CN107729801A - A kind of vehicle color identifying system based on multitask depth convolutional neural networks - Google Patents
A kind of vehicle color identifying system based on multitask depth convolutional neural networks Download PDFInfo
- Publication number
- CN107729801A CN107729801A CN201710558817.4A CN201710558817A CN107729801A CN 107729801 A CN107729801 A CN 107729801A CN 201710558817 A CN201710558817 A CN 201710558817A CN 107729801 A CN107729801 A CN 107729801A
- Authority
- CN
- China
- Prior art keywords
- mrow
- color
- vehicle
- msub
- car plate
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
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
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24143—Distances to neighbourhood prototypes, e.g. restricted Coulomb energy networks [RCEN]
-
- 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)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
A kind of vehicle color identifying system based on multitask depth convolutional neural networks, include the vision-based detection subsystem of high-definition camera above road travel line, traffic Cloud Server and vehicle color；The vision-based detection subsystem of vehicle color includes vehicle location detection module, License Plate detection module, car plate background color identification module, Colorimetry module, vehicle color correction module and vehicle color identification module, vehicle location detection module, License Plate detection module and vehicle color identification module share same Faster R CNN depth convolutional neural networks, road vehicle is gone out using depth convolutional neural networks Fast Segmentation, and the car plate further gone out with vehicle image using depth convolutional neural networks Fast Segmentation on road, then these vehicles and car plate spatial positional information shared in road image are provided again.The multitask depth convolutional neural networks that a kind of accuracy of detection of present invention offer is higher, robustness is higher are in vehicle color vision detection system.
Description
Technical field
Know the present invention relates to artificial intelligence, Digital Image Processing, convolutional neural networks and computer vision in vehicle color
The not application of aspect, belongs to intelligent transportation field.
Background technology
Color is a kind of important external appearance characteristic of vehicle.In real world, due to light source color temperature, light intensity, bat be present
Take the photograph the influence of angle, many uncertain factors such as setting of video camera so that the result finally obtained is compared with ideal conditions, car
Color can have a certain degree of colour cast；Existing published vehicle color identification method changes for vehicle attitude and vehicle
Residing photoenvironment change is very sensitive, when the photoenvironment change residing for vehicle, existing vehicle color identification method
The colour recognition degree of accuracy drastically decline, it is impossible to accurately identify vehicle color.
" localization method, the vehicle body of characteristic area are deep disclosed in the Chinese patent application of Application No. 200810041097.5
The recognition methods of light color and body color " provides the recognition methods of a kind of vehicle color identification method, vehicle color weight.
Comprise the following steps：
1st, the patent builds the energy function of complexity, it is maximum to search for energy according to the textural characteristics and architectural feature of image
Point；
2nd, according to the maximum point location vehicle color of energy and the identification region of vehicle color weight；
3rd, the pixel color and colour darkness in identification region, and carry out the face that statistics finally gives identification region
Color and colour darkness.
But the patent is in sample collection stage early stage, to the vehicle color identification under different light conditions at
Reason；Different characteristic attributes is obtained by multiple color spaces in selected characteristic vector；Then make in training pattern
Trained with the grader of multiple types；Chinese herbaceous peony cover area is only have selected when fixation and recognition region, for possible reflective
Phenomenon is not dealt with so that final vehicle color identification and the identification of the vehicle color depth produces certain deviation.
" a kind of vehicle body color in vehicle video image disclosed in the Chinese patent application of Application No. 200810240292.0
Recognition methods and system " provides a kind of vehicle body color in vehicle video image recognition methods.The patent is in training pattern
Multiple step format training is taken, is comprised the following steps：
1st, rough segmentation is carried out to vehicle body sample using cluster according to color template and obtains the sample of certain color or a variety of phases
The mixing sample of nearly color；
2nd, mixing sample is finely divided using arest neighbors sorting technique；
3rd, the model obtained according to training is slightly identified to vehicle color；
4th, careful identification is carried out using arest neighbors sorting technique.
But the patent equally not yet considers caused by vehicle color to change under different light conditions；In selected characteristic vector
When be also to use by tri- kinds of color spaces of HSV, YIQ, YCbCr and step by step；Then employed in training pattern
Cluster and arest neighbors sorting technique are combined training pattern；Vehicle color cognitive phase and it is undeclared be to take which kind of strategy to knowing
How the color of each pixel in other region is handled；And the patent only illustrates vehicle color identification method, not to car
The identification of colour darkness is explained.
At present, the identification of body color generally comprises two main modules：One is the detection in region to be identified and fixed
Position, the determination of body color reference zone, another is to carry out color classification and identification to the image of reference zone.
The detection in region to be identified and various ways are located,《Vehicle color identification method》103544480A、《Based on face
The body color recognition methods of Color Histogram》105160691A and《A kind of body color recognition methods and device》105354530A
In, detection and positioning licence plate first then according to car plate positional information, determine that vehicle color identifies reference zone.《Vehicle color
Recognition methods and device》In 102737221B, according to the reference of the texture of image and structural information positioning vehicle colour recognition
Region, carry out main identification region again afterwards and assist in identifying the positioning in region.《A kind of body color recognition methods》
In 105005766A, according to the method for moving object detection in video, determine the boundary rectangle of moving target as colour recognition
Reference zone.《A kind of method of automatic identification road gate video and the vehicle color in picture》In 104680195A, do not have
There is the positioning method for being expressly recited color candidate region, only specify that candidate region is polylith, is concentrated mainly on hood.
The vision detection technology in deep learning epoch before above-mentioned vision detection technology belongs to, there is accuracy of detection and detection
Key issues of change of the problem of robustness is not high, especially illumination and video camera imaging condition, does not solve also very well.In addition,
Above-mentioned several patent contents are only disclosed some technical overviews, still have many ins and outs and key issue in practical application still
It is not directed to, in particular for《The law on road traffic safety》Various detailed problems solution.
Recent years, deep learning in the technology of computer vision field obtained rapid development, and deep learning can utilize
Substantial amounts of training sample and hidden layer successively in depth learn the abstracted information of image, more comprehensively directly obtain characteristics of image.
Digital picture is described with matrix, and convolutional neural networks describe the whole of image preferably from local message block
Body structure, therefore solve problem using convolutional neural networks mostly in computer vision field, deep learning method.Around
Improve accuracy of detection and detection time, depth convolutional neural networks technology is from R-CNN, Faster R-CNN to Fasterer R-
CNN.Be embodied in further precision improvement, acceleration, end-to-end and more practical, almost cover from be categorized into detection, point
Cut, position every field.Depth learning technology applies to vehicle color vision-based detection and will be one to have very much actual application value
Research field.
The vision system of people has color constancy, can obtain object under the photoenvironment and image-forming condition of some changes
The invariant features of surface color.But bayonet socket Imaging for Monitoring equipment does not have this " regulation " function, different photoenvironments will
It can cause a certain degree of deviation be present between the color of the image of collection and object true colors.After this deviation will influence
The degree of accuracy of continuous vehicle color analysis and robustness.So seeking suitable color correction algorithm, photoenvironment etc. is eliminated to face
The influence that color shows, image after treatment is set correctly to reflect that the true colors of object have turned into a current research heat
Point.
National standard GA 36-2014 define the various details of automotive number plate, wherein, large-scale motor vehicles for civilian use：Yellow bottom is black
Word；Small Civil automobile：Blue bottom wrongly written or mispronounced character；People's Armed Police's special purpose vehicle：White background red " WJ ", surplus；Other foreign nationality's automobiles：Black matrix wrongly written or mispronounced character；
Make, consulate's foreign nationality's automobile：Black matrix wrongly written or mispronounced character and hollow " making " word mark；Trial Run License Plate：White background The Scarlet Letter, there is " examination " word mark before digital
Will；Temporary licence：White background The Scarlet Letter, there is " interim " two word before digital；Automobile benefit licence plate：White gravoply, with black engraved characters.Between the character of car plate
At intervals of 12mm.These regulations on car plate, the especially regulation in terms of color, a ginseng is brought to vehicle color identification
Examine standard.Due under same illumination condition, the aberration of identical depth occurs in vehicle color and car plate color；By detecting
The aberration of car plate color corrects vehicle color, and this is of great significance for lifting the discrimination tool of vehicle color.
The content of the invention
In order to overcome, the accuracy of detection of the vision-based detection mode of existing vehicle color is relatively low, detection robustness is not high not
Foot, the present invention provide the multitask depth convolutional neural networks that a kind of accuracy of detection is higher, robustness is higher and regarded in vehicle color
Feel detecting system.
