CN114677501A - License plate detection method based on two-dimensional Gaussian bounding box overlapping degree measurement - Google Patents
License plate detection method based on two-dimensional Gaussian bounding box overlapping degree measurement Download PDFInfo
- Publication number
- CN114677501A CN114677501A CN202210595586.5A CN202210595586A CN114677501A CN 114677501 A CN114677501 A CN 114677501A CN 202210595586 A CN202210595586 A CN 202210595586A CN 114677501 A CN114677501 A CN 114677501A
- Authority
- CN
- China
- Prior art keywords
- license plate
- frame
- feature
- dimensional gaussian
- bounding box
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 40
- 238000005259 measurement Methods 0.000 title claims abstract description 24
- 238000012549 training Methods 0.000 claims abstract description 42
- 238000009826 distribution Methods 0.000 claims abstract description 30
- 238000000034 method Methods 0.000 claims abstract description 25
- 238000012795 verification Methods 0.000 claims description 13
- 238000012360 testing method Methods 0.000 claims description 11
- 238000000605 extraction Methods 0.000 claims description 9
- 238000005070 sampling Methods 0.000 claims description 6
- 230000004927 fusion Effects 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000010276 construction Methods 0.000 claims description 3
- 239000000284 extract Substances 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 230000001373 regressive effect Effects 0.000 claims description 3
- 238000011524 similarity measure Methods 0.000 claims description 3
- 230000005764 inhibitory process Effects 0.000 claims 1
- 238000002372 labelling Methods 0.000 claims 1
- 238000011156 evaluation Methods 0.000 abstract description 8
- 238000013135 deep learning Methods 0.000 description 3
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000012805 post-processing Methods 0.000 description 2
- 230000001629 suppression Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- 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
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Abstract
The invention belongs to the technical field of license plate detection, and relates to a license plate detection method based on two-dimensional Gaussian bounding box overlapping degree measurement, which is characterized in that a facing bounding box is converted into a two-dimensional Gaussian distribution representation, KL divergence is used as a measurement standard to calculate the distribution distance between a predicted bounding box and a real bounding box, and the distribution distance is used as the regression loss of the bounding box, and the method is micro, solves the problem that the measurement loss of a smooth L1 is inconsistent with an evaluation mode, can improve the network training efficiency and the generalization capability of the license plate detection based on the facing bounding box, is only used in a network training stage, does not increase the calculated amount of an inference process, can be used for inclined deformation license plate detection, and can also be used for target detection tasks of various scenes such as scene text detection, supermarket commodity detection and the like.
Description
Technical Field
The invention belongs to the technical field of license plate detection, and relates to a license plate detection method based on two-dimensional Gaussian bounding box overlapping degree measurement.
Background
The license plate number used by the vehicle is a unique identity mark of the vehicle and also an important certificate for the vehicle to run on the road, and the detection and identification of the license plate number are main means for identifying the identity of the vehicle. Therefore, license plate detection is an important task of statistical analysis of intelligent traffic data and a preposed task of urban vehicle management, and has a wide application scene, such as parking lot and community access verification, road traffic monitoring and the like, however, in an actual application scene, the license plate is influenced by moving speed such as illumination, visual angle change and the like, so that the detection result is poor.
The existing license plate detection methods mainly comprise two types, one is that the license plate detection method based on the traditional method carries out license plate detection by using gradient and color space, and the method has higher detection speed, but model difference is built for the license plate characteristics in a complex background and scene, and the license plate which is shielded, blurred and deformed is inaccurately positioned; secondly, a license plate detection method based on deep learning is provided, the method has strong adaptability to different scenes by learning a large amount of training data, in order to better describe the position of the inclined license plate, a boundary frame of the license plate is defined by using a frame-oriented method in many deep learning methods, and the error is measured by using the loss of smooth L1 in training, but the loss evaluation mode is not consistent with the evaluation mode, namely the overlapping degree of a prediction frame and a real frame is not in a linear relation with the loss; in general target detection (horizontal box), the inconsistency of Loss and evaluation is usually solved by using an overlap Loss (IoU Loss), however, the use of the overlap Loss towards the box brings the problem that the Loss function is not derivable, and the training process is influenced.
