CN110378236A - Testing vehicle register identification model construction, recognition methods and system based on deep learning - Google Patents
Testing vehicle register identification model construction, recognition methods and system based on deep learning Download PDFInfo
- Publication number
- CN110378236A CN110378236A CN201910537407.0A CN201910537407A CN110378236A CN 110378236 A CN110378236 A CN 110378236A CN 201910537407 A CN201910537407 A CN 201910537407A CN 110378236 A CN110378236 A CN 110378236A
- Authority
- CN
- China
- Prior art keywords
- vehicle
- picture
- network
- face part
- testing
- 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
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Abstract
The invention belongs to technical field of computer vision, are related to a kind of testing vehicle register identification model construction, recognition methods and system based on deep learning.Firstly, the present invention carries out the model training of vehicle detection using extensive road monitoring picture, training uses more loss functions joint training strategy stage by stage.Then, component extraction is carried out to the vehicle face image detected, and situation is extracted according to vehicle face part and is classified using feature extraction and converged network or general category network.Finally, the identity characteristic vector of vehicle face is extracted and filtered using multitask network, the feature vector for being analysed to image in characteristics of image and information of vehicles library carries out similarity measurement, obtains testing vehicle register identification result.Deep learning network frame proposed by the present invention can be directed to demand, promote the ability in feature extraction in different aspect of network model, to realize optimal model formulation ability, facilitate the recognition feature vector for extracting and having significant discrimination.
Description
Technical field
The invention belongs to technical field of computer vision, more specifically, are related to the vehicle identification based on deep learning and know
Other model construction, recognition methods and system.
Background technique
In recent years, it had obtained redefining the development with matter based on neural network, the theory on computer vision of machine learning,
Not strong enough the traditional images feature extraction scheme of hand-designed effect limitation, robustness is breached, is taken out by multilayer convolution pondization
It takes high-dimensional feature to be fitted a large amount of training datas, returns out the featured aspects for being most suitable for solving the problems, such as.Thus computer
Visual field emerges large quantities of high-efficient, strong applicability, the application example being easily adapted, and Car license recognition is exactly wherein to develop more
The mature higher example of application value, the vehicle of the violation of fines being widely used at traffic block port video monitoring and enterprise garage
Screening etc..But license plate is easy to be altered by law-breaker, block, dismantles, forges as vehicle appearance important feature, therefore we
Identification to license plate cannot be depended only on to the identification of vehicle.
The AI service for having had already appeared detection vehicle, vehicle color at present, proposes the feature of other emphasis shapes of vehicle
Take the important evidence that can be used as vehicle identification.But for the vehicle of same color and vehicle model, investigate on a large scale,
Screening and identification, almost without other distinguishing characteristics other than license plate.The individual of vehicle that can not be tampered to certain shapes
Identification is carried out, as again the vehicle in the similar profile vehicle in magnanimity target screens and identifies, unlicensed or fake license plate vehicle
Identification etc..
Summary of the invention
It is insufficient for the fine granularity existing in the prior art extracted to vehicle characteristics, to testing vehicle register identification inaccuracy
Problem, the invention proposes a kind of testing vehicle register identification model construction, recognition methods and system based on deep learning, using such as
Lower technical solution is realized:
Testing vehicle register identification model building method based on deep learning, includes the following steps:
Step 1: acquiring multiple scene pictures, scene picture is pre-processed, obtain pretreatment pictures；
Step 2: be labeled to obtain tally set to pretreatment pictures, the label of the tally set include vehicle, vehicle face,
Vehicle and vehicle identity information；
Step 3: utilizing pretreatment pictures and tally set training network；
The network includes vehicle detection network, vehicle face part extraction network, vehicle classification network and testing vehicle register identification
Network；
The vehicle detection network is for exporting vehicle target picture；
Vehicle face part extracts network and is used to export vehicle face part picture using vehicle target picture；
The vehicle classification network is used to export vehicle classification result using vehicle target picture and vehicle face part picture；
The testing vehicle register identification network is used to obtain same type vehicle face part picture according to vehicle classification result, then sharp
Vehicle identification feature vector is exported with same type vehicle face part picture；
Obtain testing vehicle register identification model.
Further, the preparation method of the vehicle detection network are as follows:
Vehicle in every picture of pretreatment pictures is labeled and is pre-processed, vehicle detection data are obtained
Collection obtains vehicle detection network using vehicle detection data set training Mobilenet-SSD network.
Further, the preparation method of the vehicle classification network includes the following steps:
Step a: being labeled the vehicle face in vehicle target picture, obtains vehicle face part detection data collection, utilizes vehicle face
Component detection data collection training FPN-SSD network obtains vehicle face part and extracts network, exports vehicle face part picture；
Step b: the vehicle for including to vehicle target picture and vehicle face part picture is labeled, and obtains vehicle classification number
According to collection, multichannel convolutional neural networks are established, obtain vehicle identification using vehicle classification data set training multichannel convolutional neural networks
Identify network；
Wherein, the multichannel convolutional neural networks include adding two layers of full connection for five layers of convolutional layer of vehicle target picture
Five tunnel independence convolutional layers of floor and three-layer coil product and two layers of fully-connected network for four Zhong Che face part pictures.
