CN110210354A - A kind of detection of haze weather traffic mark with know method for distinguishing - Google Patents
A kind of detection of haze weather traffic mark with know method for distinguishing Download PDFInfo
- Publication number
- CN110210354A CN110210354A CN201910433953.XA CN201910433953A CN110210354A CN 110210354 A CN110210354 A CN 110210354A CN 201910433953 A CN201910433953 A CN 201910433953A CN 110210354 A CN110210354 A CN 110210354A
- Authority
- CN
- China
- Prior art keywords
- traffic mark
- image
- defogging
- cnn
- haze 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 30
- 238000001514 detection method Methods 0.000 title claims abstract description 23
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 36
- 238000013528 artificial neural network Methods 0.000 claims abstract description 10
- 238000012549 training Methods 0.000 claims description 22
- 230000008569 process Effects 0.000 claims description 13
- 238000005065 mining Methods 0.000 claims description 6
- 230000006978 adaptation Effects 0.000 claims description 3
- 239000000654 additive Substances 0.000 claims description 3
- 230000000996 additive effect Effects 0.000 claims description 3
- 238000009412 basement excavation Methods 0.000 claims description 3
- 238000005259 measurement Methods 0.000 claims description 3
- 230000001373 regressive effect Effects 0.000 claims description 3
- 230000001629 suppression Effects 0.000 claims description 2
- 238000005516 engineering process Methods 0.000 description 6
- 238000012545 processing Methods 0.000 description 6
- 238000013135 deep learning Methods 0.000 description 5
- 238000011160 research Methods 0.000 description 4
- 230000005540 biological transmission Effects 0.000 description 3
- 238000013136 deep learning model Methods 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 235000014676 Phragmites communis Nutrition 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 239000003595 mist Substances 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 230000004888 barrier function Effects 0.000 description 1
- 230000033228 biological regulation Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008450 motivation Effects 0.000 description 1
- 230000035772 mutation Effects 0.000 description 1
- 210000004218 nerve net Anatomy 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000009394 selective breeding Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
Classifications
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/73—Deblurring; Sharpening
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
- G06V20/582—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads of traffic signs
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of detections of haze weather traffic mark and knowledge method for distinguishing, firstly, the image defogging model AOD_Net established to the haze weather traffic scene picture obtained in advance according to convolutional neural networks carries out defogging；In conjunction with defogging treated image, the characteristic information of picture is extracted according to convolutional neural networks, is screened candidate region by three-stage cascade convolutional neural networks, is exported two classification results and area coordinate；Finally, the region of traffic mark is input in Densenet neural network, specific classification results are predicted, to mark the result information of detection identification on output picture.The haze picture taken is handled using this method, the rate and rate of precision of the defogging detection identification of this method are higher, can satisfy the demand that current traffic system auxiliary drives substantially.
Description
Technical field
The invention belongs to the technical field of computer vision of deep learning, and in particular to a kind of haze weather traffic mark inspection
It surveys and knows method for distinguishing.
Background technique
The many companies of recent domestic and scientific research institution are dedicated to the research of unmanned technology, unmanned technology day
It is beneficial mature, the detection of traffic mark and the development of identification technology to automatic driving car on road when driving avoiding barrier and
It has observed traffic rules and regulations vital effect.Currently based on the environment such as weather, road it is good under the conditions of traffic mark examine
Survey and identification technology relative maturity, but it is directed to the variability of China's current traffic environment, especially haze etc. is boisterous
Occur, the detection of traffic mark is lower with accuracy of identification in the Intelligent unattended driving based on Vehicular video, and Intelligent unattended is caused to drive
Sail that there are major safety risks.And at present both at home and abroad for the detection and identification of the traffic mark in Vehicular video under haze weather
The research of technology encounters bottleneck.