The technical solution adopted for the present invention to solve the technical problems is：
A kind of vehicle color identifying system based on multitask depth convolutional neural networks, including installed in road travel line
The vision-based detection subsystem of the high-definition camera of top, traffic Cloud Server and vehicle color；
Described high-definition camera is used to obtain the video data on road, configures in the top of lane, passes through network
Vedio data on road is transferred to described traffic Cloud Server；
Described traffic Cloud Server is used to receive from the video data on the road that described high-definition camera is obtained,
And the vision detection system for being submitted to described vehicle color carries out vehicle color identification；
The vision-based detection subsystem of described vehicle color include vehicle location detection module, License Plate detection module,
Car plate background color identification module, Colorimetry module, vehicle color correction module and vehicle color identification module, described car
Detection and localization module, described License Plate detection module and described vehicle color identification module share same Faster
R-CNN depth convolutional neural networks, road vehicle is gone out using depth convolutional neural networks Fast Segmentation, and with vehicle figure
As the car plate further gone out using depth convolutional neural networks Fast Segmentation on road, these vehicles are then provided again and car plate exists
Shared spatial positional information in road image.
Further, vehicle and License Plate Segmentation and positioning are made up of two models, and a model is that the selectivity for generating RoI is searched
Rope network；Another model is Faster R-CNN vehicles and car plate target detection network；Realized after two Classification and Identification networks
One multi-level, successively progressive multi-task learning network of multi-tag, multiple features fusion；
Described selective search network, i.e. RPN；RPN networks export rectangle using any scalogram picture as input
The set of target Suggestion box, each frame include 4 position coordinates variables and a score；The target of described target Suggestion box refers to
Be Vehicle Object and car plate object；
It is the estimated probability of target/non-targeted to each Suggestion box, is the classification realized with the softmax layers of two classification
Layer；K Suggestion box is parameterized by the corresponding k Suggestion box for being referred to as anchor；
Each anchor is centered on current sliding window mouth center, and a kind of corresponding yardstick and length-width ratio, uses 3 kinds of yardsticks
With 3 kinds of length-width ratios, so just there is k=9 anchor in each sliding position；
RPN networks are trained, a binary label is distributed to each anchor, is to mark the anchor with this
Target；Then distribute positive label and give this two class anchor：(I) have with some real target bounding box, i.e. Ground Truth, GT
The ratio between highest IoU, i.e. Interse-ction-over-Union, common factor union, overlapping anchor；(II) with any GT bags
Enclosing box has the overlapping anchor of the IoU more than 0.7；Notice that a GT bounding box may distribute positive label to multiple anchor；
The IoU ratios that the negative label of distribution gives all GT bounding boxs are below 0.3 anchor；Anon-normal non-negative anchor is to training mesh
No any effect is marked, then is abandoned；
The multitask loss in Faster R-CNN is followed, minimizes object function；The loss function of one image is determined
Justice is：
Here, i is anchor index, piIt is the prediction probability that anchor is the i-th target, if anchor is
Just, GT labelsIt is exactly 1, if anchor is negative,It is exactly 0；tiIt is a vector, represents 4 parameters of the bounding box of prediction
Change coordinate,It is the coordinate vector of GT bounding boxs corresponding with positive anchor；λ is a balance weight, here λ=10, NclsIt is
The normalized value of cls items is mini-batch size, here Ncls=256, NregThe normalized value for being reg items is anchor positions
The quantity put, Nreg~2,400, Classification Loss function LclsThree classifications, i.e. vehicle target object, car plate destination object vs.
The logarithm loss of road background：
In formula, LclsFor Classification Loss function, PiIt is the prediction probability of the i-th target for anchor；Surrounded for real target
The prediction probability of i-th target of box；
For returning loss function Lreg, defined to minor function：
In formula, LregTo return loss function, R is the loss function of robust, and smooth L are calculated with formula (4)1；
In formula, smoothL1For smooth L1Loss function, x are variable；
Faster R-CNN networks, characteristic pattern is obtained after depth convolutional neural networks in input picture, according to feature
Figure and RPN networks can then obtain corresponding RoIs, finally then pass through RoI ponds layer；Wherein RoI, i.e. area-of-interest, refer to
Be exactly vehicle target object, car plate destination object；
For Faster R-CNN networks, input is N number of Feature Mapping and R RoI；N number of Feature Mapping comes from finally
One convolutional layer, the size of each Feature Mapping is w × h × c；
Each RoI is a tuple (n, r, c, h, w), wherein, n is the index of Feature Mapping, n ∈ (0,1,2 ..., N-
1), r, c are top left co-ordinates, and h, w are height and width respectively；
Export the Feature Mapping then obtained by maximum pond；RoI in artwork is mapped with the block in characteristic pattern；Will
Characteristic pattern down-sampling is fixed size, is then passed to full connection again.
Further, the selective search network and Fast R-CNN are all stand-alone trainings, are trained and calculated using 4 steps
Method, learn shared feature by alternative optimization；The first step, according to above-mentioned training RPN, network ImageNet pre-training
Model initialization, and end-to-end finely tune suggests task for region；Second step, the Suggestion box generated using the RPN of the first step, by
Fast R-CNN train an individually detection network, and this detection network is equally initial by the model of ImageNet pre-training
Change, at this time two networks are also without shared convolutional layer；3rd step, trained with detection netinit RPN, but it is fixed shared
Convolutional layer, and only finely tune the exclusive layers of RPN, present two network share convolutional layers；4th step, keep shared convolution
Layer is fixed, fine setting Fast R-CNN fc, i.e., full articulamentum；So, two network share identical convolutional layers, a system is formed
One network；
By the processing of above-mentioned two network, vehicle target object in a frame video image, car plate target pair are detected
As and to they size and locus confined, that is, obtained vehicle target object, the size of car plate destination object
And locus；Its rv,cvIt is the top left co-ordinate of vehicle target object in the picture, hv,wvIt is that vehicle target object exists respectively
The projected size of the plane of delineation, i.e. height and width；Its rp,cpIt is the top left co-ordinate of car plate in the picture, hp,wpIt is car plate respectively
In the projected size of the plane of delineation, i.e. height and width；
The progressive cascade connection between each task is make use of in Faster R-CNN networks, that is, it is precisely fixed to be followed successively by vehicle
Position, vehicle, brand and the identification of car system, car plate precise positioning, Car license recognition and color recognizing for vehicle id, acetes chinensis, vehicle color
Correction, a colour recognition.
Further, described car plate background color identification module is used to license plate image is handled to obtain under existing environmental condition
Car plate background color；License plate image is subjected to grey level histogram processing, is in car plate at the peak valley in grey level histogram
The interval of character and character, that is, the background color of car plate；Pixel RGB color component in interval is averaged, finally
Obtain the car plate background color under existing environmental condition.
Described Colorimetry module is used to carry on the back the car plate background color of national regulations and the car plate under existing environmental condition
Calculating is compared in scape color, obtains the aberration under existing environmental condition；First, with the car plate background color ratio under existing environmental condition
To the several types of the car plate background color of national regulations, a kind of car plate background color of immediate national regulations is obtained, and
As the car plate background color under standard light；The calculating of aberration is carried out on CIE1976Lab color spaces；In order to fast
Speed is realized from RGB color to the conversion of Lab color spaces, using rapid translating mode, as shown in formula (5)；
In formula, R, G, B are respectively the color component in RGB color, and L is the lightness point of CIE1976Lab color spaces
Amount, a and b are the chromatic component of CIE1976Lab color spaces；
Car plate background color under the car plate background color of national regulations and existing environmental condition is all calculated by formula (5)
Obtain respective L, a, b value；Wherein, LNPAnd LRPRespectively under the car plate background color of national regulations and existing environmental condition
The brightness value of car plate background color, aNPAnd aRP、bNPAnd bRPRespectively under the car plate background color of national regulations and existing environmental condition
Car plate background color colourity, the aberration Δ E between bothabCIE1976Lab aberration is calculated with formula (6)；
In formula, Δ L=LNP-LRPFor luminosity equation, Δ a=aNP-aRP, Δ b=bNP-bRPFor colour difference, Δ EabIt is single for aberration
Position is NBS.
The aberration that described vehicle color correction module is used to be obtained according to detection is to the vehicle color under existing environmental condition
It is corrected, obtains the vehicle color image under national regulations ecotopia；First, it is determined that aberration Δ EabWhether threshold value is exceededIf it exceeds threshold value just carries out vehicle color correction, correction is calculated as shown in formula (7)；
In formula, LNMFor the brightness value of the vehicle color under national regulations ecotopia, LRPFor the vehicle under existing environmental condition
The brightness value of color, Δ L are the luminosity equation of the car plate background color and the car plate background color under existing environmental condition of national regulations,
aNMAnd bNMFor the chromatic value of the vehicle color under national regulations ecotopia, aRPAnd bRPShow the vehicle color under environmental condition
The colour difference of the car plate background color of chromatic value, Δ a and Δ b national regulations and the car plate background color under existing environmental condition；
Further, inverse transformation, such as formula are carried out from Lab color space to RGB color to the vehicle color after correction
(8) shown in；
In formula, R, G, B are respectively the color component in RGB color, and L is the lightness point of CIE1976Lab color spaces
Amount, a and b are the chromatic component of CIE1976Lab color spaces；
Formula (8) equation group is the formula after optimization, and floating-point operation is converted into normal multiplication of integers and the side of displacement
Formula, is written as div2^23 in formula by displacement, and expression moves right 23；The span of RGB and Lab in formula be all [0,
255], then by inverse Gamma functions the rgb value of the vehicle color under national regulations ecotopia is obtained.