Further, the preparation method of the testing vehicle register identification network are as follows:
According to vehicle classification as a result, being labeled to the vehicle identity information in the vehicle face part picture of vehicle of the same race, obtain
Chinese herbaceous peony windshield image segmentation data set training multitask convolution is utilized to Chinese herbaceous peony windshield image segmentation data set
Neural network Mask R-CNN, obtains testing vehicle register identification network.
Testing vehicle register identification method based on deep learning, includes the following steps:
Step 1: acquiring the vehicle identification feature vector and corresponding vehicle label, vehicle figure of the output of multiple scene pictures
Piece establishes information of vehicles library；
Step 2: obtaining picture to be identified, utilize the testing vehicle register identification model building method stage by stage based on deep learning
In vehicle detection network obtain the vehicle target picture of picture to be identified；
Step 3: being mentioned by the vehicle face part in the model building method of testing vehicle register identification stage by stage based on deep learning
It takes network to obtain vehicle face part picture, the vehicle face part picture of picture to be identified is inputted into the vehicle stage by stage based on deep learning
Vehicle classification network in identification model building method obtains the vehicle classification result of picture to be identified；
Step 4: being known by the vehicle identification in the model building method of testing vehicle register identification stage by stage based on deep learning
Other network obtains the vehicle identification feature vector of images to be recognized；
Step 5: the information of vehicles library that the vehicle identification feature vector of images to be recognized and step 1 are obtained carries out similarity
Testing vehicle register identification is completed in matching.
Testing vehicle register identification model construction based on deep learning, identifying system, including vehicle face detection module, vehicle classification
Module and testing vehicle register identification module；
The vehicle face detection module is used to mark vehicle using pretreated scene picture, exports vehicle target picture；
The vehicle classification module includes vehicle face part extracting sub-module and vehicle classification submodule, and vehicle face part mentions
It takes vehicle target picture of the submodule for obtaining using vehicle face detection module to mark vehicle face and exports vehicle face part picture；It is described
Vehicle classification submodule is used for using vehicle target picture and vehicle face part picture mark vehicle model information and exports vehicle classification
As a result；
The testing vehicle register identification module is used for according to vehicle classification as a result, in the vehicle face part picture of vehicle of the same race
Vehicle identity information is labeled, and is exported vehicle identification feature vector, picture to be identified is obtained, according to the vehicle based on deep learning
Personal identification method completes testing vehicle register identification.
Further, the vehicle face detection module by the vehicle in every picture to pretreated scene picture into
Rower is infused and is pre-processed, and vehicle detection data set is obtained, and utilizes vehicle detection data set training Mobilenet-SSD network
After obtain.
Further, the vehicle face part extracting sub-module is obtained by being labeled to the vehicle face in vehicle target picture
To vehicle face part detection data collection, then using being obtained after vehicle face part detection data collection training FPN-SSD network.
Further, the vehicle classification submodule passes through the vehicle that includes to vehicle target picture and vehicle face part picture
Type is labeled, and obtains vehicle classification data set, establishes multichannel convolutional neural networks, utilizes vehicle classification data set training multichannel
It is obtained after convolutional neural networks, wherein the multichannel convolutional neural networks include five layers of convolutional layer for vehicle target picture
Add the five tunnel independence convolution of two layers of full articulamentum with three-layer coil product and two layers of fully-connected network for four Zhong Che face part pictures
Layer.
Further, in the testing vehicle register identification module, using vehicle classification result to the vehicle face part of vehicle of the same race
Vehicle identity information in picture is labeled, and is obtained Chinese herbaceous peony windshield image segmentation data set and is utilized Chinese herbaceous peony windshield
Window image segmentation data set trains multitask convolutional neural networks Mask R-CNN, obtains all vehicle identification feature vectors, obtains
Picture to be identified is taken, according to the testing vehicle register identification method based on deep learning, completes testing vehicle register identification.
The present invention also has the following beneficial effects:
(1) present invention realizes that the overall situation is to greatest extent utilized and vehicle is local with merge in feature extraction stage by stage
Feature, compared with utilization local feature simple in existing method, either global characteristics realize vehicle identification, the present invention is one
A more complete extraction process and recognition methods, can be with the function of independent utility disparate modules.
(2) deep learning network frame proposed by the present invention can be directed to demand, promoted network model in different aspect
Ability in feature extraction, to realize optimal model formulation ability, facilitate extract have the identification feature of significant discrimination to
Amount.Such as it is suitble to extract the vehicle face part of multiscale target, and the extraction of the global and local feature of vehicle face and fusion can be realized
The vehicle classification network preferably utilized.The present invention is used in different phase simultaneously different depth learning network frame matches
Precision target also can more preferably realize the identification of vehicle identification.
(3) the invention proposes testing vehicle register identification, the method for having got around traditional license plate recognition and verification vehicle identification is utilized
Vehicle detection can screen out the different vehicle of vehicle appearance plus vehicle classification, by abundant in the vehicle of same disaggregated classification
Vehicle identification feature is extracted using vehicle annual test mark, interior ornaments, realizes the identification of unlicensed fake license plate vehicle.
