CN111461113A - Large-angle license plate detection method based on deformed plane object detection network - Google Patents

Large-angle license plate detection method based on deformed plane object detection network Download PDF

Info

Publication number
CN111461113A
CN111461113A CN202010138881.9A CN202010138881A CN111461113A CN 111461113 A CN111461113 A CN 111461113A CN 202010138881 A CN202010138881 A CN 202010138881A CN 111461113 A CN111461113 A CN 111461113A
Authority
CN
China
Prior art keywords
license plate
convolution
feature map
layer
network
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
Application number
CN202010138881.9A
Other languages
Chinese (zh)
Other versions
CN111461113B (en
Inventor
谢巍
王缙
张浪文
潘春文
廉胤东
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
South China University of Technology SCUT
Original Assignee
South China University of Technology SCUT
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by South China University of Technology SCUT filed Critical South China University of Technology SCUT
Priority to CN202010138881.9A priority Critical patent/CN111461113B/en
Publication of CN111461113A publication Critical patent/CN111461113A/en
Application granted granted Critical
Publication of CN111461113B publication Critical patent/CN111461113B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • 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
    • G06V20/63—Scene text, e.g. street names
    • 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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • 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
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
    • G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
    • G06V10/451—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters with interaction between the filter responses, e.g. cortical complex cells
    • G06V10/454—Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]
    • 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)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Biophysics (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Biodiversity & Conservation Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a large-angle license plate detection method based on a deformed planar object detection network, which solves the technical problem that the accuracy rate is reduced when the existing license plate detection method identifies a large-angle license plate.A deformed planar object detection network provided by the invention directly detects four corner coordinates of the license plate, then carries out affine transformation on a license plate picture, corrects the inclined license plate into a positive-angle license plate, and finally carries out character recognition on the corrected positive-angle license plate.