Described vehicle color identification module is used to the vehicle color after correction be identified, in order to effectively shared
Faster R-CNN depth convolutional neural networks, the vehicle image after color correction is put on into corresponding color label and instructed
Practice；When being identified to vehicle color, by vehicle precise positioning, car plate precise positioning, the identification of car plate background color, acetes chinensis,
After vehicle color correction process step, vehicle color identification is carried out finally by Faster R-CNN depth convolutional neural networks；
Identified with the vehicle image after color correction by Faster R-CNN depth convolutional neural networks with regard to that can obtain in standard illumination
Under the conditions of vehicle color.
Beneficial effects of the present invention are mainly manifested in：It has been inherently eliminated color caused by illumination and camera setting
Difference, effectively raise the detection robustness of vehicle color；The multitask deep learning convolutional neural networks of use, utilize each
Progressive cascade system between business, by vehicle precise positioning, vehicle, brand and the identification of car system, car plate precise positioning, Car license recognition
And color recognizing for vehicle id, same Faster R-CNN depth convolutional neural networks are shared in vehicle color identification, each being lifted
While the fixation and recognition precision of task, overall recognition time is also effectively shortened, improves the real-time of detection identification.
Brief description of the drawings
Fig. 1 is Fast R-CNN structure charts；
Fig. 2 is selective search network；
Fig. 3 is the progressive cascade connection structure charts of multitask Faster R-CNN；
Fig. 4 is multitask Faster R-CNN vehicle color vision-based detection network structures；
Fig. 5 is with the explanation figure that car plate background color method under existing environmental condition is extracted in grey level histogram；
Fig. 6 is the vehicle color identification process flow chart of multitask Faster R-CNN depth convolutional networks.
Embodiment
The invention will be further described below in conjunction with the accompanying drawings.
1~Fig. 6 of reference picture, a kind of vehicle color identifying system based on multitask depth convolutional neural networks, including peace
The vision-based detection subsystem of high-definition camera, traffic Cloud Server and vehicle color above road travel line；
Described high-definition camera is used to obtain the video data on road, configures in the top of lane, passes through network
Vedio data on road is transferred to described traffic Cloud Server；
Described traffic Cloud Server is used to receive from the video data on the road that described high-definition camera is obtained,
And the vision detection system for being submitted to described vehicle color carries out vehicle color identification；Handling process is as shown in fig. 6, head
First, segmentation positioning is carried out to video image, extracts the vehicle image in the image；Then, to the car plate figure in vehicle image
As carrying out segmentation positioning again, license plate image is extracted；Then license plate image is handled with grey level histogram, obtains car plate
Grey level histogram；Then according to the character characteristic distributions of car plate, the background color region of the car plate in grey level histogram is extracted；
Further, remove to match the car plate background colour in immediate national standard according to the car plate background colour extracted；Further,
The RGB color for the car plate background colour that national standard car plate background colour and detection are obtained is transformed into Lab color spaces respectively；
Then the aberration between both is calculated, if aberration exceedes threshold valueThe vehicle image of RGB color is transformed into Lab
On color space, chromatic aberration correction is carried out to vehicle image with this aberration, the vehicle image of no color differnece is obtained, then again by color
Correction rear vehicle image is transformed on RGB color；Finally, vehicle, brand are carried out to color correction rear vehicle image, is
Row, body color identification, finally give accurately surface description.
The People's Republic of China's automotive number plates of People's Republic of China (PRC) industry standards of public safety GA 36-2014, with
Lower abbreviation national standard, have to the color of car plate and provide in detailed below, metal material number plate is defined in national standard under the irradiation of A light sources
Chromaticity coordinate should meet the regulation of 4.4.1 in GA 666-2006, as shown in table 1；The reflective surface for defining blue background color number plate exists
The aberration of chromaticity coordinate and Standard colour board under the irradiation of D65 light sources should be no more than 8.0NBS；Luminance factor should meet GA 666-
4.4.2 regulation in 2006, as shown in table 2.Above-mentioned regulation is to the invention provides a kind of reference of Standard Colors.
Table 1
Table 2
The vision-based detection subsystem of described vehicle color include vehicle location detection module, License Plate detection module,
Car plate background color identification module, Colorimetry module, vehicle color correction module and vehicle color identification module；Described car
Detection and localization module, described License Plate detection module and described vehicle color identification module share same Faster
R-CNN depth convolutional neural networks, road vehicle is gone out using depth convolutional neural networks Fast Segmentation, and with vehicle figure
As the car plate further gone out using depth convolutional neural networks Fast Segmentation on road, these vehicles are then provided again and car plate exists
Shared spatial positional information in road image.
Vehicle and License Plate Segmentation and positioning are made up of two models, and a model is the selective search network for generating RoI；
Another model is Faster R-CNN vehicles and car plate target detection network, and detection unit structure chart is as shown in Figure 1；The present invention
In further modification has been made in Faster R-CNN networks to this model；A multilayer is realized after two Classification and Identification networks
The successively progressive multi-task learning network of secondary, multi-tag, multiple features fusion, as shown in Figure 4；
Selective search network, i.e. RPN；RPN networks are built any scalogram picture as input, output rectangular target
The set of frame is discussed, each frame includes 4 position coordinates variables and a score.For formation zone Suggestion box, at last
Small network is slided in the convolution Feature Mapping of shared convolutional layer output, this network is connected to input convolution Feature Mapping entirely
In n × n spatial window.Each sliding window is mapped on a low-dimensional vector, a sliding window of each Feature Mapping
A corresponding numerical value.This vector exports the layer of the full connection at the same level to two.
In the position of each sliding window, while k suggestion areas is predicted, so position returns layer and has 4k output,
The codes co-ordinates of i.e. k bounding box.Layer of classifying exports the score of 2k bounding box, i.e., is target/non-targeted to each Suggestion box
Estimated probability, be the classification layer realized with the softmax layers of two classification, k can also be generated with logistic recurrence
Point.K Suggestion box is parameterized by the corresponding k Suggestion box for being referred to as anchor.Each anchor is with current sliding window mouth center
Centered on, and a kind of corresponding yardstick and length-width ratio, using 3 kinds of yardsticks and 3 kinds of length-width ratios, so just have in each sliding position
K=9 anchor.For example, for the convolution Feature Mapping that size is w × h, then a total of w × h × k anchor.RPN nets
Network structure chart is as shown in Figure 2.
In order to train RPN networks, a binary label is distributed to each anchor, is to mark the anchor with this
It is not target.Then distribute positive label and give this two class anchor：(I) with some real target bounding box, i.e. Ground Truth,
GT has the ratio between highest IoU, i.e. Interse-ction-over-Union, common factor union, overlapping anchor；(II) it is and any
GT bounding boxs have the overlapping anchor of the IoU more than 0.7.Notice that a GT bounding box may give multiple anchor distribution positive mark
Label.The IoU ratios that the negative label of distribution gives all GT bounding boxs are below 0.3 anchor.Anon-normal non-negative anchor is to instruction
Practicing target does not have any effect, then abandons.
There are these to define, it then follows the multitask loss in Faster R-CNN, to minimize object function.To an image
Loss function be defined as：
Here, i is anchor index, piIt is the prediction probability that anchor is the i-th target, if anchor is
Just, GT labelsIt is exactly 1, if anchor is negative,It is exactly 0；tiIt is a vector, represents 4 parameters of the bounding box of prediction
Change coordinate,It is the coordinate vector of GT bounding boxs corresponding with positive anchor；λ is a balance weight, here λ=10, NclsIt is
The normalized value of cls items is mini-batch size, here Ncls=256, NregThe normalized value for being reg items is anchor positions
The quantity put, Nreg~2,400, Classification Loss function Lcls3 classifications, i.e. vehicle target, car plate target vs. road backgrounds
Logarithm loses：
For returning loss function Lreg, defined to minor function：
In formula, LregTo return loss function, R is the loss function of robust, and smooth L are calculated with formula (4)1；
In formula, smoothL1For smooth L1Loss function, x are variable；
Faster R-CNN network structures in input picture after depth convolutional neural networks as shown in figure 3, can obtain
To characteristic pattern, corresponding RoIs can be then obtained according to characteristic pattern and RPN networks, finally then passes through RoI ponds layer.The layer is
The only process in level spatial " pyramid " pond.Input is N number of Feature Mapping and R RoI.N number of Feature Mapping comes from most
The latter convolutional layer, the size of each Feature Mapping is w × h × c.Each RoI is a tuple (n, r, c, h, w), wherein,
N is the index of Feature Mapping, and n ∈ (0,1,2 ..., N-1), r, c are top left co-ordinates, and h, w are height and width respectively.Output then by
The Feature Mapping that maximum pond obtains.The effect of this layer mainly has two, first, by the block pair in the RoI and characteristic pattern in artwork
It should get up；It by characteristic pattern down-sampling is fixed size that another, which is, is then passed to full connection again.
