CN107403162A - The data acquisition of vehicle notification number and the system and method for classification - Google Patents
The data acquisition of vehicle notification number and the system and method for classification Download PDFInfo
- Publication number
- CN107403162A CN107403162A CN201710639834.0A CN201710639834A CN107403162A CN 107403162 A CN107403162 A CN 107403162A CN 201710639834 A CN201710639834 A CN 201710639834A CN 107403162 A CN107403162 A CN 107403162A
- Authority
- CN
- China
- Prior art keywords
- data
- notification number
- test
- view data
- module
- 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
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
- G06V20/54—Surveillance or monitoring of activities, e.g. for recognising suspicious objects of traffic, e.g. cars on the road, trains or boats
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/32—Normalisation of the pattern dimensions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/08—Detecting or categorising vehicles
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
The invention provides the system and method for a kind of vehicle notification number data acquisition and classification, including collecting part and classified part, and described classified part is connected with collecting part, wherein, described collecting part builds database to be sorted to build vehicle notification number information list to gathered data；Described classified part reclassifies the data in database are trained with acquisition disaggregated model according to notification number information in image information aspect to data, and repetitive exercise and assorting process are to build vehicle image taxonomy database.During using this kind of system, because it based on the existing information in car guard system establishes vehicle image database, and reclassified according to image information, therefore the system of the vehicle notification number data acquisition and classification can quickly establish the exploitation that vehicle image taxonomy database is used for vehicle identification product, avoid artificial data from marking.
Description
Technical field
The present invention relates to intelligent transportation field, more particularly to intelligent traffic vehicle identification field and Relational database is quick
Establish, and in particular to the system and method for a kind of vehicle notification number data acquisition and classification.
Background technology
With fast development of the deep learning in classification and test problems, its applied field in traffic and public security field
Scape increasingly diversification, application have been related to vehicle brand identification, Car license recognition, vehicle type recognition (car, goods
Car, car), colour recognition, recognition of face etc..Wherein, vehicle brand, money system and age identification function not only facilitate fake-licensed car
Examination, and the important technical just progressively developed into criminal investigation work, i.e., tie up to magnanimity according to vehicle brand and money
Suspected vehicles are searched and retrieved in video and view data, and its efficiency is significantly larger than conventional tactics of human sea.
However, in actual product R＆D process, to realize that vehicle brand, money system and age identification function can not often be kept away
Needing of exempting from faces following three outstanding problems：
(1) demand data amount is big.The common vehicle brand of in the market has hundreds of, different model vehicle under all brands
Thousands of moneys are added up.Directly affected it is well known that the quality and quantity of data set has to classification based training model, it is every kind of
The training data of vehicle will reach Zhang Caineng up to a hundred and cover its different situations under different visual angles and light.Therefore, overall number
Need to reach ten tens of thousands of even more more according to amount.
(2) data distribution is uneven.Identify in conventional vehicle brand, money system and age and found in work, pass through transportation card
The data distribution tilt phenomenon that mouth or monitor video collect is serious, is mainly reflected in following two aspects：First, same
Area, different brands data probability distributions difference are huge；Second, vehicle brand data distribution is also different between different geographical.
(3) model needs constantly to safeguard and update.Still there is new model constantly to come out, it is necessary to be done to identification model periodically every year
Renewal.The gathered data of blindness how is avoided, specific aim collection is done to the partial data lacked and is supplemented into one and needs solution badly
Certainly the problem of.
Traditional collecting method difficulty is big, efficiency is low, artificial mark cost is high, time-consuming.Analyze, need based on more than
Want a kind of vehicle brand data acquisition being easily achieved and data processing scheme.Therefore the data acquisition of vehicle notification number and classification system
System proposes a kind of method of the data acquisition based on vehicle notification number.Vehicle notification number refer to manufacturer to same type,
The title that brand, species, the vehicle of series and body model are given.Have five parts in vehicle notification number coding：Enterprise
Title code name, class of vehicle code name, major parameter code name, product serial number, enterprise make code name by oneself.Vehicle notification number includes content
Enrich and there is uniqueness.It is more directly and efficient come gathered data that index is used as by vehicle notification number.
In addition, Some vehicles bulletin number, which exists, has the problem of identical car face or tailstock morphosis, only from traffic
It can not be made a distinction in monitoring image or video.Such case is more common between the different money systems of same brand.No matter
It is in model training or service stage, to improve discrimination and Consumer's Experience just needs have identical car face or the tailstock by this kind of
Notification number merger for one kind.Obviously by the method manually marked found in thousands of kinds of car faces identical car face efficiency it is low,
Speed is slow, confidence level is not high.To solve this problem, the data acquisition of vehicle notification number and categorizing system are for vehicle bulletin number
According to the characteristics of propose a kind of vehicle notification number data automatic classification scheme, realization has identical car face or the tailstock, different notification numbers
The automatical and efficient merging of data.
The content of the invention
The purpose of the present invention be overcome it is above-mentioned in the prior art the shortcomings that, there is provided one kind be used for solve existing vehicle brand,
Data acquisition difficulty is big, redundancy is high, artificial annotating efficiency is low, different notification numbers in money system and age identification product development process
The vehicle for the problem of may having identical car face or tailstock morphosis and can not only being made a distinction from image or video information is public
The system and method for number collection and classification is accused, classifies for vehicle brand, money system and age and identifies that product establishes research and development institute
The tranining database needed.