Detailed description of the invention
Fig. 1 is the functional schematic of the testing vehicle register identification stage by stage applied to monitoring scene of the invention；
Fig. 2 is the functional framework figure of the testing vehicle register identification stage by stage applied to monitoring scene of the invention；
Fig. 3 is the classification results schematic diagram of present invention processing vehicle classification；
Fig. 4 is the structural schematic diagram that the present invention extracts vehicle identification feature vector；
Fig. 5 is the flow diagram of testing vehicle register identification of the present invention；
Fig. 6 is the identification matching result schematic diagram that the present invention handles identified vehicle image.
Specific embodiment
The following provides a specific embodiment of the present invention, it should be noted that the invention is not limited to implement in detail below
Example, all equivalent transformations made on the basis of the technical solutions of the present application each fall within protection scope of the present invention.
First the technical term occurred in this patent is explained below:
Mark: for the vision content of image, the text for reacting its content is added to image by the method for machine learning
The process of characteristic information.
Pretreatment: histogram equalization and filtering noise reduction are carried out.
Scene picture: the highway picture of the camera shooting including information of road surface, environmental information and information of vehicles.
Vehicle target picture: the picture includes coordinate, length and width and the score information of vehicle, what the score expression detected
The probability of vehicle target.
Vehicle face: from face image before the vehicle that vehicle frontal angle obtains.
Vehicle identity information: including interior pendant, window front stall decorations and vehicle annual inspection mark etc..
Vehicle face part picture: including the target components such as windshield window, side mirror, license plate, car light and logo before vehicle
Picture.
The target component of all kinds: including windshield window, side mirror, license plate, car light and logo before vehicle.
Vehicle necessary information: including unique vehicle label, vehicle pictures, the corresponding feature vector of vehicle.
Embodiment 1
Present embodiment discloses a kind of testing vehicle register identification model building method based on deep learning, including walk as follows
It is rapid:
Step 1: acquiring multiple scene pictures, scene picture is pre-processed, obtain pretreatment pictures；
Step 2: be labeled to obtain tally set to pretreatment pictures, the label of the tally set include vehicle, vehicle face,
Vehicle and vehicle identity information；
Step 3: utilizing pretreatment pictures and tally set training network；
The network includes vehicle detection network, vehicle face part extraction network, vehicle classification network and testing vehicle register identification
Network；
The vehicle detection network is for exporting vehicle target picture；
Vehicle face part extracts network and is used to export vehicle face part picture using vehicle target picture；
The vehicle classification network is used to export vehicle classification result using vehicle target picture and vehicle face part picture；
The testing vehicle register identification network is used to obtain same type vehicle face part picture according to vehicle classification result, then sharp
Vehicle identification feature vector is exported with same type vehicle face part picture；
Obtain testing vehicle register identification model.
The invention proposes a kind of the testing vehicle register identification model building method based on deep learning, existing vehicle face identification
Technology is proposed to make up the deficiency of Car license recognition, vehicle cab recognition in vehicle identification, usually has during public security investigation
Public security bayonet, the demand of the positioning screening tracking of road monitoring, due to same vehicle number are carried out to the vehicle of suspected illegal crime
Measure numerous and same car type different brands different type gap very little, and illegal traffic law such as carries out deck, blocks license plate at the side
Formula forgery or also not within minority without the vehicle of license plate.And the present invention is directed to the identification of vehicle vehicle face part, it can be by different product
Board difference style difference in appearance is more obvious, and feature is extracted in the vehicle face part for best embodying design element, becomes outside identification vehicle
One valuable help of sight and vehicle cab recognition.It has not been concerned only with the attribute information of vehicle, and and has been applied to it stage by stage
In testing vehicle register identification analytic process, the independent extraction for realizing vehicle detection, vehicle classification and vehicle identification feature,
And the building of testing vehicle register identification model from thick to thin is realized in triplicity on the basis of this.
Specifically, the preparation method of the vehicle detection network includes following sub-step:
Step 1a: vehicle is made to position, size and the classification of every picture mark vehicle target of pretreatment pictures
Detection data collection, and vehicle detection data set is divided into training set, test set, verifying collection according to the ratio of 8:1:1；
Step 1b: the vehicle detection data set training Mobilenet-SSD network obtained using step 1a obtains vehicle inspection
Survey grid network；
Step 1c: pretreated image is inputted into vehicle detection network, exports vehicle target picture.
Specifically, the preparation method of vehicle classification network includes following sub-step:
Step 2a: carrying out vehicle face mark to the vehicle target picture that vehicle detection network obtains, described to be labeled as retaining packet
Face containing vehicle comprehensively and includes most to obtain vehicle face part detection data collection without the least vehicle target picture in face part that cut-offs；
Step 2b: obtaining FPN-SSD network model with the VGG-16 network in FPN-SSD network model replacement SSD network,
Vehicle face part is obtained using vehicle face part detection data collection training FPN-SSD network and extracts network, exports vehicle face part picture；
Step 2c: carrying out vehicle mark to vehicle target picture and vehicle face part picture, obtain vehicle classification data set,
It is divided into training set, test set, verifying collection according to the ratio of 8:1:1.
Step 2d: the vehicle classification data set training establishing the convolutional neural networks structure of multichannel and being obtained using step 2,
It realizes to the vehicle face picture and vehicle face part picture independence convolution of input then fusion feature, obtains the full face Fusion Features of vehicle
Vehicle classification network exports vehicle classification result；
Specifically, in step 2d, it, will if the vehicle part picture of step 2b output does not include the target component of all kinds
The existing ResNet sorter network of vehicle classification data set training obtains vehicle classification result.