Selective search network is shared with detecting the weights of network：Selective search network and Faster R-CNN are only
Vertical training, differently to change their convolutional layer.Therefore need to allow to share convolution between two networks using a kind of
The technology of layer, rather than learn two networks respectively.A kind of 4 practical step training algorithms are used in invention, pass through alternative optimization
To learn shared feature.The first step, according to above-mentioned training RPN, the model initialization of network ImageNet pre-training, and hold
It is used for region to end fine setting and suggests task.Second step, the Suggestion box generated using the RPN of the first step, is instructed by Faster R-CNN
Practice an individually detection network, this detection network is equally by the model initialization of ImageNet pre-training, at this time
Two networks are also without shared convolutional layer.3rd step, trained with detection netinit RPN, but fixed shared convolutional layer, and
And only finely tune the exclusive layers of RPN, present two network share convolutional layers.4th step, keep shared convolutional layer to fix, finely tune
Faster R-CNN fc, i.e., full articulamentum.So, two network share identical convolutional layers, a unified network is formed.
In view of object it is multiple dimensioned the problem of, use three kinds of simple chis for each characteristic point on characteristic pattern
Degree, the area of bounding box is respectively 128 × 128,256 × 256,512 × 512 and three kind of length-width ratio, respectively 1:1、1:2、2:
1.Pass through this design, in this way it is no longer necessary to which Analysis On Multi-scale Features or multi-scale sliding window mouth predict big region, can reach section
Save the effect of a large amount of run times.
By the processing of above-mentioned two network, vehicle in a frame video image is detected and to its size and locus
Confined, that is, obtained size and the locus of vehicle, its rv,cvIt is the top left co-ordinate of vehicle in the picture, hv,
wvIt is projected size of the vehicle in the plane of delineation, i.e. height and width respectively；
Because object of interest in the present invention is vehicle, car plate, i.e. object of interest, hereinafter referred to as RoI, in order to position
With the various RoI that are partitioned on road, it is necessary in study and training convolutional neural networks, by various vehicles, car plate and road
Road background image is put on corresponding label and is trained respectively；So by Faster R-CNN depth convolutional neural networks just
It can automatically split and orient vehicle and car plate.In order to lift the positioning precision of car plate, the car after positioning will be split in the present invention
Image carries out segmentation positioning by Faster R-CNN depth convolutional neural networks to the car plate of the vehicle again；
A kind of multitask deep learning convolutional neural networks are employed in the present invention, this is due to multitask deep learning net
The image recognition of network is often better than single task deep learning network, and multitask has the association between task in learning process
Property, i.e., existence information is shared between task, and this is also the necessary condition of multitask deep learning；Multiple tasks are trained at the same time
When, network utilizes the induction bias ability of shared information enhancement system and the generalization ability of grader between task；Such as Fig. 3 institutes
Show, the progressive cascade connection between each task is taken full advantage of in the present invention, that is, is followed successively by vehicle precise positioning, vehicle, brand
With car system identification, car plate precise positioning, Car license recognition and color recognizing for vehicle id, acetes chinensis, vehicle color correction, vehicle color
Identification；
Described vehicle location detection module is oriented in road vehicle object images for segmentation, using Faster
R-CNN depth convolutional neural networks are handled the video image on road, have obtained size and the locus of vehicle, its
rv,cvIt is the top left co-ordinate of vehicle in the picture, hv,wvIt is projected size of the vehicle in the plane of delineation respectively；
Described License Plate detection module is used to split the license plate image oriented in Vehicle Object image, uses
Faster R-CNN depth convolutional neural networks are handled vehicle image, have obtained size and the locus of car plate, its
rp,cpIt is the top left co-ordinate of car plate in the picture, hp,wpIt is projected size of the car plate in the plane of delineation respectively；
Described car plate background color identification module is used to handle to obtain the car plate back of the body under existing environmental condition to license plate image
Scape color；Specific practice is license plate image to be carried out into grey level histogram processing, as shown in figure 5, the peak valley in grey level histogram
Place is the interval of the character and character in car plate, that is, the background color of car plate；Pixel RGB color component in interval is entered
Row is average, finally gives the car plate background color under existing environmental condition；
Described Colorimetry module is used to carry on the back the car plate background color of national regulations and the car plate under existing environmental condition
Calculating is compared in scape color, obtains the aberration under existing environmental condition；First, with the car plate background color ratio under existing environmental condition
To the several types of the car plate background color of national regulations, a kind of car plate background color of immediate national regulations is obtained, and
As the car plate background color under standard light；The calculating of aberration is carried out on CIE1976Lab color spaces, general next
Say, the conversion from RGB color to Lab color spaces is needed in two steps to realize；The first step is from RGB very color 24bit
XYZ color space is converted to Lab color spaces by color space conversion to XYZ color space, second step；In order to quickly realize from
RGB color to Lab color spaces conversion, present invention employs a kind of rapid translating mode, as shown in formula (5)；
Car plate background color under the car plate background color of national regulations and existing environmental condition is all calculated by formula (5)
Obtain respective L, a, b value；Wherein, LNPAnd LRPRespectively under the car plate background color of national regulations and existing environmental condition
The brightness value of car plate background color, aNPAnd aRP、bNPAnd bRPRespectively under the car plate background color of national regulations and existing environmental condition
Car plate background color colourity, the aberration Δ E between bothabFormula (6) can be used to calculate CIE1976Lab aberration；
In formula, Δ L=LNP-LRPFor luminosity equation, Δ a=aNP-aRP, Δ b=bNP-bRPFor colour difference, Δ EabIt is single for aberration
Position is NBS；
The aberration that described vehicle color correction module is used to be obtained according to detection is to the vehicle color under existing environmental condition
It is corrected, obtains the vehicle color image under national regulations ecotopia；First, it is determined that aberration Δ EabWhether threshold value is exceededIf it exceeds threshold value just carries out vehicle color correction, correction is calculated as shown in formula (7)；
In formula, LNMFor the brightness value of the vehicle color under national regulations ecotopia, LRPFor the vehicle under existing environmental condition
The brightness value of color, Δ L are the luminosity equation of the car plate background color and the car plate background color under existing environmental condition of national regulations,
aNMAnd bNMFor the chromatic value of the vehicle color under national regulations ecotopia, aRPAnd bRPShow the vehicle color under environmental condition
The colour difference of the car plate background color of chromatic value, Δ a and Δ b national regulations and the car plate background color under existing environmental condition；
Further, inverse transformation, such as formula are carried out from Lab color space to RGB color to the vehicle color after correction
(8) shown in；
In formula, R, G, B are respectively the color component in RGB color, and L is the lightness point of CIE1976Lab color spaces
Amount, a and b are the chromatic component of CIE1976Lab color spaces；
Formula (8) equation group is the formula after optimization, and floating-point operation is converted into normal multiplication of integers and the side of displacement
Formula, is written as div2^23 in formula by displacement, and expression moves right 23；The span of RGB and Lab in formula be all [0,
255], then by inverse Gamma functions the rgb value of the vehicle color under national regulations ecotopia is obtained；
Described vehicle color identification module is used to the vehicle color after correction be identified, in order to effectively shared
Faster R-CNN depth convolutional neural networks, vehicle image after color correction is put on to corresponding color mark in of the invention
Label are trained；When being identified to vehicle color, pass through vehicle precise positioning, car plate precise positioning, the identification of car plate background color, color
After difference detection, vehicle color correction process step, vehicle face is carried out finally by Faster R-CNN depth convolutional neural networks
Color identifies；This vehicle image with after color correction is identified with regard to that can obtain by Faster R-CNN depth convolutional neural networks
The color of vehicle under standard illumination condition.
The foregoing is only the preferable implementation example of the present invention, be not intended to limit the invention, it is all in spirit of the invention and
Within principle, any modification, equivalent substitution and improvements made etc., it should be included in the scope of the protection.
Claims (7)
- A kind of 1. vehicle color identifying system based on multitask depth convolutional neural networks, it is characterised in that：Including installed in The vision-based detection subsystem of high-definition camera, traffic Cloud Server and vehicle color above road travel line；Described high-definition camera is used to obtain video data on road, configures in the top of lane, by network by road Vedio data on road is transferred to described traffic Cloud Server；Described traffic Cloud Server is used to receive from the video data on the road that described high-definition camera is obtained, and will It is submitted to the vision detection system of described vehicle color and carries out vehicle color identification；The vision-based detection subsystem of described vehicle color includes vehicle location detection module, License Plate detection module, car plate Background color identification module, Colorimetry module, vehicle color correction module and vehicle color identification module, described vehicle are determined Position detection module, described License Plate detection module and described vehicle color identification module share same Faster R- CNN depth convolutional neural networks, road vehicle is gone out using depth convolutional neural networks Fast Segmentation, and use vehicle image The car plate further gone out using depth convolutional neural networks Fast Segmentation on road, then provides these vehicles and car plate in road again Shared spatial positional information in the image of road.
