CN109447069A - Collecting vehicle information recognition methods and system towards intelligent terminal - Google Patents
Collecting vehicle information recognition methods and system towards intelligent terminal Download PDFInfo
- Publication number
- CN109447069A CN109447069A CN201811281498.8A CN201811281498A CN109447069A CN 109447069 A CN109447069 A CN 109447069A CN 201811281498 A CN201811281498 A CN 201811281498A CN 109447069 A CN109447069 A CN 109447069A
- Authority
- CN
- China
- Prior art keywords
- information
- license plate
- module
- intelligent terminal
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/24—Aligning, centring, orientation detection or correction of the image
- G06V10/242—Aligning, centring, orientation detection or correction of the image by image rotation, e.g. by 90 degrees
-
- 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
- G06V10/40—Extraction of image or video features
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/14—Image acquisition
- G06V30/146—Aligning or centring of the image pick-up or image-field
- G06V30/1475—Inclination or skew detection or correction of characters or of image to be recognised
- G06V30/1478—Inclination or skew detection or correction of characters or of image to be recognised of characters or characters lines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/14—Image acquisition
- G06V30/148—Segmentation of character regions
- G06V30/153—Segmentation of character regions using recognition of characters or words
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/62—Text, e.g. of license plates, overlay texts or captions on TV images
- G06V20/625—License plates
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Traffic Control Systems (AREA)
Abstract
The present invention relates to digital image processing techniques and artificial intelligence field, especially the application in the collecting vehicle information recognition methods towards intelligent terminal and system.The recognition methods is first adjusted the image of acquisition, then carries out License Plate, then carries out projective transformation and rotation correction to image, is accurately positioned later, carries out Character segmentation and character recognition to license plate image, finally obtains license plate recognition result；Vehicle information acquisition system towards intelligent terminal includes intelligent terminal subsystem and service device terminal system, and intelligent terminal subsystem and service device terminal system pass through transmitted data on network.In order to overcome existing information collection personnel manually to acquire the mode of information of vehicles, the invention proposes collecting vehicle information recognition methods and system towards intelligent terminal, it can accurately and fast identify the license plate number to park cars and other information of vehicles, it in the working time for liberating information collection personnel, increases work efficiency.
Description
Technical field
The present invention relates to digital image processing techniques and artificial intelligence field, and especially deep learning is towards intelligent terminal
Collecting vehicle information recognition methods and system in application.
Background technique
With increasing sharply for automobile quantity, vehicles management method has been unsatisfactory for demand of today, acquires vehicle efficiency
It is not high, lead to many vehicle management problems, wherein illegal parking problem has become " persistent ailment " of society.Traffic police is now in violation of rules and regulations
Parking data is also in artificial asynchronous process；Acquisition shooting illegal parking image first, then opens penalty note manually, then uploads and disobeys
Parking information is advised to server, is finally only inquiry of the user to violation information.Period, the duty cycle is too long, and data repeatedly pass
It is defeated, it is easy to happen managerial fault.
Deep learning is in developing stage, is still in infancy in embedded end application, in arithmetic speed, calculates essence
Degree has certain limitation, and wherein calculating speed has critical status in Embedded application development to deep learning,
There are two important factor, parameter amount and calculating costs in calculating speed in embedded system.Wherein VGG16 sorter network has 1
101001000 training parameters (occupying memory 92759Kb), and calculate cost and be 15,300,000,000 (measurement standard is
The number of float32op), so huge calculating cost and training parameter, are a greatest tests to Embedded system.
For the algorithm of target detection of existing deep learning；R-cnn, yolo, SDD series.The wherein inspection of r-cnn series
Degree of testing the speed is slow, but detection accuracy is high.Yolo series rate and precision have promotion, but very low to the detectability of Small object
Under, and the ratio very little that license plate is accounted in image.SDD algorithm all has greatly improved in accuracy and speed compared to preceding two algorithm,
But in embedded systems, the demand for reaching real-time still has very long place to be worth improving.
Summary of the invention
Goal of the invention:
In order to overcome existing artificial acquisition vehicle data mode, the invention proposes the information of vehicles towards intelligent terminal
Collection and recognition method and system can accurately and fast identify the license plate number and vehicle position information of associated vehicle, liberation letter
It in the working time for ceasing collector, increases work efficiency.
Technical solution:
The present invention is achieved by the following technical solutions:
The step of collecting vehicle information recognition methods towards intelligent terminal, this method, is as follows:
Step 1: the vehicle image of acquisition is adjusted；
Step 2: license plate Primary Location is carried out with SSD algorithm of target detection to the image that step 1 obtains；
Step 3: projective transformation and rotation correction are carried out to the image after positioning, obtain horizontal license plate；
Step 4: being accurately positioned horizontal license plate, later to license plate image Character segmentation, then carries out character recognition；
Step 5: final license plate recognition result is obtained.
Step 2 specifically includes the following steps:
Step A: experience region is arranged according to license plate position to acquired image, reduces size of data；
Step B: to treated, image uses SSD algorithm to carry out car plate detection, and SSD algorithm network structure is divided into two portions
Point, front end carries out compressed data, and rear end carries out feature extraction, introduces multiple dimensioned convolutional layer and obtain various sizes of characteristic pattern, examines
Survey multiple dimensioned license plate.
Specific step is as follows for step 4:
Step a: block of pixels extra in license plate image after removal correction obtains being accurately positioned license plate position；
Step b: license plate image is divided into character block to be identified again: using domain adaptive threshold is faced, searches out character
Profile, obtain minimum circumscribed rectangle using function and profile irised out；
Step c: character block identification: histogram equalization is carried out to character block using special algorithm, is then fed into CNN nerve
Network does 65 kinds of character classification.
Collecting vehicle information identifying system towards intelligent terminal, including intelligent terminal subsystem and service device terminal system
System, intelligent terminal subsystem and service device terminal system pass through transmitted data on network；
Intelligent terminal subsystem includes data acquisition module, Car license recognition module, GPS positioning module, data transmission reception
Module；
Data acquisition module, for acquiring vehicle image and shooting time；
Car license recognition module identifies license plate number for carrying out Car license recognition to the vehicle image of acquisition；
GPS positioning module, for being positioned to vehicle；
Data transmit receiving module, for the information of vehicles of smart terminal sub-stream system acquisition to be transferred to server end subsystem
The information of the transmission of system and reception server end subsystem；For sending and receiving car owner's cellphone information；For to server
Terminal system sends query information and receives query result.
