existing colour recognition technology, it is proposed that combined using color histogram and the color convergence vector in hsv color space
Method, by Training Support Vector Machines (SVM) grader, realize the standard of vehicle color under different scenes and illumination condition
Really identification classification.This alternative embodiment is applied under a variety of traffic scenes and daily illumination condition, with good robustness.
This alternative embodiment, which provides the step of a kind of vehicle color based on video monitoring image knows method for distinguishing, this method, to be included：
Step S302：Color space conversion；
Wherein, vehicle coloured image is transformed into hsv color space by RGB color.
Step S304：Color feature extracted；
The step includes：S304-1, by tri- uniform equal interval quantizings of color component of H, S, V of image；
S304-2, according to the quantized interval of three Color Channels, tries to achieve the center color value vector of each quantized interval；
S304-3, according to the quantized interval of three Color Channels, finds out the connected region in the bianry image between each quantization back zone,
And connected region is divided into cluster and non-cluster；
S304-4, arranges color characteristic.
Step S306, SVMs (SVM) classifier training.
From above-mentioned steps, using the pixel of the quantized interval classification chart picture of Color Channel, and each quantized interval is calculated
Center color value vector, and calculate the connected region of the bianry image between each quantization back zone threshold value automatically, classification polymerization with it is non-
Polymerization.The mode of this alternative embodiment solves spatial positional information where losing certain pixel compared with color histogram method
Shortcoming, reduces the influence of color histogram null value, overcomes histogram and quantifies meticulous or can slightly reduce the defect of retrieval effectiveness excessively.
Compared to color convergence vector, interference of the subjective fixed threshold to different images is solved, the accuracy of identification is improved.
Fig. 6 is the model space schematic diagram one according to alternative embodiment of the present invention, as shown in fig. 6, RGB color：R
GB color spaces are a kind of color standards of industrial quarters, by red (R), green (G), blue (B) three Color Channels
Change and their superpositions each other, obtain miscellaneous color.RGB is to represent red, green, blue three to lead to
The color in road.
Fig. 7 is the model space schematic diagram two according to alternative embodiment of the present invention, as shown in fig. 7, hsv color space：H
SV is a kind of color space created according to the intuitive nature of color, also referred to as hexagonal pyramid model.The ginseng of color in this model
Number is respectively：Tone (H), saturation degree (S), brightness (V).
Color center value：Refer in the i-th class quantization interval, choose suitableRepresent in the i-th class quantization interval
Color value,It is an one-dimensional vector, xhi,xsi,xviThe value of tri- Color Channels of H, S, V is represented respectively.
Bianry image is connective：In bianry image, it is assumed that have m (m≤8) adjacent pixels around target pixel points, if should
Pixel grey scale belongs to a quantized interval with a certain pixel x in this m pixel, then has into the pixel with pixel x and connects
Property.
SVMs (SVM)：SVMs can be with analyze data, recognition mode, for classification and regression analysis.Give
Fixed one group of training sample, each mark is two classes, and a SVM training algorithm establishes a model, distributes new reality
Example is a class or other classes, becomes non-probability binary linearity classification.
Assuming that the given two-dimensional points for being belonging respectively to two classes are linear separabilities, solid black square is two class two-dimensionals with white hollow circle
Point, these points can be by line segmentation, and algorithm will find an optimal cut-off rule.
Illustrated with reference to the specific embodiment of alternative embodiment of the present invention；
Alternative embodiment 1：
This alternative embodiment carries out colour recognition primarily directed to the people's car target moved in video.The step of this alternative embodiment, wraps
Include：
Step S402：Read and handle frame of video.
Moving object detection is carried out by frame difference method or background modeling algorithm to every two field picture, before being extracted according to background modeling result
Scape target, the movement destination image for needing to recognize is obtained by the method for the connection domain lookup minimum enclosed rectangle in image.
Step S404：Color space conversion.