To achieve these goals, the system and method for vehicle notification number data acquisition of the invention and classification is specific such as
Under：
The vehicle notification number data acquisition and categorizing system, it is mainly characterized by, including collecting part and classified part, and
Described classified part is connected with collecting part, wherein, described collecting part to build vehicle notification number information list,
And build database to be sorted to gathered data；Described classified part to according to notification number information to the number in database
According to being trained acquisition disaggregated model, and data are reclassified in image information aspect, repetitive exercise and assorting process
To build vehicle image taxonomy database.It is preferred that described collecting part includes：
Corresponding relation building module, for establishing vehicle notification number information list, described vehicle notification number information list
Including vehicle essential information corresponding to vehicle notification number and vehicle notification number；
Data acquisition module, for the vehicle notification number in described vehicle notification number information list in car guard system
Middle gathered data, establish the database to match with the vehicle notification number information list；
Described classified part includes：
Pretreatment module, the data distribution given for the classification information that is included according to vehicle notification number in described database
Sample label, and the data in described database are divided by preset rules, at least three data set is obtained, it is described
Pretreatment module is additionally operable to pre-process the data in described database；
Disaggregated model training module, for being trained according to classification information to described data set, obtain classification based training
Model；
Disaggregated model test module, for according to described data set, being obtained to described disaggregated model training module
Classification based training model is tested, and obtains test result；
Test result statistical module, for the test result of statistical classification model measurement module output, obtain corresponding survey
Try index；
Data combiners block, for according to the test index, judging different vehicle bulletin number correspondence image information
It is whether consistent, and the vehicle notification number corresponding data consistent to image information carries out classification merging.
More preferably, described vehicle essential information includes the brand of vehicle, money system and year corresponding to this vehicle notification number
For description information.
It is particularly preferred that described database is image data base, the view data in the image data base includes described number
View data corresponding to the vehicle notification number included in described vehicle notification number information list according to acquisition module collection.
It is still further preferred that view data include vehicle frontal by when headstock view data and the tailstock picture number sailed out of of the back side
According to.
It is still further preferred that described pretreatment module is image pre-processing module, described image pre-processing module is according to classification
Described view data is divided into 3 mutual not phases by information to described view data distribution sample label, and by preset rules
The data set of friendship, respectively training set, checking collection and test set, wherein,
Described training set obtains classification based training model to carry out model training；
Described checking collects the classification based training for the unfashioned classification based training model during adjusting training, making acquisition
Model meets preparatory condition；
Described test set is testing described classification based training model；
It is described image preprocessing mould that described pretreatment module, which carries out pretreatment to the data in described database,
Block is cut out and scaled on demand to the view data in described image data base.
Capitally, the sample label of view data of the described disaggregated model training module in described training set,
Deep learning training is carried out to described training set, obtains convolutional neural networks as classification based training model, and described classification
Model training module, by unfashioned classification based training model, obtains checking and concentrated also by the view data that described checking is concentrated
View data test label, and by the test label compared with the sample label of described view data, obtain institute
The unfashioned classification based training model stated concentrates the discrimination of view data to checking, and according to discrimination adjustment not
Associated weight in the classification based training model of shaping, until described unfashioned classification based training model concentrates picture number to checking
According to discrimination reach predetermined threshold value.
Capitally, described disaggregated model test module inputs the view data in described test set described classification
Training pattern, obtains the test label of the view data in the test set, and by the sample of the test label and the view data
Label is compared, and obtains comparative result, and described test result includes test label and comparative result.
It is still further preferred that the test index in described test result statistical module includes：Test set discrimination, single class discrimination
Intersect error rate between class, it is each that described test index includes intersection error rate between test set discrimination, single class discrimination and class
Item index, wherein,
Described test set discrimination is the quantity of all view data being correctly validated and survey in described test set
The ratio of view data total quantity is concentrated in examination；
Described single class discrimination is classification based training model to the view data with certain class sample label in test set
The ratio of the total quantity of the view data of such sample label in the quantity and test set that are correctly identified；
Described intersection error rate is the view data with certain class sample label by being known after classification based training model by mistake
Wei not the probability with the view data of another kind of test label.
It is particularly preferred that described vehicle notification number data acquisition and categorizing system also have corresponding relation update module, it is corresponding
Relation update module, for being updated to vehicle notification number information table corresponding to the view data after classification merging treatment.
Above-described vehicle notification number data acquisition and categorizing system realize the data acquisition of vehicle notification number and classification
Method, it is mainly characterized by, and described method comprises the following steps：
(1) the collecting part structure vehicle notification number information list described in, and gathered data, build database；
(2) data in database are trained and reclassified by described classified part, obtain grouped data
Storehouse.
It is preferred that described step (1) is realized by described collecting part, described step (2) passes through classified part
To realize.
More preferably, described step (1) comprises the following steps：
(1.1) corresponding relation building module described in establishes vehicle notification number information list, and vehicle notification number is right with it
The vehicle essential information answered matches correspondingly one by one, and described vehicle essential information includes vehicle corresponding to this vehicle notification number
Brand, money system and age description information and the number-plate number；
(1.2) car plate of the data acquisition module according to corresponding to the vehicle notification number included in vehicle notification number information list
Number, gather the view data corresponding to the number-plate number.
It is particularly preferred that described step (2) comprises the following steps：
(2.1) pretreatment module described in pre-processes to the view data in described image data base；
(2.2) the disaggregated model training module described in is trained according to described data set, obtains classification based training model；
(2.3) the disaggregated model test module described in obtains the classification instruction that described disaggregated model training module training is drawn
Practice model, and described classification based training model is tested according to described data set, output test result；
(2.4) the test result statistical module counts test result described in, test index is obtained；
(2.5) data combiners block described in carries out vehicle notification number corresponding data according to test index and preparatory condition
Merging.
It is still further preferred that described step (2.1) comprises the following steps：
Pretreatment module described in (2.1.1) distributes sample label to the view data in image data base；
View data in image data base is divided into mutually disjoint 3 data sets by (2.1.2) by preset rules, point
Wei not training set, checking collection and test set.
Capitally, when the view data in described data set has that size differs, described image preprocessing
Module pre-processes to the view data in image data base, the view data of the data set is cut out and zoom to default chi
It is very little.
It is still further preferred that described step (2.2) is：
The sample label of view data of the described disaggregated model training module in described training set, to described
Training set carries out deep learning, obtains convolutional neural networks as classification based training model, and described disaggregated model training module
During deep learning, by the view data that described checking is concentrated by unfashioned classification based training model, output is obtained
Test label, and by the test label compared with the sample label of described view data, obtain described unshaped
Classification based training model the discrimination of view data, and the unfashioned classification according to discrimination adjustment are concentrated to checking
Associated weight in training pattern, until described unfashioned classification based training model concentrates the discrimination of view data to checking
Reach predetermined threshold value.