Description

Large-angle license plate detection method based on deformed plane object detection network
Technical Field
The invention relates to the technical field of computer vision and mode recognition, in particular to a large-angle license plate detection method based on a deformed plane object detection network.
Background
Currently, license plate detection is one of the hot research problems in the field of computer vision and pattern recognition. In many license plate detection tasks, the detected license plate is often seriously inclined, the license plate which is seriously inclined is called a large-angle license plate, and the subsequent license plate identification is influenced by the license plate which is seriously inclined, so that how to identify the large-angle license plate becomes a key point and a difficulty in the field of license plate identification.
The detection method provided by Jaderberg divides the license plate identification into two parts, firstly uses positioning to replace segmentation to position the position of each character, and then classifies each character to identify the license plate.
Disclosure of Invention
The invention aims to solve the defects in the prior art and provides a large-angle license plate detection method based on a deformed planar object detection network
The purpose of the invention can be achieved by adopting the following technical scheme:
a large-angle license plate detection method based on a deformed planar object detection network is disclosed, wherein the deformed planar object detection network is formed by sequentially connecting a backbone network and a prediction network, and the large-angle license plate detection method comprises the following steps:
s1, constructing a backbone network for large-angle license plate detection by using a convolution layer, a pooling layer and a Relu layer, and taking the backbone network as a feature extractor of a license plate image, wherein the size of a convolution kernel in the convolution layer is k x k, and feature extraction is carried out on the input license plate image by adopting the backbone network to obtain a feature map;
s2, constructing a prediction network, wherein the prediction network is provided with two parallel convolution layers which are respectively called a first convolution layer and a second convolution layer, 2 m-m convolution kernels are used for performing sliding calculation on a feature map output by a main network in the first convolution layer to obtain a first example feature map, m is the size of a convolution kernel, the first example feature map is subjected to Softmax activation function to obtain a confidence coefficient feature map, 6 n-n convolution kernels are used for performing sliding calculation on the feature map output by the main network in the second convolution layer to obtain a second example feature map, n is the size of the convolution kernel, the second example feature map is subjected to L initial activation function to obtain a confidence coefficient feature map, and finally the affine feature map and the confidence coefficient feature map are subjected to splicing operation on channel dimensions to obtain a prediction feature map FTWherein, the predicted feature map FTContaining 8 channels, the value in 8 channels is defined as viI is 1 to 8, wherein v1,v2V representing the confidence that the prediction box contains and does not contain an object, respectively3～8Respectively representing affine matrix elements, and partially decoupling confidence prediction and affine parameter prediction by designing two parallel convolution layers in a prediction network, so that the accuracy of the prediction network is improved;
s3, designing a loss function to train the backbone network and the prediction network;
s4, using the prediction feature map FTFour virtual corner points q are constructed with the (m, n) point unit as the centeriAnd i is 1-4, wherein m and n respectively represent a prediction characteristic diagram FTThe horizontal coordinates and the vertical coordinates of the midpoint unit are obtained, and then affine transformation is carried out on the four virtual corner points to obtain coordinates l of four target prediction corner pointsi,i＝1～4；
S5, according to the coordinates l of the prediction corner pointiSolving the transmission transformation equation set to obtain a homography matrix, correcting the image according to the homography matrix, and utilizing v1、v2And eliminating the corrected image with low confidence coefficient to obtain the final detection result.
Further, the specific structure of the backbone network is as follows:
the input layer is connected with the output layer in sequence as follows: convolution layer conv with convolution kernel size of 3x3 and number of 16; relu layer conv _ Relu; convolution layer conv with convolution kernel size of 3x3 and number of 16; relu layer conv _ Relu; pooling layer max _ pooling; convolution layer conv with convolution kernel size of 3x3 and number of 32; relu layer conv _ Relu; pooling layer max _ pooling; convolution layer conv with convolution kernel size of 3x3 and number of 32; relu layer conv _ Relu; pooling layer max _ pooling; convolution layer conv with convolution kernel size of 3x3 and number of 64; relu layer conv _ Relu, pooling layer max _ pooling; the convolution kernel size is 3 × 3, 128 convolution layers conv.