Preferably, the multichannel convolutional neural networks include adding two layers to connect entirely for five layers of convolutional layer of vehicle target picture
Connect five tunnel independence convolutional layers of floor with three-layer coil product and two layers of fully-connected network for four Zhong Che face part pictures.
Specifically, the preparation method of the testing vehicle register identification network are as follows:
According to vehicle classification as a result, being labeled to the vehicle identity information in the vehicle face part picture of vehicle of the same race, obtain
Chinese herbaceous peony windshield image segmentation data set training multitask convolution is utilized to Chinese herbaceous peony windshield image segmentation data set
Neural network Mask R-CNN, obtains testing vehicle register identification network.
Wherein, by the exposure mask that extracts characteristic pattern and obtained with exposure mask branch to full figure characteristic pattern and ROI feature figure into
Row filters, and splicing obtains output vehicle identification feature vector after regularization whitening processing.
In order to increase the weight of important component in vehicle face, local feature and global feature are combined, extracted as far as possible
The full face feature of vehicle carries out convolution with complete vehicle face picture after cutting all parts respectively, devises and has merged vehicle face spy
The vehicle cab recognition network of sign, as shown in Figure 4.Not shared parameter between component picture and the convolution of component picture, component picture
The also not shared parameter between full face picture, the reason is that different vehicle face district-share convolution kernels has been used to will lead to losing for information
It loses.Finally by the convolution results direct splicing of the convolution results of the regional area where component and the full face picture of vehicle, input two
The full articulamentum of 4096 neuron of layer and the full articulamentum of one layer of 228 neuron are merged, and are calculated finally by softmax
Method is classified to obtain the type of input vehicle face to 228 dimension results.
Embodiment 2
The testing vehicle register identification method based on deep learning that present embodiment discloses a kind of, includes the following steps:
Step 1: the vehicle identification feature vector for acquiring the output of multiple scene pictures establishes information of vehicles library；
Step 2: obtaining picture to be identified, utilize the vehicle identification stage by stage based on deep learning any in embodiment 1
Vehicle detection network in identification model construction method obtains the vehicle target picture of picture to be identified；
Step 3: being mentioned by the vehicle face part in the model building method of testing vehicle register identification stage by stage based on deep learning
It takes network to obtain vehicle face part picture, the vehicle face part picture of picture to be identified is inputted into the vehicle stage by stage based on deep learning
Vehicle classification network in identification model building method obtains the vehicle classification result of picture to be identified；As shown in figure 3, should
Picture classification result is KIA kx5.
Step 4: being known by the vehicle identification in the model building method of testing vehicle register identification stage by stage based on deep learning
Other network obtains the vehicle identification feature vector of images to be recognized；
Step 5: the information of vehicles library that the vehicle identification feature vector of images to be recognized and step 1 are obtained carries out similarity
Testing vehicle register identification is completed in matching.
Fig. 1 is the functional schematic of the testing vehicle register identification stage by stage applied to monitoring scene, the process identified stage by stage
The normal cognitive course for meeting the mankind, the various structures for taking full advantage of convolutional neural networks targetedly realize different phase
Target is suitble to the realization of vehicle identification at this stage.
The vehicle location of input picture is detected first, and vehicle classification then is carried out to vehicle face picture therein, obtains it
Concrete model realizes screening and filtering, finally divides glass window region to realize two identification of vehicle using Chinese herbaceous peony shelves, is matched to data
The vehicle information of vehicles registered in library.
The testing vehicle register identification method based on deep learning that the invention proposes a kind of, devises vehicle face on vehicle cab recognition
The method that the extraction and fusion of global and local feature are classified designs the vehicle face part for being suitble to extract multiscale target and extracts
Network and the vehicle classification network that vehicle face feature can be made full use of；It is extracted in testing vehicle register identification and believes containing vehicle identification
The feature of breath is filtered using multitask network implementations and has got more accurate vehicle identification feature vector, while this feature
The retrieval that splicing keeps the matching energy of vehicle identification progressive, improves the efficiency of testing vehicle register identification.
Information of vehicles is corresponded to specifically, being returned in information of vehicles library if successful match, returns to the vehicle if it fails to match
Not in information of vehicles library；Need to test a certain number of vehicle faces when being matched, so that each vehicle face can obtain and information
The vector distance of all vehicle faces in library, choosing a threshold value can allow the maximum value of sens [sens=TP/P] or acc maximum
Value [acc=(TP+TN)/total test sample number], as shown in Figure 5 in this implementation embodiment, threshold value 0.82.Wherein, TP
(True positive) indicates real example, i.e., by the correctly predicted class number that is positive of positive class；TN (True negative) indicates very negative
Example, i.e., by the correctly predicted class number that is negative of negative class；Acc (accuracy) indicates accuracy, is that the most common machine learning evaluation refers to
Mark, usually, accuracy is higher, and classifier is better；Sens (sensitive) indicates sensitivity, and expression is all positive examples
It is middle by point pair ratio, measured classifier to the recognition capability of positive example.