Server end subsystem includes information receiving module, database module, message processing module and information transfer module；
Information receiving module, the information of vehicles uploaded for receiving storage intelligent terminal subsystem, receives corresponding vehicle
The information of car owner's feedback receives information processing personnel to the processing result of exception information, receives the inquiry of intelligent terminal subsystem
Information；
Database module, for store intelligent terminal subsystem upload acquisition information of vehicles and traffic police put on record vehicle letter
Breath；
Message processing module, the vehicle in Car license recognition information and date library module for transmitting intelligent terminal subsystem
Owner information is matched, and vehicle owner information corresponding with Car license recognition information is found out；For handling abnormal prompt information；
For by the information in information processing personnel treated information update database module；
Information transfer module, for information to be issued to the car owner of corresponding vehicle by short message form, by intelligent terminal subsystem
The information of system inquiry sends intelligent terminal subsystem to.
The Car license recognition module includes license plate experience region setup module, License Plate module, VLP correction module
With license plate number identification module；
License plate experience region setup module, for carrying out resolution adjustment and cutting to the original vehicle image of acquisition；
License Plate module, for carrying out license plate Primary Location using SSD algorithm of target detection；
VLP correction module obtains horizontal license plate for carrying out projective transformation and rotation correction to the license plate after positioning；
License plate number identification module is used to carry out license plate image Character segmentation to horizontal license plate, then carries out character recognition, obtains
Take target license plate information.
License plate experience region setup module includes:
Image resolution ratio adjustment unit, for being adjusted to the resolution ratio for obtaining image；
Image cropping unit is cut for original vehicle image.
License Plate module includes:
License plate position extraction unit carries out SSD algorithm of target detection to the vehicle image after cutting, first carries out compression number
According to being re-introduced into convolutional layer and obtain various sizes of characteristic pattern, last rear end carries out feature extraction, obtains license plate image and position is believed
Breath.
VLP correction module includes:
VLP correction unit: projective transformation and rotation correction are carried out to the image after positioning.
License plate number identification module includes:
License plate removes pixel unit, for removing block of pixels extra in license plate image；
License Plate Character Segmentation unit is isolated the biggish character zone of contrast, is found two using adaptive threshold limit value
It is worth all profiles of image, all profiles is irised out with rectangle frame, then by the ratio of width to height, area, which filters out, meets character feature
Region；
Recognition of License Plate Characters unit is then fed into CNN for carrying out histogram equalization to character block using special algorithm
Neural network does 65 kinds of character classification, identifies each characters on license plate.
Advantage and effect:
By in algorithm transplanting intelligent terminal of the invention, the work of collecting vehicle information is completed in intelligent terminal subsystem,
The system identification speed is fast, and accuracy of identification is high, realizes multi-angle License Plate, and 100 reality scene image accuracy are
98%, average detected speed is 445 milliseconds.
By the modification to SSD algorithm, system improving calculating speed realizes the real-time of system, and wherein training parameter is 4
1000000000,30 times of original vgg16 network training parameter are reduced into, reducing and calculating cost is 500,000,000 6 thousand 9 million, than original
There are 27 times reduced, but too many there is no losing in precision.For the detection effect for increasing system, transfer learning plan is taken
Slightly, data of multiple angles is manually added, system is made to have certain generalization ability, can identify the vehicle data of multi-angle,
This system saves manpower and material resources, and information collection personnel can acquire data whenever and wherever possible, as long as at intelligent end
End installation this system, makes acquisition data become convenient and efficient, can be with upload server after acquisition, and collector is to the number after acquisition
According to having no right to handle herein, synchronization process mode acquires, and identifies, uploads, notice, all primary to complete, and increases management work more
Effect.
Detailed description of the invention
Fig. 1 is the collecting vehicle information recognition methods flow chart towards intelligent terminal；
Fig. 2 is that smart terminal sub-stream system identification is schemed as the result is shown；
Fig. 3 is the priori frame schematic diagram of SSD algorithm in step 2；
Fig. 4 is 8 × 8 characteristic patterns of SSD algorithm in step 2；
Fig. 5 is 4 × 4 characteristic patterns of SSD algorithm in step 2；
Fig. 6 multi-angle car plate detection result figure；
Fig. 7 is SSD algorithm network structure；
Fig. 8 is traditional convolution structure chart；
Fig. 9 is depth convolutional coding structure figure；
Figure 10 is point convolutional coding structure figure；
Figure 11 is to face domain adaptive threshold Character segmentation figure；
Figure 12 is license plate recognition result figure；
Figure 13 is the architecture diagram of the license board information acquisition system towards intelligent terminal；
Figure 14 is the flow chart of vehicle information acquisition system of the invention；
Figure 15 is the functional block diagram for knowing Car license recognition module；
Figure 16 is the information querying flow figure of intelligent terminal subsystem；
Figure 17 is map GPS parameter selection figure.
Specific embodiment
The present invention is to solve information collection personnel acquisition information of vehicles needs manually largely to be operated, such as traffic police is manual
The problem that illegal parking list either acquires the data hardly possible of illegal parking is write, this method can be completed to shoot illegal parking manually
Image, identification license plate, shows the information that parks cars in violation of rules and regulations at positioning licence plate.The APP that this system can be used as intelligent terminal facilitates letter
It ceases collector and carries out information collection work.
In the epoch that the computational efficiency of computer quickly improves, performance of the deep learning in image classification and target detection
It is very outstanding.The method that this method just uses deep learning；Wherein car plate detection uses SSD algorithm of target detection, due to pressing
The front network of contracting characteristic pattern is changed, and former VGG network becomes mobile-net network, and basic variation is by conventional roll
Product mode has become depth and has separated convolution, reduces the time of convolution algorithm, to accelerate the time of Objective extraction.
For deep learning training: it include 1680 images in car plate detection data set, wherein 1500 are used as training set,
180 are used as the test set training stage, carry out transfer learning, license plate inspection on pre-training 200,000 times models on VOC data set
Measured data collection does 200,000 training.
In the selection of frame, the present invention uses Open Framework TensorFlow due to applying in embedded end
Lite, the frame improve calculating speed to optimizing in network operations.
The license plate of multi-angle can be positioned, license plate number identification is accurate, speed is fast.Intelligent terminal of the present invention can be with
It is existing smart phone, plate or other Portable intelligent terminals with camera function.
The present invention will now be described in detail with reference to the accompanying drawings:
As shown in Figure 1, the collecting vehicle information recognition methods towards intelligent terminal, the step of this method, is as follows:
Step 1: the vehicle image of acquisition is adjusted；
Step 2: license plate Primary Location is carried out with SSD algorithm of target detection to the image that step 1 obtains；
Step 3: projective transformation and rotation correction are carried out to the image after positioning, obtain horizontal license plate；
Step 4: being accurately positioned horizontal license plate, removes excess pixel block, later to license plate image Character segmentation,
Character recognition is carried out again；
Step 5: final license plate recognition result is obtained；Final intelligent terminal shows that result is shown in Fig. 2.