It is still further preferred that described step (2.3) is：
Described disaggregated model test module inputs the view data in described test set described classification based training mould
Type, the test label of the view data in the test set is obtained, and the sample label of the test label and the view data is entered
Row compares, and obtains comparative result, and described test result includes test label and comparative result.
It is still further preferred that also had steps of after described step (2.5)：
(2.6) repeat step (2.1) is to step (2.6) until the test that described test result statistical module counts obtain
Index meets default iteration exit criteria, and described data combiners block also completes the merging for meeting preparatory condition；
(2.7) the corresponding relation update module described in merges relation according to the data provided in described step (2.6), right
Vehicle notification number information list is updated replacement.
Using the system and method for the vehicle notification number data acquisition in the present invention and classification, carried out for vehicle notification number
Data acquisition and classification, data acquisition and classification effectiveness are improved, the quality of data is improved, reduces mark cost, therefore using this
System and method in invention can overcome well in the prior art directly collection great amount of images data, data acquisition blindly, it is superfluous
The problem of remaining is high, and also overcome and using the method that manually marks data are classified in the prior art and examination, data
Treatment effeciency and with a low credibility, the problem of the manpower and high time cost of cost.The present invention is directed to the characteristic of vehicle data, uses
Vehicle notification number more directly efficiently, while by deep learning algorithm classification training pattern, judges as index gathered data
Characteristic similarity in database between different notification number headstocks or tailstock data is gone forward side by side driving head or tailstock data and corresponding car
Information merges, higher than artificial mask method efficiency and confidence level in the prior art.
Brief description of the drawings
Fig. 1 is the system block diagram of the data acquisition of vehicle notification number and categorizing system.
Fig. 2 is vehicle notification number and corresponding description information list structure figure.
Fig. 3 realizes the data acquisition of vehicle notification number for the system of the data acquisition of vehicle notification number and the classification of the present invention and divided
The flow chart of class.
Fig. 4 is that the test result statistical module in the system of the data acquisition of vehicle notification number and the classification of the present invention is surveyed
Flow chart when test result counts.
Fig. 5 is that the notification number merging module in the system of the data acquisition of vehicle notification number and the classification of the present invention is announced
Number merge when flow chart.
Embodiment
In order to more clearly describe the technology contents of the present invention, carried out with reference to specific embodiment further
Description.
In order that the technical means, the inventive features, the objects and the advantages of the present invention are easy to understand, tie below
Conjunction is specifically illustrating, and the present invention is expanded on further.
Fig. 1 is referred to, it show the system block diagram of the data acquisition of vehicle notification number and categorizing system.The vehicle notification number
Data acquisition and categorizing system press view data of the vehicle notification number as index information highly effective gathering headstock and the tailstock, and by car
Head or the consistent bulletin number of tailstock morphosis merge, to realize that vehicle brand, money system and age sort product carry
For required data basis.The vehicle notification number data acquisition and categorizing system, including collecting part and classified part, and it is described
Classified part is connected with collecting part, wherein, described collecting part to build vehicle notification number information list, and to
Gathered data builds database to be sorted；Described classified part to according to notification number information in database data carry out
Training obtains disaggregated model, and data is reclassified in image information aspect, and repetitive exercise and assorting process are to build
Vehicle image taxonomy database.
As shown in Figure 1, the vehicle notification number data acquisition and categorizing system mainly include corresponding relation building module, data
Acquisition module, image pre-processing module, disaggregated model training module, disaggregated model test module, test result statistical module, number
According to merging module.Wherein corresponding relation building module, data acquisition module belongs to collecting part, and image pre-processing module, point
Class model training module, disaggregated model test module, test result statistical module, data combiners block belong to classified part.
Described corresponding relation building module is used to establish vehicle notification number information list, described vehicle notification number information
List includes vehicle essential information corresponding to vehicle notification number and vehicle notification number.In a kind of specific embodiment, described pair
It should be related to and establish module by vehicle administration office or count the existing vehicle notification number in market, create vehicle notification number information list, and
The corresponding relation established between each notification number and vehicle essential information.Referring to Fig. 2, the vehicle created by corresponding relation building module
Notification number information list includes two row, respectively vehicle notification number and corresponding vehicle essential information, and wherein left column is existing market
In all vehicle notification number information；Every a line in right column is the vehicle essential information of corresponding vehicle notification number, the vehicle base
This information includes：Age letter under money system information, the affiliated money system of vehicle under the affiliated brand message of vehicle, the affiliated brand of vehicle
Breath.
The vehicle notification number that described data acquisition module is used in described vehicle notification number information list is in car
Gathered data in guard system, establish the database to match with the vehicle notification number information list.In a kind of specific embodiment,
Described data acquisition module is according to vehicle notification number information list by notification number collection view data.View data includes vehicle
Headstock view data and tailstock view data.In this kind of specific embodiment, pass through target in described data acquisition module
Detection algorithm obtains the image of the relevant vehicle location in original image, and carries out Car license recognition to the image of vehicle location, and
According to license plate recognition result, by the vehicle sectional drawing deposit image data base for matching notification number.
The classification information that described pretreatment module is used to be included according to vehicle notification number is to the number in described database
According to distribution sample label, and the data in described database are divided by preset rules, obtain at least three data set,
Described pretreatment module is additionally operable to pre-process the data in described database.Described pretreatment module is image
Pretreatment module, described image pre-processing module distributes sample label according to classification information to described view data, and presses
Described view data is divided into 3 mutually disjoint data sets, respectively training set, checking collection and test by preset rules
Collection, wherein, described training set obtains classification based training model to carry out model training；Described checking collection is to adjusting training
During unfashioned classification based training model, the classification based training model of acquisition is met preparatory condition；Described test set is used
To test described classification based training model；Described pretreatment module carries out pretreatment to the data in described database as institute
The image pre-processing module stated is cut out and scaled on demand to the view data in described image data base.A kind of specific
In embodiment, image pre-processing module gives each sample label by the classification information of vehicle notification number, and divides data in proportion
Storehouse is training set, checking collection and test set three parts, and sample image size is unified to pixel wide and height needed for training.
