CN108009509A - Vehicle target detection method - Google Patents
Vehicle target detection method Download PDFInfo
- Publication number
- CN108009509A CN108009509A CN201711321241.6A CN201711321241A CN108009509A CN 108009509 A CN108009509 A CN 108009509A CN 201711321241 A CN201711321241 A CN 201711321241A CN 108009509 A CN108009509 A CN 108009509A
- Authority
- CN
- China
- Prior art keywords
- vehicle
- target
- target detection
- detection method
- faster
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- 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
- G06V20/584—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads of vehicle lights or traffic lights
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of vehicle target detection method, it is intended to solves the technical problem that traditional detection method testing process is complicated, computationally intensive, detection is inaccurate.The present invention makes 2007 vehicle data collection of VOC according to ImageNet data sets；Using caffe deep learning frames, using improved Faster R CNN algorithm configuration training patterns, and introduce Inception network structures and feature extraction is carried out to image；A kind of area is added as 642Sliding window be used for detect Small object；Target detection problems in scene are converted to two classification problems of target, carry out the detection identification of vehicle target；Optimized with RPN loss functions；Classify to vehicle image characteristic use SoftMax algorithms, so as to obtain final result.The advantageous effects of the present invention are：The calculation amount of characteristics of image reduces, testing process is simplified, extract the enhancing of characteristics of image ability, the accuracy of identification of network improves, and testing result is more comprehensively.
Description
Technical field
The present invention relates to target information detection technique field, and in particular to a kind of vehicle target detection method.
Background technology
In recent years, as the rapid growth of vehicular traffic quantity, traffic monitoring face huge challenge.Vehicle target detection is made
To build a key technology of traffic video monitoring, it is subject to the extensive concern of domestic and international researcher for a long time.Target
Detection is broadly divided into based on background modeling as image procossing and an important branch of computer vision, its research method
Method and the method based on appearance features information.
Domestic and foreign scholars have carried out many trials using traditional machine learning method, by being carried to target signature
Take, such as HOG (histogram of oriented gradient), SIFT (scale invariant feature
Transform) the methods of, and the feature extracted is inputted to grader, such as support vector machines（support vector
machine）, iterator (AdaBoost) etc. carry out Classification and Identification.These features are substantially a kind of features of hand-designed, pin
To different identification problems, the feature quality extracted directly affects system performance, it is therefore desirable to which researcher is to wanting
Solve the problems, such as that field carries out in-depth study, to design the feature of better adaptability.Warp of this method to technical staff
Test and require high and computationally intensive, operating process complexity, since it is directed to some specific identification mission, it is difficult in reality
Accurately recognition effect is realized among application problem.
The content of the invention
The technical problem to be solved in the present invention is to provide a kind of vehicle target detection method, to solve traditional vehicle detection
Method is computationally intensive, operating process is complicated, detects not accurately technical barrier.
Present invention is generally directed to the detection algorithm based on vehicle target appearance features information, i.e., is examined in actual traffic picture
Survey and position the target of set vehicle.Its difficult point essentially consists in vehicle target in picture because of illumination, visual angle and vehicle
Inside etc. changes and produces change.
In order to solve the above technical problems, the present invention adopts the following technical scheme that：
A kind of vehicle identification detection method is designed, is comprised the following steps：（1）ImageNet vehicle databases are obtained, are made
VOC2007 vehicle image data sets；（2）Using convolutional neural networks frame and using Faster R-CNN algorithm configurations models come
Feature extraction is carried out to vehicle image；（3）A kind of sliding window for increasing anchor point number is set；（4）Target detection problems are turned
It is changed to two classification problems of target；（5）Optimized using RPN loss functions；（6）The characteristic use obtained to step 2
SoftMax algorithms are classified to obtain testing result.
Further, step（1）Including following processing：A. counted from 2007 data sets of VOC：Width, the height of picture, figure
The ratio of width to height of piece, width, the height of vehicle target, the ratio of width to height of target；B. basis（1）In parameter sieved from ImageNet data sets
Select satisfactory picture；C. by the picture making screened in b into VOC2007 data sets.
Further, step（2）When middle utilization Faster R-CNN build convolutional Neural learning framework extraction characteristics of image
Inception network structures are introduced, so that network has the ability of more powerful extraction characteristics of image, make the feature extracted
Express more rich.