Embodiment 3
The testing vehicle register identification model construction that present embodiment discloses a kind of based on deep learning, identifying system, such as Fig. 2 institute
Show, including vehicle face detection module, vehicle classification module and testing vehicle register identification module；
The vehicle face detection module is used to mark vehicle using pretreated scene picture, exports vehicle target picture；
The module is one of big module of this system three, is the basis of subsequent module function.Vehicle in big figure will be inputted by mainly realizing
Detection and the cutting of vehicle store work, and pretrigger module therein is responsible for the reading of picture and model load is each function
Indispensable part in energy module.
The vehicle classification module includes vehicle face part extracting sub-module and vehicle classification submodule, and vehicle face part mentions
It takes vehicle target picture of the submodule for obtaining using vehicle face detection module to mark vehicle face and exports vehicle face part picture；It is described
Vehicle classification submodule is used for using vehicle target picture and vehicle face part picture mark vehicle model information and exports vehicle classification
As a result；The presence that vehicle face part saves is in order to apply in vehicle classification, and for providing front windshield of vehicle
Necessary material of the picture as testing vehicle register identification.
The testing vehicle register identification module is used for according to vehicle classification as a result, in the vehicle face part picture of vehicle of the same race
Vehicle identity information is labeled, and is exported vehicle identification feature vector, picture to be identified is obtained, according to any in embodiment 2
Testing vehicle register identification method based on deep learning completes testing vehicle register identification.The main submodule of this module has information of vehicles note
Volume and two modules of individual identification.Information of vehicles registration module function is accomplished that each vehicle corresponds to the persistence of necessary information,
Necessary information includes unique vehicle label, vehicle pictures, the corresponding feature vector of vehicle.The responsible calling depth of individual identification
It practises platform and obtains the feature vector of detection picture using trained network model, compared with registered vehicle feature in database
To return similarity and corresponding car number.
Specifically, the vehicle face detection module is carried out by the vehicle in every picture to pretreated scene picture
It marks and is pre-processed, obtain vehicle detection data set, after vehicle detection data set training Mobilenet-SSD network
It obtains.
Specifically, the vehicle face part extracting sub-module is obtained by being labeled to the vehicle face in vehicle target picture
Vehicle face part detection data collection, then using being obtained after vehicle face part detection data collection training FPN-SSD network.
Specifically, the vehicle classification submodule passes through the vehicle that includes to vehicle target picture and vehicle face part picture
It is labeled, obtains vehicle classification data set, establish multichannel convolutional neural networks, rolled up using vehicle classification data set training multichannel
It is obtained after product neural network, wherein the multichannel convolutional neural networks include adding for five layers of convolutional layer of vehicle target picture
Five tunnel independence convolutional layers of two layers of full articulamentum and three-layer coil product and two layers of fully-connected network for four Zhong Che face part pictures.
Specifically, in the testing vehicle register identification module, using vehicle classification result to the vehicle face component diagram of vehicle of the same race
Vehicle identity information in piece is labeled, and is obtained Chinese herbaceous peony windshield image segmentation data set and is utilized Chinese herbaceous peony windshield
Image segmentation data set trains multitask convolutional neural networks Mask R-CNN, obtains all vehicle identification feature vectors, obtains
Picture to be identified completes testing vehicle register identification according to the testing vehicle register identification method based on deep learning.
As shown in fig. 6, the identification matching result schematic diagram of identified vehicle image is handled in the present embodiment, from recognition result
From the point of view of, testing vehicle register identification can correctly be matched to the identity information of corresponding vehicle.
Claims (10)
1. the testing vehicle register identification model building method based on deep learning, which comprises the steps of:
Step 1: acquiring multiple scene pictures, scene picture is pre-processed, obtain pretreatment pictures；
Step 2: pretreatment pictures being labeled to obtain tally set, the label of the tally set includes vehicle, vehicle face, vehicle
And vehicle identity information；
Step 3: utilizing pretreatment pictures and tally set training network；
The network includes vehicle detection network, vehicle face part extraction network, vehicle classification network and testing vehicle register identification network；
The vehicle detection network is for exporting vehicle target picture；
Vehicle face part extracts network and is used to export vehicle face part picture using vehicle target picture；
The vehicle classification network is used to export vehicle classification result using vehicle target picture and vehicle face part picture；
The testing vehicle register identification network is used to obtain same type vehicle face part picture according to vehicle classification result, then using same
Type vehicle face part picture exports vehicle identification feature vector；
Obtain testing vehicle register identification model.
2. the testing vehicle register identification model building method based on deep learning as described in claim 1, which is characterized in that described
The preparation method of vehicle detection network are as follows:
Vehicle in every picture of pretreatment pictures is labeled and is pre-processed, vehicle detection data set is obtained,
Vehicle detection network is obtained using vehicle detection data set training Mobilenet-SSD network.
3. the testing vehicle register identification model building method based on deep learning as described in claim 1, which is characterized in that described
The preparation method of vehicle classification network includes the following steps:
Step a: being labeled the vehicle face in vehicle target picture, obtains vehicle face part detection data collection, utilizes vehicle face part
Detection data collection training FPN-SSD network obtains vehicle face part and extracts network, exports vehicle face part picture；
Step b: the vehicle for including to vehicle target picture and vehicle face part picture is labeled, and obtains vehicle classification data
Collection, establishes multichannel convolutional neural networks, obtains vehicle identification using vehicle classification data set training multichannel convolutional neural networks and knows
Other network；
Wherein, the multichannel convolutional neural networks include for vehicle target picture five layers of convolutional layer add two layers of full articulamentum with
For the long-pending five tunnel independence convolutional layers with two layers of fully-connected network of three-layer coil of four Zhong Che face part pictures.