The size for mainly changing input picture is adjusted in step 1 to the vehicle image of acquisition, image is cut out
Cut the adjustment with resolution ratio.
Wherein, step 2 specifically includes the following steps:
Step A: experience region is arranged according to license plate position to acquired image, reduces size of data；
Step B: carrying out treated image using deep learning SSD algorithm, and SSD algorithm uses network structure, and SSD is calculated
Method network structure is divided into two parts, and front end carries out compressed data, and rear end carries out feature extraction, introduces multiple dimensioned convolutional layer and obtain
Various sizes of characteristic pattern detects multiple dimensioned license plate.
SSD in step 2 has used for reference the theory of anchor in Faster R-CNN, each unit setting scale or length
For width than different priori frames, the bounding box (bounding boxes) of prediction is on the basis of these priori frames, in certain journey
Training difficulty is reduced on degree.Under normal circumstances, multiple priori frames can be arranged in each unit, and scale and length-width ratio have differences,
As shown in Fig. 3, Fig. 4 and Fig. 5, it can be seen that each unit has used 4 different priori frames, and arithmetic result is as shown in Figure 6.
SSD algorithm is using network structure as shown in fig. 7, network is divided into two parts: using the VGG front network conduct of Standard convolution
The front network of SSD algorithm of target detection, purpose are compressive features.Rear end is to extract character network, and convolution characteristic layer is added, obtains
To the characteristic pattern of different scale, to realize that multiscale target detects.
Algorithm will be implanted in embedded end, so the requirement of real-time is not achieved in tradition VGG network, then by VGG network
Replace with Mobile-net (v1).
The core concept of Mobile-Net is the mode for separating convolution by using depth and combining with point-by-point grouping convolution
Ingenious decomposition is carried out to original Standard convolution core known to us, convolution mode is as shown in Fig. 8, Fig. 9 and Figure 10, so that the network
Weighting parameter is reduced, and is reduced compared to Standard convolution operation calculation amount, to reduce in permissible range reaching in precision
Promote the purpose of network operations speed.
The decomposable convolution of depth, Standard convolution can be resolved into one convolution of a depth convolution sum (1 × 1 by it
Convolution kernel).Each convolution kernel is applied to each channel by depth convolution, and 1 × 1 convolution is used to combine the defeated of channel convolution
Out.
Fig. 9 and Figure 10 is inputted based on the improved Mobile-Net convolution filter model hypothesis of Standard convolution filter
The size of Feature Mapping F is that accepted standard convolution K is (DK,DK, M, N), the calculation amount of corresponding Standard convolution are as follows: DK×M×N
×DF×DF；Wherein, M is the port number of input, and N is the port number of output.
But such as by Standard convolution (DK,DK, M, N) and it is split as the point-by-point convolution of depth convolution sum: depth convolution is responsible for filtering and is made
With having a size of (DK,DK,1,M)；Point-by-point convolution is responsible for ALT-CH alternate channel, having a size of (1,1, M, N).
The calculation amount of depth convolution is Dk·Dk·M·DF, point-by-point convolutional calculation amount MNDF·DF, so being rolled up with depth
Long-pending and point-by-point convolution replaces the amount of calculation of traditional convolution: Dk·Dk·M·DF+M·N·DF·DF。
The calculation amount of two kinds of networks is compared, it will be seen that Mobile-Net can be with due to the decomposition to Standard convolution
Calculation amount is substantially reduced, comparing calculation formula is as follows:
Therefore, we are with the basic network before Mobile-Net network, come replace being made of in script SSD VGG-16
Basic network, to accelerate network, object detection results are more preferable.
Projective transformation and rotation correction are carried out to the image after positioning in step 3:
Projective transformation model can be modeled by following Bilinear Equations:
x1=s (x0,y0)=c1x0+c2y0+c3x0y0+c4 (2)
y1=t (x0,y0)=c5x0+c6y0+c7x0y0+c8 (3)
Wherein x1And y1It is each of original image pixel coordinate point, x0And y0Be in new figure transformed image with original
Scheme corresponding coordinate points.Pass through mapping relations s (x0,y0) and t (x0,y0), to complete projective transformation.It needs to resolve 8 in total
Parameter (c1,c2......c8).For that purpose it is necessary at least know 4 points in original image；Four coordinate points are obtained in SSD algorithm,
c1,c2......c8The equation coefficient in formula 2 and formula 3 is respectively represented；
Wherein, the functional expression of c value:
Specific step is as follows for step 4:
Step a: block of pixels extra in license plate image after removal correction obtains being accurately positioned license plate position；
Step b: license plate image is divided into character block to be identified again: using domain adaptive threshold is faced, searches out character
Profile, obtain minimum circumscribed rectangle using function and profile irised out；
Due to being the minimum circumscribed rectangle profile sought under multiple groups threshold value, it should select the band of position of best match.It does
Method is first to seek the central point of multiple groups rectangular area averagely, then make ratio at a distance from average central point with every group of central point
Compared with finding and obtained with the central point with mean center point minimum range using the rectangle frame of this central point as final output
The boundary coordinate of character block；
Since license plate background is uneven, objective body often shows as locally brighter or dark than background, can not determine
Global threshold operation, needs to find a suitable threshold value by its neighborhood and is split, using the calculating core on adjacent boundary 11 × 11,
Multiple adaptive threshold, threshold parameter C value range are that (- 26,20) step-length is 2；
Wherein, adaptive thresholding value function in Opencv:
adaptiveThreshold(src,dst,maxValue,adaptiveMethod,thresholdType,
blockSize,C)；
Parameter declaration:
Src- indicates source (input) image；Dst- indicates target (output) image；The change of mono- double type of maxval-
Amount indicates the pixel value upper limit；AdaptiveMethod- indicates the integer variable of the type of adaptive approach to be used, it can
To be one of following two value；ADAPTIVE_THRESH_MEAN_C- threshold value is the average value of neighborhood；ADAPTIVE_THRESH_
GAUSSIAN_C- threshold value be weight be Gauss window neighborhood value weighted sum；ThresholdType- indicates threshold to be used
Value Types (just, anti-)；BlockSize- indicates the size for calculating the neighborhood of pixels of threshold value；C- constant (from average value or adds
Subtracted in weight average value) double categorical variable.
Threshold method takes ADAPTIVE_THRESH_MEAN_C, i.e. field average value threshold value in the present invention.