The trunk network completely adopts convolution kernels with the size of 3x3, and on the premise of obtaining the same receptive field size, a plurality of convolution kernels with the size of 3x3 have less parameter quantity and more nonlinearity than one convolution layer with the large size, so that the fitting capacity of the deformed plane object detection network is stronger, and the parameter quantity of the deformed plane object detection network is less.
Further, the loss function is composed of two parts, the first part is a corner coordinate distance loss function, the second part is a confidence coefficient loss function, wherein the calculation process of the corner coordinate distance loss function is as follows:
carrying out affine transformation on the virtual corner points:
Figure BDA0002398311750000031
the virtual corner points are affine transformed instead of transmission transformed because division operations involved in transmission transformation may generate smaller values in denominators, which in turn leads to unstable values.
For four real angular points p of license plateiCentralizing to obtain centralized coordinate Cmn(pi)：
Figure BDA0002398311750000032
Wherein N issRepresenting the down-sampling multiple of the deformed planar object detection network,
calculating true corner pointsCoordinate piAnd Vmn(qi) Distance error between the two points to obtain a corner coordinate distance loss function:
Figure BDA0002398311750000041
wherein, the confidence coefficient loss function calculation formula is as follows:
Figure BDA0002398311750000042
wherein the content of the first and second substances,
Figure BDA0002398311750000043
is an objective indication function for indicating a predicted feature map FTWhether the (m, n) cell contains a target;
and a cross entropy loss function is adopted in the confidence coefficient loss function, so that the training difficulty of the deformed plane object detection network is reduced.
The loss function is obtained by adding a corner coordinate distance loss function and a confidence coefficient loss function, and the expression is as follows:
Figure BDA0002398311750000044
further, in step S5, the coordinates of the prediction corner point are transformed by using a transmission transform to obtain a corrected image, which includes the following steps:
firstly, solving an affine equation set for k, k is more than or equal to 4 groups of non-collinear matching characteristic points to obtain a homography matrix H, then carrying out transmission transformation on an original image according to the homography matrix to obtain a corrected image, and cutting a target area according to a prediction coordinate.
Compared with the prior art, the invention has the following advantages and effects:
1. the large-angle license plate detection method based on the deformed planar object detection network is formed by improving the YO L O method and the STN method, has the characteristics of license plate correction and high running speed, and can greatly improve the running speed on terminal equipment
2. The license plate detection method disclosed by the invention has the advantages that the four corner coordinates of the large-angle license plate are directly returned through constructing a deformed plane object detection network, then the transmission transformation is carried out on the large-angle license plate, the large-angle license plate is corrected into a positive-angle license plate, and the license plate identification accuracy under various non-limited scenes is effectively improved.
Drawings
FIG. 1 is a flow chart of a large-angle license plate detection method based on a deformed planar object detection network disclosed by the embodiment of the invention;
FIG. 2 is a diagram of exemplary processing steps of a large-angle license plate detection method based on a deformed planar object detection network according to an embodiment of the present disclosure;
FIG. 3 is a diagram of a backbone network structure of a deformed planar object detection network according to an embodiment of the present invention;
fig. 4 is a diagram of a prediction network structure in a deformed planar object detection network according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Examples
The embodiment discloses a large-angle license plate detection method based on a deformed plane object detection network, which specifically comprises the following steps:
s1, the deformation plane object detection network is composed of a trunk network and a prediction network in series connection, the trunk network for large-angle license plate detection is constructed by using a convolution layer, a pooling layer and a Relu layer and is used as a feature extractor of a license plate image, wherein the size of a convolution kernel in the convolution layer is 3x3, the trunk network performs feature extraction on an input license plate image to obtain a trunk network output feature map, and the specific structure of the trunk network is as follows:
the input layer is connected with the output layer in sequence as follows: convolution layer conv with convolution kernel size of 3x3 and number of 16; relu layer conv _ Relu; convolution layer conv with convolution kernel size of 3x3 and number of 16; relu layer conv _ Relu; pooling layer max _ pooling; convolution layer conv with convolution kernel size of 3x3 and number of 32; relu layer conv _ Relu; pooling layer max _ pooling; convolution layer conv with convolution kernel size of 3x3 and number of 32; relu layer conv _ Relu; pooling layer max _ pooling; convolution layer conv with convolution kernel size of 3x3 and number of 64; relu layer conv _ Relu, pooling layer max _ pooling; the convolution kernel size is 3 × 3, 128 convolution layers conv.
S2, constructing a prediction network, wherein the prediction network is provided with two parallel convolution layers which are respectively called a first convolution layer and a second convolution layer, (i) in the first convolution layer, two m × m convolution kernels are used for performing sliding calculation on a feature map output by a main network to obtain a first example feature map, wherein m is the size of a convolution kernel, the first example feature map is subjected to Softmax activation function to obtain a confidence coefficient feature map, and (ii) in the second convolution layer, six n × n convolution kernels are used for performing sliding calculation on the feature map output by the main network to obtain a second example feature map, wherein n is the size of the convolution kernel, the second example feature map is subjected to L initial activation function to obtain an affine feature map, and finally the affine feature map and the confidence coefficient feature map are subjected to splicing operation on channel dimensions to obtain a final prediction feature map FT. Predicted feature map FTContaining 8 channels, the value in 8 channels is defined as viI is 1 to 8, wherein v1,v2V representing the confidence that the prediction box contains and does not contain an object, respectively3～8Respectively representing affine matrix elements, the function of the prediction network being to partially decouple the confidence prediction from the affine parameter prediction.
And S3, training the backbone network and the prediction network by using the training loss function.
The loss function is composed of two parts, the first part is a corner coordinate distance loss function, and the second part is a confidence coefficient loss function.
Carrying out affine transformation on the virtual corner points:
Figure BDA0002398311750000061
for four real angular points p of license plateiCentralizing to obtain centralized coordinate Cmn(pi)：
Figure BDA0002398311750000062
Wherein N issRepresenting the deformed planar object detection network down-sampling multiple.
Calculating the real corner point coordinate piAnd Vmn(qi) Distance error between the two points to obtain a corner coordinate distance loss function:
Figure BDA0002398311750000071
the confidence coefficient loss function calculation formula is as follows:
Figure BDA0002398311750000072
wherein the content of the first and second substances,
Figure BDA0002398311750000073
is an objective indication function for indicating a predicted feature map FTWhether the (m, n) cell contains a target;
the loss function is obtained by adding the corner coordinate distance loss function and the confidence coefficient loss function, and the expression is as follows:
Figure BDA0002398311750000074
s4, predicting feature map FTTaking the middle (m, n) point unit as the center, and constructing four virtual angular points qiAnd i is 1-4, wherein m and n respectively represent a prediction characteristic diagram FTThe abscissa and ordinate of the midpoint unit, and then for four virtualCorner point affine transformation Vmn(qi) Obtaining the coordinates l of four predicted angular points of the targeti,i＝1～4。
S5, according to the coordinates l of the prediction corner pointiSolving the transmission transformation equation set to obtain a homography matrix, correcting the image according to the homography matrix, and utilizing v1,v2And eliminating the corrected image with low confidence coefficient to obtain the final detection result.
Network training and testing are carried out on a Dell server with 256 memories, the server is configured with 4 Nvidia1080Ti GPUs and Intel Xeon CPU E5-2660V3 CPUs, the running main frequency is 2.6GHz, an operating system is Ubuntu 16.04.4L TS, an image processing library Opencv2.4.9, a program development language is python, based on a tensoflow deep learning framework, 1600 pictures are used for network training, the learning rate is set to be 0.0001, network iteration is carried out for 50000 times, in 30000 times, the attenuation parameter is 0.01, 200 pictures are used as a test set, the inclination angle range of a license plate is-20 degrees to +20 degrees, the positioning accuracy of a license plate detection result and the accuracy of a license plate correct labeling area IoU is set to be more than 0.7 and is tested correctly, the positioning accuracy of the method is 95.5 percent and is 5 percent higher than the accuracy of an open source project Easyy PR, and the validity of the method is verified.
The above embodiments are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be construed as equivalents thereof, and all such changes, modifications, substitutions, combinations, and simplifications are intended to be included in the scope of the present invention.