RGB color known to target image from us is converted into hsv color space, because HSV can be more accurate
Reaction people visual idea, be a kind of intuitively color model for user.
If (R, G, B) is red, green, blue coordinate of the pixel of coloured image in RGB color, their value is
Real number between 0 to 255.Max is R, G, the maximum in B, and min is R, G, reckling in B.
(H, S, V) is respectively hue, saturation, intensity coordinate of the image pixel in hsv color space, here, we by they
Value adjust to 0 to 255 between real number.It is calculated as follows：
V=max (4)
Image will pixel-by-pixel be handled by formula (1) (2) (3) (4), so as to be changed by RGB color
Into hsv color space.
Step S406：Color feature extracted；
1. by three uniform equal interval quantizings of color component；
HSV color spaces are homogeneous spaces, and the coordinate system that it is used is also even coordinate system, in this space, of the invention H,
S and V span is adjusted to 0 to 255.By three uniform equal interval quantizings of color component, by H, S, and V points
Not uniformly it is not divided into 5 parts at equal intervals, is [(k-1) × 51, k × 51] (k=1,2,3,4,5) per a interval.Then quantify
H afterwards, S, V are respectively QH, QS, QV.Then single color component is synthesized, as one-dimensional characteristic vector
Qi=[QHi,QSi,QVi] (i=1,2 ..., 125), such H, S, tri- components of V are just distributed on a n dimensional vector n
Come, according to the image after quantization, the one-dimensional color histogram of feature of 125 dimensions can be obtained.
2. according to the quantized interval of three Color Channels, try to achieve the center color value vector of each quantized interval；
If image I color average value isFormula (5) is shown in calculating：
Wherein xhi, xsi, xviIt is the H of i-th pixel, S respectively, V value, n is image I pixel number.
Reflect the colouring information of entire image.Define the weights W of jth point pixel in i-th of quantized intervalij, determined by formula (6)：
WhereinFor three color vectors H, S, V of jth point pixel in i-th of quantized interval.WijReflection
Each pixel degree close with the color average of whole graph coloring in image, weights are bigger closer on the contrary, then gap is bigger.
The process of color characteristic is obtained, exactly to each pixel is sorted out in image process.The present invention by image in itself
The weights of information, i.e. pixel, calculate the cluster centre of each quantized interval, that is, obtain belonging to i-th quantized interval
Center color valueI=1,2 ..., ni.Calculated and obtained by formula (7)：
WhereinIt is the one-dimensional vector for including three Color Channels,It is i-th
The triple channel color vector of j-th of pixel, n in quantized intervaliIt is pixel sum in i-th of quantized interval.So draw
Center color value can accurately represent the triple channel color value of each quantized interval.
3. according to the quantized interval of three Color Channels, the connected region in the bianry image between each quantization back zone is found out, and
Connected region is divided into cluster and non-cluster.
In bianry image comprising black (pixel value is 0) and in vain (pixel value is 255) two kinds of colors, respectively as not locating
Color in quantized interval and the color in quantized interval.
Complete scanning is carried out once to pixel all in entire image I first, if the color value of the pixel is in i-th of amount
Change in interval, then the bianry image P of i-th of quantized intervaliColor at the pixel is white, and the two of other quantized intervals
It is black to be worth color of the image at the pixel.
The simple extraction algorithm for finding multiple connected regions in bianry image is as follows：
Scanning sequency first to image uses first left and then right, first up and then down method, is 25 until running into first pixel value
First point of 5 pixel, as connected region, then using the point as starting point, is tracked in its profile, mark boundaries
Pixel.When profile is completely closed, scanning returns to a position, until finding new connected domain again.Often find one
Connected domain, records the size of its connected domain.Fig. 8 is the model space schematic diagram three according to alternative embodiment of the present invention, such as
Shown in embodiment in Fig. 8, (a) is processing image I, in (b) with it is black be dominant hue for the binary map in quantized interval
Picture, with vain for the profile for connected region in respective binary image of dominant hue.It is excessive between quantized interval, only list herein
The corresponding bianry image in preceding 4 intervals.