BlockSize size takes 11 (N × N), C value range (- 26,20).
Mathematical formulae expression:
1. finding out average gray in field to be avr_gray. and then subtract C, constitute with reference to the average value avr_ compared
newgray；
Avr_newgray=avr_gray-C (7)
2. each pixel is made comparisons with avr_newgray in field, it is greater than then threshold value output 255, otherwise exports 0.
Different C are worth output result as shown in figure 11.
Step c: character block identification: histogram equalization is carried out to character block using equalizeHist algorithm, is then sent
Enter CNN neural network, does 65 kinds of character classification.
Using CNN convolutional neural networks, structure is similar with Handwritten Digit Recognition network structure, and using multilayer convolution, drop is adopted
Sample, full articulamentum are finally completed 65 kinds of character classifications.
Network inputs are 32*14；
First layer uses 3 × 3 convolution, exports 32 characteristic patterns, relu activation primitive layer, 2 × 2 maximum pond layer,
Step-length is 2.
The second layer uses 3 × 3 convolution, exports 64 characteristic patterns, relu activation primitive, 2 × 2 maximum pond layer, step
A length of 2.
Third layer uses 2 × 2 convolution, exports 128 characteristic patterns, relu activation primitive.
Again by two layers of full articulamentum, 256 neurons and 65 neurons are sequentially output, 65 points of character are finally completed
Class.
Characters on license plate category set, wherein including 65 kinds, Chinese character is 31 classes, and english character is 24 classes, number 10.Divided
Class training, obtains final characters on license plate disaggregated model.
Recognition result is as shown in figure 12 with the time.Character recognition accuracy rate is 98%.
As shown in figure 13, the collecting vehicle information identifying system towards intelligent terminal, including intelligent terminal subsystem kimonos
Business device terminal system, intelligent terminal subsystem and service device terminal system pass through transmitted data on network；
Intelligent terminal subsystem includes data acquisition module, Car license recognition module, GPS positioning module, data transmission reception
Module；Data acquisition module, for acquiring vehicle image and shooting time；
Wherein, the system time of shooting time source intelligent terminal subsystem,
Car license recognition module identifies license plate number for carrying out Car license recognition to the vehicle image of acquisition；
GPS positioning module, for being positioned to vehicle；
Data transmit receiving module, for the information of vehicles of smart terminal sub-stream system acquisition to be transferred to server end subsystem
The information of the transmission of system and reception server end subsystem, for sending and receiving car owner's cellphone information；For to server
Terminal system sends query information and receives query result.
Information of vehicles is passed to Car license recognition module by data acquisition module, and Car license recognition module identifies in vehicle image
License plate number, then the license plate number and GPS positioning and original vehicle image and acquisition time that Car license recognition goes out are passed through into number
Server end subsystem is transmitted to according to transmission receiving module.
Server end subsystem includes information receiving module, database module, message processing module and information transfer module；
Information receiving module, the information of vehicles uploaded for receiving storage intelligent terminal subsystem, receives corresponding vehicle
The information of car owner's feedback receives information processing personnel to the processing result of exception information, receives the inquiry of intelligent terminal subsystem
Information；
Database module, for store intelligent terminal subsystem upload acquisition information of vehicles and traffic police put on record vehicle letter
Breath；
Message processing module, the vehicle in Car license recognition information and date library module for transmitting intelligent terminal subsystem
Owner information is matched, and vehicle owner information corresponding with Car license recognition information is found out；For handling abnormal prompt information；
For by the information in information processing personnel treated information update database module；
Information transfer module, for information to be issued to the car owner of corresponding vehicle by short message form, by intelligent terminal subsystem
The information of system inquiry sends intelligent terminal subsystem to.
Wherein, for handling abnormal prompt information the case where includes that 1. car owners have objection to judgement result, then by feedback letter
Breath sends back server end subsystem；2. the information of vehicles in information of vehicles and database module in traffic police's record information table is not
Then server end subsystem carries out abnormal prompt for matching；3. Car license recognition is correct in intelligent terminal subsystem, information of vehicles and number
According to information matches success in library, but under this license plate number it is two vehicles.
Wherein, database module, inside include two tables, are storage acquisition vehicle information table and traffic police's record information respectively
Table；Storage acquisition vehicle information table includes the vehicle image of acquisition, license plate number, acquisition time, acquisition vehicle position；It hands over
Guard case information table includes vehicle image, vehicle license plate number, owner name, gender, car owner's contact method etc..
Vehicle owner information in the Car license recognition information and date library module of intelligent terminal subsystem transmitting is matched,
Exactly storage acquisition vehicle information table and traffic police's record information table to be matched, successful matching then obtains car owner's contact method,
It matches unsuccessful, then carries out abnormal prompt.
The information receiving module of server end subsystem receives the information of vehicles that intelligent terminal subsystem uploads, and information receives
Module is stored in information of vehicles is acquired in the storage acquisition vehicle information table of database module, and message processing module is by database
In traffic police's record information table in the acquisition license plate number and database module in storage acquisition vehicle information table in module
License plate number carries out pairing lookup, finds corresponding owner information, sends information transfer module for owner information, information transmits mould
Information of vehicles is issued car owner by block, and car owner determines information of vehicles: if car owner to determine result have no objection, it is specified that
It is indicated in time without feedback information, then updates database module information；If car owner has objection to judgement result, by feedback letter
Breath sends back server end subsystem, and the information receiving module of server end subsystem receives car owner's feedback information, and information receives
Module passes to message processing module, and message processing module is transmitted to information transfer module, and information transfer module is sent to information
Treatment people, for information processing personnel again by from the data newly verified, feedback result sends final information to server end subsystem
System, and final information is sent to car owner, the database module of server end subsystem more new database, last process terminates.
The server if the information of vehicles in information of vehicles and database module in traffic police's record information table mismatches
Terminal system carries out abnormal prompt, and sends information to information processing personnel to dealing of abnormal data, and information processing personnel are logical
It crosses original information of vehicles and carries out artificial treatment abnormal data, information is inconsistent, and the information is marked, passes server subsystem back,
Update the corresponding informance of database module；Information unanimously then verifies the data of acquisition again, and server end subsystem determines again
Feedback result meets acquisition reason and then sends final information to car owner and more new database, do not meet acquisition reason and then send most
Whole information is to car owner and more new database.