Therefore, in the process of license plate orientation frame training based on deep learning, the problem of inconsistent loss and evaluation modes of the existing method seriously influences the training efficiency and precision, so that an effective method is urgently needed for improving the license plate detection precision.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides a license plate detection method based on two-dimensional Gaussian bounding box overlapping degree measurement, which is used for solving the problems of loss and inconsistent evaluation modes in the deformed and deformed license plate training process, can be used for a deformed license plate training task in any scene, and can be used for efficiently detecting the license plate with high precision and training efficiency.
In order to achieve the purpose, the invention firstly constructs a data set and designs a concise backbone network, avoids any complex post-processing step, is used for predicting the orientation frame of the rotary license plate, then approximates the predicted orientation frame based on a five-point positioning method into two-dimensional Gaussian distribution, measures the overlapping degree of the two frames by calculating the distance between the two-dimensional Gaussian distribution between the predicted orientation frame and the real orientation frame, and further realizes the measurement of the quality of the orientation frame, and the specific process is as follows:
(1) and (3) data set construction: collecting images containing conventional, inclined and distorted license plates in scenes such as a community entrance, a side parking lot and the like, constructing an image data set containing the license plates, marking the positions of the license plates, and dividing the data set into a training set, a verification set and a test set, wherein the training set is used for training, the verification set is used for model verification in the training process, and the test set is used for final precision test;
(2) extracting features of the backbone network: initializing the size and the numerical range of the picture in the data set, and inputting the processed image into a backbone network for convolution feature extraction;
(3) multi-scale feature hierarchy: the image after the feature extraction is subjected to multi-scale feature hierarchical feature image set obtained through a multi-scale feature hierarchical structure based on feature fusion;
(4) positioning the license plate: connecting two convolution layers with convolution kernel of 3 x 3 and one average pooling layer after different multi-scale convolution layers according to the multi-scale feature hierarchical feature graph set obtained in the step (3), setting an anchor frame at each feature point of the feature graph, and respectively using two full-connection layers with the same structure but not sharing parametersLearning classification and position deviation information by a subnetwork to realize classification and position regression of the target frame; the position deviation information is the deviation of an anchor frame corresponding to a target frame and a characteristic point of the license plate, the classification of the category is that whether the license plate is judged by z, and the position is an orientation frame of the license plate
In which
Represents the center point of the license plate,
indicating the length and width of the license plate toward the frame,
represents an angle with the horizontal direction;
(5) and (3) converting towards a frame: orientation frame based on five-point representation
Converting the orientation frame expressed based on the two-dimensional Gaussian distribution into a two-dimensional Gaussian distribution form of a predicted orientation frame and a real orientation frame;
(6) orientation frame similarity measure: using the KL divergence as a metric between the two distributions to measure the distance between the predicted and true orientation boxes;
(7) training a network structure to obtain trained model parameters: image size 640 using training set in dataset
480
3, sequentially inputting the images into the network, and outputting the classification confidence of the license plate by using the IOU threshold as a measurement standard of a sample distribution strategy
And the location of the regressive coordinate
B is the number of samples selected by one-time training, Class is 2, namely whether the license plate is the license plate or not, N is the number of output and predicted license plate targets, and 5 is the coordinate of the center point of the license plate facing to the frame, the length, the width and the angle of the frame; adopting Focal loss to calculate errors of a prediction category and a real category, adopting step-oriented frame similarity measurement to calculate errors of a predicted license plate position and a real license plate position, updating parameters through back propagation, saving model parameters with the best results on a verification set after 200 times of complete training set training iteration, and obtaining trained license plate detection network parameters;
(8) outputting license plate positions and types: filling the short edge of the picture after scaling (resize) the long edge of the picture to 640 under the condition of keeping the proportion of the long edge and the short edge of the picture unchanged, so that the image size is 640
480, as the input of the license plate detection network, the classification confidence of the license plate and the coordinate position of the license plate can be output, a threshold value is set to filter the license plate with low confidence, and a non-maximum suppression (NMS) is used to delete the redundant frame output by the network, so as to obtain the position of the license plate.