Further, step（3）The area of middle setting sliding window is 642, Faster R-CNN have preset 9 anchor points,
Three kinds of window areas 128 are corresponded to respectively2、2562、5122With three kinds of window the ratio of width to height 1：1、1：2、2：1 independent assortment, and increase newly
A kind of window area 64 is added2, 3 anchor points have thus been newly increased, i.e. 12 anchor points altogether, can more detect Small object.
Further, step（4）The method that the middle target detection problems by scene are converted to two classification problems of target
It is：By the num output in program：21 are changed to num output：2, i.e. foreground and background, wherein prospect are target vehicle.
Further, step（5）Optimization method be：A. distribute "Yes" label give two class reference zones, be respectively with
There is the reference zone of the ratio between highest intersection in some true value region and has the ginseng of the ratio between the intersection more than 0.7 with any true value region
Examination district domain；B. the label for distributing "No" gives the reference zone that the ratio between the intersection in all true value regions is below 0.3；C. it is non-just non-
Negative region is not involved in RPN model trainings；D. optimized using multitask loss function, its function expression is：L({pi},
{ti})=（1/Ncls）*∑Lcls(pi,yi)+(λ/Nreg)*∑yiLreg(ti,ti ’)；Wherein, λ represents coefficient of balance, for controlling
Two loss function proportions；NclsRepresent the number of input sample in mini-batch；NregRepresent all reference zones of input
Number；I represents the index of reference zone in a mini-batch；LclsPresentation class loss function, selects two classes（Target
With it is non-targeted）Log-likelihood loss function, piRepresent the true tag of the reference zone；LregRepresent that frame returns loss
Function, selecting smooth 1- norms, smooth 1- norms are more robust for discrete group point, control the two-stage of gradient as loss function
Fly so that being not easy to run during training；tiIt is a vector, represents 4 parametrization coordinates of the candidate region of prediction, ti ’It is one
Vector, represents 4 parametrization coordinates in true value region.
Compared with prior art, advantageous effects of the invention are：
1. feature of present invention extractability is strong, using deep learning caffe frames, Faster R-CNN algorithm configurations are taken to train
Model, and Inception web results are introduced, make network that there is more powerful extraction image spy's card ability.
2. recognition accuracy of the present invention is high, the target detection problems in scene will be converted to two classification problems of target,
The Yes/No that abstract target detection is changed into having to elephant judges, so as to improve the accuracy of identification of network.
3. accuracy of identification of the present invention is high, a kind of sliding window area 64 is set2, anchor point is increased to 12 by 9, by original
Originally small target deteection easy to identify is not allowed to come out.
4. calculation amount of the present invention is small, operating process is simplified, and detection result is comprehensive, is instructed using Faster R-CNN algorithm configurations
Practice model, SoftMax algorithms classify, incorporate testing process, greatly reduce redundant computation, and make classification rich in bullet
Property, testing result is more comprehensively.
Brief description of the drawings
Fig. 1 is the general frame figure of the present invention；
Fig. 2 is the basic structure of Faster R-CNN algorithms in the present invention；
Fig. 3 is the network structure of the faster_rcnn_test.pt in present invention VGG-16 models；
Fig. 4 is the part-structure figure of Faster R-CNN networks before addition Inception model structures；
Fig. 5 is the part-structure figure of improvement Faster R-CNN networks after addition Inception models；
Fig. 6 is Inception（3a）Model structure；
Fig. 7 is RPN network structures；
Fig. 8 is the part picture that the vehicle data that the present invention makes is concentrated；
Fig. 9 is the Precision-Recall curve maps of training under different models；
Figure 10 is vehicle testing result figure under Faster R-CNN models；
Figure 11 is the present invention to improved Faster R-CNN models vehicle testing result figure.
Embodiment
Illustrate the embodiment of the present invention with reference to the accompanying drawings and examples, but following embodiments are used only in detail
Describe the bright present invention in detail, do not limit the scope of the invention in any way.
Program that is involved or relying on is the conventional program or simple program of the art in following embodiments, ability
Field technique personnel can make conventional selection or accommodation according to concrete application scene；Involved method, step or
Algorithm, is then conventional method, step or algorithm unless otherwise instructed.
Embodiment 1：A kind of vehicle target detection method, referring to Fig. 1, it is suitable to be filtered out first from ImageNet data sets
Training sample, extraction characteristics of image simultaneously the vehicle target in sample is labeled.Then sample is inputted into RPN network trainings
Until network convergence, then the convolutional layer network parameter that RPN network trainings obtain is input to improved Faster R-CNN networks
It is trained until the parameter of each network layer, is finally input to the detection and knowledge that vehicle target is carried out in model by network convergence
Not.