As shown in figure 14, the information collection process of whole system, comprises the concrete steps that:
Intelligent terminal subsystem carries out collecting vehicle information, shows vehicle-related information and acquisition in intelligent terminal subsystem
Information of vehicles is transmitted to server end subsystem by reason, intelligent terminal subsystem, server end subsystem by information of vehicles with
Database information pairing sends information of vehicles to car owner if successful matching obtains the contact method of car owner, car owner to information into
Row judgement, it is believed that without mistake, it is specified that time in without feedback information expression, then server end subsystem more new database；Such as
Fruit information of vehicles and database information pairing are unsuccessful, then carry out abnormal prompt, and server end subsystem prompting message handles people
Member carries out artificial treatment abnormal data to dealing of abnormal data, then sends back server end subsystem, and server end subsystem is logical
It is whether consistent in data base querying information of vehicles to cross license plate number, it is as a result correct then to verify acquisition data again, feedback result is sentenced
Disconnected correctness simultaneously sends final information to car owner, more new database；If server end subsystem is by license plate number in data
The incorrect then mark information of library inquiry information of vehicles, more new database；
If car owner thinks that information gets the wrong sow by the ear, information is fed back into server end subsystem, server end subsystem
Abnormal prompt is generated, information processing personnel carry out artificial treatment abnormal data to dealing of abnormal data, to returning after abnormality processing
Return server end subsystem, and more new database.
Wherein, there are following several situations in abnormality processing at this time,
(1) Car license recognition is correct in intelligent terminal subsystem, but without this information of vehicles in server end subsystem database,
To need the data marked.
(2) Car license recognition is correct in intelligent terminal subsystem, information matches success in information of vehicles and database, but this
It is two vehicles under license plate number, this information is the data for needing to mark, similar to (1).
(3) Car license recognition is correct in intelligent terminal subsystem, information matches success, car owner couple in information of vehicles and database
This acquisition reason is not accepted, and is query information.
Information processing personnel are simply divided into three kinds to exception information processing；Information of vehicles mismatches, and needs flag data,
Wrong data is acquired, manual identified is carried out to license plate first, by correct license plate number in server end subsystem database
In information of vehicles is inquired, check whether vehicle consistent, inconsistent data includes, and database is handed over without this information of vehicles, database
Guard case information of vehicles is not met with acquisition information of vehicles, this data of information processing worker labels, and more new database；Data one
In the case of cause, as feedback operation, information processing personnel are to acquisition information Authenticate afresh, according to former vehicle-related information (vehicle
Image, vehicle position information, vehicle acquisition time) Authenticate afresh, send result to car owner and more new database.
For Car license recognition module:
As shown in figure 15, the Car license recognition module includes license plate experience region setup module, License Plate module, vehicle
Board correction module and license plate number identification module；
License plate experience region setup module, for carrying out resolution adjustment and cutting to the original vehicle image of acquisition；
License Plate module, for carrying out license plate Primary Location using SSD algorithm of target detection；
VLP correction module obtains horizontal license plate for carrying out projective transformation and rotation correction to the license plate after positioning；
License plate number identification module is used to carry out license plate image Character segmentation to horizontal license plate, then carries out character recognition, obtains
Take target license plate information.
Wherein, license plate experience region setup module includes:
Image resolution ratio adjustment unit, for being adjusted to the resolution ratio for obtaining image；
Image cropping unit is cut for original vehicle image.
License Plate module includes:
License plate position extraction unit carries out SSD algorithm of target detection to the vehicle image after cutting, first carries out compression number
According to being re-introduced into convolutional layer and obtain various sizes of characteristic pattern, last rear end carries out feature extraction, obtains license plate image and position is believed
Breath.
VLP correction module includes:
VLP correction unit: projective transformation and rotation correction are carried out to the image after positioning.
License plate number identification module includes:
License plate removes pixel unit, for removing block of pixels extra in license plate image；
License Plate Character Segmentation unit is isolated the biggish character zone of contrast, is found two using adaptive threshold limit value
It is worth all profiles of image, all profiles is irised out with rectangle frame, then by the ratio of width to height, area, which filters out, meets character feature
Region；
The license plate image for removing extra block of pixels is adjusted to wide 136 pixel, high 36 pixel size.It is adaptive using domain is faced
Threshold is answered, the screening of character zone is completed, finds the profile of bianry image character, outlines each word with minimum circumscribed rectangle
Symbol, by Feature Selection, specific wide and high, the ratio of Gao Yukuan filters out the rectangle frame for meeting character feature, only filters out
Six characters other than Chinese character.Chinese character subsequent processing is recorded to rectangle frame according to sequence from left to right
Rectangle frame coordinate, size parameter.
The segmentation of Chinese character, due to Chinese character form of diverse, such as character " river ", " Shanxi ", regional area separation
It is larger, it is easy that three vertical lines, or " Shanxi " is divided into break up the family up and down, becomes 2 characters or 3 characters, solution is pair at this time
Chinese character region uses longitudinal dilatation and lateral expansion, keeps Connection operator complete, then go to seek profile and rectangle frame, filters out
Chinese character region.
By 7 character zones, permutation and combination is finished from left to right, completes the segmentation of character.
Recognition of License Plate Characters unit is then fed into CNN for carrying out histogram equalization to character block using special algorithm
Neural network does 65 kinds of character classification, identifies each characters on license plate.Special algorithm is using equalizeHist function to word
It accords with block and carries out histogram equalization.
As shown in figure 16, the information querying flow figure of intelligent terminal subsystem:
Information collection personnel point opens the APP in intelligent terminal subsystem, starts to be inquired；
Firstly, input inquiry condition, querying condition may is that license plate number, time, position etc.；
Then, querying condition is inputted, querying condition is uploaded to server end subsystem by intelligent terminal subsystem, service
Device terminal system screens information, then match information is exported, information back to smart phone terminal system, smart phone
Terminal system shows final information；
Finally, querying flow terminates.
As shown in figure 14, for intelligent terminal selection smart phone:
Intelligent identifying system of the invention is embedded into smart phone, specifically used steps are as follows:
S1, information collection personnel are acquired data to be collected by smart phone, and identify license plate number；
S2, mobile phone end obtain camera site by GPS positioning；
S3, data (including vehicle image, location information, license plate number, shooting time acquire reason) transmission by acquisition
To server end subsystem；
S4, server end subsystem obtain car owner's contact method by license plate number, and send collected vehicle letter
Breath；
After S5, car owner receive information, if car owner is without demur to this information, indicate to receive secondary information of vehicles, server
Terminal system more new database；
If S6, car owner can raise an objection to this information objection by feedback system, after receiving feedback, server end
Subsystem generates abnormal prompt and car owner's objection information is issued information processing personnel, information processing personnel to dealing of abnormal data,
Artificial treatment abnormal data is carried out, to returning to server end subsystem, and more new database after abnormality processing；
If information of vehicles matches unsuccessful with database information in step S5, abnormal prompt, server terminal are carried out
System alert information processing personnel carry out artificial treatment abnormal data to dealing of abnormal data, then send back server end subsystem
System, whether server end subsystem is consistent in data base querying information of vehicles by license plate number, and as a result correct then verify again is adopted
Collect data, whether to feedback result correct judgment and sends final information to car owner, more new database；If server end subsystem
System is by license plate number in the incorrect then mark information of data base querying information of vehicles, more new database.