Further, the step (1) of marking the license plate is to mark four corner points of the license plate, and the center point, the length and the width of the orientation frame of the license plate and the included angle between the orientation frame and the horizontal direction are calculated according to the positions of the four corner points, so as to supervise the positioning of the license plate.
Further, in the step (2), the backbone network uses the pre-trained VGG16 in the ImageNet data set as a feature extraction network, and extracts C3, C4 and C5 respectively and inputs the extracted C3, C4 and C5 into a subsequent network.
Further, in the step (3), the multi-scale feature hierarchical structure based on feature fusion performs semantic expansion on shallow features through a top-down path, and combines top-level feature mapping with bottom-level information through a bottom-up path, and each layer of the feature hierarchy can capture rich semantic representation and bottom-level information at the same time; the top-down path uses the up-sampling module and the convolution layer to amplify the features and fuse the features with corresponding sizes, the bottom-up path uses the down-sampling module and the convolution layer to scale the features and fuse the features with corresponding sizes, and finally, features of different layers are combined to form a strong feature representation for the multi-scale target.
Further, the orientation box represented based on the two-dimensional Gaussian distribution in the step (5) is
Wherein
By (a)
) Expressed, the formula is as follows:
Further, the KL divergence calculation formula in step (6) is:
compared with the prior art, the invention has the beneficial effects that: the method is characterized in that the orientation bounding box is converted into a two-dimensional Gaussian distribution representation, KL divergence is used as a measurement standard to calculate the distribution distance between the predicted bounding box and the real bounding box, the distribution distance is used as the regression loss of the bounding box, the method is micro, the problem that the smooth L1 measurement loss is inconsistent with an evaluation mode is solved, the network training efficiency and the generalization capability of the license plate detection based on the orientation bounding box can be improved, the method is only used in a network training stage, the calculation amount of an inference process is not increased, the method can be used for not only oblique deformation license plate detection but also target detection tasks of various scenes such as scene text detection and supermarket commodity detection, compared with the existing method, the detection precision is improved from 93.2% to 97.5%, and the supervision capability and detection of the network are improved.
Drawings
Fig. 1 is a schematic diagram of a backbone network architecture according to the present invention.
Fig. 2 is a schematic block diagram of the working flow of the present invention.
Detailed Description
The invention is further described below by way of examples and with reference to the accompanying drawings, without limiting the scope of the invention in any way.