, will using the form and evaluation algorithms instrument of 2007 data sets of VOC when the present embodiment makes training sample
ImageNet data sets are converted to the form of 2007 data sets of VOC, comprise the following steps that：（1）Unite from 2007 data sets of VOC
Count the width, height, the ratio of width to height of the width of picture, height, the ratio of width to height and target；（2）According to the value that the first step is drawn from ImageNet data
Concentration filters out satisfactory picture；（3）By gained picture making in previous step into VOC2007 data sets.
For the present embodiment when building convolutional neural networks frame, the configuration of model utilizes Faster R-CNN algorithms, its base
This structure such as Fig. 2, can be divided into 4 main contents：
（1）Conv layers.As a kind of CNN network objectives detection method, Faster R-CNN are first by one group of basis
The characteristic pattern of convolutional layer, active coating and pond layer extraction image, this feature figure are shared on follow-up RPN layers and full articulamentum；
（2）Region Proposal Networks.That is RPN networks, for generating suggestion areas.The layer is sentenced by Softmax
Disconnected anchor point belongs to prospect or background, recycles bounding box to return and corrects the accurate region of anchor point acquisition；
（3）RoI Pooling.This layer collects the characteristic pattern and characteristic area of input, and provincial characteristics is extracted after these comprehensive information
Figure, is sent into follow-up full articulamentum and judges target classification；
（4）Classification.Using the classification of provincial characteristics figure zoning, while side is examined with the recurrence of boundary's frame again
Survey the final exact position of frame.
It is the network structure for the faster_rcnn_test.pt that the present embodiment is used in VGG-16 models shown in Fig. 3.By
Figure understands image of the network for a secondary arbitrary size P*Q, fixed size M*N is scaled the images to first, then by M*N images
It is sent into network and forms 13 ConV Layers+13 Relu Layers+4 Pooling Layers；RPN networks first pass around
3x3 convolution, then generate prospect anchor point respectively and return offset with bounding box, calculate characteristic area；And Pooling layers of RoI is then
The follow-up full articulamentum of characteristic area feeding is extracted from characteristic pattern and Softmax networks are classified using region.
Inception structures are added in VGG-16 by the present embodiment, in addition add 64 in the preceding layer of Inception
The convolution kernel of 1*1 sizes, as the effects of these cores with the function of Inception models is.Fig. 4 and Fig. 5 is to add respectively
Before Inception model structures Faster is improved after the part-structure of Faster R-CNN networks and addition Inception models
The part-structure of R-CNN networks.In order to which the structure of Inception models can be better described, it is individually taken out such as Fig. 6
Shown, the network before contrast adds and after addition can be seen that：
（1）Improved Faster R-CNN convolutional neural networks have more powerful extraction image spy than Faster R-CNN networks
The ability of sign, the feature representation extracted are more rich.On the one hand the different number size dimensions of improved Web vector graphic differ
Convolution kernel come increase extraction feature diversity；On the other hand, improved network uses a kind of parallel structure, this can
Fully to integrate the feature of different convolution kernels extraction, so as to get feature representation it is more rich detailed.
（2）It is more excellent to improve performance of the convolutional neural networks of Faster R-CNN than Faster R-CNN networks in time
More.By contrast it can be seen that the number of parameters of the improved network part-structure is less, this allows for network in backpropagation
The time it takes cost is lower when undated parameter and parameter calculate, that is to say, that equally in training or when test, changes
Convolutional neural networks into Faster R-CNN more save the time.
RPN network structures such as Fig. 7, it is actual to be divided into two lines road, above one classified by softmax before anchor point obtains
Scape and background, below one be used for calculate for anchor point bounding box return offset, to obtain accurate region.It is and last
Proposal layers be then responsible for comprehensive prospect anchor point and bounding box and return offset obtaining proposals, while reject it is too small and
Proposals beyond border.RPN networks use different area and the cunning of aspect ratio centered on each point on characteristic pattern
Dynamic window carrys out the feature in acquisition characteristics figure specific region.Faster R-CNN have preset 9 anchor points, correspond to three kinds of windows respectively
Area 1282、2562、5122With three kinds of window the ratio of width to height 1：1、1：2、2：1 independent assortment, the present embodiment have newly increased a kind of window
Open area 642, 3 anchor points have thus been newly increased, i.e. 12 anchor points altogether, small target deteection easy to identify can not have been allowed to go out originally
Come.