- 2. the vehicle color identifying system as claimed in claim 1 based on multitask depth convolutional neural networks, its feature exist In：Vehicle and License Plate Segmentation and positioning are made up of two models, and a model is the selective search network for generating RoI；It is another Individual model is Faster R-CNN vehicles and car plate target detection network；Realized after two Classification and Identification networks one it is multi-level, The successively progressive multi-task learning network of multi-tag, multiple features fusion；Described selective search network, i.e. RPN；RPN networks export rectangular target using any scalogram picture as input The set of Suggestion box, each frame include 4 position coordinates variables and a score；The target of described target Suggestion box refers to Vehicle Object and car plate object；It is the estimated probability of target/non-targeted to each Suggestion box, is the classification layer realized with the softmax layers of two classification；K Suggestion box is parameterized by the corresponding k Suggestion box for being referred to as anchor；Each anchor is centered on current sliding window mouth center, and a kind of corresponding yardstick and length-width ratio, uses 3 kinds of yardsticks and 3 Kind length-width ratio, so just has k=9 anchor in each sliding position；RPN networks are trained, give each anchor to distribute a binary label, with this come to mark the anchor be mesh Mark；Then distribute positive label and give this two class anchor：(I) have most with some real target bounding box, i.e. Ground Truth, GT The ratio between high IoU, i.e. Interse-ction-over-Union, common factor union, overlapping anchor；(II) surrounded with any GT Box has the overlapping anchor of the IoU more than 0.7；Notice that a GT bounding box may distribute positive label to multiple anchor；Point With negative label give all GT bounding boxs IoU ratios be below 0.3 anchor；Anon-normal non-negative anchor is to training objective There is no any effect, then abandon；The multitask loss in Faster R-CNN is followed, minimizes object function；The loss function of one image is defined as：<mrow> <mi>L</mi> <mrow> <mo>(</mo> <mo>{</mo> <msub> <mi>p</mi> <mi>i</mi> </msub> <mo>}</mo> <mo>,</mo> <mo>{</mo> <msub> <mi>t</mi> <mi>i</mi> </msub> <mo>}</mo> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mn>1</mn> <msub> <mi>N</mi> <mrow> <mi>c</mi> <mi>l</mi> <mi>s</mi> </mrow> </msub> </mfrac> <munder> <mo>&Sigma;</mo> <mi>i</mi> </munder> <msub> <mi>L</mi> <mrow> <mi>c</mi> <mi>l</mi> <mi>s</mi> </mrow> </msub> <mrow> <mo>(</mo> <msub> <mi>p</mi> <mi>i</mi> </msub> <mo>,</mo> <msubsup> <mi>p</mi> <mi>i</mi> <mo>*</mo> </msubsup> <mo>)</mo> </mrow> <mo>+</mo> <mi>&lambda;</mi> <mfrac> <mn>1</mn> <msub> <mi>N</mi> <mrow> <mi>r</mi> <mi>e</mi> <mi>g</mi> </mrow> </msub> </mfrac> <munder> <mo>&Sigma;</mo> <mi>i</mi> </munder> <msubsup> <mi>p</mi> <mi>i</mi> <mo>*</mo> </msubsup> <msub> <mi>L</mi> <mrow> <mi>r</mi> <mi>e</mi> <mi>g</mi> </mrow> </msub> <mrow> <mo>(</mo> <msub> <mi>t</mi> <mi>i</mi> </msub> <mo>,</mo> <msubsup> <mi>t</mi> <mi>i</mi> <mo>*</mo> </msubsup> <mo>)</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>1</mn> <mo>)</mo> </mrow> </mrow>Here, i is anchor index, piIt is the prediction probability that anchor is the i-th target, if anchor is just, GT marks LabelIt is exactly 1, if anchor is negative,It is exactly 0；tiIt is a vector, represents 4 parametrization coordinates of the bounding box of prediction,It is the coordinate vector of GT bounding boxs corresponding with positive anchor；λ is a balance weight, here λ=10, NclsIt is cls items Normalized value is mini-batch size, here Ncls=256, NregIt is that the normalized value of reg items is the number of anchor positions Amount, Nreg~2,400, Classification Loss function LclsIt is three classifications, i.e., vehicle target object, car plate destination object vs. roads are carried on the back The logarithm loss of scape：<mrow> <msub> <mi>L</mi> <mrow> <mi>c</mi> <mi>l</mi> <mi>s</mi> </mrow> </msub> <mrow> <mo>(</mo> <msub> <mi>p</mi> <mi>i</mi> </msub> <mo>,</mo> <msubsup> <mi>p</mi> <mi>i</mi> <mo>*</mo> </msubsup> <mo>)</mo> </mrow> <mo>=</mo> <mo>-</mo> <mi>l</mi> <mi>o</mi> <mi>g</mi> <mo>&lsqb;</mo> <msubsup> <mi>p</mi> <mi>i</mi> <mo>*</mo> </msubsup> <msub> <mi>p</mi> <mi>i</mi> </msub> <mo>+</mo> <mrow> <mo>(</mo> <mn>1</mn> <mo>-</mo> <msubsup> <mi>p</mi> <mi>i</mi> <mo>*</mo> </msubsup> <mo>)</mo> </mrow> <mrow> <mo>(</mo> <mn>1</mn> <mo>-</mo> <msub> <mi>p</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mo>&rsqb;</mo> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>2</mn> <mo>)</mo> </mrow> </mrow>In formula, LclsFor Classification Loss function, PiIt is the prediction probability of the i-th target for anchor；For real target bounding box The prediction probability of i-th target；For returning loss function Lreg, defined to minor function：<mrow> <msub> <mi>L</mi> <mrow> <mi>r</mi> <mi>e</mi> <mi>g</mi> </mrow> </msub> <mo>=</mo> <mrow> <mo>(</mo> <msub> <mi>t</mi> <mi>i</mi> </msub> <mo>,</mo> <msubsup> <mi>t</mi> <mi>i</mi> <mo>*</mo> </msubsup> <mo>)</mo> </mrow> <mo>=</mo> <mi>R</mi> <mrow> <mo>(</mo> <msub> <mi>t</mi> <mi>i</mi> </msub> <mo>-</mo> <msubsup> <mi>t</mi> <mi>i</mi> <mo>*</mo> </msubsup> <mo>)</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>3</mn> <mo>)</mo> </mrow> </mrow>In formula, LregTo return loss function, R is the loss function of robust, and smooth L are calculated with formula (4)1；<mrow> <msub> <mi>smooth</mi> <mrow> <mi>L</mi> <mn>1</mn> </mrow> </msub> <mrow> <mo>(</mo> <mi>x</mi> <mo>)</mo> </mrow> <mo>=</mo> <mfenced open = "{" close = ""> <mtable> <mtr> <mtd> <mrow> <mn>0.5</mn> <msup> <mi>x</mi> <mn>2</mn> </msup> </mrow> </mtd> <mtd> <mrow> <mi>i</mi> <mi>f</mi> <mo>|</mo> <mi>x</mi> <mo>|</mo> <mo><</mo> <mn>1</mn> </mrow> </mtd> </mtr> <mtr> <mtd> <mrow> <mo>|</mo> <mi>x</mi> <mo>|</mo> <mo>-</mo> <mn>0.5</mn> </mrow> </mtd> <mtd> <mrow> <mi>o</mi> <mi>t</mi> <mi>h</mi> <mi>e</mi> <mi>r</mi> <mi>w</mi> <mi>i</mi> <mi>s</mi> <mi>e</mi> <mo>,</mo> </mrow> </mtd> </mtr> </mtable> </mfenced> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>4</mn> <mo>)</mo> </mrow> </mrow>In formula, smoothL1For smooth L1Loss function, x are variable；Faster R-CNN networks, characteristic pattern is obtained after depth convolutional neural networks in input picture, according to characteristic pattern with And RPN networks can then obtain corresponding RoIs, finally then pass through RoI ponds layer；Wherein RoI, i.e. area-of-interest, refer to just It is vehicle target object, car plate destination object；For Faster R-CNN networks, input is N number of Feature Mapping and R RoI；N number of Feature Mapping comes from last Convolutional layer, the size of each Feature Mapping is w × h × c；Each RoI is a tuple (n, r, c, h, w), wherein, n is the index of Feature Mapping, n ∈ (0,1,2 ..., N-1), r, C is top left co-ordinate, and h, w are height and width respectively；Export the Feature Mapping then obtained by maximum pond；RoI in artwork is mapped with the block in characteristic pattern；By feature Figure down-sampling is fixed size, is then passed to full connection again.
- 3. the vehicle color identifying system as claimed in claim 2 based on multitask depth convolutional neural networks, its feature exist In：The selective search network and Fast R-CNN are all stand-alone trainings, using 4 step training algorithms, pass through alternative optimization To learn shared feature；The first step, according to above-mentioned training RPN, the model initialization of network ImageNet pre-training, and hold It is used for region to end fine setting and suggests task；Second step, the Suggestion box generated using the RPN of the first step, is trained by Fast R-CNN One single detection network, this detection network are equally by the model initialization of ImageNet pre-training at this time two Individual network is also without shared convolutional layer；3rd step, trained with detection netinit RPN, but fixed shared convolutional layer, and Only layer exclusive fine setting RPN, present two network share convolutional layers；4th step, keep shared convolutional layer to fix, finely tune Fast R-CNN fc, i.e., full articulamentum；So, two network share identical convolutional layers, a unified network is formed；By the processing of above-mentioned two network, vehicle target object in a frame video image, car plate destination object are detected simultaneously Their size and locus are confined, that is, obtained vehicle target object, the size of car plate destination object and sky Between position；Its rv,cvIt is the top left co-ordinate of vehicle target object in the picture, hv,wvIt is vehicle target object respectively in image The projected size of plane, i.e. height and width；Its rp,cpIt is the top left co-ordinate of car plate in the picture, hp,wpIt is that car plate is being schemed respectively The projected size of image plane, i.e. height and width；The progressive cascade connection between each task is make use of in Faster R-CNN networks, that is, is followed successively by vehicle precise positioning, car Type, brand and the identification of car system, car plate precise positioning, Car license recognition and color recognizing for vehicle id, acetes chinensis, vehicle color correction, Colour recognition.