Example (b):
in this embodiment, a simple backbone network is designed first, so as to avoid any cumbersome post-processing step, and the orientation frame of the license plate is predicted to be approximately a two-dimensional gaussian distribution, and the overlap of two frames is measured by calculating the distance between the two-dimensional gaussian distributions between the predicted orientation frame and the actual orientation frame, so as to measure the quality of the orientation frame, and the specific implementation includes the following steps:
(1) and (3) data set construction: collecting images containing conventional, inclined and distorted license plates of scenes such as a community entrance, a side-position parking lot and the like, constructing an image data set containing the license plates, marking the positions of the license plates, mainly marking four angular points of the license plates, calculating the center point, the length and the width of the license plates facing a frame and the included angle between the center point, the length and the width and the horizontal direction of the license plates through the positions of the four angular points, monitoring the positioning of the license plates, and dividing the data set into a training set, a verification set and a test set, wherein the training set is used for training, the verification set is used for model verification in the training process, and the test set is used for final precision test;
(2) extracting basic characteristics from the backbone network: taking an image in a data set as an input, firstly initializing the size and the numerical range line of the image, then inputting the processed image into a backbone network for convolution feature extraction, wherein the backbone network uses pre-trained VGG16 in the ImageNet data set as a feature extraction network, and respectively extracts C3, C4 and C5 and inputs the C3, C4 and C5 into a subsequent network;
(3) multi-scale feature hierarchy: aiming at license plates with different sizes, a multi-scale feature hierarchical structure based on feature fusion is used after a feature extraction network, the structure semantically expands shallow features through a top-down path, top-level feature mapping is combined with bottom-level information through a bottom-up path, each layer of a feature hierarchy can capture rich semantic representation and bottom-level information at the same time, the top-down path amplifies the features through an up-sampling module and a convolution layer, the features with corresponding sizes are fused, the bottom-up path scales the features through a down-sampling module and the convolution layer and is fused with the features with corresponding sizes, and finally, a strong multi-scale feature hierarchical feature map set can be formed for a multi-scale target by combining the features with different levels;
(4) positioning the license plate: and (4) according to the multi-scale feature hierarchical feature map set obtained in the step (3), connecting two convolution layers with convolution kernels of 3 x 3 and an average pooling layer after different multi-scale convolution layers, setting an anchor frame at each feature point of the feature map, and then respectively learning classification and position deviation information by using two full-connection layer sub-networks with the same structure but without sharing parameters, wherein the position deviation information is the deviation of a compact target frame of the license plate and the anchor frame corresponding to the feature point, so that the tasks of classification and position regression of the target frame are completed. WhereinClassifying whether the vehicle is a license plate or not, and determining the position information as the orientation frame of the license plate
Wherein
The center point of the license plate is represented,
indicating the length and width of the license plate toward the frame,
represents an angle with the horizontal direction;
(5) and (3) converting towards a frame: orientation frame conversion is based on five-point representation
Conversion to an orientation box based on a two-dimensional Gaussian distribution representation
Wherein
Can be prepared by
) The expression is made as follows:
converting the predicted orientation frame and the real orientation frame into a two-dimensional Gaussian distribution form through the formula;
(6) orientation frame similarity measure: after obtaining the form of the two-dimensional gaussian distribution of the predicted orientation box and the true orientation box, the distance between the two boxes needs to be measured, using KL divergence as the measure between the two distributions, which is formulated as follows:
(7) training a network structure to obtain trained model parameters: image size 640 using training set in dataset
480
3, inputting the images into the network in sequence, and inputting the whole network
Wherein B is the number of samples selected by one training, and the high-quality level candidate frame module outputs the classification confidence of the license plate by using the IOU threshold as the measurement standard of the sample distribution strategy
And regressive coordinate position
Wherein Class is 2, namely whether the license plate is detected, N is the number of output predicted license plate targets, and 5 is the coordinate of the center point of the orientation frame of the license plate, the length, the width and the angle of the orientation frame; calculating the error between the predicted license plate position and the real license plate position by adopting the Focal loss to calculate the error between the predicted license plate position and the real license plate position and adopting the similarity measurement of the orientation frames of the steps (5) and (6)Updating parameters through back propagation, saving the model parameters with the best results on the verification set after 200 rounds of complete training set training iteration, and taking the model parameters as the parameters of the final model training to obtain the trained license plate detection network parameters;
(8) testing the network and outputting the position and the type of the license plate: during the test, the image long side was scaled (resize) to 640 with the scale of the picture long and short sides unchanged, and then the short sides of the picture were filled in such that the image size was 640
480 as input to 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, and a non-maximum suppression (NMS) is used for deleting redundant frames output by the network to obtain the position of the license plate.