It is the part picture for the vehicle data concentration that the present embodiment makes as shown in Figure 8, in order to further improve detection
The test problems of target, are converted into two classification problems by precision and accuracy, that is, it is vehicle or other targets to judge target, is instructed
Experienced strategy uses fine-tuning technologies, and trained sample is the data set of the VOC2007 forms made at the beginning.Fig. 9 is
The Precision-Recall curve maps of training under different models.For different improved methods, grader experimental result is called together
Return rate curve as shown in the figure.Wherein, Faster R-CNN are the testing result of original Faster R-CNN models, detect vehicle target
MAP values be 81.5%.Method-4 is model inspection as a result, the mAP values of detection vehicle target reach after increase anchor point number
To 84.7%, the mAP values compared to former Faster R-CNN improve 3%.Method-3 is to detect single mesh using two sorting techniques
Target testing result, its mAP value reach 86.6%, and the mAP values compared to the vehicle target of 20 classification improve 6%.Method-2 is
The testing result of Inception model is introduced, the mAP values for detecting vehicle target have reached 88.1%, compared to former Faster R-
The mAP values of CNN improve 7%.Method-1 is the testing result of three improved Faster R-CNN models, detects car
The mAP values of target can reach 90.4%, and the mAP values compared to former Faster R-CNN improve 9%.
The present embodiment selects the vehicle pictures 000003,000004,000005 and 000006 under complex scene respectively in original
Tested on Faster R-CNN network models and the improved Faster R-CNN network models of this implementation.In Faster R-
Tested on CNN network models, its test result is as shown in Figure 10.Carried out on improved Faster R-CNN network models
Test, its test result are as shown in figure 11.It can be seen that the knot tested on improved Faster R-CNN network models
Fruit is more accurate, resolution ratio higher.
The present invention is described in detail above in conjunction with drawings and examples, still, those of skill in the art
Member can also carry out each design parameter in above-described embodiment it is understood that on the premise of present inventive concept is not departed from
Change, forms multiple specific embodiments, is the common excursion of the present invention, is no longer described in detail one by one herein.
Claims (6)
1. a kind of vehicle target detection method, it is characterised in that comprise the following steps：
Step 1, obtain ImageNet vehicle databases, makes 2007 vehicle image data sets of VOC；
Step 2, build convolutional neural networks frame；
Step 3, set a kind of sliding window for increasing anchor point number；
Step 4, two classification problems for target detection problems being converted to target；
Step 5, optimized using loss function；
Step 6, the characteristic use SoftMax algorithms obtained to step 2 are classified to obtain testing result.
2. vehicle target detection method according to claim 1, it is characterised in that the step 1 includes following processing and walks
Suddenly：
（1）Following parameter is counted based on 2007 data sets of VOC：The width of picture and high, the ratio of width to height of picture, the width of vehicle target
With high, vehicle target the ratio of width to height；
（2）The picture for meeting 2007 data set requirements of VOC is filtered out from ImageNet data sets according to above-mentioned parameter；
（3）By the picture making of previous step screening into VOC2007 data sets.
3. vehicle target detection method according to claim 1, it is characterised in that the step 2 includes following processing and walks
Suddenly：Convolutional neural networks frame is built using Faster R-CNN algorithms, and following improvement is made to Faster R-CNN algorithms：
Introduce Inception network structures and increase the convolution kernel of 64 1*1 sizes, make network that there is more powerful extraction characteristics of image
Ability.
4. vehicle target detection method according to claim 1, it is characterised in that the step 3 includes following processing and walks
Suddenly：Increase a kind of window area 642, so that anchor point number is increased to 12 default 9 by Faster R-CNN.
5. the vehicle target according to claim 1 for being carried out feature extraction to image based on convolutional neural networks frame is detected
Method, it is characterised in that the step 4 includes following processing：By the num output in program：21 are changed to num output：
2, multi-target detection is changed into two class target detection of vehicle and background.