The concept of deep learning is derived from the research of artificial neural network, is one kind of unsupervised learning, motivation is to build
Vertical, simulation human brain carries out the neural network of analytic learning, which can form more abstract high level by combination low-level feature
Attribute classification or feature are indicated, to find that the distributed nature of data indicates.Deep learning image recognition, speech recognition and from
Right Language Processing etc. all has been greatly developed.Deep learning in 1898 is attempted to answer by LeCun and his colleague for the first time
For field of image recognition, and very ten-strike is achieved, using a kind of deep neural network-with convolutional coding structure
Convolutional neural networks (Convolution Neural Networks, CNN).In recent years, field of image recognition researchers'
Under unremitting effort and innovation, more and more deep learning models and its mutation are published at applied to the achievement of image recognition
It is many.
There is presently no the precedents that full utilization deep learning combines image defogging with traffic mark detection technique, lead to
The model of depth learning technology building is crossed, picture feature information is extracted, picture defogging is completed and traffic mark detection identifies
Processing, the method for carrier vehicle video processing, under severe haze weather, the lower situation of visibility, auxiliary driver drive
It sails, or is used for the deep learning model continued to optimize to assist automatic Pilot field.
Summary of the invention
Goal of the invention: the present invention provides a kind of detection of haze weather traffic mark and knows method for distinguishing, can solve in mist
Under haze weather condition, visibility reduces, road environment is poor, carrier vehicle video frame picture quality is poor, feature Fuzzy occurs, color
The problems such as distortion, solves the problems such as effect is undesirable.
Summary of the invention: a kind of haze weather traffic mark detection of the present invention with know method for distinguishing, including following step
It is rapid:
(1) the image defogging model that the haze weather traffic scene picture obtained in advance is established according to convolutional neural networks
AOD_Net carries out defogging；
(2) defogging treated image is combined, the characteristic information of picture is extracted according to convolutional neural networks, passes through three-level grade
Join convolutional neural networks and screen candidate region, exports two classification results and area coordinate；
(3) region of traffic mark is input in Densenet neural network, predicts specific classification results, thus
On output picture, the result information of detection identification is marked.
Image defogging model described in step (1) is by K-estimate module and clean image generation module
Composition；K-estimate module is estimated to minimize the reconstructed error of image pixel fields, clean image using 5 convolutional layers
Generation module is made of an element-wise multiplication layer and two element-wise additive layers.
Three-stage cascade convolutional neural networks building process described in step (2) are as follows:
The cascade CNN multi-task learning of the unification that frame uses includes three phases, passes through a shallow CNN in the first stage
Quickly generate candidate window；Second stage refines window by more complicated CNN, rejects a large amount of non-traffic mark window
Mouthful；Phase III refines result using more powerful CNN, exports traffic mark coordinate position and two classification results.
The step (2) the following steps are included:
(21) one image pyramid is constructed, below for its size is adjusted to different scales for given picture
The input of three phases cascade frame:
First stage: using full convolutional network to obtain candidate window and its bounding box regression vector, estimation is then utilized
Bounding box regression vector candidate vector is demarcated, then, the time of high superposed is merged using non-maximum suppression (NMS)
Option；
Second stage: feeding back to another CNN for all candidate targets, further refuses a large amount of false candidates object,
It is calibrated using bounding box recurrence, and merges NMS candidate target；
Phase III: being more fully described traffic mark, which will export 5 traffic mark positions；
(23) three tasks train CNN detector: traffic mark classification, non-traffic mark classification, bounding box return and
Traffic mark positioning.
The step (22) the following steps are included:
(221) traffic mark is classified: learning objective being defined as two classification problems, for each sample, uses cross entropy
Loss；
It is wherein piIndicate the sample that network generates for the probability of traffic mark, symbolIndicate true tag；
(222) bounding box returns: for each candidate window, predicting that it is inclined between immediate ground truth
Shifting amount, packet expand frame left side top, height and width, learning objective and are represented as a regression problem, use Europe to each sample
A few Reed losses:
It wherein, is true value coordinate, including the upper left corner, height and width by the regressive object that network obtains；
(223) traffic mark coordinate setting: it is similar with bounding box recurrence task, by traffic mark coordinate measurement problem representation
For regression problem, by European minimization of loss:
(224) multi-source training: due to using different tasks in each CNN, so will appear not in learning process
The training image of same type only calculates the sample of background areaOther two losses are set as 0, global learning target
It can indicate are as follows:
It is wherein N training sample number, αiThe importance of expression task utilizes stochastic gradient descent in this case
Training neural network；
(225) online difficult sample Mining Strategy: it is different from the difficult sample excavation after traditional original classification device training,
Online difficulty sample Mining Strategy is used for adaptation training process.
