CN115050028A - Sample vehicle license plate detection method in severe weather - Google Patents
Sample vehicle license plate detection method in severe weather Download PDFInfo
- Publication number
- CN115050028A CN115050028A CN202210671981.7A CN202210671981A CN115050028A CN 115050028 A CN115050028 A CN 115050028A CN 202210671981 A CN202210671981 A CN 202210671981A CN 115050028 A CN115050028 A CN 115050028A
- Authority
- CN
- China
- Prior art keywords
- license plate
- sample
- network
- distance
- severe weather
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 48
- 238000012549 training Methods 0.000 claims abstract description 33
- 238000012360 testing method Methods 0.000 claims abstract description 11
- 238000000034 method Methods 0.000 claims description 12
- 238000000605 extraction Methods 0.000 claims description 9
- 239000013598 vector Substances 0.000 claims description 9
- 238000012795 verification Methods 0.000 claims description 8
- 238000010276 construction Methods 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 5
- 230000008447 perception Effects 0.000 claims description 5
- 238000013507 mapping Methods 0.000 claims description 3
- 238000000691 measurement method Methods 0.000 claims description 3
- 238000012544 monitoring process Methods 0.000 claims description 3
- 230000003014 reinforcing effect Effects 0.000 claims description 2
- 230000005764 inhibitory process Effects 0.000 claims 1
- 238000005259 measurement Methods 0.000 abstract description 7
- 230000000694 effects Effects 0.000 abstract description 4
- 238000012512 characterization method Methods 0.000 abstract 2
- 238000010586 diagram Methods 0.000 description 2
- 230000001629 suppression Effects 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000001373 regressive effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- 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
-
- 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
- G06V10/20—Image preprocessing
- G06V10/22—Image preprocessing by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention belongs to the technical field of license plate detection, and particularly relates to a small sample vehicle license plate detection method in severe weather, wherein a succinct main network is designed for extracting positive and negative sample frames, a plurality of embedded space representations of positive samples and embedded space representations of negative samples are constructed for regular weather data during training, adding new positive sample characterization and negative sample embedding space characterization of severe weather image during test, pushing the distance between the target suggestion frame and the positive sample embedding space, enlarging the distance between the target suggestion frame and the negative sample embedding space, and finally judging whether the target candidate frame is a license plate or not according to the measurement, thereby achieving the effect of detecting the license plate, not only being used for detecting the small sample vehicle license plate in severe weather, but also being used for detecting various small sample targets, the small sample data detection precision can reach 96.5% in severe weather, and the detection and classification effects are greatly improved.
Description
Technical Field
The invention belongs to the technical field of license plate detection, and particularly relates to a small sample vehicle license plate detection method in severe weather.
Background
In recent years, with the development of economy, the number of automobiles in a city is increased sharply, great challenges are brought to urban traffic management, and license plate detection is taken as a preposed task of license plate identification and plays a significant role in intelligent traffic data statistics and analysis. The application scenes of the method are quite wide, such as the application scenes of a community entrance, a highway, a toll station, city traffic violation and the like, however, in practical application, the license plate is influenced by extremely severe weather such as foggy days, rainy days, snowy days and the like, so that the detection result is poor.
Under extreme severe weather, the image quality is seriously influenced, the detail characteristics are seriously lost, the information of the license plate is difficult to extract, the conditions of wrong detection and missing detection are easily caused in the license plate detection, the data under the severe weather is seriously insufficient, the generalization of a deep learning network is limited to a great extent, and the existing license plate detection method is difficult to achieve ideal detection performance under the severe weather; on the other hand, license plate detection needs to be simultaneously positioned and classified, so that the model adaptation process is more complex, the model has excessive self-adaptation (similar to overfitting during training by using a small amount of data samples) and unstable risks, and finally the target domain detection performance is reduced.
At present, for data collection and labeling in a severe scene, huge manpower and material resources are consumed, only a small amount of representative sample data is collected, the effort is negligible, meanwhile, inevitable noise caused by a large amount of samples can be reduced, and the study on the positioning and classification of a license plate under the condition of only a small amount of data samples in severe weather is very challenging.
Therefore, license plate detection in severe weather is limited by training samples, and the existing method is easy to over-fit on a small amount of data and limited in detection performance, so that an effective means is urgently needed to improve the detection performance of the small sample vehicle license plate in severe weather.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides a small sample vehicle license plate detection method in severe weather, which is used for solving the problem of small sample vehicle license plate detection in severe weather, can be used for a license plate detection task in a small sample scene, and can efficiently detect the precision of the license plate and the generalization performance of a model.