Described disaggregated model training module is used to be trained described data set according to classification information, obtains classification
Training pattern, in a kind of specific embodiment, disaggregated model training module uses the training set divided in image pre-processing module
Collect with checking, according to sample label, using deep learning Algorithm for Training disaggregated model.
Described disaggregated model test module is used for according to described data set, and described disaggregated model training module is obtained
The classification based training model taken is tested, and obtains test result.In a kind of specific embodiment, disaggregated model test module is to figure
As the test set divided in pretreatment module does classification prediction, to the data distribution label in test set, and described classification mould
The test label of data in type test module output test set, and the sample that test label is allocated with the data in test set
This label compares, and obtains comparative result.Test result includes test label, sample label and the comparative result of the data.
Described test result statistical module is used for the test result of statistical classification model measurement module output, obtains corresponding
Test index.In a kind of specific embodiment, test index includes：Intersect between test set discrimination, single class discrimination and class
Error rate, described test index include intersecting error rate indices between test set discrimination, single class discrimination and class, its
In, in quantity and test set of the described test set discrimination for all view data being correctly validated in described test set
The ratio of view data total quantity；Described single class discrimination is classification based training model to having certain class sample mark in test set
The ratio of the total quantity of the view data of such sample label in quantity and test set that the view data of label is correctly identified；
Described intersection error rate be the view data with certain class sample label by be misidentified as after classification based training model with
The probability of the view data of another kind of test label.
Described data combiners block is used for according to the test index, judges different vehicle bulletin number correspondence image
Whether information is consistent, and the vehicle notification number corresponding data consistent to image information carries out classification merging.It is embodied in one kind
Example in, data combiners block analysis test result statistical module in every statistical result, to meet merging condition data and
Corresponding notification number carries out classification merging, is arranged according to the connectivity between amalgamation result and merges relation list.
In a kind of specific embodiment, described vehicle essential information includes the product of vehicle corresponding to this vehicle notification number
Board, money system and age description information.
In a kind of specific embodiment, described database is image data base, the view data in the image data base
Corresponding to the vehicle notification number included in described vehicle notification number information list including described data collecting module collected
View data, view data include vehicle frontal by when headstock view data and the tailstock view data sailed out of of the back side.
In a kind of specific embodiment, picture number of the described disaggregated model training module in described training set
According to sample label, deep learning training is carried out to described training set, obtains convolutional neural networks as classification based training model,
And described disaggregated model training module also by the view data that described checking is concentrated by unfashioned classification based training model,
The test label for the view data that checking is concentrated is obtained, and the sample label of the test label and described view data is carried out
Compare, obtain the discrimination that described unfashioned classification based training model concentrates view data to checking, and according to the discrimination
Associated weight in the described unfashioned classification based training model of adjustment, until described unfashioned classification based training model is to testing
Card concentrates the discrimination of view data to reach predetermined threshold value.
In a kind of specific embodiment, described disaggregated model test module is by the view data in described test set
The described classification based training model of input, the test label of the view data in the test set is obtained, and by the test label with being somebody's turn to do
The sample label of view data is compared, and obtains comparative result, and described test result includes test label and comparative result.
Described vehicle notification number data acquisition and categorizing system also have corresponding relation update module, described corresponding relation
Update module is used to be updated vehicle notification number information table corresponding to the view data after classification merging treatment.
In a kind of specific embodiment, merging that described corresponding relation update module exports according to notification number merging module
Relation list regenerates vehicle brand, money system and the age description information for merging class, and updates it and merge each public affairs in class with this
Corresponding relation between announcement number.
Above-described vehicle notification number data acquisition and categorizing system realize the data acquisition of vehicle notification number and classification
Method comprises the following steps：
(1) the collecting part structure vehicle notification number information list described in, and gathered data, build database；
(1.1) corresponding relation building module described in establishes vehicle notification number information list, and vehicle notification number is right with it
The vehicle essential information answered matches correspondingly one by one, and described vehicle essential information includes vehicle corresponding to this vehicle notification number
Brand, money system and age description information and the number-plate number；
(1.2) car plate of the data acquisition module according to corresponding to the vehicle notification number included in vehicle notification number information list
Number, gather the view data corresponding to the number-plate number.
(2) it is trained by described training department's decilog according to the data in storehouse, obtains classification based training model；
(2.1) pretreatment module described in pre-processes to the view data in described image data base；
Pretreatment module described in (2.1.1) distributes sample label to the view data in image data base；
View data in image data base is divided into mutually disjoint 3 data sets by (2.1.2) by preset rules, point
Wei not training set, checking collection and test set；
(2.2) the disaggregated model training module described in is trained according to described data set, obtains classification based training model；
(2.3) the disaggregated model test module described in obtains the classification instruction that described disaggregated model training module training is drawn
Practice model, and described classification based training model is tested according to described data set, output test result；
(2.4) the test result statistical module counts test result described in, test index is obtained；
(2.5) the notification number merging module described in carries out the merging of vehicle notification number according to test index and preparatory condition；
(2.6) repeat step (2.1) is to step (2.6) until the test that described test result statistical module counts obtain
Index meets default iteration exit criteria, and described data combiners block also completes the merging for meeting preparatory condition；
(2.7) the corresponding relation update module described in merges relation according to the data provided in described step (2.6), right
Vehicle notification number information list is updated replacement.
Described step (1) realizes that described step (2) is realized by classified part by described collecting part.
In a kind of preferably embodiment, when the view data in described data set has that size differs,
Described image pre-processing module pre-processes to the view data in image data base, cuts the view data of the data set
Cut out and zoom to pre-set dimension.
In a kind of specific embodiment, image of the described disaggregated model training module in described training set
The sample label of data, deep learning is carried out to described training set, obtain convolutional neural networks as classification based training model, and
Described disaggregated model training module passes through the view data that described checking is concentrated unfashioned during deep learning
Classification based training model, the test label of output is obtained, and the sample label of the test label and described view data is carried out
Compare, obtain the discrimination that described unfashioned classification based training model concentrates view data to checking, and according to the discrimination
Associated weight in the described unfashioned classification based training model of adjustment, until described unfashioned classification based training model is to testing
Card concentrates the discrimination of view data to reach predetermined threshold value.