The embodiment aims at the license plate which is obliquely deformed towards the boundary box, the distribution distance between the predicted boundary box and the real boundary box is calculated by converting the orientation of the boundary box into a two-dimensional Gaussian distribution representation and using KL divergence as a measurement standard, and the distribution distance is used as the regression loss of the boundary box, so that the problem that the measurement loss of the smooth L1 is inconsistent with an evaluation mode is solved, the network training efficiency and the generalization capability of the license plate detection based on the orientation of the boundary box can be improved, and the network training stage is only used, and the calculation amount of an inference process is not increased.
It is noted that the present embodiment is intended to aid in further understanding of the present invention, but those skilled in the art will understand that: various substitutions and modifications are possible without departing from the spirit and scope of this disclosure and the 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 (6)
1. A license plate detection method based on two-dimensional Gaussian bounding box overlapping degree measurement is characterized by comprising the following steps:
(1) and (3) data set construction: collecting images containing conventional, inclined and distorted license plates in parking lots at the entrance and exit of a community and at the side positions, constructing an image data set containing the license plates, marking the positions of the license plates, and dividing the data set into a training set, a verification set and a test set, wherein the training set is used for training, the verification set is used for model verification in the training process, and the test set is used for final precision test;
(2) extracting features of the backbone network: initializing the size and the numerical range of the picture in the data set, and inputting the processed image into a backbone network for convolution feature extraction;
(3) multi-scale feature hierarchy: the image after the feature extraction is subjected to multi-scale feature hierarchical feature image set obtained through a multi-scale feature hierarchical structure based on feature fusion;
(4) positioning the license plate: connecting two convolution layers with convolution kernels of 3 x 3 and an average pooling layer after different multi-scale convolution layers according to the multi-scale feature hierarchical feature graph set obtained in the step (3), setting an anchor frame at each feature point of the feature graph, and respectively learning classification and position deviation information by using two full-connection layer sub-networks with the same structure but without sharing parameters to realize classification and position regression of the target frame; the position deviation information is the deviation of an anchor frame corresponding to a target frame and a characteristic point of the license plate, the classification of the category is that whether the license plate is judged by z, and the position is an orientation frame of the license plate
In which
The center point of the license plate is represented,
indicating the length and width of the license plate toward the frame,
represents an angle with the horizontal direction;
(5) and (3) converting towards a frame: orientation frame based on five-point representation
Converting the orientation frame expressed based on the two-dimensional Gaussian distribution into a two-dimensional Gaussian distribution form of a predicted orientation frame and a real orientation frame;
(6) orientation frame similarity measure: using the KL divergence as a metric between the two distributions to measure the distance between the predicted orientation box and the true orientation box;
(7) training a network structure to obtain trained model parameters: image size 640 using training set in dataset
480
3, sequentially inputting the images into the network, and outputting the classification confidence of the license plate by using the IOU threshold as a measurement standard of a sample distribution strategy
And regressive coordinate position
B is the number of samples selected by one-time training, Class is 2, namely whether the license plate is the license plate or not, N is the number of output and predicted license plate targets, and 5 is the coordinate of the center point of the license plate facing to the frame, the length, the width and the angle of the frame; adopting Focal loss to calculate errors of a prediction category and a real category, adopting step-oriented frame similarity measurement to calculate errors of a predicted license plate position and a real license plate position, updating parameters through back propagation, saving model parameters with the best results on a verification set after 200 times of complete training set training iteration, and obtaining trained license plate detection network parameters;
(8) outputting license plate positions and types: under the condition of keeping the proportion of the long side and the short side of the picture unchanged, the long side of the picture is zoomed (resize) to 640, then the short side of the picture is filled, the size of the picture is 640 multiplied by 480, the picture is used as the input of a license plate detection network, namely, the classification confidence coefficient 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 coefficient, and redundant frames output by the network are deleted by using non-maximum inhibition, so that the position of the license plate is obtained.