In order to achieve the purpose, the specific process for detecting the license plate with the small sample comprises the following steps:
(1) and (3) data set construction: collecting images containing conventional, inclined and distorted license plates in scenes such as traffic monitoring in conventional weather, parking lots and the like, constructing a license plate data set as basic data for training, dividing the license plate data set into a training set, a verification set and a test set, and collecting not less than 200 license plate images in severe weather as small sample data;
(2) shared backbone network feature extraction: initializing the size and numerical range of the picture, and inputting the processed image into a backbone network for convolution feature extraction; then, reinforcing the extracted features through a feature pyramid network to obtain a multi-scale license plate convolution feature map set;
(3) positioning the license plate: connecting two layers of convolution kernels to form 3 after different multi-scale convolution layers according to the multi-scale license plate convolution feature map set obtained in the step (2)
3, setting an anchor frame at each feature point of the feature map, and outputting candidate frames after learning classification and position offset information by using two full-connection layer sub-networks with the same structure but without shared parameters respectively;
(4) sample embedding space construction: compared with a sample label, defining a candidate box with IoU >0.7 as positive sample information, defining a candidate box with 0.2< IoU <0.3 as negative sample information, constructing a multilayer perception embedded code by using three full-connection layers, inputting a convolution feature extracted by the step (2) of the positive sample candidate box and the negative sample candidate box to corresponding regions, outputting positive sample embedding and negative sample embedding, and respectively bringing the positive sample embedding and the negative sample embedding into a positive sample embedding space and a negative sample embedding space;
(5) sample distance metric: respectively carrying out distance calculation on the embedded coded vectors corresponding to the candidate frames output in the step (3) and the positive sample embedding space and the negative sample embedding space, and outputting the probability that each candidate frame is the license plate by using the two obtained distances through a probability measurement method;
(6) training a network structure to obtain a trained license plate detection network: using images of the training set in the dataset, picture size 512
512
3, inputting the images into the network in sequence according to the number of samples required by one training and inputting the whole network
Where B is the number of samples required for a training session and the output regression coordinate position is scaled using the IoU threshold as a measure of the sample distribution strategy
N is the number of output predicted license plate targets, 4 represents four dimensions, and comprises coordinates (x, y) of the center point of a horizontal frame and length (w, h) of a license plate; and (5) calculating whether the candidate frame is the license plate or not by adopting the same method as the step (5), updating parameters through back propagation, and storing the model with the best result on the verification set after 500 rounds of complete training set training iterationThe parameters are used as parameters of the final model training to obtain a trained license plate detection network;
(7) testing the network and outputting the position and the type of the license plate: scaling (resize) the image long side to 512 with the picture long and short side scale unchanged, and then filling the short side of the picture so that the image size is 512
512, the classification confidence of the license plate and the coordinate position of the license plate are output as the input of a license plate detection network, the license plate with low confidence is filtered by setting a threshold value, redundant frames output by the network are deleted by using non-maximum suppression (NMS) to obtain a license plate detection frame, and finally the license plate characteristics corresponding to the license plate detection frame are input to a license plate recognition module to obtain the license plate number.
As a further technical scheme of the invention, the positions of the license plate need to be marked in the images in the license plate data set and the small sample data in the step (1), namely four angular points of the license plate are marked, and a point set representing the position of the license plate is calculated according to the positions of the four angular points, wherein the point set comprises a center point of the license plate, the four angular points and center points of four edges.
As a further technical scheme of the invention, ResNet50 pre-trained in an ImageNet data set is used as a feature extraction network in the main network in the step (2), the extraction modules 3, 4 and 5 are respectively input into a subsequent network, a feature pyramid network is arranged behind ResNet50, and the feature pyramid network performs reinforced utilization on the features extracted by ResNet50 to obtain a convolution feature map set with stronger expressive force and containing multi-scale license plate information, so that license plates with different sizes at a short distance and a long distance are better captured.
As a further technical solution of the present invention, the position deviation information in step (3) is a deviation between a compact target frame of a license plate and an anchor frame corresponding to the feature point, thereby completing a task of classification of the target frame and position information, where the classification is whether the license plate is present or not, and the position information is a minimum external moment (orientation frame) of the license plate.