The utility model has the advantages that compared with prior art, beneficial effects of the present invention: the method is completed for the processing of carrier vehicle video
The processing of picture defogging and detection identification, in different degrees of haze weather, for combining picture defogging and handing over
The detection identification of logical mark has relatively good as a result, meeting most demand, can be used for that driver is assisted to drive, or will not
The deep learning model of disconnected optimization is for assisting automatic Pilot field.
Detailed description of the invention
Fig. 1 is flow chart of the method for the present invention；
Fig. 2 is defogging prototype network module map；
Fig. 3 is K-estimate module network structure chart.
Specific embodiment
The present invention is described in further detail below in conjunction with the accompanying drawings, as shown in Figure 1, the present invention the following steps are included:
1, the image defogging model AOD_Net established according to a kind of convolutional neural networks, it is based on the figure redefined
Defogging is carried out to the picture taken as defogging model, and according to described image defogging model.
As shown in Fig. 2, according to the image defogging model redefined, image defogging model removes the picture taken
Mist Processing Algorithm and operation are described in detail below:
The image defogging model does not estimate transmission matrix and atmosphere light respectively as previous most models, but logical
It crosses lightweight CNN and directly generates clearly image；
Physical model: atmospherical scattering model is always the classical description of blurred picture generating process:
I (x)=J (x) t (x)+A (1-t (x))
Wherein I (x) is the blurred picture observed, and J (x) is scene brightness (clean image) to be restored, wherein having
Two key parameters: A is global atmosphere light, and t (x) is transmission matrix；
T (x)=e-βd(x)
The scattering coefficient β, d (x) of atmosphere are the distance between object and video camera；
AOD-Net is constructed based on different ideas, the physical model can by it is a kind of end to end in a manner of indicate,
All parameters are all estimated that AOD-Net does not need the image after directly exporting defogging any in a unified model
Intermediate steps estimate parameter, and different from the end-to-end study from blurred picture to transmission matrix, AOD-Net's is complete end-to-end
Formula has erected bridge between blurred picture and clear image；
The network core thought is that two t (x), A parameters are unified for a formula, and the K (x) in following formula, is directly most
The reconstructed error of smallization image pixel fields:
J (x)=K (x) I (x)-K (x)+b
Wherein, b is constant biasing, default value 1；
AOD-Net consists of two parts: a K-estimate module, it estimates K (x) using 5 convolutional layers, then
It is a clean image generation module, it is by an element-wise multiplication layer and two element-wise
Additive layer composition restores image for calculating to generate；K-estimate module is the key component of AOD-Net, and it is deep to be responsible for estimation
Degree and opposite haze grade form multiple dimensioned spy by merging different size of filter as shown in figure 3, having used 5 convolutional layers
Sign, using the parallel-convolution of different filter sizes, the feature of thick scale network has been connect with the middle layer of thin scale network
Come, the concat1 layer of AODNet connects conv1 and conv2 layers of characteristic, and concat2 will be in conv2 and conv3
It connects, concat3 connects the value from " conv1 ", " conv2 ", " conv3 " and " conv4 ", this multiple dimensioned
Design can capture the characteristic of different scale, and centre connection can also compensate for the information loss in convolution process；
In the training process, weight is initialized using Gaussian random variable, it is non-linear using the generation of ReLU neuron,
Momentum and attenuation parameter are respectively set to 0.9 and 0.0001, using simple mean square error (MSE) loss function；According to
The upper parameter and network structure, are trained this network with data set.
2, contain traffic mark image in conjunction with what defogging model exported, convolutional neural networks extract the figure containing traffic mark
The characteristic information of piece, the multitask concatenated convolutional network integration Densenet nerve net completed by the training of traffic mark data set
Network, output category result and area coordinate, to mark the result information of detection identification on output picture.