S7, car owner is sent to feedback result；
S8, process terminate.
Identify that the detailed process of license plate number uses the collecting vehicle information of the invention towards intelligent terminal in step S1
Recognition methods is first adjusted the vehicle image of acquisition, then carries out license plate Primary Location with SSD algorithm of target detection, right
Image after positioning carries out projective transformation and rotation correction, obtains horizontal license plate, is accurately positioned to horizontal license plate, right later
License plate image Character segmentation, then character recognition is carried out, obtain final license plate recognition result.
By Car license recognition, license plate number is obtained, vehicle image acquisition has been completed, by license plate number, has been recorded in mobile phone
Client database.
The location information of vehicle is recorded in mobile phone by GPS positioning and time clock feature by the mobile phone end in step S2
The database at end can accomplish accurate positioning function, as shown in figure 17 be map GPS parameter selection figure；
Wherein,
Low-power consumption mode:
GPS is closed, network is opened: obtaining total data；
GPS is opened, network is opened: GPS occurred once, other are network positions.
Only equipment mode:
GPS is closed, network is closed: no data；
GPS is closed, network is opened: most of no datas；Occasional uses network positions, obtains total data；
GPS is opened, network is opened: most of to use GPS positioning, occasional network positions can obtain total data；
GPS is opened, network is closed: only using GPS positioning, there is longitude and latitude data, zero-address data.
High-precision fixed bit pattern:
GPS is opened, network is opened: can be using GPS, no satellite Shi Huiyong network when having satellite；
Locating periodically；Such as this is set as 2000 milliseconds.2000 milliseconds are represented to relocate once；
Network timeout；Such as this is set as 30000 milliseconds, representing 30000 milliseconds, to represent data network unavailable；
Single positioning；Only do primary positioning；
GPS is preferential；GPS and network select GPS when existing simultaneously；
Inverse geocoding；This coding mode can navigate to street number, such as " Shenyang City, Shen Liaoning Western Road, Tiexi District 111
Number "；
Open caching；Record position information；
Improve positioning accuracy for the first time；Improve the precision of positioning for the first time；
Use sensor；By receiving the satellite-signal in space, by satellite conjunction, to the current location of shooting image
It is positioned.
Step S3 is specifically by license plate number and vehicle image, and stand information and temporal information compression are packaged, transmission
To server end subsystem, in server end subsystem by data decompression, this purpose is to reduce data transfer throughput, saves transmission
Time saves time cost.
After server receives data, clothes are stored in again by license plate number as this data major key to data decompression
It is engaged in device terminal system database.
Step S4 obtains car owner's contact method, car owner, place, vehicle is notified by way of short message by license plate number
Information collection time, license plate number acquire information of vehicles reason, inform feedback operation mode.
If step S5 car owner is without demur to this information, the collected information of vehicles reason for receiving oneself, the vehicle are indicated
Information is by the database of typing server end subsystem.
If step S6 car owner to this information objection, can give in a feedback manner, after receiving feedback, server end subsystem
Prompting message treatment people, information processing personnel's artificial treatment abnormal data, verifies data, again with server end subsystem
GPS and image information in the database of storage is as foundation, if it is determined that and it is former as a result, to notify that car owner's original sentence determines result correct,
This information of vehicles is transmitted to the database of server end subsystem, more new database；If negating former data, by feedback result
Car owner is fed back to, and more new database, process terminate.
The APP of the intelligence system can acquire information of vehicles, and system can be with the license plate of Quick Test Vehicle, and accurately identifies
Number out shows result.System is 445 milliseconds in the average time for completing whole flow process, can accomplish real-time, character completely
Accuracy of identification is 98%, which greatly improves working efficiency for auxiliary information collector's data acquisition.
Claims (10)
1. the collecting vehicle information recognition methods towards intelligent terminal, it is characterised in that: the step of this method is as follows:
Step 1: the vehicle image of acquisition is adjusted；
Step 2: license plate Primary Location is carried out with SSD algorithm of target detection to the image that step 1 obtains；
Step 3: projective transformation and rotation correction are carried out to the image after positioning, obtain horizontal license plate；
Step 4: being accurately positioned horizontal license plate, later to license plate image Character segmentation, then carries out character recognition；
Step 5: final license plate recognition result is obtained.
2. the collecting vehicle information recognition methods according to claim 1 towards intelligent terminal, it is characterised in that: step 2
Specifically includes the following steps:
Step A: experience region is arranged according to license plate position to acquired image, reduces size of data；
Step B: to treated, image uses SSD algorithm to carry out car plate detection, and SSD algorithm network structure is divided into two parts,
Front end carries out compressed data, and rear end carries out feature extraction, introduces multiple dimensioned convolutional layer and obtains various sizes of characteristic pattern, detects more
Scale license plate.
3. the collecting vehicle information recognition methods according to claim 1 towards intelligent terminal, it is characterised in that: step 4
Specific step is as follows:
Step a: block of pixels extra in license plate image after removal correction obtains being accurately positioned license plate position；
Step b: license plate image is divided into character block to be identified again: using domain adaptive threshold is faced, searches out the wheel of character
Exterior feature obtains minimum circumscribed rectangle using function and profile is irised out；
Step c: character block identification: histogram equalization is carried out to character block using special algorithm, is then fed into CNN nerve net
Network does 65 kinds of character classification.
4. the collecting vehicle information identifying system towards intelligent terminal, it is characterised in that: including intelligent terminal subsystem and service
Device terminal system, intelligent terminal subsystem and service device terminal system pass through transmitted data on network；
Intelligent terminal subsystem includes data acquisition module, Car license recognition module, GPS positioning module, data transmission receiving module；
Data acquisition module, for acquiring vehicle image and shooting time；
Car license recognition module identifies license plate number for carrying out Car license recognition to the vehicle image of acquisition；
GPS positioning module, for being positioned to vehicle；
Data transmit receiving module, for by the information of vehicles of smart terminal sub-stream system acquisition be transferred to server end subsystem with
And receive the information of the transmission of server end subsystem；For sending and receiving car owner's cellphone information；For to server terminal
System sends query information and receives query result.