- 4. the vehicle color identification method based on multitask depth convolutional neural networks as described in one of claims 1 to 3, its It is characterised by：Described car plate background color identification module is used to handle to obtain the car plate back of the body under existing environmental condition to license plate image Scape color；License plate image is subjected to grey level histogram processing, is character and word in car plate at the peak valley in grey level histogram The interval of symbol, that is, the background color of car plate；Pixel RGB color component in interval is averaged, finally gives existing ring Car plate background color under the conditions of border.
- 5. the vehicle color identification method as claimed in claim 4 based on multitask depth convolutional neural networks, its feature exist In：Described Colorimetry module is used for the car plate background color of national regulations and the car plate background color under existing environmental condition Calculating is compared, obtains the aberration under existing environmental condition；First, national standard is compared with the car plate background color under existing environmental condition The several types of defined car plate background color, obtain a kind of car plate background color of immediate national regulations, and as Car plate background color under standard light；The calculating of aberration is carried out on CIE1976Lab color spaces；In order to quickly realize From RGB color to the conversion of Lab color spaces, using rapid translating mode, as shown in formula (5)；<mrow> <mfenced open = "{" close = ""> <mtable> <mtr> <mtd> <mrow> <mi>L</mi> <mo>=</mo> <mi>Y</mi> <mn>1</mn> <mo>=</mo> <mn>0.2126</mn> <mo>&times;</mo> <mi>R</mi> <mo>+</mo> <mn>0.7152</mn> <mo>&times;</mo> <mi>G</mi> <mo>+</mo> <mn>0.0722</mn> <mo>&times;</mo> <mi>B</mi> </mrow> </mtd> </mtr> <mtr> <mtd> <mrow> <mi>a</mi> <mo>=</mo> <mn>1.4749</mn> <mo>&times;</mo> <mrow> <mo>(</mo> <mn>0.2213</mn> <mo>&times;</mo> <mi>R</mi> <mo>-</mo> <mn>0.3390</mn> <mo>&times;</mo> <mi>G</mi> <mo>+</mo> <mn>0.1177</mn> <mo>&times;</mo> <mi>B</mi> <mo>)</mo> </mrow> <mo>+</mo> <mn>128</mn> </mrow> </mtd> </mtr> <mtr> <mtd> <mrow> <mi>b</mi> <mo>=</mo> <mn>0.6245</mn> <mo>&times;</mo> <mrow> <mo>(</mo> <mn>0.1949</mn> <mo>&times;</mo> <mi>R</mi> <mo>+</mo> <mn>0.6057</mn> <mo>&times;</mo> <mi>G</mi> <mo>-</mo> <mn>0.8006</mn> <mo>&times;</mo> <mi>B</mi> <mo>)</mo> </mrow> <mo>+</mo> <mn>128</mn> </mrow> </mtd> </mtr> </mtable> </mfenced> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>5</mn> <mo>)</mo> </mrow> </mrow>In formula, R, G, B are respectively the color component in RGB color, and L is the lightness component of CIE1976Lab color spaces, a With the chromatic component that b is CIE1976Lab color spaces；Car plate background color under the car plate background color of national regulations and existing environmental condition is all calculated by formula (5) Respective L, a, b value；Wherein, LNPAnd LRPCar plate respectively under the car plate background color of national regulations and existing environmental condition The brightness value of background color, aNPAnd aRP、bNPAnd bRPCar respectively under the car plate background color of national regulations and existing environmental condition The colourity of board background color, the aberration Δ E between bothabCIE1976Lab aberration is calculated with formula (6)；<mrow> <msub> <mi>&Delta;E</mi> <mrow> <mi>a</mi> <mi>b</mi> </mrow> </msub> <mo>=</mo> <msqrt> <mrow> <msup> <mi>&Delta;L</mi> <mn>2</mn> </msup> <mo>+</mo> <msup> <mi>&Delta;a</mi> <mn>2</mn> </msup> <mo>+</mo> <msup> <mi>&Delta;b</mi> <mn>2</mn> </msup> </mrow> </msqrt> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>6</mn> <mo>)</mo> </mrow> </mrow>In formula, Δ L=LNP-LRPFor luminosity equation, Δ a=aNP-aRP, Δ b=bNP-bRPFor colour difference, Δ EabFor aberration, unit is NBS。
- 6. the vehicle color identifying system as claimed in claim 5 based on multitask depth convolutional neural networks, it is characterised in that： Described vehicle color correction module is used to be corrected the vehicle color under existing environmental condition according to the aberration that detection obtains, and obtains Vehicle color image under to national regulations ecotopia；First, it is determined that aberration Δ EabWhether threshold value is exceeded Such as Fruit just carries out vehicle color correction more than threshold value, and correction is calculated as shown in formula (7)；<mrow> <mfenced open = "{" close = ""> <mtable> <mtr> <mtd> <mrow> <msub> <mi>L</mi> <mrow> <mi>N</mi> <mi>M</mi> </mrow> </msub> <mo>=</mo> <msub> <mi>L</mi> <mrow> <mi>R</mi> <mi>P</mi> </mrow> </msub> <mo>+</mo> <mi>&Delta;</mi> <mi>L</mi> </mrow> </mtd> </mtr> <mtr> <mtd> <mrow> <msub> <mi>a</mi> <mrow> <mi>N</mi> <mi>M</mi> </mrow> </msub> <mo>=</mo> <msub> <mi>a</mi> <mrow> <mi>R</mi> <mi>P</mi> </mrow> </msub> <mo>+</mo> <mi>&Delta;</mi> <mi>a</mi> </mrow> </mtd> </mtr> <mtr> <mtd> <mrow> <msub> <mi>b</mi> <mrow> <mi>N</mi> <mi>M</mi> </mrow> </msub> <mo>=</mo> <msub> <mi>b</mi> <mrow> <mi>R</mi> <mi>P</mi> </mrow> </msub> <mo>+</mo> <mi>&Delta;</mi> <mi>b</mi> </mrow> </mtd> </mtr> </mtable> </mfenced> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>7</mn> <mo>)</mo> </mrow> </mrow>In formula, LNMFor the brightness value of the vehicle color under national regulations ecotopia, LRPFor the vehicle color under existing environmental condition Brightness value, Δ L is the car plate background color of national regulations and the luminosity equation of the car plate background color under existing environmental condition, aNMWith bNMFor the chromatic value of the vehicle color under national regulations ecotopia, aRPAnd bRPThe colourity of vehicle color under existing environmental condition The colour difference of the car plate background color of value, Δ a and Δ b national regulations and the car plate background color under existing environmental condition；Further, inverse transformation is carried out from Lab color space to RGB color to the vehicle color after correction, such as formula (8) institute Show；In formula, R, G, B are respectively the color component in RGB color, and L is the lightness component of CIE1976Lab color spaces, a With the chromatic component that b is CIE1976Lab color spaces；Formula (8) equation group is the formula after optimization, and floating-point operation is converted into normal multiplication of integers and the mode of displacement, public Displacement is written as div2^23 in formula, expression moves right 23；The span of RGB and Lab in formula are all [0,255], then are passed through Cross the rgb value of the vehicle color obtained against Gamma functions under national regulations ecotopia.