Multitask frame is cascaded according to depth, and the depth convolutional network including well-designed three phases can be with
Rough arrive subtly predicts traffic mark and coordinate position, and in learning process, excavates plan according to a kind of online difficult sample
Slightly, performance can be improved automatically in the case where not needing artificial selection sample, algorithm is described in detail below:
The cascade CNN multi-task learning of the unification that frame uses includes three phases, and in the first stage, it is shallow by one
CNN quickly generates candidate window；Window is refined by more complicated CNN, rejects a large amount of non-traffic mark window；Make
Result is refined with more powerful CNN, exports traffic mark coordinate position；
For given image, for its size is adjusted to different scales, an image pyramid is constructed first, this
It is the input of following three phases cascade frame；
First stage: utilize full convolutional network, referred to as Proposal Network (P-Net), with obtain candidate window and
Then its bounding box regression vector is demarcated candidate vector using the bounding box regression vector of estimation, then, using it is non-most
It is big to inhibit (NMS) to merge the candidate item of high superposed；
Second stage: feeding back to another CNN, referred to as Refine Network (R-Net) for all candidate targets, it into
One step refuses a large amount of false candidates object, is calibrated using bounding box recurrence, and merge NMS candidate target；
Phase III: this stage is similar to second stage, but in this stage, and target is that traffic mark is more fully described
Know, which will export 5 traffic mark positions；
CNN detector is trained using three tasks: traffic mark/non-traffic mark classification, bounding box returns and traffic
Mark location；
1) traffic mark is classified: learning objective being defined as two classification problems, for each sample, is damaged using cross entropy
It loses；
It is wherein piIndicate the sample that network generates for the probability of traffic mark, symbolIndicate true tag；
2) bounding box returns: for each candidate window, predicting its offset between immediate groundtruth
Amount, packet expand frame left side top, height and width, learning objective and are represented as a regression problem, several using Europe to each sample
Reed loss:
It wherein, is true value coordinate, including the upper left corner, height and width by the regressive object that network obtains；
3) traffic mark coordinate setting: it is similar with bounding box recurrence task, be by traffic mark coordinate measurement problem representation
Regression problem, by European minimization of loss:
4) multi-source training: due to using different tasks in each CNN, so will appear inhomogeneity in learning process
The training image of type only calculates the sample of background areaOther two losses are set as 0, and global learning target can be with
It indicates are as follows:
It is wherein N training sample number, αiThe importance of expression task utilizes stochastic gradient descent in this case
Training neural network；
5) online difficult sample Mining Strategy: it is different from the difficult sample excavation after traditional original classification device training,
Line difficulty sample Mining Strategy is used for adaptation training process；
Particularly, it is each it is small quantities of in, all samples are ranked up in the loss that the forward-propagating stage calculates, and selects
Preceding 70%, as difficult sample, then only calculates the gradient of back-propagating stage difficulty sample, this has ignored that in training
Little simple sample is helped to enhancing detector a bit；
Four kinds of different data marks have been used in training process:
(i) negatives: intersected region of the ratio less than 0.3 of simultaneously (IoU) and any ground truth；
(ii) positives:IoU is higher than 0.65 to groundtruth；
(iii) part face: the IOU with ground truth is between 0.4~0.65；
(iv) traffic mark coordinate: 5 coordinate positions of traffic mark label, two classification task of traffic mark use
Positives, negatives, bounding box, which returns, uses positives, part face, and traffic mark coordinate uses traffic mark
Know characteristic point.
3, the region of traffic mark is input in Densenet neural network, predicts specific classification results, thus
On output picture, the result information of detection identification is marked.
The above embodiments merely illustrate the technical concept and features of the present invention, and its object is to allow person skilled in the art
Scholar cans understand the content of the present invention and implement it accordingly, and it is not intended to limit the scope of the present invention.It is all according to the present invention
Equivalent change or modification made by Spirit Essence, should be covered by the protection scope of the present invention.