In a kind of specific embodiment, described disaggregated model test module is by the picture number in described test set
According to the described classification based training model of input, obtain the test label of the view data in the test set, and by the test label with
The sample label of the view data is compared, and obtains comparative result, and described test result includes test label and compares knot
Fruit.
Referring to Fig. 3, in a kind of embodiment, vehicle is realized according to the data acquisition of vehicle notification number and categorizing system
The method of notification number data acquisition and classification is divided into following steps：
(1) establish and the one-to-one vehicle brand of vehicle notification number, money system and age description information table.It is initial in system
The notification number of headstock and the tailstock is identical with vehicle brand, money system and age description information mapping table during change.
(2) it is to index in various regions vehicle administration office system using vehicle notification number before data acquisition module carries out data acquisition
The number-plate number that each vehicle notification number corresponds to vehicle is gathered, license plate number quantity corresponding to each notification number is no less than 10.Press
Vehicle frontal image is retrieved in the traffic block port system of various regions according to license board information, vehicle back side image is retrieved in the alert system of electricity.
In a kind of preferably embodiment, the headstock of vehicle or tailstock amount of images are no less than under each notification number
400.And the algorithm of target detection in described data acquisition module is DPM (Deformable Parts Model) algorithm,
The band of position of all vehicles in acquired original image is intercepted by the DPM algorithms, and car plate knowledge is carried out to all interception vehicles
Not, recognition result and former retrieval are considered the corresponding vehicle pictures of the notification number and are stored in data using the number-plate number is consistent
Storehouse, image data base is built, what is stored in the image data base is the headstock image with the one-to-one vehicle of vehicle notification number
Data or tailstock view data.The recognition result vehicle sectional drawing inconsistent with the retrieval number-plate number not preserves.
(3) when image pre-processing module carries out sample label distribution to the view data of image data base, according to bulletin
Number information carries out sample label distribution, and N number of notification number is N class sample labels, and corresponding vehicle is considered as same under each notification number
Class, distribute identical sample label information.Sample is reclassified according to merging relation list during successive iterations,
Notification number sample after merging is accordingly to be regarded as one kind and distributes sample label information.
And use image pre-processing module by headstock or tailstock data in the ratio per class 50%, 20%, 30%, by headstock
Or parking stall data are divided into mutually disjoint training set, checking collection and test set, and the picture size in each data set is united
One cuts out and zooms to the pixel wide and height needed for training, is used for training module.
(4) data in training set are trained using deep learning method, training AlexNet convolutional neural networks are made
For classification based training model, to realize the Classification and Identification to the data of input.Make training set as training classification based training model
With by checking collection to test discrimination of the classification based training model in iterative process is trained, and according to discrimination adjustment mould
Type weight, when the discrimination drawn in the unfashioned classification based training model of data input that checking is concentrated is intended to convergence, i.e.,
Complete the training of classification based training model.
(5) the disaggregated model test module described in utilizes test set, and the classification based training model trained in step 4 is surveyed
Examination, obtains the test label exported after the data input classification based training model in test set, and with the sample label of the data
It is compared, the former label and prediction label and comparative result of data is saved as into test result.
(6) every discrimination of test set is obtained by described test statisticses module, referring to Fig. 4.Wherein test set
The ratio of correct sample size and test set total sample number is identified in discrimination R, i.e. test set.Statistics is per a kind of discrimination
rn(1≤n≤N), i.e., correct number of samples and such sample are identified in all test samples of current n-th class (1≤n≤N)
The ratio of sum.Statistics is mistakenly identified as the probability e of other classes per one kindnm(1≤n≤N, 1≤m≤N and m ≠ n), i.e., current n-th
The sample size and the n-th class total sample number of m classes (1≤m≤N and m ≠ n) are identified as in the test sample of class (1≤n≤N)
Ratio.Statistical result is saved as into form to preserve and export.
(7) relation is merged according to merging condition judgment notification number, detailed process is as follows：
(7.1) judge whether test set discrimination R is more than the first threshold of setting, current data set need not enter if meeting
Row union operation, otherwise need to merge operation, into step (7.2)；
(7.2) comprised the following steps that referring to Fig. 5, notification number merging module：
(7.2.1) is judged current class n discrimination r by class one by onenWhether the Second Threshold of setting is more than, if eligible
Then current class enters if ineligible and judged in next step without merging with other classes；
(7.2.2) judges that current class n is mistakenly identified as the probability e of m classes one by onenmWith the discrimination r of the n-th classnRatio be
It is no to be more than the 3rd threshold value of setting, if eligible n-th class with m classes are similar enough needs to merge, n-th if ineligible
Class is with m classes without merging；
(7.3) arranged according to the connectivity between merging relation and merge relation list, the son of each major class after merging
Identical car face or tailstock morphosis are respectively provided between class.Redistribute label corresponding to each notification number after merging.Merge into one
The notification number of class has same label, and in a kind of preferably embodiment, label value is the minimum value of label in its subclass.Not
The class being merged still keeps original label constant.
(8) whether comprising needing the class that merges in the output result of judgment step 7, if having repeat step 3 to step 7 until
Discrimination R of the disaggregated model on test set meets default iteration exit criteria R ＞ tR(tRFor the 4th threshold value), or without meeting
The class of merging condition, then complete data and merge.4th threshold value t in actual useR=0.9.
(9) according to the notification number and vehicle product that headstock or the tailstock are established in the amalgamation result renewal step 1 between notification number
Board, money system and age description information mapping table, modification process are as follows：
(9.1) whether being made up of per a kind of multiple notification number subclasses after merging is judged one by one, if only one in current class
Individual notification number subclass, then its corresponding vehicle brand, money system and age description information do not make an amendment, otherwise enter in next step；
(9.2) vehicle brand, money system and age information corresponding to notification number all in current class are extracted, such as：Currently
Class is merged by k notification number to be formed, then description information corresponding to its notification number is：Brand xx- moneys system xx- ages xx1, brand
Xx- moneys system xx- age xx2 ... brand xx- moneys system xx- ages xxk；
(9.3) splice the description information of each notification number, generate the new description information of current class, such as：Brand xx- moneys system
Xx- ages xx1 or brand xx- moneys system xx- ages xx2 ... or brand xx- moneys system xx- ages xxk；
By each vehicle notification number pair in current class in the mapping table that vehicle notification number and vehicle essential information are formed
The basic description information of vehicle answered is revised as the new description information of generation in (9.3).So far complete headstock or tailstock notification number and
The renewal of vehicle brand, money system and age description information mapping table.