4. the testing vehicle register identification model building method based on deep learning as described in claim 1, which is characterized in that described
The preparation method of testing vehicle register identification network are as follows:
According to vehicle classification as a result, being labeled to the vehicle identity information in the vehicle face part picture of vehicle of the same race, vehicle is obtained
Front windshield glass window image segmentation data set utilizes Chinese herbaceous peony windshield image segmentation data set training multitask convolutional Neural
Network Mask R-CNN, obtains testing vehicle register identification network.
5. the testing vehicle register identification method based on deep learning, which comprises the steps of:
Step 1: the vehicle identification feature vector and corresponding vehicle label, vehicle pictures for acquiring the output of multiple scene pictures are built
Vertical information of vehicles library；
Step 2: obtaining picture to be identified, known using any vehicle identification stage by stage based on deep learning of Claims 1-4 4
Vehicle detection network in other model building method obtains the vehicle target picture of picture to be identified；
Step 3: passing through any testing vehicle register identification model building method stage by stage based on deep learning of Claims 1-4
In vehicle face part extract network and obtain vehicle face part picture, the vehicle face part picture of picture to be identified is inputted into claim 1
Vehicle classification network into 4 any model building methods of testing vehicle register identification stage by stage based on deep learning is obtained wait know
The vehicle classification result of other picture；
Step 4: passing through any testing vehicle register identification model building method stage by stage based on deep learning of Claims 1-4
In testing vehicle register identification network obtain the vehicle identification feature vector of images to be recognized；
Step 5: the information of vehicles library that the vehicle identification feature vector of images to be recognized and step 1 are obtained carries out similarity
Match, completes testing vehicle register identification.
6. the testing vehicle register identification model construction based on deep learning, identifying system, which is characterized in that detect mould including vehicle face
Block, vehicle classification module and testing vehicle register identification module；
The vehicle face detection module is used to mark vehicle using pretreated scene picture, exports vehicle target picture；
The vehicle classification module includes vehicle face part extracting sub-module and vehicle classification submodule, and vehicle face part extracts son
The vehicle target picture that module is used to obtain using vehicle face detection module marks vehicle face and exports vehicle face part picture；The vehicle
Classification submodule is used for using vehicle target picture and vehicle face part picture mark vehicle model information and exports vehicle classification result；
The testing vehicle register identification module is used for according to vehicle classification as a result, to the vehicle in the vehicle face part picture of vehicle of the same race
Identity information is labeled, and is exported vehicle identification feature vector, picture to be identified is obtained, according to base as claimed in claim 5
In the testing vehicle register identification method of deep learning, testing vehicle register identification is completed.
7. the testing vehicle register identification model construction based on deep learning, identifying system, feature exist as claimed in claim 6
In the vehicle face detection module is labeled and is carried out pre- by the vehicle in every picture to pretreated scene picture
Processing, obtains vehicle detection data set, obtains after training Mobilenet-SSD network using vehicle detection data set.
8. the testing vehicle register identification model construction based on deep learning, identifying system, feature exist as claimed in claim 6
In the vehicle face part extracting sub-module obtains the detection of vehicle face part by being labeled to the vehicle face in vehicle target picture
Data set, then using being obtained after vehicle face part detection data collection training FPN-SSD network.
9. the testing vehicle register identification model construction based on deep learning, identifying system, feature exist as claimed in claim 6
In the vehicle classification submodule is labeled by the vehicle for including to vehicle target picture and vehicle face part picture, is obtained
To vehicle classification data set, multichannel convolutional neural networks are established, utilize vehicle classification data set training multichannel convolutional neural networks
After obtain, wherein the multichannel convolutional neural networks include adding two layers of full connection for five layers of convolutional layer of vehicle target picture
Five tunnel independence convolutional layers of floor and three-layer coil product and two layers of fully-connected network for four Zhong Che face part pictures.