The threshold value of i-th of quantized interval is mi, calculated by formula (8)：
For the number of pixel in all connected domains of i-th of quantized interval, niFor i-th of quantized interval
In all connected domains quantity.
If mik≥mi, then it is assumed that the connected region polymerize, and is otherwise judged to not polymerize.So judge whether polymerization completely by binary map
The connected domain distribution of picture is determined, rather than subjective setting.
Step S408：Arrange color characteristic；
By the uniform equal interval quantizing of three color components, in the center color value vector and bianry image of trying to achieve each quantized interval
After the cluster and non-cluster area of connected region, ifFor the center color value vector of i-th of quantized interval, αiFor i-th of amount
Change the cluster area of connected domain in interval, βiFor the non-cluster area of connected domain in i-th of quantized interval.In this way, a width complete graph
The feature of picture isWillSplit, feature is to represent
For：
(<CCh1,CCs1,CCv1,α1,β1>,<CCh2,CCs2,CCv2,α2,β2>,…,<CChn,CCsn,CCvn,αn,βn>)
N is n=125 in the number of quantized interval, the present invention, can obtain 625 dimension color feature vectors of image.
Step S310：SVMs (SVM) classifier training；
It can be seen that, this alternative embodiment uses C_SVC C class support vector classifications, and n classes (n >=2) are grouped, here
It is divided into：Black, blue, palm fibre is green, red, ash, and silver is in vain, yellow, totally 9 class.Allow to be carried out with exceptional value penalty factor endless
Full classification.
SVM core type is：The linear kernels of LINEAR.Do not have any to map to high-order spatial linear and return in original
Completed in beginning feature space, convergence speed is very fast.Its Kernel Function is：K(xi,yi)=<xi,yi>.Maximum iteration
For 500000, maximum precision is 1e-10.
Alternative embodiment 2：
This alternative embodiment is to be directed to the step of target to be identified carries out colour recognition, this alternative embodiment in still image to include：
Step S502：Utilize sliding window scanned picture.
Image to be identified is read, agglomerate to be detected is obtained by the sliding window of different scale, agglomerate is placed through training
SVM classifier in obtain whether the result of target to be identified, if the agglomerate be target to be identified, marked interception
Out used for subsequent operation.
Step S504：Color space conversion.
RGB color known to target image from us is converted into hsv color space, because HSV can be more accurate
Reaction people visual idea, be a kind of intuitively color model for user.
If (R, G, B) is red, green, blue coordinate of the pixel of coloured image in RGB color, their value is
Real number between 0 to 255.Max is R, G, the maximum in B, and min is R, G, reckling in B.
(H, S, V) is respectively hue, saturation, intensity coordinate of the image pixel in hsv color space, here, we by they
Value adjust to 0 to 255 between real number.It is calculated as follows：
V=max (4)
Image will pixel-by-pixel be handled by formula (1) (2) (3) (4), so as to be changed by RGB color
Into hsv color space.
Step S506：Color feature extracted
1. by three uniform equal interval quantizings of color component；
HSV color spaces are homogeneous spaces, and the coordinate system that it is used is also even coordinate system, in this space, of the invention H,
S and V span is adjusted to 0 to 255.By three uniform equal interval quantizings of color component, by H, S, and V points
Not uniformly it is not divided into 5 parts at equal intervals, is [(k-1) × 51, k × 51] (k=1,2,3,4,5) per a interval.Then quantify
H afterwards, S, V are respectively QH, QS, QV.Then single color component is synthesized, as one-dimensional characteristic vector
Qi=[QHi,QSi,QVi] (i=1,2 ..., 125), such H, S, tri- components of V are just distributed on a n dimensional vector n
Come, according to the image after quantization, the one-dimensional color histogram of feature of 125 dimensions can be obtained.