6. vehicle target detection method according to claim 1, it is characterised in that the step 5 includes following processing and walks
Suddenly：
（1）Distributing an expression to each reference zone is/is not the binary label of target；
（2）The label that distribution represents to be is to two class reference zones：
（a）There is the reference zone of the ratio between highest intersection with some true value region；
（b）There is the reference zone of the ratio between the intersection more than 0.7 with any true value region；
（3）Distribution represents that no label gives the reference zone that the ratio between the intersection in all true value regions is below 0.3, non-just non-negative
Reference zone be not involved in RPN model trainings；
（4）Optimized using Region Proposal Network loss functions, its loss function is defined as：L({pi},
{ti})=（1/Ncls）*∑Lcls(pi,yi)+(λ/Nreg)*∑yiLreg(ti,ti ’)；Wherein, λ represents coefficient of balance, for controlling
Two loss function proportions；NclsRepresent the number of input sample in mini-batch；NregRepresent all reference zones of input
Number；I represents the index of reference zone in a mini-batch；LclsPresentation class loss function, selects two classes（Target
With it is non-targeted）Log-likelihood loss function, piRepresent the true tag of the reference zone；LregRepresent that frame returns loss
Function, selecting smooth 1- norms, smooth 1- norms are more robust for discrete group point, control the two-stage of gradient as loss function
Fly so that being not easy to run during training；tiIt is a vector, represents 4 parametrization coordinates of the candidate region of prediction, ti ’It is one
Vector, represents 4 parametrization coordinates in true value region.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711321241.6A CN108009509A (en) | 2017-12-12 | 2017-12-12 | Vehicle target detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711321241.6A CN108009509A (en) | 2017-12-12 | 2017-12-12 | Vehicle target detection method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108009509A true CN108009509A (en) | 2018-05-08 |
Family
ID=62058654
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711321241.6A Pending CN108009509A (en) | 2017-12-12 | 2017-12-12 | Vehicle target detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108009509A (en) |
Cited By (33)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108681718A (en) * | 2018-05-20 | 2018-10-19 | 北京工业大学 | A kind of accurate detection recognition method of unmanned plane low target |
| CN108694401A (en) * | 2018-05-09 | 2018-10-23 | 北京旷视科技有限公司 | Object detection method, apparatus and system |
| CN108932471A (en) * | 2018-05-23 | 2018-12-04 | 浙江科技学院 | A kind of vehicle checking method |
| CN109214333A (en) * | 2018-08-31 | 2019-01-15 | 厦门中控智慧信息技术有限公司 | Convolutional neural networks structure, face character recognition methods, device and terminal device |
| CN109300114A (en) * | 2018-08-30 | 2019-02-01 | 西南交通大学 | The minimum target components of high iron catenary support device hold out against missing detection method |
| CN109345529A (en) * | 2018-09-30 | 2019-02-15 | 福州大学 | Based on the secondary target detection network wire clamp of modified, grading ring fault recognition method |
| CN109447018A (en) * | 2018-11-08 | 2019-03-08 | 天津理工大学 | A kind of road environment visual perception method based on improvement Faster R-CNN |
| CN109492416A (en) * | 2019-01-07 | 2019-03-19 | 南京信息工程大学 | A kind of guard method of big data image and system based on safety zone |
| CN109583512A (en) * | 2018-12-14 | 2019-04-05 | 北京旷视科技有限公司 | Image processing method, apparatus and system |
| CN109635749A (en) * | 2018-12-14 | 2019-04-16 | 网易（杭州）网络有限公司 | Image processing method and device based on video flowing |
| CN109657622A (en) * | 2018-12-21 | 2019-04-19 | 广东工业大学 | A kind of detection of traffic lights and recognition methods, device and equipment |
| CN109815802A (en) * | 2018-12-18 | 2019-05-28 | 中国海洋大学 | A kind of monitor video vehicle detection and recognition method based on convolutional neural networks |
| CN109829469A (en) * | 2018-11-08 | 2019-05-31 | 电子科技大学 | A kind of vehicle checking method based on deep learning |