As a further technical scheme of the invention, the distance calculation formula in the step (5) is as follows:
whereinXFor the embedded vector to which the candidate box corresponds,Efor the embedded vectors stored in the sample space,
is a temperature coefficient ofXAndYa distance coefficient between, the probability metric being a mapping of the distance to the probability using a gaussian function
The probability is:
wherein
The distance that the candidate box is embedded from the positive sample,
distance of embedding for candidate box and negative sample, and addingβValue to ensure that the distance subtraction is not negative.
Aiming at the problem of small sample number in severe weather, the invention fully utilizes the negative sample in small sample detection by introducing a new measurement learning framework based on the negative sample and the positive sample and a new inference scheme based on the negative sample and the positive sample, and simultaneously learns an embedding space on a training image by using the positive sample and the negative sample, wherein the distance corresponds to the similarity measurement of the license plate and the positive sample and the negative sample.
Compared with the prior art, the invention has the beneficial effects that:
(1) the performance on the degraded license plate is greatly improved on the basis of increasing little calculation amount by constructing the embedding space of the positive and negative samples;
(2) the small sample target detection method can be used for detecting the small sample vehicle license plate in severe weather and can also be used for multiple small sample target detection tasks, the small sample data detection precision in severe weather can reach 96.5%, and the detection and classification effects are greatly improved.
Drawings
Fig. 1 is a schematic diagram of a small sample vehicle license plate detection network structure framework in severe weather.
Fig. 2 is a block diagram of a flow of the sample vehicle license plate detection in severe weather according to the present invention.
Detailed Description
The invention will be further described by way of examples, without in any way limiting the scope of the invention, with reference to the accompanying drawings.
Example (b):
the embodiment is used for extracting positive and negative sample frames by designing a simple backbone network, a plurality of positive sample embedding space representations and negative sample embedding space representations are constructed for conventional weather data during training, new severe weather image positive sample representations and new severe weather image negative sample embedding space representations are added during testing, the optimization target is to push the distance between the target suggestion frame and the positive sample embedding space, the distance between the target suggestion frame and the negative sample embedding space is enlarged, and finally whether the target candidate frame is a license plate is judged through the measurement, so that the license plate detection effect is achieved, and the method specifically comprises the following steps:
(1) and (3) data set construction: collecting images containing conventional, inclined and distorted license plates in scenes such as traffic monitoring in conventional weather, parking lots and the like, constructing a data set of the license plates as training basic data, and dividing the data set into a training set, a verification set and a test set; collecting 3000 license plate data in severe weather with 200 pieces of information as small sample data, dividing the small sample data into a training set, a verification set and a test set, wherein the collected images need to mark the positions of license plates, mainly marking four angular points of the license plates, and calculating a point set representing the positions of the license plates according to the positions of the four angular points, wherein the point set comprises the center points of the license plates, the four angular points and the center points of four edges;
(2) extracting the features of the shared backbone network: initializing the size and numerical range of a picture, inputting the processed image into a backbone network for convolution feature extraction, wherein the backbone network uses ResNet50 pre-trained in an ImageNet data set as a feature extraction network, respectively extracting a module 3, a module 4 and a module 5, inputting the extracted features into a subsequent network, adding a feature pyramid network behind ResNet50, and performing enhanced utilization on the features extracted by ResNet50 to obtain a convolution feature map set with stronger expressive force and containing multi-scale license plate information, so that license plates with different sizes in a short distance and a long distance can be better captured;
(3) positioning the license plate: connecting two layers of convolution kernels to form 3 after different multi-scale convolution layers according to the multi-scale license plate convolution feature map set obtained in the step (2)
3, setting an anchor frame at each feature point of the feature map, and then respectively using two full-connection layer sub-networks with the same structure but without shared parameters to learn classification and position offset information and then outputting candidate frames, wherein the position offset information is the deviation of a compact target frame of the license plate and the anchor frame corresponding to the feature point so as to complete the tasks of classification and position information of the target frames, wherein the classification is whether the license plate is present, and the position information is the minimum external moment (facing frame) of the license plate;