Claims (5)
1. a kind of haze weather traffic mark detection and knowledge method for distinguishing, which comprises the following steps:
(1) the image defogging model AOD_ that the haze weather traffic scene picture obtained in advance is established according to convolutional neural networks
Net carries out defogging；
(2) defogging treated image is combined, the characteristic information of picture is extracted according to convolutional neural networks, is rolled up by three-stage cascade
Product neural network screens candidate region, exports two classification results and area coordinate；
(3) region of traffic mark is input in Densenet neural network, predicts specific classification results, thus defeated
Out on picture, the result information of detection identification is marked.
2. a kind of haze weather traffic mark detection and knowledge method for distinguishing according to claim 1, which is characterized in that step
(1) the image defogging model described in is made of K-estimate module and clean image generation module；K-
Estimate module is estimated to minimize the reconstructed error of image pixel fields, clean image using 5 convolutional layers
Generation module is made of an element-wise multiplication layer and two element-wise additive layers.
3. a kind of haze weather traffic mark detection and knowledge method for distinguishing according to claim 1, which is characterized in that step
(2) the three-stage cascade convolutional neural networks building process described in are as follows:
The cascade CNN multi-task learning of the unification that frame uses includes three phases, quick by a shallow CNN in the first stage
Generate candidate window；Second stage refines window by more complicated CNN, rejects a large amount of non-traffic mark window；
Phase III refines result using more powerful CNN, exports traffic mark coordinate position and two classification results.
4. a kind of haze weather traffic mark detection and knowledge method for distinguishing according to claim 1, which is characterized in that the step
Suddenly (2) the following steps are included:
(21) one image pyramid is constructed in order its size is adjusted to different scales for given picture, three below
The input of multi-stage cascade frame:
First stage: using full convolutional network to obtain candidate window and its bounding box regression vector, the side of estimation is then utilized
Boundary's frame regression vector demarcates candidate vector, then, the candidate of high superposed is merged using non-maximum suppression (NMS)
?；
Second stage: feeding back to another CNN for all candidate targets, further refuses a large amount of false candidates object, uses
Bounding box recurrence is calibrated, and merges NMS candidate target；
Phase III: being more fully described traffic mark, which will export 5 traffic mark positions；
(22) three tasks train CNN detector: traffic mark classification, the classification of non-traffic mark, bounding box returns and traffic
Mark location.
5. a kind of haze weather traffic mark detection and knowledge method for distinguishing according to claim 4, which is characterized in that the step
Suddenly (22) the following steps are included:
(221) traffic mark is classified: learning objective being defined as two classification problems, for each sample, uses intersection entropy loss；
It is wherein piIndicate the sample that network generates for the probability of traffic mark, symbolIndicate true tag；
(222) bounding box returns: for each candidate window, predict its offset between immediate groundtruth,
Packet expands frame left side top, height and width, learning objective and is represented as a regression problem, uses euclidean to each sample
Loss:
It wherein, is true value coordinate, including the upper left corner, height and width by the regressive object that network obtains；
(223) traffic mark coordinate setting: it is similar with bounding box recurrence task, traffic mark coordinate measurement problem representation is back
Return problem, by European minimization of loss:
(224) multi-source training: due to using different tasks in each CNN, so will appear inhomogeneity in learning process
The training image of type only calculates the sample of background areaOther two losses are set as 0, and global learning target can be with
It indicates are as follows:
It is wherein N training sample number, αiThe importance of expression task utilizes stochastic gradient descent training mind in this case
Through network；
(225) online difficult sample Mining Strategy: it is different from the difficult sample excavation after traditional original classification device training, online
Difficult sample Mining Strategy is used for adaptation training process.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910433953.XA CN110210354A (en) | 2019-05-23 | 2019-05-23 | A kind of detection of haze weather traffic mark with know method for distinguishing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910433953.XA CN110210354A (en) | 2019-05-23 | 2019-05-23 | A kind of detection of haze weather traffic mark with know method for distinguishing |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110210354A true CN110210354A (en) | 2019-09-06 |
Family
ID=67788445
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910433953.XA Pending CN110210354A (en) | 2019-05-23 | 2019-05-23 | A kind of detection of haze weather traffic mark with know method for distinguishing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110210354A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110689081A (en) * | 2019-09-30 | 2020-01-14 | 中国科学院大学 | Weak supervision target classification and positioning method based on bifurcation learning |
| CN110717534A (en) * | 2019-09-30 | 2020-01-21 | 中国科学院大学 | Target classification and positioning method based on network supervision |
| CN110837800A (en) * | 2019-11-05 | 2020-02-25 | 畅加风行(苏州)智能科技有限公司 | Port severe weather-oriented target detection and identification method |
| CN111161160A (en) * | 2019-12-04 | 2020-05-15 | 新奇点企业管理集团有限公司 | Method and device for detecting obstacle in foggy weather, electronic equipment and storage medium |
| CN111291624A (en) * | 2020-01-16 | 2020-06-16 | 国网山西省电力公司电力科学研究院 | Excavator target identification method and system |
| CN111523493A (en) * | 2020-04-27 | 2020-08-11 | 东南数字经济发展研究院 | Target detection algorithm for foggy weather image |
| CN112837281A (en) * | 2021-01-27 | 2021-05-25 | 湘潭大学 | Pin defect identification method, device and equipment based on cascade convolutional neural network |
| CN112949389A (en) * | 2021-01-28 | 2021-06-11 | 西北工业大学 | Haze image target detection method based on improved target detection network |
| CN113392804A (en) * | 2021-07-02 | 2021-09-14 | 昆明理工大学 | Multi-angle-based traffic police target data set scene construction method and system |
| CN115063304A (en) * | 2022-05-19 | 2022-09-16 | 湖南师范大学 | End-to-end multi-size fusion-based pyramid neural network image defogging method and system |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107316001A (en) * | 2017-05-31 | 2017-11-03 | 天津大学 | Small and intensive method for traffic sign detection in a kind of automatic Pilot scene |
| CN107609485A (en) * | 2017-08-16 | 2018-01-19 | 中国科学院自动化研究所 | The recognition methods of traffic sign, storage medium, processing equipment |
| US10262218B2 (en) * | 2017-01-03 | 2019-04-16 | Qualcomm Incorporated | Simultaneous object detection and rigid transform estimation using neural network |
-
2019
- 2019-05-23 CN CN201910433953.XA patent/CN110210354A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10262218B2 (en) * | 2017-01-03 | 2019-04-16 | Qualcomm Incorporated | Simultaneous object detection and rigid transform estimation using neural network |
| CN107316001A (en) * | 2017-05-31 | 2017-11-03 | 天津大学 | Small and intensive method for traffic sign detection in a kind of automatic Pilot scene |
| CN107609485A (en) * | 2017-08-16 | 2018-01-19 | 中国科学院自动化研究所 | The recognition methods of traffic sign, storage medium, processing equipment |
Non-Patent Citations (4)
| Title |
|---|
| CHERRY1307: "AOD-NET：An All-in-One Network for Dehazing and Beyond", 《HTTPS://BLOG.CSDN.NET/CHERRY1307/ARTICLE/DETAILS/88395074》 * |
| ZHANG K , ET AL: "Joint Face Detection and Alignment Using Multitask Cascaded Convolutional Networks", 《IEEE SIGNAL PROCESSING LETTERS》 * |
| 李晨，等: "基于卷积神经网络的交通标志实时检测与识别", 《数字技术与应用》 * |
| 贾永红，等: "《数字图像处理技巧》", 31 January 2017 * |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110717534B (en) * | 2019-09-30 | 2020-09-15 | 中国科学院大学 | Target classification and positioning method based on network supervision |
| CN110717534A (en) * | 2019-09-30 | 2020-01-21 | 中国科学院大学 | Target classification and positioning method based on network supervision |