- 7. the vehicle color identification method based on multitask depth convolutional neural networks as described in one of claims 1 to 3, its It is characterised by：Described vehicle color identification module is used to the vehicle color after correction be identified, in order to effectively shared Faster R-CNN depth convolutional neural networks, the vehicle image after color correction is put on into corresponding color label and instructed Practice；When being identified to vehicle color, by vehicle precise positioning, car plate precise positioning, the identification of car plate background color, acetes chinensis, After vehicle color correction process step, vehicle color identification is carried out finally by Faster R-CNN depth convolutional neural networks； Identified with the vehicle image after color correction by Faster R-CNN depth convolutional neural networks with regard to that can obtain in standard illumination Under the conditions of vehicle color.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710558817.4A CN107729801B (en) | 2017-07-11 | 2017-07-11 | Vehicle color recognition system based on multitask deep convolution neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710558817.4A CN107729801B (en) | 2017-07-11 | 2017-07-11 | Vehicle color recognition system based on multitask deep convolution neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107729801A true CN107729801A (en) | 2018-02-23 |
| CN107729801B CN107729801B (en) | 2020-12-18 |
Family
ID=61201088
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710558817.4A Active CN107729801B (en) | 2017-07-11 | 2017-07-11 | Vehicle color recognition system based on multitask deep convolution neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107729801B (en) |
Cited By (35)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108334955A (en) * | 2018-03-01 | 2018-07-27 | 福州大学 | Copy of ID Card detection method based on Faster-RCNN |
| CN108389220A (en) * | 2018-02-05 | 2018-08-10 | 湖南航升卫星科技有限公司 | Remote sensing video image motion target real-time intelligent cognitive method and its device |
| CN108399386A (en) * | 2018-02-26 | 2018-08-14 | 阿博茨德（北京）科技有限公司 | Information extracting method in pie chart and device |
| CN108416377A (en) * | 2018-02-26 | 2018-08-17 | 阿博茨德（北京）科技有限公司 | Information extracting method in block diagram and device |
| CN108509978A (en) * | 2018-02-28 | 2018-09-07 | 中南大学 | The multi-class targets detection method and model of multi-stage characteristics fusion based on CNN |
| CN108564088A (en) * | 2018-04-17 | 2018-09-21 | 广东工业大学 | Licence plate recognition method, device, equipment and readable storage medium storing program for executing |
| CN108615049A (en) * | 2018-04-09 | 2018-10-02 | 华中科技大学 | A kind of vehicle part detection model compression method and system |
| CN108734219A (en) * | 2018-05-23 | 2018-11-02 | 北京航空航天大学 | A kind of detection of end-to-end impact crater and recognition methods based on full convolutional neural networks structure |
| CN108830908A (en) * | 2018-06-15 | 2018-11-16 | 天津大学 | A kind of magic square color identification method based on artificial neural network |
| CN108846343A (en) * | 2018-06-05 | 2018-11-20 | 北京邮电大学 | Multi-task collaborative analysis method based on three-dimensional video |
| CN109145798A (en) * | 2018-08-13 | 2019-01-04 | 浙江零跑科技有限公司 | A kind of Driving Scene target identification and travelable region segmentation integrated approach |
| CN109145759A (en) * | 2018-07-25 | 2019-01-04 | 腾讯科技（深圳）有限公司 | Vehicle attribute recognition methods, device, server and storage medium |
| CN109241321A (en) * | 2018-07-19 | 2019-01-18 | 杭州电子科技大学 | The image and model conjoint analysis method adapted to based on depth field |
| CN109447064A (en) * | 2018-10-09 | 2019-03-08 | 温州大学 | A kind of duplicate rows License Plate Segmentation method and system based on CNN |
| CN109543753A (en) * | 2018-11-23 | 2019-03-29 | 中山大学 | Licence plate recognition method based on adaptive fuzzy repair mechanism |
| CN109583305A (en) * | 2018-10-30 | 2019-04-05 | 南昌大学 | A kind of advanced method that the vehicle based on critical component identification and fine grit classification identifies again |
| CN109657590A (en) * | 2018-12-11 | 2019-04-19 | 合刃科技（武汉）有限公司 | A kind of method, apparatus and storage medium detecting information of vehicles |
| CN109961057A (en) * | 2019-04-03 | 2019-07-02 | 罗克佳华科技集团股份有限公司 | A kind of vehicle location preparation method and device |
| CN110047102A (en) * | 2019-04-18 | 2019-07-23 | 北京字节跳动网络技术有限公司 | Methods, devices and systems for output information |
| CN110084190A (en) * | 2019-04-25 | 2019-08-02 | 南开大学 | Unstructured road detection method in real time under a kind of violent light environment based on ANN |
| CN110222604A (en) * | 2019-05-23 | 2019-09-10 | 复钧智能科技(苏州)有限公司 | Target identification method and device based on shared convolutional neural networks |
| CN110334709A (en) * | 2019-07-09 | 2019-10-15 | 西北工业大学 | Detection method of license plate based on end-to-end multitask deep learning |
| CN110738225A (en) * | 2018-07-19 | 2020-01-31 | 杭州海康威视数字技术股份有限公司 | Image recognition method and device |
| CN110751155A (en) * | 2019-10-14 | 2020-02-04 | 西北工业大学 | Novel target detection method based on Faster R-CNN |
| CN111126515A (en) * | 2020-03-30 | 2020-05-08 | 腾讯科技（深圳）有限公司 | Model training method based on artificial intelligence and related device |
| CN111444911A (en) * | 2019-12-13 | 2020-07-24 | 珠海大横琴科技发展有限公司 | Training method and device of license plate recognition model and license plate recognition method and device |
| CN111507210A (en) * | 2020-03-31 | 2020-08-07 | 华为技术有限公司 | Traffic signal lamp identification method and system, computing device and intelligent vehicle |
| CN111767915A (en) * | 2019-04-02 | 2020-10-13 | 顺丰科技有限公司 | License plate detection method, device, equipment and storage medium |
| CN111860539A (en) * | 2020-07-20 | 2020-10-30 | 济南博观智能科技有限公司 | License plate color recognition method, device and medium |
| CN112215245A (en) * | 2020-11-05 | 2021-01-12 | 中国联合网络通信集团有限公司 | Image identification method and device |
| CN112507801A (en) * | 2020-11-14 | 2021-03-16 | 武汉中海庭数据技术有限公司 | Lane road surface digital color recognition method, speed limit information recognition method and system |
| CN113239836A (en) * | 2021-05-20 | 2021-08-10 | 广州广电运通金融电子股份有限公司 | Vehicle body color identification method, storage medium and terminal |
| CN113673467A (en) * | 2021-08-30 | 2021-11-19 | 武汉长江通信智联技术有限公司 | Vehicle color identification method under white light condition |
| CN113723408A (en) * | 2021-11-02 | 2021-11-30 | 上海仙工智能科技有限公司 | License plate recognition method and system and readable storage medium |
| CN114240929A (en) * | 2021-12-28 | 2022-03-25 | 季华实验室 | Color difference detection method and device |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104134067A (en) * | 2014-07-07 | 2014-11-05 | 河海大学常州校区 | Road vehicle monitoring system based on intelligent visual Internet of Things |
| CN105046196A (en) * | 2015-06-11 | 2015-11-11 | 西安电子科技大学 | Front vehicle information structured output method base on concatenated convolutional neural networks |
| CN105354572A (en) * | 2015-12-10 | 2016-02-24 | 苏州大学 | Automatic identification system of number plate on the basis of simplified convolutional neural network |
-
2017
- 2017-07-11 CN CN201710558817.4A patent/CN107729801B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104134067A (en) * | 2014-07-07 | 2014-11-05 | 河海大学常州校区 | Road vehicle monitoring system based on intelligent visual Internet of Things |
| CN105046196A (en) * | 2015-06-11 | 2015-11-11 | 西安电子科技大学 | Front vehicle information structured output method base on concatenated convolutional neural networks |
| CN105354572A (en) * | 2015-12-10 | 2016-02-24 | 苏州大学 | Automatic identification system of number plate on the basis of simplified convolutional neural network |
Non-Patent Citations (3)
| Title |
|---|
| ITEYE_9281: "《从RGB到Lab色彩空间的转换》", 《BLOG.CSDN.NET/ITEYE_9281/ARTICLE/DETAILS/81643572》 * |
| SHAOQING REN等: "《Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks》", 《ARXIV:1506.01497V3》 * |
| 任玉涛: "《车脸主要信息识别技术研究》", 《中国优秀硕士学位论文全文数据库 信息科技Ⅱ辑》 * |
Cited By (50)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108389220A (en) * | 2018-02-05 | 2018-08-10 | 湖南航升卫星科技有限公司 | Remote sensing video image motion target real-time intelligent cognitive method and its device |
| CN108416377B (en) * | 2018-02-26 | 2021-12-10 | 阿博茨德（北京）科技有限公司 | Information extraction method and device in histogram |
| CN108399386A (en) * | 2018-02-26 | 2018-08-14 | 阿博茨德（北京）科技有限公司 | Information extracting method in pie chart and device |
| CN108416377A (en) * | 2018-02-26 | 2018-08-17 | 阿博茨德（北京）科技有限公司 | Information extracting method in block diagram and device |
| CN108399386B (en) * | 2018-02-26 | 2022-02-08 | 阿博茨德（北京）科技有限公司 | Method and device for extracting information in pie chart |
| CN108509978A (en) * | 2018-02-28 | 2018-09-07 | 中南大学 | The multi-class targets detection method and model of multi-stage characteristics fusion based on CNN |
| CN108509978B (en) * | 2018-02-28 | 2022-06-07 | 中南大学 | Multi-class target detection method and model based on CNN (CNN) multi-level feature fusion |