2. according to the quantized interval of three Color Channels, try to achieve the center color value vector of each quantized interval；
If image I color average value isFormula (5) is shown in calculating：
Wherein xhi, xsi, xviIt is the H of i-th pixel, S respectively, V value, n is image I pixel number.
Reflect the colouring information of entire image.Define the weights W of jth point pixel in i-th of quantized intervalij, determined by formula (6)：
Wherein,For three color vectors H, S, V of jth point pixel in i-th of quantized interval.WijInstead
The pixel of each in the image degree close with the color average of whole graph coloring is reflected, weights are bigger closer on the contrary, then gap is bigger.
The process of color characteristic is obtained, exactly to each pixel is sorted out in image process.The present invention by image in itself
The weights of information, i.e. pixel, calculate the cluster centre of each quantized interval, that is, obtain belonging to i-th quantized interval
Center color valueI=1,2 ..., ni.Calculated and obtained by formula (7)：
WhereinIt is the one-dimensional vector for including three Color Channels,It is i-th
The triple channel color vector of j-th of pixel, n in quantized intervaliIt is pixel sum in i-th of quantized interval.So draw
Center color value can accurately represent the triple channel color value of each quantized interval.
3. according to the quantized interval of three Color Channels, the connected region in the bianry image between each quantization back zone is found out, and
Connected region is divided into cluster and non-cluster.
In bianry image comprising black (pixel value is 0) and in vain (pixel value is 255) two kinds of colors, respectively as not locating
Color in quantized interval and the color in quantized interval.
Complete scanning is carried out once to pixel all in entire image I first, if the color value of the pixel is in i-th of amount
Change in interval, then the bianry image P of i-th of quantized intervaliColor at the pixel is white, and the two of other quantized intervals
It is black to be worth color of the image at the pixel.
The simple extraction algorithm for finding multiple connected regions in bianry image is as follows：
Scanning sequency first to image uses first left and then right, first up and then down method, is 25 until running into first pixel value
First point of 5 pixel, as connected region, then using the point as starting point, is tracked in its profile, mark boundaries
Pixel.When profile is completely closed, scanning returns to a position, until finding new connected domain again.Often find one
Connected domain, records the size of its connected domain.As shown in the embodiment depicted in fig. 8, (a) is processing image I, in (b)
With it is black be dominant hue for the bianry image in quantized interval, with the black wheel for connected region in respective binary image for dominant hue
It is wide.It is excessive between quantized interval, the corresponding bianry image in preceding 4 intervals is only listed herein.
The threshold value of i-th of quantized interval is mi, calculated by formula (8)：
For the number of pixel in all connected domains of i-th of quantized interval, niFor i-th of quantized interval
In all connected domains quantity.
If mik≥mi, then it is assumed that the connected region polymerize, and is otherwise judged to not polymerize.So judge whether polymerization completely by binary map
The connected domain distribution of picture is determined, rather than subjective setting.
4. arrange color characteristic
By the uniform equal interval quantizing of three color components, in the center color value vector and bianry image of trying to achieve each quantized interval
After the cluster and non-cluster area of connected region, ifFor the center color value vector of i-th of quantized interval, αiFor i-th of amount
Change the cluster area of connected domain in interval, βiFor the non-cluster area of connected domain in i-th of quantized interval.In this way, a width complete graph
The feature of picture isWillSplit, feature is to represent
For：
(<CCh1,CCs1,CCv1,α1,β1>,<CCh2,CCs2,CCv2,α2,β2>,…,<CChn,CCsn,CCvn,αn,βn>)
N is n=125 in the number of quantized interval, the present invention, can obtain 625 dimension color feature vectors of image.
Step S508：SVMs (SVM) classifier training；
This alternative embodiment uses C_SVC C class support vector classifications, and n classes (n >=2) packet is divided into here：
Black, blue, palm fibre is green, red, ash, and silver is in vain, yellow, totally 9 class.Allow not exclusively to be classified with exceptional value penalty factor.