Claims (4)

1. A large-angle license plate detection method based on a deformed planar object detection network is characterized in that the large-angle license plate detection method comprises the following steps:
s1, constructing a backbone network for large-angle license plate detection by using a convolution layer, a pooling layer and a Relu layer, and taking the backbone network as a feature extractor of a license plate image, wherein the size of a convolution kernel in the convolution layer is k x k, and feature extraction is carried out on the input license plate image by adopting the backbone network to obtain a feature map;
s2, constructing a prediction network, wherein the prediction network is provided with two parallel convolution layers which are respectively called a first convolution layer and a second convolution layer, 2 m-m convolution kernels are used for performing sliding calculation on a feature map output by a main network in the first convolution layer to obtain a first example feature map, m is the size of a convolution kernel, the first example feature map is subjected to Softmax activation function to obtain a confidence coefficient feature map, 6 n-n convolution kernels are used for performing sliding calculation on the feature map output by the main network in the second convolution layer to obtain a second example feature map, n is the size of the convolution kernel, the second example feature map is subjected to L initial activation function to obtain a confidence coefficient feature map, and finally the affine feature map and the confidence coefficient feature map are subjected to splicing operation on channel dimensions to obtain a prediction feature map FTWherein, the predicted feature map FTContaining 8 channels, the value in 8 channels is defined as viI is 1 to 8, wherein v1,v2V representing the confidence that the prediction box contains and does not contain an object, respectively3～8Respectively representing affine matrix elements;
s3, designing a loss function to train the backbone network and the prediction network;
s4, using the prediction feature map FTFour virtual corner points q are constructed with the (m, n) point unit as the centeriAnd i is 1-4, wherein m and n respectively represent a prediction characteristic diagram FTThe horizontal coordinates and the vertical coordinates of the midpoint unit are obtained, and then affine transformation is carried out on the four virtual corner points to obtain coordinates l of four target prediction corner pointsi,i＝1～4；
S5, according to the coordinates l of the prediction corner pointiSolving the transmission transformation equation set to obtain a homography matrix, correcting the image according to the homography matrix, and utilizing v1、v2And eliminating the corrected image with low confidence coefficient to obtain the final detection result.
2. The large-angle license plate detection method based on the deformed planar object detection network as claimed in claim 1, wherein the specific structure of the backbone network is as follows:
the input layer is connected with the output layer in sequence as follows: convolution layer conv with convolution kernel size of 3x3 and number of 16; relu layer conv _ Relu; convolution layer conv with convolution kernel size of 3x3 and number of 16; relu layer conv _ Relu; pooling layer max _ pooling; convolution layer conv with convolution kernel size of 3x3 and number of 32; relu layer conv _ Relu; pooling layer max _ pooling; convolution layer conv with convolution kernel size of 3x3 and number of 32; relu layer conv _ Relu; pooling layer max _ pooling; convolution layer conv with convolution kernel size of 3x3 and number of 64; relu layer conv _ Relu, pooling layer max _ pooling; the convolution kernel size is 3 × 3, 128 convolution layers conv.
3. The method according to claim 1, wherein the loss function comprises two parts, the first part is a corner coordinate distance loss function, the second part is a confidence coefficient loss function, and the calculation process of the corner coordinate distance loss function is as follows:
carrying out affine transformation on the virtual corner points:
Figure FDA0002398311740000021
for four real angular points p of license plateiCentralizing to obtain centralized coordinate Cmn(pi)：
Figure FDA0002398311740000022
Wherein N issRepresenting the down-sampling multiple of the deformed planar object detection network,
calculating the real corner point coordinate piAnd Vmn(qi) Distance error between the two points to obtain a corner coordinate distance loss function:
Figure FDA0002398311740000031
wherein, the confidence coefficient loss function calculation formula is as follows:
Figure FDA0002398311740000032
wherein the content of the first and second substances,
Figure FDA0002398311740000033
is an objective indication function for indicating a predicted feature map FTWhether the (m, n) cell contains a target;
the loss function is obtained by adding a corner coordinate distance loss function and a confidence coefficient loss function, and the expression is as follows:
Figure FDA0002398311740000034
4. the method for detecting the large-angle license plate based on the deformed planar object detection network of claim 1, wherein in the step S5, the coordinates of the prediction corner points are transformed by using the transmission transformation to obtain the corrected image, and the process is as follows:
firstly, solving an affine equation set for k, k is more than or equal to 4 groups of non-collinear matching characteristic points to obtain a homography matrix H, then carrying out transmission transformation on an original image according to the homography matrix to obtain a corrected image, and cutting a target area according to a prediction coordinate.
CN202010138881.9A 2020-03-03 2020-03-03 Large-angle license plate detection method based on deformed plane object detection network Active CN111461113B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010138881.9A CN111461113B (en) 2020-03-03 2020-03-03 Large-angle license plate detection method based on deformed plane object detection network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010138881.9A CN111461113B (en) 2020-03-03 2020-03-03 Large-angle license plate detection method based on deformed plane object detection network