| CN108334955A (en) * | 2018-03-01 | 2018-07-27 | 福州大学 | Copy of ID Card detection method based on Faster-RCNN |
| CN108615049A (en) * | 2018-04-09 | 2018-10-02 | 华中科技大学 | A kind of vehicle part detection model compression method and system |
| CN108564088A (en) * | 2018-04-17 | 2018-09-21 | 广东工业大学 | Licence plate recognition method, device, equipment and readable storage medium storing program for executing |
| CN108734219B (en) * | 2018-05-23 | 2022-02-01 | 北京航空航天大学 | End-to-end collision pit detection and identification method based on full convolution neural network structure |
| CN108734219A (en) * | 2018-05-23 | 2018-11-02 | 北京航空航天大学 | A kind of detection of end-to-end impact crater and recognition methods based on full convolutional neural networks structure |
| CN108846343A (en) * | 2018-06-05 | 2018-11-20 | 北京邮电大学 | Multi-task collaborative analysis method based on three-dimensional video |
| CN108830908A (en) * | 2018-06-15 | 2018-11-16 | 天津大学 | A kind of magic square color identification method based on artificial neural network |
| CN110738225A (en) * | 2018-07-19 | 2020-01-31 | 杭州海康威视数字技术股份有限公司 | Image recognition method and device |
| CN109241321A (en) * | 2018-07-19 | 2019-01-18 | 杭州电子科技大学 | The image and model conjoint analysis method adapted to based on depth field |
| CN109145759A (en) * | 2018-07-25 | 2019-01-04 | 腾讯科技（深圳）有限公司 | Vehicle attribute recognition methods, device, server and storage medium |
| CN109145798B (en) * | 2018-08-13 | 2021-10-22 | 浙江零跑科技股份有限公司 | Driving scene target identification and travelable region segmentation integration method |
| CN109145798A (en) * | 2018-08-13 | 2019-01-04 | 浙江零跑科技有限公司 | A kind of Driving Scene target identification and travelable region segmentation integrated approach |
| CN109447064B (en) * | 2018-10-09 | 2019-07-30 | 温州大学 | A kind of duplicate rows License Plate Segmentation method and system based on CNN |
| CN109447064A (en) * | 2018-10-09 | 2019-03-08 | 温州大学 | A kind of duplicate rows License Plate Segmentation method and system based on CNN |
| CN109583305A (en) * | 2018-10-30 | 2019-04-05 | 南昌大学 | A kind of advanced method that the vehicle based on critical component identification and fine grit classification identifies again |
| CN109583305B (en) * | 2018-10-30 | 2022-05-20 | 南昌大学 | Advanced vehicle re-identification method based on key component identification and fine-grained classification |
| CN109543753B (en) * | 2018-11-23 | 2024-01-05 | 中山大学 | License plate recognition method based on self-adaptive fuzzy repair mechanism |
| CN109543753A (en) * | 2018-11-23 | 2019-03-29 | 中山大学 | Licence plate recognition method based on adaptive fuzzy repair mechanism |
| CN109657590A (en) * | 2018-12-11 | 2019-04-19 | 合刃科技（武汉）有限公司 | A kind of method, apparatus and storage medium detecting information of vehicles |
| CN111767915A (en) * | 2019-04-02 | 2020-10-13 | 顺丰科技有限公司 | License plate detection method, device, equipment and storage medium |
| CN109961057A (en) * | 2019-04-03 | 2019-07-02 | 罗克佳华科技集团股份有限公司 | A kind of vehicle location preparation method and device |
| CN109961057B (en) * | 2019-04-03 | 2021-09-03 | 罗克佳华科技集团股份有限公司 | Vehicle position obtaining method and device |
| CN110047102A (en) * | 2019-04-18 | 2019-07-23 | 北京字节跳动网络技术有限公司 | Methods, devices and systems for output information |
| CN110084190B (en) * | 2019-04-25 | 2024-02-06 | 南开大学 | Real-time unstructured road detection method under severe illumination environment based on ANN |
| CN110084190A (en) * | 2019-04-25 | 2019-08-02 | 南开大学 | Unstructured road detection method in real time under a kind of violent light environment based on ANN |
| CN110222604B (en) * | 2019-05-23 | 2023-07-28 | 复钧智能科技(苏州)有限公司 | Target identification method and device based on shared convolutional neural network |
| CN110222604A (en) * | 2019-05-23 | 2019-09-10 | 复钧智能科技(苏州)有限公司 | Target identification method and device based on shared convolutional neural networks |
| CN110334709A (en) * | 2019-07-09 | 2019-10-15 | 西北工业大学 | Detection method of license plate based on end-to-end multitask deep learning |
| CN110751155A (en) * | 2019-10-14 | 2020-02-04 | 西北工业大学 | Novel target detection method based on Faster R-CNN |
| CN111444911A (en) * | 2019-12-13 | 2020-07-24 | 珠海大横琴科技发展有限公司 | Training method and device of license plate recognition model and license plate recognition method and device |
| CN111444911B (en) * | 2019-12-13 | 2021-02-26 | 珠海大横琴科技发展有限公司 | Training method and device of license plate recognition model and license plate recognition method and device |
| CN111126515A (en) * | 2020-03-30 | 2020-05-08 | 腾讯科技（深圳）有限公司 | Model training method based on artificial intelligence and related device |
| CN111507210B (en) * | 2020-03-31 | 2023-11-21 | 华为技术有限公司 | Traffic signal lamp identification method, system, computing equipment and intelligent vehicle |
| CN111507210A (en) * | 2020-03-31 | 2020-08-07 | 华为技术有限公司 | Traffic signal lamp identification method and system, computing device and intelligent vehicle |
| CN111860539A (en) * | 2020-07-20 | 2020-10-30 | 济南博观智能科技有限公司 | License plate color recognition method, device and medium |
| CN111860539B (en) * | 2020-07-20 | 2024-05-10 | 济南博观智能科技有限公司 | License plate color recognition method, device and medium |
| CN112215245A (en) * | 2020-11-05 | 2021-01-12 | 中国联合网络通信集团有限公司 | Image identification method and device |
| CN112507801A (en) * | 2020-11-14 | 2021-03-16 | 武汉中海庭数据技术有限公司 | Lane road surface digital color recognition method, speed limit information recognition method and system |
| CN113239836A (en) * | 2021-05-20 | 2021-08-10 | 广州广电运通金融电子股份有限公司 | Vehicle body color identification method, storage medium and terminal |
| CN113673467A (en) * | 2021-08-30 | 2021-11-19 | 武汉长江通信智联技术有限公司 | Vehicle color identification method under white light condition |
| CN113723408B (en) * | 2021-11-02 | 2022-02-25 | 上海仙工智能科技有限公司 | License plate recognition method and system and readable storage medium |
| CN113723408A (en) * | 2021-11-02 | 2021-11-30 | 上海仙工智能科技有限公司 | License plate recognition method and system and readable storage medium |
| CN114240929A (en) * | 2021-12-28 | 2022-03-25 | 季华实验室 | Color difference detection method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107729801B (en) | 2020-12-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107729801A (en) | A kind of vehicle color identifying system based on multitask depth convolutional neural networks | |
| Björklund et al. | Robust license plate recognition using neural networks trained on synthetic images | |
| CN106599773B (en) | Deep learning image identification method and system for intelligent driving and terminal equipment | |
| CN103824081B (en) | Method for detecting rapid robustness traffic signs on outdoor bad illumination condition | |
| CN109190444B (en) | Method for realizing video-based toll lane vehicle feature recognition system | |
| CN108108761A (en) | A kind of rapid transit signal lamp detection method based on depth characteristic study | |
| CN109344825A (en) | A kind of licence plate recognition method based on convolutional neural networks | |
| CN106326858A (en) | Road traffic sign automatic identification and management system based on deep learning | |
| CN108171112A (en) | Vehicle identification and tracking based on convolutional neural networks | |
| CN106845487A (en) | A kind of licence plate recognition method end to end | |
| CN105005989B (en) | A kind of vehicle target dividing method under weak contrast | |
| CN104517103A (en) | Traffic sign classification method based on deep neural network | |
| CN106022232A (en) | License plate detection method based on deep learning | |
| CN105809138A (en) | Road warning mark detection and recognition method based on block recognition | |
| CN105354568A (en) | Convolutional neural network based vehicle logo identification method | |
| CN108647700A (en) | Multitask vehicle part identification model based on deep learning, method and system | |
| CN105809121A (en) | Multi-characteristic synergic traffic sign detection and identification method | |
| CN108304785A (en) | Road traffic sign detection based on self-built neural network and recognition methods | |
| Priese et al. | Ideogram identification in a realtime traffic sign recognition system | |
| Huang et al. | An in-car camera system for traffic sign detection and recognition | |
| CN110348396B (en) | Deep learning-based method and device for recognizing character traffic signs above roads | |
| CN112464731B (en) | Traffic sign detection and identification method based on image processing | |
| CN111860509A (en) | Coarse-to-fine two-stage non-constrained license plate region accurate extraction method | |
| Fleyeh | Traffic and road sign recognition | |
| CN109993806A (en) | A kind of color identification method, device and electronic equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CP01 | Change in the name or title of a patent holder |
Address after: 310012 1st floor, building 1, 223 Yile Road, Hangzhou City, Zhejiang Province Patentee after: Yinjiang Technology Co.,Ltd. Address before: 310012 1st floor, building 1, 223 Yile Road, Hangzhou City, Zhejiang Province Patentee before: ENJOYOR Co.,Ltd. |
|
| CP01 | Change in the name or title of a patent holder |