Using the system and method for the vehicle notification number data acquisition in the present invention and classification, carried out for vehicle notification number
Data acquisition and classification, data acquisition and classification effectiveness are improved, the quality of data is improved, reduces mark cost, therefore using this
System and method in invention can overcome well in the prior art directly collection great amount of images data, data acquisition blindly, it is superfluous
The problem of remaining is high, and also overcome and using the method that manually marks data are classified in the prior art and examination, data
Treatment effeciency and with a low credibility, the problem of the manpower and high time cost of cost.The present invention is directed to the characteristic of vehicle data, uses
Vehicle notification number more directly efficiently, while by deep learning algorithm classification training pattern, judges as index gathered data
Similitude between different notification number headstocks or tailstock data simultaneously carries out data conjunction, than artificial mask method efficiency in the prior art
It is higher with confidence level.
In this description, the present invention is described with reference to its specific embodiment.But it is clear that it can still make
Various modifications and alterations are without departing from the spirit and scope of the present invention.Therefore, specification and drawings are considered as illustrative
It is and nonrestrictive.
Claims (10)
1. a kind of vehicle notification number data acquisition and categorizing system, it is characterised in that including collecting part and classified part, and institute
The classified part stated is connected with collecting part, wherein, described collecting part to build vehicle notification number information list, and
Database to be sorted is built to gathered data；Described classified part to according to notification number information to the data in database
Be trained acquisition disaggregated model, and data reclassified in image information aspect, repetitive exercise and assorting process with
Build vehicle image taxonomy database.
2. vehicle notification number data acquisition according to claim 1 and categorizing system, it is characterised in that described collection portion
Dividing includes：
Corresponding relation building module, for establishing vehicle notification number information list, described vehicle notification number information list includes
Vehicle essential information corresponding to vehicle notification number and vehicle notification number；
Data acquisition module, adopted for the vehicle notification number in described vehicle notification number information list in car guard system
Collect data, establish the database to match with the vehicle notification number information list；
Described classified part includes：
Pretreatment module, the data distribution sample given for the classification information that is included according to vehicle notification number in described database
Label, and the data in described database are divided by preset rules, obtain at least three data set, described pre- place
Reason module is additionally operable to pre-process the data in described database；
Disaggregated model training module, for being trained according to classification information to described data set, obtain classification based training model；
Disaggregated model test module, for the classification according to described data set, obtained to described disaggregated model training module
Training pattern is tested, and obtains test result；
Test result statistical module, for the test result of statistical classification model measurement module output, obtain corresponding test and refer to
Mark；
Data combiners block, for according to the test index, whether judging different vehicle bulletin number correspondence image information
Unanimously, and to image information, consistent vehicle notification number corresponding data carries out classification merging.
3. vehicle notification number data acquisition according to claim 2 and categorizing system, it is characterised in that described vehicle base
This information includes the brand of vehicle, money system and age description information corresponding to this vehicle notification number；
Described database is image data base, and the view data in the image data base is adopted including described data acquisition module
View data corresponding to the vehicle notification number included in the described vehicle notification number information list of collection；
Described view data include vehicle frontal by when headstock view data and the tailstock view data sailed out of of the back side；
Described pretreatment module is image pre-processing module, and described image pre-processing module is according to classification information to described
View data distributes sample label, and described view data is divided into 3 mutually disjoint data sets by preset rules, point
Not Wei training set, checking collection and test set, wherein,
Described training set obtains classification based training model to carry out model training；
Described checking collects the classification based training model for the unfashioned classification based training model during adjusting training, making acquisition
Meet preparatory condition；
Described test set is testing described classification based training model；
It is described image pre-processing module pair that described pretreatment module, which carries out pretreatment to the data in described database,
View data in described image data base is cut out and scaled on demand.
4. vehicle notification number data acquisition according to claim 3 and categorizing system, it is characterised in that described classification mould
The sample label of view data of the type training module in described training set, deep learning instruction is carried out to described training set
Practice, obtain convolutional neural networks as classification based training model, and described disaggregated model training module is also by described checking collection
In view data by unfashioned classification based training model, obtain the test label for the view data that checking is concentrated, and should
Test label obtains described unfashioned classification based training model to testing compared with the sample label of described view data
Card concentrates the discrimination of view data, and the correlative weight in described unfashioned classification based training model is adjusted according to the discrimination
Weight, until described unfashioned classification based training model concentrates the discrimination of view data to reach predetermined threshold value checking；
Described disaggregated model test module inputs the view data in described test set on described classification based training model, obtains
The test label of the view data in the test set is taken, and the sample label of the test label and the view data is compared
Compared with acquisition comparative result, described test result includes test label and comparative result；
Test index in described test result statistical module includes：Intersect between test set discrimination, single class discrimination and class
Error rate, wherein,
Described test set discrimination is the quantity and test set of all view data being correctly validated in described test set
The ratio of middle view data total quantity；
Described single class discrimination is that classification based training model is carried out to the view data with certain class sample label in test set
The ratio of the total quantity of the view data of such sample label in the quantity and test set that correctly identify；
Described intersection error rate is the view data with certain class sample label by being misidentified as after classification based training model
The probability of view data with another kind of test label.
5. vehicle notification number data acquisition according to claim 2 and categorizing system, it is characterised in that described vehicle is public
Accuse number collection and categorizing system also has corresponding relation update module, described corresponding relation update module is used to close to sorting out
And vehicle notification number information table corresponding to the view data after handling is updated.