| CN109934255A (en) * | 2019-01-22 | 2019-06-25 | 小黄狗环保科技有限公司 | A kind of Model Fusion method for delivering object Classification and Identification suitable for beverage bottle recycling machine |
| CN109977958A (en) * | 2019-03-25 | 2019-07-05 | 中国科学技术大学 | A kind of offline handwritten form mathematical formulae identification reconstructing method |
| CN109977780A (en) * | 2019-02-26 | 2019-07-05 | 广东工业大学 | A kind of detection and recognition methods of the diatom based on deep learning algorithm |
| CN110069993A (en) * | 2019-03-19 | 2019-07-30 | 同济大学 | A kind of target vehicle detection method based on deep learning |
| CN110120035A (en) * | 2019-04-17 | 2019-08-13 | 杭州数据点金科技有限公司 | A kind of tire X-ray defect detection method differentiating defect grade |
| CN110458113A (en) * | 2019-08-14 | 2019-11-15 | 旭辉卓越健康信息科技有限公司 | A kind of non-small face identification method cooperated under scene of face |
| CN110705338A (en) * | 2018-07-10 | 2020-01-17 | 浙江宇视科技有限公司 | Vehicle detection method and device and monitoring equipment |
| CN110751155A (en) * | 2019-10-14 | 2020-02-04 | 西北工业大学 | Novel target detection method based on Faster R-CNN |
| CN110751091A (en) * | 2019-10-18 | 2020-02-04 | 江西理工大学 | Convolutional neural network model for static image behavior recognition |
| CN110929632A (en) * | 2019-11-19 | 2020-03-27 | 复旦大学 | Complex scene-oriented vehicle target detection method and device |
| CN111144383A (en) * | 2020-01-15 | 2020-05-12 | 河南理工大学 | Method for detecting vehicle deflection angle |
| CN111339341A (en) * | 2018-12-19 | 2020-06-26 | 顺丰科技有限公司 | Model training method and device, positioning method and device, and equipment |
| CN112016449A (en) * | 2020-08-27 | 2020-12-01 | 西华大学 | Vehicle identification and detection method based on deep learning |
| CN112329559A (en) * | 2020-10-22 | 2021-02-05 | 空间信息产业发展股份有限公司 | Method for detecting homestead target based on deep convolutional neural network |
| CN112364687A (en) * | 2020-09-29 | 2021-02-12 | 上善智城(苏州)信息科技有限公司 | Improved Faster R-CNN gas station electrostatic sign identification method and system |
| CN112528705A (en) * | 2019-09-17 | 2021-03-19 | 同方威视技术股份有限公司 | Image recognition system and method based on semantics |
| CN113159083A (en) * | 2020-12-09 | 2021-07-23 | 贵州大学 | Tobacco leaf grading method based on Faster R-CNN network |
| WO2021244079A1 (en) * | 2020-06-02 | 2021-12-09 | 苏州科技大学 | Method for detecting image target in smart home environment |
| ES2908944R1 (en) * | 2018-07-24 | 2022-05-13 | Fund Centro Tecnoloxico De Telecomunicacions De Galicia | A COMPUTER IMPLEMENTED METHOD AND SYSTEM FOR DETECTING SMALL OBJECTS IN AN IMAGE USING CONVOLUTIONAL NEURAL NETWORKS |
| CN115294537A (en) * | 2022-08-10 | 2022-11-04 | 青岛文达通科技股份有限公司 | Vehicle attribute identification method and system based on feature association |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160307074A1 (en) * | 2014-11-21 | 2016-10-20 | Adobe Systems Incorporated | Object Detection Using Cascaded Convolutional Neural Networks |
| CN107220603A (en) * | 2017-05-18 | 2017-09-29 | 惠龙易通国际物流股份有限公司 | Vehicle checking method and device based on deep learning |
-
2017
- 2017-12-12 CN CN201711321241.6A patent/CN108009509A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160307074A1 (en) * | 2014-11-21 | 2016-10-20 | Adobe Systems Incorporated | Object Detection Using Cascaded Convolutional Neural Networks |
| CN107220603A (en) * | 2017-05-18 | 2017-09-29 | 惠龙易通国际物流股份有限公司 | Vehicle checking method and device based on deep learning |
Non-Patent Citations (4)
| Title |
|---|
| QINGHUI ZHANG等: "Multiple Objects Detection based on Improved Faster R-CNN", 《PROCEEDINGS OF THE 9TH INTERNATIONAL CONFERENCE ON SIGNAL PROCESSING SYSTEMS》 * |
| SHAOQING REN等: "Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
| YONGZHENG XU等: "Car Detection from Low-Altitude UAV Imagery with the Faster R-CNN", 《JOURNAL OF ADVANCED TRANSPORTATION》 * |
| 曹诗雨 等: "基于Fast R-CNN的车辆目标检测", 《中国图象图形学报》 * |
Cited By (42)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11250591B2 (en) * | 2018-05-09 | 2022-02-15 | Beijing Kuangshi Technology Co., Ltd. | Target detection method, system, and non-volatile storage medium |
| CN108694401A (en) * | 2018-05-09 | 2018-10-23 | 北京旷视科技有限公司 | Object detection method, apparatus and system |