2. The method for detecting the license plate based on the two-dimensional Gaussian bounding box overlap degree measurement as claimed in claim 1, wherein the step (1) of labeling the license plate is to label four corner points of the license plate, and the center point, the length and the width of the license plate facing the frame and the included angle with the horizontal direction are calculated according to the positions of the four corner points, so as to supervise the positioning of the license plate.
3. The two-dimensional Gaussian bounding box overlap measurement-based license plate detection method of claim 2, wherein in the step (2), the backbone network respectively extracts C3, C4 and C5 and inputs the extracted C3, C4 and C5 into subsequent networks by using pre-trained VGG16 in ImageNet data set as a feature extraction network.
4. The license plate detection method based on two-dimensional Gaussian bounding box overlap degree measurement according to claim 3, characterized in that in step (3), the multi-scale feature hierarchical structure based on feature fusion performs semantic expansion on shallow features through a top-down path, and combines top-level feature mapping with bottom-level information through a bottom-up path, and each layer of feature hierarchy can capture rich semantic representation and bottom-level information at the same time; the top-down path uses the up-sampling module and the convolution layer to amplify the features and fuse the features with corresponding sizes, the bottom-up path uses the down-sampling module and the convolution layer to scale the features and fuse the features with corresponding sizes, and finally, features of different layers are combined to form a strong feature representation for the multi-scale target.
5. According to the claimsSolving 4 the license plate detection method based on the two-dimensional Gaussian boundary frame overlapping degree measurement, wherein in the step (5), the orientation frame represented based on the two-dimensional Gaussian distribution is
Wherein
By (a)
) Expressed, the formula is as follows:
the two-dimensional Gaussian distribution forms of the predicted orientation frame and the real orientation frame are respectively
。
6. The vehicle license plate detection method based on two-dimensional Gaussian bounding box overlap degree measurement according to claim 5, wherein the KL divergence calculation formula in the step (6) is as follows:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210595586.5A CN114677501A (en) | 2022-05-30 | 2022-05-30 | License plate detection method based on two-dimensional Gaussian bounding box overlapping degree measurement |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210595586.5A CN114677501A (en) | 2022-05-30 | 2022-05-30 | License plate detection method based on two-dimensional Gaussian bounding box overlapping degree measurement |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114677501A true CN114677501A (en) | 2022-06-28 |
Family
ID=82079454
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210595586.5A Pending CN114677501A (en) | 2022-05-30 | 2022-05-30 | License plate detection method based on two-dimensional Gaussian bounding box overlapping degree measurement |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114677501A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114882490A (en) * | 2022-07-08 | 2022-08-09 | 松立控股集团股份有限公司 | Unlimited scene license plate detection and classification method based on point-guided positioning |
| CN115063786A (en) * | 2022-08-18 | 2022-09-16 | 松立控股集团股份有限公司 | High-order distant view fuzzy license plate detection method |
| CN115171079A (en) * | 2022-09-08 | 2022-10-11 | 松立控股集团股份有限公司 | Vehicle detection method based on night scene |
| CN115512341A (en) * | 2022-09-15 | 2022-12-23 | 粤丰科盈智能投资(广东)有限公司 | Target detection method and device based on Gaussian distribution fitting and computer medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20210149022A1 (en) * | 2019-11-14 | 2021-05-20 | Toyota Research Institute, Inc. | Systems and methods for 3d object detection |
| CN113609895A (en) * | 2021-06-22 | 2021-11-05 | 上海中安电子信息科技有限公司 | Road traffic information acquisition method based on improved Yolov3 |
| 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