10. the testing vehicle register identification model construction based on deep learning, identifying system, feature exist as claimed in claim 6
In in the testing vehicle register identification module, using vehicle classification result to the vehicle body in the vehicle face part picture of vehicle of the same race
Part information is labeled, and is obtained Chinese herbaceous peony windshield image segmentation data set and is utilized Chinese herbaceous peony windshield image segmentation data
Collect training multitask convolutional neural networks Mask R-CNN, obtains all vehicle identification feature vectors, obtain picture to be identified, press
According to the testing vehicle register identification method as claimed in claim 5 based on deep learning, testing vehicle register identification is completed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910537407.0A CN110378236B (en) | 2019-06-20 | 2019-06-20 | Vehicle identity recognition model construction and recognition method and system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910537407.0A CN110378236B (en) | 2019-06-20 | 2019-06-20 | Vehicle identity recognition model construction and recognition method and system based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110378236A true CN110378236A (en) | 2019-10-25 |
| CN110378236B CN110378236B (en) | 2023-04-07 |
Family
ID=68249055
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910537407.0A Active CN110378236B (en) | 2019-06-20 | 2019-06-20 | Vehicle identity recognition model construction and recognition method and system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110378236B (en) |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110991506A (en) * | 2019-11-22 | 2020-04-10 | 高新兴科技集团股份有限公司 | Vehicle brand identification method, device, equipment and storage medium |
| CN111079543A (en) * | 2019-11-20 | 2020-04-28 | 浙江工业大学 | Efficient vehicle color identification method based on deep learning |
| CN111126393A (en) * | 2019-12-22 | 2020-05-08 | 上海眼控科技股份有限公司 | Vehicle appearance refitting judgment method and device, computer equipment and storage medium |
| CN111126224A (en) * | 2019-12-17 | 2020-05-08 | 成都通甲优博科技有限责任公司 | Vehicle detection method and classification recognition model training method |
| CN111144372A (en) * | 2019-12-31 | 2020-05-12 | 上海眼控科技股份有限公司 | Vehicle detection method, device, computer equipment and storage medium |
| CN111310844A (en) * | 2020-02-26 | 2020-06-19 | 广州华工邦元信息技术有限公司 | Vehicle identification model construction method and device and identification method and device |
| CN111339834A (en) * | 2020-02-04 | 2020-06-26 | 浙江大华技术股份有限公司 | Method for recognizing vehicle traveling direction, computer device, and storage medium |
| CN111340796A (en) * | 2020-03-10 | 2020-06-26 | 创新奇智(成都)科技有限公司 | Defect detection method and 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 |
| CN112087420A (en) * | 2020-07-24 | 2020-12-15 | 西安电子科技大学 | Network killing chain detection method, prediction method and system |
| CN112115800A (en) * | 2020-08-25 | 2020-12-22 | 南京航空航天大学 | Vehicle combination recognition system and method based on deep learning target detection |
| CN112257869A (en) * | 2020-09-29 | 2021-01-22 | 北京北大千方科技有限公司 | Fake-licensed car analysis method and system based on random forest and computer medium |
| CN112257625A (en) * | 2020-10-29 | 2021-01-22 | 上海工程技术大学 | Vehicle weight recognition method based on vehicle front face features |
| CN112528921A (en) * | 2020-12-21 | 2021-03-19 | 山东雨润环保机械设备有限公司 | Construction site dust identification system and method based on machine vision |
| CN113298021A (en) * | 2021-06-11 | 2021-08-24 | 宿州学院 | Mining area transport vehicle head and tail identification method and system based on convolutional neural network |
| CN113780130A (en) * | 2021-08-31 | 2021-12-10 | 东南大学 | Vehicle type identification method based on magnetic field data |
| WO2022241807A1 (en) * | 2021-05-20 | 2022-11-24 | 广州广电运通金融电子股份有限公司 | Method for recognizing color of vehicle body of vehicle, and storage medium and terminal |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2993618A1 (en) * | 2014-09-04 | 2016-03-09 | Xerox Corporation | Domain adaptation for image classification with class priors |
| CN107092876A (en) * | 2017-04-12 | 2017-08-25 | 湖南源信光电科技股份有限公司 | The low-light (level) model recognizing method combined based on Retinex with S SIFT features |
| CN107729818A (en) * | 2017-09-21 | 2018-02-23 | 北京航空航天大学 | A kind of multiple features fusion vehicle recognition methods again based on deep learning |
| CN108319952A (en) * | 2017-01-16 | 2018-07-24 | 浙江宇视科技有限公司 | A kind of vehicle characteristics extracting method and device |
| CN108460328A (en) * | 2018-01-15 | 2018-08-28 | 浙江工业大学 | A kind of fake-licensed car detection method based on multitask convolutional neural networks |
| CN109033175A (en) * | 2018-06-25 | 2018-12-18 | 高新兴科技集团股份有限公司 | A kind of method and system to scheme to search vehicle |
-
2019
- 2019-06-20 CN CN201910537407.0A patent/CN110378236B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2993618A1 (en) * | 2014-09-04 | 2016-03-09 | Xerox Corporation | Domain adaptation for image classification with class priors |
| CN108319952A (en) * | 2017-01-16 | 2018-07-24 | 浙江宇视科技有限公司 | A kind of vehicle characteristics extracting method and device |
| CN107092876A (en) * | 2017-04-12 | 2017-08-25 | 湖南源信光电科技股份有限公司 | The low-light (level) model recognizing method combined based on Retinex with S SIFT features |
| CN107729818A (en) * | 2017-09-21 | 2018-02-23 | 北京航空航天大学 | A kind of multiple features fusion vehicle recognition methods again based on deep learning |