SVM core type is：The linear kernels of LINEAR.Do not have any to map to high-order spatial linear and return in original
Completed in beginning feature space, convergence speed is very fast.Its Kernel Function is：K(xi,yi)=<xi,yi>.Maximum iteration
For 500000, maximum precision is 1e-10.
Alternative embodiment 3：
The step of this alternative embodiment is and carries out colour recognition, this alternative embodiment for the objective fuzzy image for intercepting out is wrapped
Include：
Step S602：Read picture and carry out image enhancement processing.
Target image to be identified is read, image enhancement processing is carried out to it according to picture quality, mainly includes the operation of defogging deblurring,
The illumination and contrast of appropriate adjustment image.
Step S604：Color space conversion.
RGB color known to target image from us is converted into hsv color space, because HSV can be more accurate
Reaction people visual idea, be a kind of intuitively color model for user.
If (R, G, B) is red, green, blue coordinate of the pixel of coloured image in RGB color, their value is
Real number between 0 to 255.Max is R, G, the maximum in B, and min is R, G, reckling in B.
(H, S, V) is respectively hue, saturation, intensity coordinate of the image pixel in hsv color space, here, we by they
Value adjust to 0 to 255 between real number.It is calculated as follows：
V=max (4)
Image will pixel-by-pixel be handled by formula (1) (2) (3) (4), so as to be changed by RGB color
Into hsv color space.
Step S606：Color feature extracted
1. by three uniform equal interval quantizings of color component；
HSV color spaces are homogeneous spaces, and the coordinate system that it is used is also even coordinate system, in this space, of the invention H,
S and V span is adjusted to 0 to 255.By three uniform equal interval quantizings of color component, by H, S, and V points
Not uniformly it is not divided into 5 parts at equal intervals, is [(k-1) × 51, k × 51] (k=1,2,3,4,5) per a interval.Then quantify
H afterwards, S, V are respectively QH, QS, QV.Then single color component is synthesized, as one-dimensional characteristic vector
Qi=[QHi,QSi,QVi] (i=1,2 ..., 125), such H, S, tri- components of V are just distributed on a n dimensional vector n
Come, according to the image after quantization, the one-dimensional color histogram of feature of 125 dimensions can be obtained.
2. according to the quantized interval of three Color Channels, try to achieve the center color value vector of each quantized interval；
If image I color average value isFormula (5) is shown in calculating：
Wherein, xhi, xsi, xviIt is the H of i-th pixel, S respectively, V value, n is image I pixel number.
Reflect the colouring information of entire image.Define the weights W of jth point pixel in i-th of quantized intervalij, determined by formula (6)：
Wherein,For three color vectors H, S, V of jth point pixel in i-th of quantized interval.WijInstead
The pixel of each in the image degree close with the color average of whole graph coloring is reflected, weights are bigger closer on the contrary, then gap is bigger.
The process of color characteristic is obtained, exactly to each pixel is sorted out in image process.The present invention by image in itself
The weights of information, i.e. pixel, calculate the cluster centre of each quantized interval, that is, obtain belonging to i-th quantized interval
Center color valueI=1,2 ..., ni.Calculated and obtained by formula (7)：
Wherein,It is the one-dimensional vector for including three Color Channels,It is i-th
The triple channel color vector of j-th of pixel, n in individual quantized intervaliIt is pixel sum in i-th of quantized interval.So draw
Center color value can accurately represent the triple channel color value of each quantized interval.
3. according to the quantized interval of three Color Channels, the connected region in the bianry image between each quantization back zone is found out, and
Connected region is divided into cluster and non-cluster.
In bianry image comprising black (pixel value is 0) and in vain (pixel value is 255) two kinds of colors, respectively as not locating
Color in quantized interval and the color in quantized interval.