5. the collecting vehicle information identifying system according to claim 4 towards intelligent terminal, it is characterised in that: server
Terminal system includes information receiving module, database module, message processing module and information transfer module；
Information receiving module, the information of vehicles uploaded for receiving storage intelligent terminal subsystem, receives the car owner of corresponding vehicle
The information of feedback receives information processing personnel to the processing result of exception information, receives the query information of intelligent terminal subsystem；
Database module, acquisition information of vehicles and traffic police for storing the upload of intelligent terminal subsystem are put on record information of vehicles；
Message processing module, the vehicle vehicle in Car license recognition information and date library module for transmitting intelligent terminal subsystem
Main information is matched, and vehicle owner information corresponding with Car license recognition information is found out；For handling abnormal prompt information；For
By the information in information processing personnel treated information update database module；
Information transfer module looks into intelligent terminal subsystem for information to be issued to the car owner of corresponding vehicle by short message form
The information of inquiry sends intelligent terminal subsystem to.
6. the collecting vehicle information identifying system according to claim 4 towards intelligent terminal, it is characterised in that: described
Car license recognition module includes license plate experience region setup module, License Plate module, VLP correction module and license plate number identification
Module；
License plate experience region setup module, for carrying out resolution adjustment and cutting to the original vehicle image of acquisition；
License Plate module, for carrying out license plate Primary Location using SSD algorithm of target detection；
VLP correction module obtains horizontal license plate for carrying out projective transformation and rotation correction to the license plate after positioning；
License plate number identification module is used to carry out license plate image Character segmentation to horizontal license plate, then carries out character recognition, obtains mesh
Mark license board information.
7. the collecting vehicle information identifying system according to claim 6 towards intelligent terminal, it is characterised in that: license plate warp
Testing region setup module includes:
Image resolution ratio adjustment unit, for being adjusted to the resolution ratio for obtaining image；
Image cropping unit is cut for original vehicle image.
8. the collecting vehicle information identifying system according to claim 6 towards intelligent terminal, it is characterised in that: license plate is fixed
Position module include:
License plate position extraction unit carries out SSD algorithm of target detection to the vehicle image after cutting, first carries out compressed data, then
It introduces convolutional layer and obtains various sizes of characteristic pattern, last rear end carries out feature extraction, obtains license plate image and location information.
9. the collecting vehicle information identifying system according to claim 6 towards intelligent terminal, it is characterised in that:
VLP correction module includes:
VLP correction unit: projective transformation and rotation correction are carried out to the image after positioning.
10. the collecting vehicle information identifying system according to claim 6 towards intelligent terminal, it is characterised in that: license plate
Number identification module includes:
License plate removes pixel unit, for removing block of pixels extra in license plate image；
License Plate Character Segmentation unit is isolated the biggish character zone of contrast, is found binary map using adaptive threshold limit value
As all profiles, all profiles are irised out with rectangle frame, then by the ratio of width to height, area filters out the area for meeting character feature
Domain；
Recognition of License Plate Characters unit is then fed into CNN nerve for carrying out histogram equalization to character block using special algorithm
Network does 65 kinds of character classification, identifies each characters on license plate.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811281498.8A CN109447069A (en) | 2018-10-31 | 2018-10-31 | Collecting vehicle information recognition methods and system towards intelligent terminal |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811281498.8A CN109447069A (en) | 2018-10-31 | 2018-10-31 | Collecting vehicle information recognition methods and system towards intelligent terminal |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109447069A true CN109447069A (en) | 2019-03-08 |
Family
ID=65549926
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811281498.8A Pending CN109447069A (en) | 2018-10-31 | 2018-10-31 | Collecting vehicle information recognition methods and system towards intelligent terminal |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109447069A (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110176011A (en) * | 2019-05-28 | 2019-08-27 | 北京百度网讯科技有限公司 | Vehicle abnormality based reminding method and device |
| CN110390330A (en) * | 2019-07-25 | 2019-10-29 | 网链科技集团有限公司 | Electric bicycle license plate identifying system and method |
| CN110414574A (en) * | 2019-07-10 | 2019-11-05 | 厦门美图之家科技有限公司 | A kind of object detection method calculates equipment and storage medium |
| CN110555371A (en) * | 2019-07-19 | 2019-12-10 | 华瑞新智科技（北京）有限公司 | Wild animal information acquisition method and device based on unmanned aerial vehicle |
| CN110852324A (en) * | 2019-08-23 | 2020-02-28 | 上海撬动网络科技有限公司 | Deep neural network-based container number detection method |
| CN111435445A (en) * | 2019-12-24 | 2020-07-21 | 珠海大横琴科技发展有限公司 | Training method and device of character recognition model and character recognition method and device |
| CN111898497A (en) * | 2020-07-16 | 2020-11-06 | 济南博观智能科技有限公司 | License plate detection method, system, equipment and readable storage medium |
| CN112908017A (en) * | 2019-11-19 | 2021-06-04 | 北京沃东天骏信息技术有限公司 | Method, device and equipment for detecting vehicle parking state |
| CN112950954A (en) * | 2021-02-24 | 2021-06-11 | 电子科技大学 | Intelligent parking license plate recognition method based on high-position camera |
| CN114627653A (en) * | 2022-05-12 | 2022-06-14 | 浙江电马云车科技有限公司 | 5G intelligent barrier gate management system based on binocular recognition |
| CN114934467A (en) * | 2022-07-08 | 2022-08-23 | 江苏永达电力金具有限公司 | Parking space barrier gate control method, parking space barrier gate system and medium |
| CN115035573A (en) * | 2022-05-27 | 2022-09-09 | 哈尔滨工程大学 | Lip segmentation method based on fusion strategy |