Publications (2)

Publication Number Publication Date
CN111461113A true CN111461113A (en) 2020-07-28
CN111461113B CN111461113B (en) 2023-02-14

Family

ID=71682458

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010138881.9A Active CN111461113B (en) 2020-03-03 2020-03-03 Large-angle license plate detection method based on deformed plane object detection network

Country Status (1)

Country Link
CN (1) CN111461113B (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112215222A (en) * 2020-10-12 2021-01-12 上海眼控科技股份有限公司 License plate recognition method, device, equipment and storage medium
CN112528994A (en) * 2020-12-18 2021-03-19 南京师范大学 Free-angle license plate detection method, license plate identification method and identification system
CN113435446A (en) * 2021-07-07 2021-09-24 南京云创大数据科技股份有限公司 Inclined license plate correction method based on deep learning
CN114332447A (en) * 2022-03-14 2022-04-12 浙江大华技术股份有限公司 License plate correction method, license plate correction device and computer readable storage medium
CN114708234A (en) * 2022-04-12 2022-07-05 北京优创新港科技股份有限公司 Method and device for identifying number of detonators on automatic bayonet coding all-in-one machine

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190266737A1 (en) * 2016-12-01 2019-08-29 Boe Technology Group Co., Ltd. Image matching method, image matching apparatus, image matching system, and storage medium
CN110427937A (en) * 2019-07-18 2019-11-08 浙江大学 A kind of correction of inclination license plate and random length licence plate recognition method based on deep learning

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190266737A1 (en) * 2016-12-01 2019-08-29 Boe Technology Group Co., Ltd. Image matching method, image matching apparatus, image matching system, and storage medium
CN110427937A (en) * 2019-07-18 2019-11-08 浙江大学 A kind of correction of inclination license plate and random length licence plate recognition method based on deep learning

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
WEI XIE ET AL.: "Control of non-linear systems based on interval observer design", 《IET CONTROL THEORY & APPLICATIONS》 *

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112215222A (en) * 2020-10-12 2021-01-12 上海眼控科技股份有限公司 License plate recognition method, device, equipment and storage medium
CN112528994A (en) * 2020-12-18 2021-03-19 南京师范大学 Free-angle license plate detection method, license plate identification method and identification system
CN112528994B (en) * 2020-12-18 2024-03-29 南京师范大学 Free angle license plate detection method, license plate recognition method and recognition system
CN113435446A (en) * 2021-07-07 2021-09-24 南京云创大数据科技股份有限公司 Inclined license plate correction method based on deep learning
CN113435446B (en) * 2021-07-07 2023-10-31 南京云创大数据科技股份有限公司 Deep learning-based inclined license plate correction method
CN114332447A (en) * 2022-03-14 2022-04-12 浙江大华技术股份有限公司 License plate correction method, license plate correction device and computer readable storage medium
CN114332447B (en) * 2022-03-14 2022-08-09 浙江大华技术股份有限公司 License plate correction method, license plate correction device and computer readable storage medium
CN114708234A (en) * 2022-04-12 2022-07-05 北京优创新港科技股份有限公司 Method and device for identifying number of detonators on automatic bayonet coding all-in-one machine
CN114708234B (en) * 2022-04-12 2022-12-02 北京优创新港科技股份有限公司 Method and device for identifying number of detonators on automatic bayonet coding all-in-one machine

Also Published As

Publication number Publication date
CN111461113B (en) 2023-02-14

Similar Documents

Publication Publication Date Title
CN111461113B (en) Large-angle license plate detection method based on deformed plane object detection network
US10977523B2 (en) Methods and apparatuses for identifying object category, and electronic devices
CN109685152B (en) Image target detection method based on DC-SPP-YOLO
CN108898086B (en) Video image processing method and device, computer readable medium and electronic equipment
CN108960211B (en) Multi-target human body posture detection method and system
CN109118473B (en) Angular point detection method based on neural network, storage medium and image processing system
CN109740606B (en) Image identification method and device
WO2023193401A1 (en) Point cloud detection model training method and apparatus, electronic device, and storage medium
CN111444948B (en) Image feature extraction and matching method
CN112084849A (en) Image recognition method and device
CN113822278A (en) License plate recognition method for unlimited scene
US20200005078A1 (en) Content aware forensic detection of image manipulations
CN114140623A (en) Image feature point extraction method and system
CN114743189A (en) Pointer instrument reading identification method and device, electronic equipment and storage medium
CN112085709A (en) Image contrast method and equipment
CN111353580A (en) Training method of target detection network, electronic device and storage medium
CN111027412A (en) Human body key point identification method and device and electronic equipment
CN112364974A (en) Improved YOLOv3 algorithm based on activation function
CN111104941B (en) Image direction correction method and device and electronic equipment
CN108992033B (en) Grading device, equipment and storage medium for vision test
CN111523586A (en) Noise-aware-based full-network supervision target detection method
CN113221855B (en) Small target detection method and system based on scale sensitive loss and feature fusion
US20230401670A1 (en) Multi-scale autoencoder generation method, electronic device and readable storage medium
CN116778164A (en) Semantic segmentation method for improving deep V < 3+ > network based on multi-scale structure
CN111104965A (en) Vehicle target identification method and device

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