| CN108460328A (en) * | 2018-01-15 | 2018-08-28 | 浙江工业大学 | A kind of fake-licensed car detection method based on multitask convolutional neural networks |
| CN109033175A (en) * | 2018-06-25 | 2018-12-18 | 高新兴科技集团股份有限公司 | A kind of method and system to scheme to search vehicle |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111079543B (en) * | 2019-11-20 | 2022-02-15 | 浙江工业大学 | Efficient vehicle color identification method based on deep learning |
| CN111079543A (en) * | 2019-11-20 | 2020-04-28 | 浙江工业大学 | Efficient vehicle color identification method based on deep learning |
| CN110991506A (en) * | 2019-11-22 | 2020-04-10 | 高新兴科技集团股份有限公司 | Vehicle brand identification method, device, equipment and storage medium |
| CN111126224A (en) * | 2019-12-17 | 2020-05-08 | 成都通甲优博科技有限责任公司 | Vehicle detection method and classification recognition model training method |
| CN111126393A (en) * | 2019-12-22 | 2020-05-08 | 上海眼控科技股份有限公司 | Vehicle appearance refitting judgment method and device, computer equipment and storage medium |
| CN111144372A (en) * | 2019-12-31 | 2020-05-12 | 上海眼控科技股份有限公司 | Vehicle detection method, device, computer equipment and storage medium |
| CN111339834A (en) * | 2020-02-04 | 2020-06-26 | 浙江大华技术股份有限公司 | Method for recognizing vehicle traveling direction, computer device, and storage medium |
| CN111310844A (en) * | 2020-02-26 | 2020-06-19 | 广州华工邦元信息技术有限公司 | Vehicle identification model construction method and device and identification method and device |
| CN111340796A (en) * | 2020-03-10 | 2020-06-26 | 创新奇智(成都)科技有限公司 | Defect detection method and device, electronic equipment and storage medium |
| CN112087420A (en) * | 2020-07-24 | 2020-12-15 | 西安电子科技大学 | Network killing chain detection method, prediction method and system |
| CN112087420B (en) * | 2020-07-24 | 2022-06-14 | 西安电子科技大学 | Network killing chain detection method, prediction method and system |
| CN112016433A (en) * | 2020-08-24 | 2020-12-01 | 高新兴科技集团股份有限公司 | Vehicle color identification method based on deep neural network |
| CN112115800A (en) * | 2020-08-25 | 2020-12-22 | 南京航空航天大学 | Vehicle combination recognition system and method based on deep learning target detection |
| CN112257869A (en) * | 2020-09-29 | 2021-01-22 | 北京北大千方科技有限公司 | Fake-licensed car analysis method and system based on random forest and computer medium |
| CN112257625A (en) * | 2020-10-29 | 2021-01-22 | 上海工程技术大学 | Vehicle weight recognition method based on vehicle front face features |
| CN112528921A (en) * | 2020-12-21 | 2021-03-19 | 山东雨润环保机械设备有限公司 | Construction site dust identification system and method based on machine vision |
| WO2022241807A1 (en) * | 2021-05-20 | 2022-11-24 | 广州广电运通金融电子股份有限公司 | Method for recognizing color of vehicle body of vehicle, and storage medium and terminal |
| CN113298021A (en) * | 2021-06-11 | 2021-08-24 | 宿州学院 | Mining area transport vehicle head and tail identification method and system based on convolutional neural network |
| CN113780130A (en) * | 2021-08-31 | 2021-12-10 | 东南大学 | Vehicle type identification method based on magnetic field data |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110378236B (en) | 2023-04-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110378236A (en) | Testing vehicle register identification model construction, recognition methods and system based on deep learning | |
| CN107729818B (en) | Multi-feature fusion vehicle re-identification method based on deep learning | |
| CN104166841B (en) | The quick detection recognition methods of pedestrian or vehicle is specified in a kind of video surveillance network | |
| Hwang et al. | Multispectral pedestrian detection: Benchmark dataset and baseline | |
| CN105488453B (en) | A kind of driver based on image procossing does not fasten the safety belt detection recognition method | |
| CN106682586A (en) | Method for real-time lane line detection based on vision under complex lighting conditions | |
| US20090309966A1 (en) | Method of detecting moving objects | |
| CN103473571B (en) | Human detection method | |
| CN109948416A (en) | A kind of illegal occupancy bus zone automatic auditing method based on deep learning | |
| CN107085696A (en) | A kind of vehicle location and type identifier method based on bayonet socket image | |
| CN109033175A (en) | A kind of method and system to scheme to search vehicle | |
| CN102915433B (en) | Character combination-based license plate positioning and identifying method | |
| CN110119726A (en) | A kind of vehicle brand multi-angle recognition methods based on YOLOv3 model | |
| CN111860291A (en) | Multi-mode pedestrian identity recognition method and system based on pedestrian appearance and gait information | |
| CN102663411A (en) | Recognition method for target human body | |
| CN105184301B (en) | A kind of method that vehicle heading is differentiated using four-axle aircraft | |
| CN106295532A (en) | A kind of human motion recognition method in video image | |
| CN110009058A (en) | A kind of parking lot Vehicle License Plate Recognition System and method | |
| CN105893971A (en) | Traffic signal lamp recognition method based on Gabor and sparse representation | |
| CN106919939B (en) | A kind of traffic signboard tracks and identifies method and system | |
| CN111814751A (en) | Vehicle attribute analysis method and system based on deep learning target detection and image recognition | |
| CN114049572A (en) | Detection method for identifying small target | |
| CN109684986A (en) | A kind of vehicle analysis method and system based on automobile detecting following | |
| CN111444843A (en) | Multimode driver and vehicle illegal behavior monitoring method and system | |
| Srivastava et al. | Enhanced object detection with deep convolutional neural networks |
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 |