(4) sample embedding space construction: defining a candidate box with a sample label IoU >0.7 as positive sample information, defining a candidate box with a sample label 0.2< IoU <0.3 as negative sample information, and constructing a multilayer perception embedding code by using three layers of full connection layers, wherein the input of the multilayer perception embedding code is convolution characteristics of the positive sample candidate box and the negative sample candidate box to corresponding regions (extracted from the step (2)), the output of the multilayer perception embedding code is positive sample embedding and negative sample embedding, and the output results are respectively contained in positive sample embedding space and negative sample embedding space;
(5) sample distance metric: and (4) respectively carrying out distance measurement on the embedded coded vectors corresponding to the candidate frames output in the step (3) and the positive sample embedding space and the negative sample embedding space, comprehensively utilizing the two distances, and outputting the probability that each candidate frame is the license plate by a probability measurement method, wherein the distance measurement formula is shown as the following formula:
whereinXFor the embedded vector to which the candidate box corresponds,Efor the embedded vectors stored in the sample space,
is a temperature coefficient ofXAndYcoefficient of distance between, then mapping the distance to a probability using a gaussian function
The probability is defined as:
wherein
The distance that the candidate box is embedded from the positive sample,
distance of embedding for candidate box and negative sample, and addingβA value to ensure that the distance subtraction is not negative;
(6) training a network structure to obtain a trained license plate detection network: using images of the training set in the dataset, picture size 512
512
3, according to the batch size (B), B is the number of samples required by one training, and the samples are input into the network in sequence, so the input of the whole network
And using the IOU threshold as a measure of the sample distribution strategy, the high quality level candidate box module outputs the regressive coordinate position
N is the number of target of output prediction license plate, 4 represents four dimensions, and comprises the coordinates (x, y) of the center point of the horizontal frame of the license plate and the length and width (w, h); calculating whether the candidate frame is a license plate or not by adopting a sample distance measurement and probability estimation method in the step (5), updating parameters through back propagation, storing model parameters with the best results on a verification set after 500 rounds of complete training set training iteration, and taking the model parameters as parameters of a final model training to obtain a trained license plate detection network;
(7) the testing network outputs the position and the type of the license plate: during the test, the image long side was scaled (resize) to 512 keeping the picture long and short side scale unchanged, and then the short side of the picture was filled in such a way that the image size was 512
512, as the input of the network, the classification confidence of the license plate and the coordinate position of the license plate can be output, the threshold value is set to filter the license plate with low confidence, the redundant frame output by the network is deleted by using the non-maximum suppression (NMS) to obtain the license plate detection frame, and finally the license plate characteristics corresponding to the license plate detection frame are input to the license plate recognition module to obtain the license plate number.
Example 2:
in this embodiment, 1000 images of the ccpd _ weather data in the ccpd data set are selected for precision detection, and the detection result is 96.5%, where the training set, the test set, and the verification set respectively include 200, 500, and 300.
Algorithms and modules not described in detail herein are all well known or commonly used in the art.
It is noted that the disclosed embodiments are intended to aid in further understanding of the invention, but those skilled in the art will appreciate that: various substitutions and modifications are possible without departing from the spirit and scope of the invention and appended claims. Therefore, the invention should not be limited to the embodiments disclosed, but the scope of the invention is defined by the appended claims.
Claims (5)
1. A sample car license plate detection method in severe weather is characterized by comprising the following specific steps:
(1) and (3) data set construction: collecting traffic monitoring in normal weather and images containing normal, inclined and distorted license plates in a parking lot, constructing a license plate data set as basic data for training, dividing the license plate data set into a training set, a verification set and a test set, and collecting not less than 200 license plate images in severe weather as small sample data;
(2) shared backbone network feature extraction: initializing the size and numerical range of the picture, and inputting the processed image into a backbone network for convolution feature extraction; then, reinforcing the extracted features through a feature pyramid network to obtain a multi-scale license plate convolution feature map set;
(3) positioning the license plate: connecting two layers of convolution kernels to form 3 after different multi-scale convolution layers according to the multi-scale license plate convolution feature map set obtained in the step (2)