6. a kind of system based on any one of claim 1 to 5 realizes the side of the data acquisition of vehicle notification number and classification
Method, it is characterised in that described method comprises the following steps：
(1) the collecting part structure vehicle notification number information list described in, and gathered data, build database；
(2) data in database are trained and reclassified by described classified part, obtain taxonomy database.
7. the method according to claim 6 for realizing the data acquisition of vehicle notification number and classification, it is characterised in that
Described step (1) realized by described collecting part, wherein, described label corresponds with vehicle notification number,
Described collecting part includes：
Corresponding relation building module, for establishing vehicle notification number information list, described vehicle notification number information list includes
Vehicle essential information corresponding to vehicle notification number and vehicle notification number；
Data acquisition module, adopted for the vehicle notification number in described vehicle notification number information list in car guard system
Collect data, establish the database to match with the vehicle notification number information list；
Described step (2) realizes that described classified part includes by classified part：
Pretreatment module, the data distribution sample given for the classification information that is included according to vehicle notification number in described database
Label, and the data in described database are divided by preset rules, obtain at least three data set, described pre- place
Reason module is additionally operable to pre-process the data in described database；
Disaggregated model training module, for being trained according to described data set, obtain classification based training model；
Disaggregated model test module, for according to described data set, testing the classification based training model obtained, and obtain
Take test result；
Test result statistical module, for the test result of statistical classification model measurement module output, obtain corresponding test and refer to
Mark；
Data combiners block, for according to the test index, whether judging different vehicle bulletin number correspondence image information
Unanimously, and to image information, consistent vehicle notification number corresponding data carries out classification merging；
Described database is image data base, and the view data in the image data base is adopted including described data acquisition module
View data corresponding to the vehicle notification number of collection, and vehicle notification number is the car included in described vehicle notification number information list
Notification number；Described view data include vehicle frontal by when headstock view data and the tailstock picture number sailed out of of the back side
According to；
Described step (1) comprises the following steps：
(1.1) corresponding relation building module described in establishes vehicle notification number information list, and vehicle notification number is corresponding
Vehicle essential information matches correspondingly one by one, and described vehicle essential information includes the product of vehicle corresponding to this vehicle notification number
Board, money system and age description information and the number-plate number；
(1.2) license plate number of the data acquisition module according to corresponding to the vehicle notification number included in vehicle notification number information list
Code, gathers the view data corresponding to the number-plate number；
Described step (2) comprises the following steps：
(2.1) pretreatment module described in pre-processes to the view data in described image data base；
(2.2) the disaggregated model training module described in is trained according to described data set, obtains classification based training model；
(2.3) the disaggregated model test module described in obtains the classification based training mould that described disaggregated model training module training is drawn
Type, and described classification based training model is tested according to described data set, output test result；
(2.4) the test result statistical module counts test result described in, test index is obtained；
(2.5) data combiners block described in carries out the conjunction of vehicle notification number corresponding data according to test index and preparatory condition
And.
8. the method according to claim 7 for realizing the data acquisition of vehicle notification number and classification, it is characterised in that
Described pretreatment module is image pre-processing module, and described image pre-processing module is according to classification information to described
View data distributes sample label, and described view data is divided into 3 mutually disjoint data sets by preset rules, point
Not Wei training set, checking collection and test set, wherein,
Described training set obtains classification based training model to carry out model training；
Described checking collects the classification based training model for the unfashioned classification based training model during adjusting training, making acquisition
Meet preparatory condition；
Described test set is testing described classification based training model；
It is described image pre-processing module pair that described pretreatment module, which carries out pretreatment to the data in described database,
View data in described image data base is cut out and scaled on demand,
Described step (2.1) comprises the following steps：
Pretreatment module described in (2.1.1) distributes sample label to the view data in image data base；
View data in image data base is divided into mutually disjoint 3 data sets by (2.1.2) by preset rules, is respectively
Training set, checking collection and test set；
When view data in described data set has that size differs, described image pre-processing module is to picture number
Pre-processed according to the view data in storehouse, the view data of the data set is cut out and zoom to pre-set dimension.
9. the method according to claim 7 for realizing the data acquisition of vehicle notification number and classification, it is characterised in that described
Step (2.2) is：
The sample label of view data of the described disaggregated model training module in described training set, to described training
Collection carries out deep learning, obtains convolutional neural networks as classification based training model, and described disaggregated model training module is in depth
Spend in learning process, the view data that described checking is concentrated obtains the survey of output by unfashioned classification based training model
Test-object label, and by the test label compared with the sample label of described view data, obtain described unfashioned point
Class training pattern concentrates the discrimination of view data to checking, and described unfashioned classification based training is adjusted according to the discrimination
Associated weight in model, until described unfashioned classification based training model concentrates the discrimination of view data to reach checking
Predetermined threshold value；
Described step (2.3) is：
Described disaggregated model test module inputs the view data in described test set on described classification based training model, obtains
The test label of the view data in the test set is taken, and the sample label of the test label and the view data is compared
Compared with acquisition comparative result, described test result includes test label and comparative result.