| CN108681718B (en) * | 2018-05-20 | 2021-08-06 | 北京工业大学 | Unmanned aerial vehicle low-altitude target accurate detection and identification method |
| CN108681718A (en) * | 2018-05-20 | 2018-10-19 | 北京工业大学 | A kind of accurate detection recognition method of unmanned plane low target |
| CN108932471B (en) * | 2018-05-23 | 2020-06-26 | 浙江科技学院 | Vehicle detection method |
| CN108932471A (en) * | 2018-05-23 | 2018-12-04 | 浙江科技学院 | A kind of vehicle checking method |
| CN110705338A (en) * | 2018-07-10 | 2020-01-17 | 浙江宇视科技有限公司 | Vehicle detection method and device and monitoring equipment |
| ES2908944R1 (en) * | 2018-07-24 | 2022-05-13 | Fund Centro Tecnoloxico De Telecomunicacions De Galicia | A COMPUTER IMPLEMENTED METHOD AND SYSTEM FOR DETECTING SMALL OBJECTS IN AN IMAGE USING CONVOLUTIONAL NEURAL NETWORKS |
| CN109300114A (en) * | 2018-08-30 | 2019-02-01 | 西南交通大学 | The minimum target components of high iron catenary support device hold out against missing detection method |
| CN109214333A (en) * | 2018-08-31 | 2019-01-15 | 厦门中控智慧信息技术有限公司 | Convolutional neural networks structure, face character recognition methods, device and terminal device |
| CN109345529A (en) * | 2018-09-30 | 2019-02-15 | 福州大学 | Based on the secondary target detection network wire clamp of modified, grading ring fault recognition method |
| CN109345529B (en) * | 2018-09-30 | 2021-09-24 | 福州大学 | Network wire clamp and grading ring fault identification method based on improved secondary target detection |
| CN109829469A (en) * | 2018-11-08 | 2019-05-31 | 电子科技大学 | A kind of vehicle checking method based on deep learning |
| CN109447018A (en) * | 2018-11-08 | 2019-03-08 | 天津理工大学 | A kind of road environment visual perception method based on improvement Faster R-CNN |
| CN109447018B (en) * | 2018-11-08 | 2021-08-03 | 天津理工大学 | Road environment visual perception method based on improved Faster R-CNN |
| CN109635749A (en) * | 2018-12-14 | 2019-04-16 | 网易（杭州）网络有限公司 | Image processing method and device based on video flowing |
| CN109583512A (en) * | 2018-12-14 | 2019-04-05 | 北京旷视科技有限公司 | Image processing method, apparatus and system |
| CN109583512B (en) * | 2018-12-14 | 2021-05-25 | 北京旷视科技有限公司 | Image processing method, device and system |
| CN109815802A (en) * | 2018-12-18 | 2019-05-28 | 中国海洋大学 | A kind of monitor video vehicle detection and recognition method based on convolutional neural networks |
| CN111339341A (en) * | 2018-12-19 | 2020-06-26 | 顺丰科技有限公司 | Model training method and device, positioning method and device, and equipment |
| CN109657622A (en) * | 2018-12-21 | 2019-04-19 | 广东工业大学 | A kind of detection of traffic lights and recognition methods, device and equipment |
| CN109492416A (en) * | 2019-01-07 | 2019-03-19 | 南京信息工程大学 | A kind of guard method of big data image and system based on safety zone |
| CN109934255A (en) * | 2019-01-22 | 2019-06-25 | 小黄狗环保科技有限公司 | A kind of Model Fusion method for delivering object Classification and Identification suitable for beverage bottle recycling machine |
| CN109934255B (en) * | 2019-01-22 | 2023-05-30 | 小黄狗环保科技有限公司 | Model fusion method suitable for classification and identification of delivered objects of beverage bottle recycling machine |
| CN109977780A (en) * | 2019-02-26 | 2019-07-05 | 广东工业大学 | A kind of detection and recognition methods of the diatom based on deep learning algorithm |
| CN110069993B (en) * | 2019-03-19 | 2021-10-08 | 同济大学 | Target vehicle detection method based on deep learning |
| CN110069993A (en) * | 2019-03-19 | 2019-07-30 | 同济大学 | A kind of target vehicle detection method based on deep learning |
| CN109977958A (en) * | 2019-03-25 | 2019-07-05 | 中国科学技术大学 | A kind of offline handwritten form mathematical formulae identification reconstructing method |
| CN110120035A (en) * | 2019-04-17 | 2019-08-13 | 杭州数据点金科技有限公司 | A kind of tire X-ray defect detection method differentiating defect grade |
| CN110458113A (en) * | 2019-08-14 | 2019-11-15 | 旭辉卓越健康信息科技有限公司 | A kind of non-small face identification method cooperated under scene of face |
| CN112528705A (en) * | 2019-09-17 | 2021-03-19 | 同方威视技术股份有限公司 | Image recognition system and method based on semantics |
| CN110751155A (en) * | 2019-10-14 | 2020-02-04 | 西北工业大学 | Novel target detection method based on Faster R-CNN |