Complete scanning is carried out once to pixel all in entire image I first, if the color value of the pixel is in i-th of amount
Change in interval, then the bianry image P of i-th of quantized intervaliColor at the pixel is white, and the two of other quantized intervals
It is black to be worth color of the image at the pixel.
The simple extraction algorithm for finding multiple connected regions in bianry image is as follows：
Scanning sequency first to image uses first left and then right, first up and then down method, is 25 until running into first pixel value
First point of 5 pixel, as connected region, then using the point as starting point, is tracked in its profile, mark boundaries
Pixel.When profile is completely closed, scanning returns to a position, until finding new connected domain again.Often find one
Connected domain, records the size of its connected domain.As shown in the embodiment depicted in fig. 8, (a) is processing image I, in (b)
With it is black be dominant hue for the bianry image in quantized interval, with the black wheel for connected region in respective binary image for dominant hue
It is wide.It is excessive between quantized interval, the corresponding bianry image in preceding 4 intervals is only listed herein.
The threshold value of i-th of quantized interval is mi, calculated by formula (8)：
For the number of pixel in all connected domains of i-th of quantized interval, niFor i-th of quantized interval
In all connected domains quantity.
If mik≥mi, then it is assumed that the connected region polymerize, and is otherwise judged to not polymerize.So judge whether polymerization completely by binary map
The connected domain distribution of picture is determined, rather than subjective setting.
4. arrange color characteristic
By the uniform equal interval quantizing of three color components, in the center color value vector and bianry image of trying to achieve each quantized interval
After the cluster and non-cluster area of connected region, ifFor the center color value vector of i-th of quantized interval, αiFor i-th of amount
Change the cluster area of connected domain in interval, βiFor the non-cluster area of connected domain in i-th of quantized interval.In this way, a width complete graph
The feature of picture isWillSplit, feature is to represent
For：
(<CCh1,CCs1,CCv1,α1,β1>,<CCh2,CCs2,CCv2,α2,β2>,…,<CChn,CCsn,CCvn,αn,βn>)
N is n=125 in the number of quantized interval, the present invention, can obtain 625 dimension color feature vectors of image.
Step S608：SVMs (SVM) classifier training
This alternative embodiment uses C_SVC C class support vector classifications, and n classes (n >=2) packet is divided into here：
Black, blue, palm fibre is green, red, ash, and silver is in vain, yellow, totally 9 class.Allow not exclusively to be classified with exceptional value penalty factor.
SVM core type is：The linear kernels of LINEAR.Do not have any to map to high-order spatial linear and return in original
Completed in beginning feature space, convergence speed is very fast.Its Kernel Function is：K(xi,yi)=<xi,yi>.Maximum iteration
For 500000, maximum precision is 1e-10.
Alternatively, the specific example in the present embodiment may be referred to the example described in above-described embodiment and optional embodiment,
The present embodiment will not be repeated here.
Obviously, those skilled in the art should be understood that above-mentioned each module of the invention or each step can use general calculating
Device realizes that they can be concentrated on single computing device, or be distributed on the network that multiple computing devices are constituted,
Alternatively, they can be realized with the executable program code of computing device, it is thus possible to be stored in storage device
In performed by computing device, and in some cases, can be to perform shown or described step different from order herein
Suddenly, they are either fabricated to each integrated circuit modules respectively or be fabricated to the multiple modules or step in them single
Integrated circuit modules are realized.So, the present invention is not restricted to any specific hardware and software combination.
The preferred embodiments of the present invention are these are only, are not intended to limit the invention, for those skilled in the art,
The present invention can have various modifications and variations.Within the spirit and principles of the invention, any modification for being made, equivalent substitution,
Improve etc., it should be included in the scope of the protection.