3, setting an anchor frame at each feature point of the feature map, and outputting candidate frames after learning classification and position offset information by using two full-connection layer sub-networks with the same structure but without shared parameters respectively;
(4) sample embedding space construction: compared with a sample label, defining a candidate box with IoU >0.7 as positive sample information, defining a candidate box with 0.2< IoU <0.3 as negative sample information, constructing a multilayer perception embedded code by using three full-connection layers, inputting a convolution feature extracted by the step (2) of the positive sample candidate box and the negative sample candidate box to corresponding regions, outputting positive sample embedding and negative sample embedding, and respectively bringing the positive sample embedding and the negative sample embedding into a positive sample embedding space and a negative sample embedding space;
(5) sample distance metric: respectively carrying out distance calculation on the embedded coded vectors corresponding to the candidate frames output in the step (3) and the positive sample embedding space and the negative sample embedding space, and outputting the probability that each candidate frame is the license plate by using the two obtained distances through a probability measurement method;
(6) training a network structure to obtain a trained license plate detection network: using images of the training set in the dataset, picture size 512
512
3, inputting the images into the network in sequence according to the number of samples required by one training and inputting the whole network
Where B is the number of samples required for a training session and the output regression coordinate position is scaled using the IoU threshold as a measure of the sample distribution strategy
N is the number of target of output prediction license plate, 4 represents four dimensions, and comprises the coordinates (x, y) of the center point of the horizontal frame of the license plate and the length and width (w, h); calculating whether the candidate frame is the license plate or not by adopting the same method as the step (5), updating parameters through back propagation, saving the model parameters with the best result on the verification set after 500 rounds of complete training set training iteration, and taking the model parameters as the parameters of the final model training to obtain a trained license plate detection network;
(7) testing network and outputting license plate position and classification: scaling the long side of the picture to 512 with the proportion of the long side and the short side of the picture unchanged, and then filling the short side of the picture to make the size of the picture 512
512, the classification confidence of the license plate and the coordinate position of the license plate are output as the input of a license plate detection network, the license plate with low confidence is filtered by setting a threshold, redundant frames output by the network are deleted by using non-maximum inhibition to obtain a license plate detection frame, and finally the license plate characteristics corresponding to the license plate detection frame are input to a license plate recognition module to obtain the license plate number.
2. The method for detecting the small sample vehicle license plate under the severe weather as claimed in claim 1, wherein the license plate data set in step (1) and the images in the small sample data are marked with the positions of the license plate, that is, with four angular points of the license plate, and a point set representing the positions of the license plate is calculated according to the positions of the four angular points, wherein the point set comprises a center point of the license plate, the four angular points and center points of four edges.
3. The method for detecting the self-sample license plate under the severe weather as claimed in claim 2, wherein in the step (2), the ResNet50 pre-trained in an ImageNet data set is used as a feature extraction network by the main network, the modules 3, 4 and 5 are respectively extracted and input into a subsequent network, the feature pyramid network is arranged behind the ResNet50, and the feature pyramid network performs reinforced utilization on the features extracted by the ResNet50 to obtain a convolution feature map set with stronger expressive force and containing multi-scale license plate information, so that license plates with different sizes in a short distance and a long distance can be better captured.
4. The method for detecting the small sample vehicle license plate under the bad weather according to claim 3, wherein the position deviation information in the step (3) is a deviation between a compact target frame of the license plate and an anchor frame corresponding to the feature point, so as to complete a task of classification of the target frame and position information, wherein the classification is whether the license plate is used or not, and the position information is a minimum external moment of the license plate.
5. The method for detecting the sample own-vehicle license plate under the severe weather as claimed in claim 4, wherein the distance calculation formula in the step (5) is as follows:
whereinXFor the embedded vector to which the candidate box corresponds,Efor the embedded vectors stored in the sample space,
is a temperature coefficient ofXAndYa distance coefficient between, the probability metric being a mapping of the distance to the probability using a gaussian function
The probability is:
wherein
The distance that the candidate box is embedded from the positive sample,