| CN110751091A (en) * | 2019-10-18 | 2020-02-04 | 江西理工大学 | Convolutional neural network model for static image behavior recognition |
| CN110929632A (en) * | 2019-11-19 | 2020-03-27 | 复旦大学 | Complex scene-oriented vehicle target detection method and device |
| CN111144383B (en) * | 2020-01-15 | 2023-03-28 | 河南理工大学 | Method for detecting vehicle deflection angle |
| CN111144383A (en) * | 2020-01-15 | 2020-05-12 | 河南理工大学 | Method for detecting vehicle deflection angle |
| WO2021244079A1 (en) * | 2020-06-02 | 2021-12-09 | 苏州科技大学 | Method for detecting image target in smart home environment |
| CN112016449A (en) * | 2020-08-27 | 2020-12-01 | 西华大学 | Vehicle identification and detection method based on deep learning |
| CN112364687A (en) * | 2020-09-29 | 2021-02-12 | 上善智城(苏州)信息科技有限公司 | Improved Faster R-CNN gas station electrostatic sign identification method and system |
| CN112329559A (en) * | 2020-10-22 | 2021-02-05 | 空间信息产业发展股份有限公司 | Method for detecting homestead target based on deep convolutional neural network |
| CN113159083A (en) * | 2020-12-09 | 2021-07-23 | 贵州大学 | Tobacco leaf grading method based on Faster R-CNN network |
| CN115294537A (en) * | 2022-08-10 | 2022-11-04 | 青岛文达通科技股份有限公司 | Vehicle attribute identification method and system based on feature association |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108009509A (en) | Vehicle target detection method | |
| Yang et al. | Real-time face detection based on YOLO | |
| CN107133616B (en) | Segmentation-free character positioning and identifying method based on deep learning | |
| CN106650806B (en) | A kind of cooperating type depth net model methodology for pedestrian detection | |
| WO2024021413A1 (en) | Image segmentation method combining super-pixels and multi-scale hierarchical feature recognition | |
| CN103218831B (en) | A kind of video frequency motion target classifying identification method based on profile constraint | |
| CN109614985A (en) | A kind of object detection method based on intensive connection features pyramid network | |
| CN109446992A (en) | Remote sensing image building extracting method and system, storage medium, electronic equipment based on deep learning | |
| CN110210362A (en) | A kind of method for traffic sign detection based on convolutional neural networks | |
| CN109800628A (en) | A kind of network structure and detection method for reinforcing SSD Small object pedestrian detection performance | |
| CN110533695A (en) | A kind of trajectory predictions device and method based on DS evidence theory | |
| CN109241982A (en) | Object detection method based on depth layer convolutional neural networks | |
| CN109523520A (en) | A kind of chromosome automatic counting method based on deep learning | |
| CN101667245B (en) | Human face detection method by cascading novel detection classifiers based on support vectors | |
| CN109711288A (en) | Remote sensing ship detecting method based on feature pyramid and distance restraint FCN | |
| CN106096561A (en) | Infrared pedestrian detection method based on image block degree of depth learning characteristic | |
| CN109543632A (en) | A kind of deep layer network pedestrian detection method based on the guidance of shallow-layer Fusion Features | |
| CN106408030A (en) | SAR image classification method based on middle lamella semantic attribute and convolution neural network | |
| CN107909081A (en) | The quick obtaining and quick calibrating method of image data set in a kind of deep learning | |
| CN105427309A (en) | Multiscale hierarchical processing method for extracting object-oriented high-spatial resolution remote sensing information | |
| CN103778435A (en) | Pedestrian fast detection method based on videos | |
| CN111368660A (en) | Single-stage semi-supervised image human body target detection method | |
| CN112052772A (en) | Face shielding detection algorithm | |
| CN110097091A (en) | It is trained be distributed with inference data it is inconsistent under the conditions of image fine granularity recognition methods | |
| CN109993803A (en) | The intellectual analysis and evaluation method of city tone |
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: 20180508 |
|
| RJ01 | Rejection of invention patent application after publication |