Claims (10)
1. a kind of recognition methods of vehicle color, it is characterised in that including：
It is the second color space by the first color space conversion of the vehicle image got, wherein, first face
The colour space and second color space are made up of three Color Channels respectively；
Color Channel in second color space is carried out to quantify the center color value vector for obtaining Color Channel
With the connected region in bianry image, wherein, the connected region includes cluster connected region and non-cluster connected region
Domain；
Obtain being used to represent the car according to the connected region in bianry image described in the center color value vector sum
The multidimensional color feature vector of color of image；
Classification based training is carried out to the multidimensional color characteristic.
2. according to the method described in claim 1, it is characterised in that the Color Channel of first color space includes：It is red
Color, green, blueness；The parameter of three Color Channels of second color space includes：Tone, saturation degree,
Brightness.
3. method according to claim 2, it is characterised in that carried out to the Color Channel in second color space
Quantifying to obtain the connected region in the center color value vector sum bianry image of Color Channel includes：
By the parameter of three Color Channels of second color space：Hue, saturation, intensity is uniform at equal intervals
Quantify；
The center color value of each quantized interval is obtained according to the quantized interval after quantization；
Obtained according to the quantized interval after quantization cluster connected region in the bianry image between each quantization back zone and
Non-cluster connected region.
4. according to the method described in claim 1, it is characterised in that obtained according to the quantized interval after quantization after each quantization
Connected region in interval bianry image includes：
Obtain the pixel in the quantized interval after quantifying；
The connected region area of pixel in same quantized interval is more than to the connection of average connected region area
Region division is cluster connected region；
The connected region area of pixel in same quantized interval is less than to the connection of average connected region area
Region division is non-cluster connected region.
5. according to the method described in claim 1, it is characterised in that the multidimensional color characteristic is carried out in the following manner
Classification based training：Support vector machines.
6. a kind of identifying device of vehicle color, it is characterised in that including：
Modular converter, for being the second color space by the first color space conversion of the vehicle image got, its
In, first color space and second color space are made up of three Color Channels respectively；
Quantization modules, for carrying out quantifying to obtain in Color Channel to the Color Channel in second color space
Connected region in heart color value vector sum bianry image, wherein, the connected region include cluster connected region and
Non-cluster connected region；
Processing module, for being used according to the connected region in bianry image described in the center color value vector sum
In the multidimensional color feature vector for representing the vehicle image color；
Sort module, for carrying out classification based training to the multidimensional color characteristic.
7. device according to claim 6, it is characterised in that the Color Channel of first color space includes：It is red
Color, green, blueness；The parameter of three Color Channels of second color space includes：Tone, saturation degree,
Brightness.
8. device according to claim 7, it is characterised in that the quantization modules include：
Quantifying unit, for by the parameter of three Color Channels of second color space：Tone, saturation degree,
Brightness uniformity equal interval quantizing；
First processing units, the center color value for obtaining each quantized interval according to the quantized interval after quantization；
Second processing unit, for being obtained according to the quantized interval after quantization in the bianry image between each quantization back zone
Cluster connected region and non-cluster connected region.
9. device according to claim 8, it is characterised in that the second processing unit includes：
Subelement is obtained, for obtaining the pixel in the quantized interval after quantifying；
First divides subelement, average for the connected region area of the pixel in same quantized interval to be more than
The connected region of connected region area is divided into cluster connected region；
Second divides subelement, average for the connected region area of the pixel in same quantized interval to be less than
The connected region of connected region area is divided into non-cluster connected region.