10. the method according to claim 7 for realizing the data acquisition of vehicle notification number and classification, it is characterised in that described
The data acquisition of vehicle notification number and categorizing system also have：
Corresponding relation update module, for being carried out to vehicle notification number information table corresponding to the view data after classification merging treatment
Renewal；
Also had steps of after described step (2.5)：
(2.6) repeat step (2.1) is to step (2.6) until the test index that described test result statistical module counts obtain
Meet default iteration exit criteria, described data combiners block also completes the merging for meeting preparatory condition；
(2.7) the corresponding relation update module described in merges relation according to the data provided in described step (2.6), to vehicle
Notification number information list is updated replacement；
Described test index includes intersecting error rate indices between test set discrimination, single class discrimination and class, wherein,
Described test set discrimination is the quantity and test set of all view data being correctly validated in described test set
The ratio of middle view data total quantity；
Described single class discrimination is that classification based training model is carried out to the view data with certain class sample label in test set
The ratio of the total quantity of the view data of such sample label in the quantity and test set that correctly identify；
Described intersection error rate is the view data with certain class sample label by being misidentified as after classification based training model
The probability of view data with another kind of label.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710450924 | 2017-06-15 | ||
| CN2017104509245 | 2017-06-15 |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107403162A true CN107403162A (en) | 2017-11-28 |
| CN107403162B CN107403162B (en) | 2020-11-03 |
Family
ID=60402480
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710639834.0A Active CN107403162B (en) | 2017-06-15 | 2017-07-31 | System and method for collecting and classifying vehicle notice number data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107403162B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110786847A (en) * | 2018-08-02 | 2020-02-14 | 深圳市理邦精密仪器股份有限公司 | Electrocardiogram signal library building method and analysis method |
| US10818042B1 (en) | 2020-01-14 | 2020-10-27 | Capital One Services, Llc | Vehicle information photo overlay |
| US10832400B1 (en) | 2020-01-14 | 2020-11-10 | Capital One Services, Llc | Vehicle listing image detection and alert system |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2013148971A (en) * | 2012-01-17 | 2013-08-01 | Oki Electric Ind Co Ltd | Vehicle type discrimination device |
| CN103927877A (en) * | 2014-04-01 | 2014-07-16 | 内蒙古银安科技开发有限责任公司 | Motor vehicle image information storage reading method and system |
| CN106022285A (en) * | 2016-05-30 | 2016-10-12 | 北京智芯原动科技有限公司 | Vehicle type identification method and vehicle type identification device based on convolutional neural network |
| CN106295541A (en) * | 2016-08-03 | 2017-01-04 | 乐视控股（北京）有限公司 | Vehicle type recognition method and system |
| CN106446949A (en) * | 2016-09-26 | 2017-02-22 | 成都通甲优博科技有限责任公司 | Vehicle model identification method and apparatus |
-
2017
- 2017-07-31 CN CN201710639834.0A patent/CN107403162B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2013148971A (en) * | 2012-01-17 | 2013-08-01 | Oki Electric Ind Co Ltd | Vehicle type discrimination device |
| CN103927877A (en) * | 2014-04-01 | 2014-07-16 | 内蒙古银安科技开发有限责任公司 | Motor vehicle image information storage reading method and system |
| CN106022285A (en) * | 2016-05-30 | 2016-10-12 | 北京智芯原动科技有限公司 | Vehicle type identification method and vehicle type identification device based on convolutional neural network |
| CN106295541A (en) * | 2016-08-03 | 2017-01-04 | 乐视控股（北京）有限公司 | Vehicle type recognition method and system |
| CN106446949A (en) * | 2016-09-26 | 2017-02-22 | 成都通甲优博科技有限责任公司 | Vehicle model identification method and apparatus |
Non-Patent Citations (2)
| Title |
|---|
| STEFAN BADURA ET AL.: "Advanced scale-space, invariant, low detailed feature recognition from images - car brand recognition", 《PROCEEDINGS OF THE INTERNATIONAL MULTICONFERENCE ON COMPUTER SCIENCE AND INFORMATION TECHNOLOGY》 * |
| 李熙莹 等: "基于LLC与加权SPM的车辆品牌型号识别", 《计算机工程》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110786847A (en) * | 2018-08-02 | 2020-02-14 | 深圳市理邦精密仪器股份有限公司 | Electrocardiogram signal library building method and analysis method |
| CN110786847B (en) * | 2018-08-02 | 2022-11-04 | 深圳市理邦精密仪器股份有限公司 | Electrocardiogram signal library building method and analysis method |
| US10818042B1 (en) | 2020-01-14 | 2020-10-27 | Capital One Services, Llc | Vehicle information photo overlay |
| US10832400B1 (en) | 2020-01-14 | 2020-11-10 | Capital One Services, Llc | Vehicle listing image detection and alert system |
| US11587224B2 (en) | 2020-01-14 | 2023-02-21 | Capital One Services, Llc | Vehicle listing image detection and alert system |
| US11620769B2 (en) | 2020-01-14 | 2023-04-04 | Capital One Services, Llc | Vehicle information photo overlay |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107403162B (en) | 2020-11-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108765404B (en) | A kind of road damage testing method and device based on deep learning image classification | |
| CN101329734B (en) | License plate character recognition method based on K-L transform and LS-SVM | |
| CN102184405B (en) | Image acquisition-analysis method | |
| CN102110122B (en) | Method and device for establishing sample picture index table, method and device for filtering pictures and method and device for searching pictures | |
| CN110808945B (en) | Network intrusion detection method in small sample scene based on meta-learning | |
| CN111401418A (en) | Employee dressing specification detection method based on improved Faster r-cnn | |
| CN104952073B (en) | Scene Incision method based on deep learning | |
| CN108830254B (en) | Fine-grained vehicle type detection and identification method based on data balance strategy and intensive attention network | |
| CN101678404A (en) | Method of handling transmittals including a graphic classification of the signatures associated with the transmittals | |
| CN111310850A (en) | License plate detection model construction method and system and license plate detection method and system | |
| CN114067143A (en) | Vehicle weight recognition method based on dual sub-networks | |
| CN108764302A (en) | A kind of bill images sorting technique based on color characteristic and bag of words feature | |
| CN112711983B (en) | Nuclear analysis system, method, electronic device, and readable storage medium | |
| CN107403162A (en) | The data acquisition of vehicle notification number and the system and method for classification | |
| CN101655909A (en) | Device and method for calculating matching degree | |
| CN115719475B (en) | Three-stage trackside equipment fault automatic detection method based on deep learning | |
| JP2010170207A (en) | Vehicle monitoring system | |
| CN110991442A (en) | High-accuracy highway license plate cloud identification method | |
| CN105426926B (en) | A kind of couple of AMOLED carries out the method and device of detection classification | |
| CN106570514A (en) | Automobile wheel hub classification method based on word bag model and support vector machine | |
| CN116863418A (en) | Road traffic limiting vehicle species division method based on CA_MixNet | |
| CA3012927A1 (en) | Counting objects in images based on approximate locations | |
| CN113680685A (en) | Garbage classification treatment system based on recoverability | |
| CN115497293B (en) | Dynamic traffic accident hidden trouble point identification method | |
| CN110321949B (en) | Distributed vehicle tracking method and system based on observation terminal network |
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 |