distance of embedding for candidate box and negative sample, and addingβValue to ensure that the distance subtraction is not negative.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210671981.7A CN115050028B (en) | 2022-06-15 | 2022-06-15 | Small sample license plate detection method in severe weather |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210671981.7A CN115050028B (en) | 2022-06-15 | 2022-06-15 | Small sample license plate detection method in severe weather |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115050028A true CN115050028A (en) | 2022-09-13 |
| CN115050028B CN115050028B (en) | 2024-03-29 |
Family
ID=83161420
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210671981.7A Active CN115050028B (en) | 2022-06-15 | 2022-06-15 | Small sample license plate detection method in severe weather |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115050028B (en) |
Citations (7)
| 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 |
| US20180260415A1 (en) * | 2017-03-10 | 2018-09-13 | Xerox Corporation | Instance-level image retrieval with a region proposal network |
| CN110674689A (en) * | 2019-08-19 | 2020-01-10 | 浙江省北大信息技术高等研究院 | Vehicle re-identification method and system based on feature embedding space geometric constraint |
| WO2020181685A1 (en) * | 2019-03-12 | 2020-09-17 | 南京邮电大学 | Vehicle-mounted video target detection method based on deep learning |
| CN112132130A (en) * | 2020-09-22 | 2020-12-25 | 福州大学 | Real-time license plate detection method and system for whole scene |
| CN113159153A (en) * | 2021-04-13 | 2021-07-23 | 华南理工大学 | License plate recognition method based on convolutional neural network |
| CN113822278A (en) * | 2021-11-22 | 2021-12-21 | 松立控股集团股份有限公司 | License plate recognition method for unlimited scene |
-
2022
- 2022-06-15 CN CN202210671981.7A patent/CN115050028B/en active Active
Patent Citations (7)
| 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 |
| US20180260415A1 (en) * | 2017-03-10 | 2018-09-13 | Xerox Corporation | Instance-level image retrieval with a region proposal network |
| WO2020181685A1 (en) * | 2019-03-12 | 2020-09-17 | 南京邮电大学 | Vehicle-mounted video target detection method based on deep learning |
| CN110674689A (en) * | 2019-08-19 | 2020-01-10 | 浙江省北大信息技术高等研究院 | Vehicle re-identification method and system based on feature embedding space geometric constraint |
| CN112132130A (en) * | 2020-09-22 | 2020-12-25 | 福州大学 | Real-time license plate detection method and system for whole scene |
| CN113159153A (en) * | 2021-04-13 | 2021-07-23 | 华南理工大学 | License plate recognition method based on convolutional neural network |
| CN113822278A (en) * | 2021-11-22 | 2021-12-21 | 松立控股集团股份有限公司 | License plate recognition method for unlimited scene |
Non-Patent Citations (2)
| Title |
|---|
| 何霞;汤一平;陈朋;王丽冉;袁公萍;: "多任务分段紧凑特征的车辆检索方法", 中国图象图形学报, no. 12, 16 December 2018 (2018-12-16) * |
| 秦丽娟;赵宇辉;: "基于Faster RCNN的车牌检测算法", 电子世界, no. 24, 30 December 2019 (2019-12-30) * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115050028B (en) | 2024-03-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111368687B (en) | Sidewalk vehicle illegal parking detection method based on target detection and semantic segmentation | |
| CN109977812B (en) | Vehicle-mounted video target detection method based on deep learning | |
| CN111160311B (en) | Yellow river ice semantic segmentation method based on multi-attention machine system double-flow fusion network | |
| CN110796168B (en) | Vehicle detection method based on improved YOLOv3 | |
| CN107563372B (en) | License plate positioning method based on deep learning SSD frame | |
| CN111178213B (en) | Aerial photography vehicle detection method based on deep learning | |
| CN113033604B (en) | Vehicle detection method, system and storage medium based on SF-YOLOv4 network model | |
| CN111709416B (en) | License plate positioning method, device, system and storage medium | |
| CN112837315B (en) | Deep learning-based transmission line insulator defect detection method | |
| CN102289686B (en) | Method for identifying classes of moving targets based on transfer learning | |
| CN114677507A (en) | Street view image segmentation method and system based on bidirectional attention network | |
| CN110659601B (en) | Depth full convolution network remote sensing image dense vehicle detection method based on central point | |
| CN115063786A (en) | High-order distant view fuzzy license plate detection method | |
| CN112990065A (en) | Optimized YOLOv5 model-based vehicle classification detection method | |
| CN115223063A (en) | Unmanned aerial vehicle remote sensing wheat new variety lodging area extraction method and system based on deep learning | |
| CN114913498A (en) | Parallel multi-scale feature aggregation lane line detection method based on key point estimation | |
| CN114677501A (en) | License plate detection method based on two-dimensional Gaussian bounding box overlapping degree measurement | |
| CN116665153A (en) | Road scene segmentation method based on improved deep bv3+ network model | |
| CN115909241A (en) | Lane line detection method, system, electronic device and storage medium | |
| CN106529391B (en) | A kind of speed limit road traffic sign detection of robust and recognition methods | |
| CN114863122A (en) | Intelligent high-precision pavement disease identification method based on artificial intelligence | |
| Guillermo et al. | A Comparative Study on Satellite Image Analysis for Road Traffic Detection using YOLOv3-SPP, Keras RetinaNet and Full Convolutional Network | |
| CN116824520A (en) | Vehicle track prediction method and system based on ReID and graph convolution network | |
| CN115861997B (en) | License plate detection and recognition method for key foreground feature guided knowledge distillation | |
| CN115050028B (en) | Small sample license plate detection method in severe weather |
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 |