10. device according to claim 6, it is characterised in that carried out in the following manner to the multidimensional color characteristic
Classification based training：Support vector machines.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610235077.6A CN107301421A (en) | 2016-04-15 | 2016-04-15 | The recognition methods of vehicle color and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610235077.6A CN107301421A (en) | 2016-04-15 | 2016-04-15 | The recognition methods of vehicle color and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107301421A true CN107301421A (en) | 2017-10-27 |
Family
ID=60136675
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610235077.6A Pending CN107301421A (en) | 2016-04-15 | 2016-04-15 | The recognition methods of vehicle color and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107301421A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109389111A (en) * | 2018-10-24 | 2019-02-26 | 浙江宇视科技有限公司 | Image acquisition and processing method, device and computer readable storage medium |
| CN111476813A (en) * | 2020-04-28 | 2020-07-31 | 兰州交通大学 | Image change detection method, image change detection device, electronic equipment and storage medium |
| CN111860533A (en) * | 2019-04-30 | 2020-10-30 | 深圳数字生命研究院 | Image recognition method and device, storage medium and electronic device |
| 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 |
| CN116542968A (en) * | 2023-06-29 | 2023-08-04 | 中国铁路设计集团有限公司 | Intelligent counting method for steel bars based on template matching |
-
2016
- 2016-04-15 CN CN201610235077.6A patent/CN107301421A/en active Pending
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109389111A (en) * | 2018-10-24 | 2019-02-26 | 浙江宇视科技有限公司 | Image acquisition and processing method, device and computer readable storage medium |
| CN111860533A (en) * | 2019-04-30 | 2020-10-30 | 深圳数字生命研究院 | Image recognition method and device, storage medium and electronic device |
| CN111860533B (en) * | 2019-04-30 | 2023-12-12 | 深圳数字生命研究院 | Image recognition method and device, storage medium and electronic device |
| CN111476813A (en) * | 2020-04-28 | 2020-07-31 | 兰州交通大学 | Image change detection method, image change detection device, electronic equipment and storage medium |
| 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 |
| CN116542968A (en) * | 2023-06-29 | 2023-08-04 | 中国铁路设计集团有限公司 | Intelligent counting method for steel bars based on template matching |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109740478B (en) | Vehicle detection and identification method, device, computer equipment and readable storage medium | |
| Wang et al. | Fuzzy-based algorithm for color recognition of license plates | |
| CN107301421A (en) | The recognition methods of vehicle color and device | |
| CN109101934A (en) | Model recognizing method, device and computer readable storage medium | |
| CN106384117B (en) | A kind of vehicle color identification method and device | |
| CN106548160A (en) | A kind of face smile detection method | |
| CN105160691A (en) | Color histogram based vehicle body color identification method | |
| CN109918971B (en) | Method and device for detecting number of people in monitoring video | |
| Huang et al. | Foreground/background segmentation of color images by integration of multiple cues | |
| Li et al. | Research on vehicle license plate location based on neural networks | |
| CN109255326B (en) | Traffic scene smoke intelligent detection method based on multi-dimensional information feature fusion | |
| CN105138987B (en) | A kind of vehicle checking method based on converging channels feature and estimation | |
| CN110245691A (en) | A kind of intelligent identification Method of vehicle appearance color discoloration repacking | |
| CN109711379B (en) | Complex environment traffic signal lamp candidate area extraction and identification method | |
| CN110287798B (en) | Vector network pedestrian detection method based on feature modularization and context fusion | |
| CN105069816B (en) | A kind of method and system of inlet and outlet people flow rate statistical | |
| CN111507426A (en) | No-reference image quality grading evaluation method and device based on visual fusion characteristics | |
| KR101048582B1 (en) | Method and device for detecting faces in color images | |
| CN103530640A (en) | Unlicensed vehicle detection method based on AdaBoost and SVM (support vector machine) | |
| CN105825168A (en) | Golden snub-nosed monkey face detection and tracking algorithm based on S-TLD | |
| CN107944403A (en) | Pedestrian's attribute detection method and device in a kind of image | |
| Ma et al. | An all-weather lane detection system based on simulation interaction platform | |
| CN112818775A (en) | Forest road rapid identification method and system based on regional boundary pixel exchange | |
| CN109284759A (en) | One kind being based on the magic square color identification method of support vector machines (svm) | |
| CN1237485C (en) | Method for covering face of news interviewee using quick face detection |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20171027 |