| CN117152732A (en) * | 2023-10-18 | 2023-12-01 | 北京精英智通科技股份有限公司 | Multi-feature-assisted license plate recognition method and system |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3182334A1 (en) * | 2015-12-17 | 2017-06-21 | Xerox Corporation | License plate recognition using coarse-to-fine cascade adaptations of convolutional neural networks |
| CN107563372A (en) * | 2017-07-20 | 2018-01-09 | 济南中维世纪科技有限公司 | A kind of license plate locating method based on deep learning SSD frameworks |
| CN107679531A (en) * | 2017-06-23 | 2018-02-09 | 平安科技（深圳）有限公司 | Licence plate recognition method, device, equipment and storage medium based on deep learning |
| CN108009526A (en) * | 2017-12-25 | 2018-05-08 | 西北工业大学 | A kind of vehicle identification and detection method based on convolutional neural networks |
| CN108510750A (en) * | 2018-04-25 | 2018-09-07 | 济南浪潮高新科技投资发展有限公司 | A method of the unmanned plane inspection parking offense based on neural network model |
-
2018
- 2018-10-31 CN CN201811281498.8A patent/CN109447069A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3182334A1 (en) * | 2015-12-17 | 2017-06-21 | Xerox Corporation | License plate recognition using coarse-to-fine cascade adaptations of convolutional neural networks |
| CN107679531A (en) * | 2017-06-23 | 2018-02-09 | 平安科技（深圳）有限公司 | Licence plate recognition method, device, equipment and storage medium based on deep learning |
| CN107563372A (en) * | 2017-07-20 | 2018-01-09 | 济南中维世纪科技有限公司 | A kind of license plate locating method based on deep learning SSD frameworks |
| CN108009526A (en) * | 2017-12-25 | 2018-05-08 | 西北工业大学 | A kind of vehicle identification and detection method based on convolutional neural networks |
| CN108510750A (en) * | 2018-04-25 | 2018-09-07 | 济南浪潮高新科技投资发展有限公司 | A method of the unmanned plane inspection parking offense based on neural network model |
Non-Patent Citations (2)
| Title |
|---|
| EINSTEIN_LIU: "实现MobileNet-SSD网络并进行优化", 《HTTPS://BLOG.CSDN.NET/LIU_XIAO_CHENG/ARTICLE/DETAILS/82781082》 * |
| 卜凯: "基于Android平台的车牌识别系统的研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110176011A (en) * | 2019-05-28 | 2019-08-27 | 北京百度网讯科技有限公司 | Vehicle abnormality based reminding method and device |
| CN110414574A (en) * | 2019-07-10 | 2019-11-05 | 厦门美图之家科技有限公司 | A kind of object detection method calculates equipment and storage medium |
| CN110555371A (en) * | 2019-07-19 | 2019-12-10 | 华瑞新智科技（北京）有限公司 | Wild animal information acquisition method and device based on unmanned aerial vehicle |
| CN110390330A (en) * | 2019-07-25 | 2019-10-29 | 网链科技集团有限公司 | Electric bicycle license plate identifying system and method |
| CN110852324A (en) * | 2019-08-23 | 2020-02-28 | 上海撬动网络科技有限公司 | Deep neural network-based container number detection method |
| CN112908017A (en) * | 2019-11-19 | 2021-06-04 | 北京沃东天骏信息技术有限公司 | Method, device and equipment for detecting vehicle parking state |
| CN111435445A (en) * | 2019-12-24 | 2020-07-21 | 珠海大横琴科技发展有限公司 | Training method and device of character recognition model and character recognition method and device |
| CN111898497B (en) * | 2020-07-16 | 2024-05-10 | 济南博观智能科技有限公司 | License plate detection method, system, device and readable storage medium |
| CN111898497A (en) * | 2020-07-16 | 2020-11-06 | 济南博观智能科技有限公司 | License plate detection method, system, equipment and readable storage medium |
| CN112950954A (en) * | 2021-02-24 | 2021-06-11 | 电子科技大学 | Intelligent parking license plate recognition method based on high-position camera |
| CN112950954B (en) * | 2021-02-24 | 2022-05-20 | 电子科技大学 | Intelligent parking license plate recognition method based on high-position camera |
| CN114627653A (en) * | 2022-05-12 | 2022-06-14 | 浙江电马云车科技有限公司 | 5G intelligent barrier gate management system based on binocular recognition |
| CN114627653B (en) * | 2022-05-12 | 2022-08-02 | 浙江电马云车科技有限公司 | 5G intelligent barrier gate management system based on binocular recognition |
| CN115035573A (en) * | 2022-05-27 | 2022-09-09 | 哈尔滨工程大学 | Lip segmentation method based on fusion strategy |
| CN114934467A (en) * | 2022-07-08 | 2022-08-23 | 江苏永达电力金具有限公司 | Parking space barrier gate control method, parking space barrier gate system and medium |
| CN114934467B (en) * | 2022-07-08 | 2024-04-30 | 江苏永达电力金具有限公司 | Parking space barrier control method, parking space barrier system and medium |
| CN117152732A (en) * | 2023-10-18 | 2023-12-01 | 北京精英智通科技股份有限公司 | Multi-feature-assisted license plate recognition method and system |
| CN117152732B (en) * | 2023-10-18 | 2024-02-27 | 北京精英智通科技股份有限公司 | Multi-feature-assisted license plate recognition method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109447069A (en) | Collecting vehicle information recognition methods and system towards intelligent terminal | |
| CN104700099B (en) | The method and apparatus for recognizing traffic sign | |
| CN106778595B (en) | Method for detecting abnormal behaviors in crowd based on Gaussian mixture model | |
| CN108491797A (en) | A kind of vehicle image precise search method based on big data | |
| CN108898047B (en) | Pedestrian detection method and system based on blocking and shielding perception | |
| CN112308092A (en) | Light-weight license plate detection and identification method based on multi-scale attention mechanism | |
| CN105608456B (en) | A kind of multi-direction Method for text detection based on full convolutional network | |
| CN107480649B (en) | Fingerprint sweat pore extraction method based on full convolution neural network | |
| CN111898523A (en) | Remote sensing image special vehicle target detection method based on transfer learning | |
| CN107704857A (en) | A kind of lightweight licence plate recognition method and device end to end | |
| CN107944450A (en) | A kind of licence plate recognition method and device | |
| CN108268871A (en) | A kind of licence plate recognition method end to end and system based on convolutional neural networks | |
| CN110992366B (en) | Image semantic segmentation method, device and storage medium | |
| CN110516633A (en) | A kind of method for detecting lane lines and system based on deep learning | |
| CN114943893B (en) | Feature enhancement method for land coverage classification | |
| CN110838230B (en) | Mobile video monitoring method, monitoring center and system | |
| CN112668375B (en) | Tourist distribution analysis system and method in scenic spot | |
| CN111008574A (en) | Key person track analysis method based on body shape recognition technology | |
| CN112488046A (en) | Lane line extraction method based on high-resolution images of unmanned aerial vehicle | |
| CN107944354A (en) | A kind of vehicle checking method based on deep learning | |
| CN109086765B (en) | Licence plate recognition method, device, medium, server and automobile data recorder | |
| CN109800756A (en) | A kind of text detection recognition methods for the intensive text of Chinese historical document | |
| CN111626357B (en) | Image identification method based on neural network model | |
| CN115497015A (en) | River floating pollutant identification method based on convolutional neural network | |
| CN115457277A (en) | Intelligent pavement disease identification and detection method and system |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190308 |