| CN110689081A (en) * | 2019-09-30 | 2020-01-14 | 中国科学院大学 | Weak supervision target classification and positioning method based on bifurcation learning |
| CN110837800A (en) * | 2019-11-05 | 2020-02-25 | 畅加风行(苏州)智能科技有限公司 | Port severe weather-oriented target detection and identification method |
| CN111161160A (en) * | 2019-12-04 | 2020-05-15 | 新奇点企业管理集团有限公司 | Method and device for detecting obstacle in foggy weather, electronic equipment and storage medium |
| CN111291624A (en) * | 2020-01-16 | 2020-06-16 | 国网山西省电力公司电力科学研究院 | Excavator target identification method and system |
| CN111523493A (en) * | 2020-04-27 | 2020-08-11 | 东南数字经济发展研究院 | Target detection algorithm for foggy weather image |
| CN112837281A (en) * | 2021-01-27 | 2021-05-25 | 湘潭大学 | Pin defect identification method, device and equipment based on cascade convolutional neural network |
| CN112837281B (en) * | 2021-01-27 | 2022-10-28 | 湘潭大学 | Pin defect identification method, device and equipment based on cascade convolution neural network |
| CN112949389A (en) * | 2021-01-28 | 2021-06-11 | 西北工业大学 | Haze image target detection method based on improved target detection network |
| CN113392804A (en) * | 2021-07-02 | 2021-09-14 | 昆明理工大学 | Multi-angle-based traffic police target data set scene construction method and system |
| CN113392804B (en) * | 2021-07-02 | 2022-08-16 | 昆明理工大学 | Multi-angle-based traffic police target data set scene construction method and system |
| CN115063304A (en) * | 2022-05-19 | 2022-09-16 | 湖南师范大学 | End-to-end multi-size fusion-based pyramid neural network image defogging method and system |
| CN115063304B (en) * | 2022-05-19 | 2023-08-25 | 湖南师范大学 | Multi-size fused pyramid neural network image defogging method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110210354A (en) | A kind of detection of haze weather traffic mark with know method for distinguishing | |
| CN110111335B (en) | Urban traffic scene semantic segmentation method and system for adaptive countermeasure learning | |
| Hu et al. | Pavement crack detection method based on deep learning models | |
| CN110163187B (en) | F-RCNN-based remote traffic sign detection and identification method | |
| CN111310773B (en) | Efficient license plate positioning method of convolutional neural network | |
| CN109902806A (en) | Method is determined based on the noise image object boundary frame of convolutional neural networks | |
| CN111563909B (en) | Semantic segmentation method for complex street view image | |
| CN108229338A (en) | A kind of video behavior recognition methods based on depth convolution feature | |
| CN107609525A (en) | Remote Sensing Target detection method based on Pruning strategy structure convolutional neural networks | |
| CN104517103A (en) | Traffic sign classification method based on deep neural network | |
| CN106096602A (en) | A kind of Chinese licence plate recognition method based on convolutional neural networks | |
| CN110796009A (en) | Method and system for detecting marine vessel based on multi-scale convolution neural network model | |
| CN107016357A (en) | A kind of video pedestrian detection method based on time-domain convolutional neural networks | |
| CN113807464B (en) | Unmanned aerial vehicle aerial image target detection method based on improved YOLO V5 | |
| Liu et al. | Deep multi-level fusion network for multi-source image pixel-wise classification | |
| CN110009095A (en) | Road driving area efficient dividing method based on depth characteristic compression convolutional network | |
| CN111695448A (en) | Roadside vehicle identification method based on visual sensor | |
| CN112434723B (en) | Day/night image classification and object detection method based on attention network | |
| CN111832453B (en) | Unmanned scene real-time semantic segmentation method based on two-way deep neural network | |
| CN110310241A (en) | A kind of more air light value traffic image defogging methods of fusion depth areas segmentation | |
| CN110197152A (en) | A kind of road target recognition methods for automated driving system | |
| CN112686276A (en) | Flame detection method based on improved RetinaNet network | |
| CN107038442A (en) | A kind of car plate detection and global recognition method based on deep learning | |
| CN111310593B (en) | Ultra-fast lane line detection method based on structure perception | |
| CN111540203B (en) | Method for adjusting green light passing time based on fast-RCNN |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190906 |