- 2022-05-30 CN CN202210595586.5A patent/CN114677501A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20210149022A1 (en) * | 2019-11-14 | 2021-05-20 | Toyota Research Institute, Inc. | Systems and methods for 3d object detection |
| CN113609895A (en) * | 2021-06-22 | 2021-11-05 | 上海中安电子信息科技有限公司 | Road traffic information acquisition method based on improved Yolov3 |
| 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 |
|---|
| XUE YANG,AND ETC: "Learning High-Precision Bounding Box for Rotated Object Detection via Kullback-Leibler Divergence", 《ARXIV：2106.01883V5》 * |
| 张筱晗等: "基于中心点的遥感图像多方向船舰目标检测", 《光子学报》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114882490A (en) * | 2022-07-08 | 2022-08-09 | 松立控股集团股份有限公司 | Unlimited scene license plate detection and classification method based on point-guided positioning |
| CN114882490B (en) * | 2022-07-08 | 2022-09-20 | 松立控股集团股份有限公司 | Unlimited scene license plate detection and classification method based on point-guided positioning |
| CN115063786A (en) * | 2022-08-18 | 2022-09-16 | 松立控股集团股份有限公司 | High-order distant view fuzzy license plate detection method |
| CN115171079A (en) * | 2022-09-08 | 2022-10-11 | 松立控股集团股份有限公司 | Vehicle detection method based on night scene |
| CN115171079B (en) * | 2022-09-08 | 2023-04-07 | 松立控股集团股份有限公司 | Vehicle detection method based on night scene |
| CN115512341A (en) * | 2022-09-15 | 2022-12-23 | 粤丰科盈智能投资(广东)有限公司 | Target detection method and device based on Gaussian distribution fitting and computer medium |
| CN115512341B (en) * | 2022-09-15 | 2023-10-27 | 粤丰科盈智能投资(广东)有限公司 | Target detection method, device and computer medium based on Gaussian distribution fitting |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114677501A (en) | License plate detection method based on two-dimensional Gaussian bounding box overlapping degree measurement | |
| CN111368687B (en) | Sidewalk vehicle illegal parking detection method based on target detection and semantic segmentation | |
| CN111178213B (en) | Aerial photography vehicle detection method based on deep learning | |
| CN112016605B (en) | Target detection method based on corner alignment and boundary matching of bounding box | |
| CN113822278B (en) | License plate recognition method for unlimited scene | |
| CN110688902B (en) | Method and device for detecting vehicle area in parking space | |
| CN112906816B (en) | Target detection method and device based on optical differential and two-channel neural network | |
| CN110956207B (en) | Method for detecting full-element change of optical remote sensing image | |
| CN113343985B (en) | License plate recognition method and device | |
| CN111340026A (en) | Training method of vehicle annual payment identification model and vehicle annual payment identification method | |
| CN115063786A (en) | High-order distant view fuzzy license plate detection method | |
| CN113111722A (en) | Automatic driving target identification method based on improved Mask R-CNN | |
| CN112634369A (en) | Space and or graph model generation method and device, electronic equipment and storage medium | |
| CN109829469A (en) | A kind of vehicle checking method based on deep learning | |
| CN112634368A (en) | Method and device for generating space and OR graph model of scene target and electronic equipment | |
| König et al. | What's Cracking? A Review and Analysis of Deep Learning Methods for Structural Crack Segmentation, Detection and Quantification | |
| Mijić et al. | Traffic sign detection using yolov3 | |
| CN113609895A (en) | Road traffic information acquisition method based on improved Yolov3 | |
| CN114882490B (en) | Unlimited scene license plate detection and classification method based on point-guided positioning | |
| CN116665153A (en) | Road scene segmentation method based on improved deep bv3+ network model | |
| CN115346206B (en) | License plate detection method based on improved super-resolution deep convolution feature recognition | |
| CN113313008B (en) | Target and identification tracking method based on YOLOv3 network and mean shift | |
| CN115223114A (en) | End-to-end vehicle attitude estimation method based on bidirectional fusion feature pyramid | |
| CN115035164A (en) | Moving target identification method and device | |
| CN114463720A (en) | Lane line detection method based on line segment intersection-to-parallel ratio loss function |
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 |
Application publication date: 20220628 |
|
| RJ01 | Rejection of invention patent application after publication |