CN109766769A - A kind of road target detection recognition method based on monocular vision and deep learning - Google Patents
A kind of road target detection recognition method based on monocular vision and deep learning Download PDFInfo
- Publication number
- CN109766769A CN109766769A CN201811547843.8A CN201811547843A CN109766769A CN 109766769 A CN109766769 A CN 109766769A CN 201811547843 A CN201811547843 A CN 201811547843A CN 109766769 A CN109766769 A CN 109766769A
- Authority
- CN
- China
- Prior art keywords
- road
- deep learning
- loss function
- network
- detection
- 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 65
- 238000000034 method Methods 0.000 title claims abstract description 36
- 238000013135 deep learning Methods 0.000 title claims abstract description 25
- 238000012549 training Methods 0.000 claims abstract description 27
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 12
- 230000006870 function Effects 0.000 claims description 35
- 230000008569 process Effects 0.000 claims description 8
- 238000011478 gradient descent method Methods 0.000 claims description 6
- 238000011897 real-time detection Methods 0.000 claims description 6
- 230000008859 change Effects 0.000 claims description 5
- 230000009191 jumping Effects 0.000 claims description 4
- 238000013508 migration Methods 0.000 claims description 4
- 230000005012 migration Effects 0.000 claims description 4
- 238000000605 extraction Methods 0.000 claims description 3
- 230000006855 networking Effects 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 230000008439 repair process Effects 0.000 claims 1
- 230000000694 effects Effects 0.000 abstract description 8
- 238000002474 experimental method Methods 0.000 abstract description 3
- 239000011248 coating agent Substances 0.000 description 4
- 238000000576 coating method Methods 0.000 description 4
- 230000010354 integration Effects 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 238000005259 measurement Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000005303 weighing Methods 0.000 description 3
- 239000012141 concentrate Substances 0.000 description 2
- 238000013480 data collection Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 208000018883 loss of balance Diseases 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000004321 preservation Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Landscapes
- Image Analysis (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention belongs to road vehicle detection technique fields, disclose a kind of road target detection recognition method based on monocular vision and deep learning；Depth convolutional neural networks are applied to vehicle detection, learn to comprehensive and abundant vehicle characteristics in still image, realize the quickly and accurately road vehicle detection based on characteristics of image, to block, be truncated, light variation, the common disturbed condition such as shade have stronger robustness, region recommendation step cumbersome in two-period form convolutional neural networks detector is abandoned, the possible position for surrounding frame and size are determined by convolution experiment, and it is respectively sorted in training stage shared confidence level, number of parameters is reduced significantly.The present invention is predicted in multiple characteristic patterns of forecast period combination different levels, increase the semantic meaning representation ability of network model, the accuracy of final detection result and speed are superior to the detection effect of two-period form reference detector Faster R-CNN under same experiment condition.
Description
Technical field
The invention belongs to road vehicle detection technique field more particularly to a kind of roads based on monocular vision and deep learning
Road target detection recognition methods.
Background technique
Currently, the prior art commonly used in the trade is such thatVehicle detection is branch's research direction of target detection,
There is vital status in the urban construction such as intelligent traffic administration system, smart city, car networking application.As computer regards
Feel that the rapid development of research, target detection research achieve important breakthrough in recent years, the detection algorithm based on deep learning has become
For mainstream detection algorithm, vehicle detection research is also gradually drawn close to deep learning direction.And widely available high quality and low cost
Road camera is on the one hand to realize that Intelligent System of Vehicle detection provides data basis, on the other hand but also improving vehicle detection
Speed and the demand of accuracy are more urgent.Complexity and vehicle diversity in view of domestic road traffic, under real scene
Road vehicle detection has higher requirements to the robustness of algorithm, generalization and real-time.Exist in city and largely occupies blind way
With the illegal parking event on pavement, the problem of Urban Governance is belonged to, and about tricycle in municipal administration's dynamic traffic system
The regulation of No entry main road.Traffic condition is complicated under real roads camera, and wagon flow is intensive, and a variety of models are unevenly distributed weighing apparatus and make
At unbalance between vehicle sample class, concentrate vehicle condition difference very big with public data, and open library data are derived from state's outer course more
Road picture has relatively big difference with China's road vehicle situation, such as seldom there is electric vehicle target, and many target shapes
It has differences.
In conclusion problem of the existing technology is:
Traffic condition is complicated under real roads camera, and wagon flow is intensive, and a variety of models are unevenly distributed weighing apparatus and cause vehicle sample
It is unbalance between class, concentrate vehicle condition difference very big with public data, and seldom there are tricycle samples.When this causes model training
Cannot balanced study variety classes vehicle feature, the vehicle detection effect that certain classifications can occur in practical applications is good, and
The problem of the vehicle detection effect difference of other classifications.In addition it if sample without certain class vehicle, also can not just train to obtain
It is able to detect the model of such vehicle.
Solve the difficulty and meaning of above-mentioned technical problem:
The fact that above-mentioned technical problem is objective reality, a large amount of readily available scene image datas are from the point of view of training angle
Incomplete, need largely to work the artificial or semi-artificial data set improved for training, while using the training of innovation
Method trains detection model, so that model has same and extensive discernment for all kinds of targets.Solve above-mentioned technology
Problem for lift scheme for the verification and measurement ratio and recall rate of road target complicated in real world, while greatly increasing model
Availability have conclusive significance.
Summary of the invention
In view of the problems of the existing technology, the present invention provides a kind of road mesh based on monocular vision and deep learning
Detection recognition method is marked, depth convolutional neural networks are applied to vehicle detection, learn to comprehensive and abundant vehicle in still image
Feature realizes the quickly and accurately road vehicle detection based on characteristics of image, to block, be truncated, light variation, shade etc. it is normal
The disturbed condition seen has stronger robustness, and pointedly improves tricycle sample data in the database, with obtain can
Detect the model of tricycle.
The invention is realized in this way a kind of road target detection recognition method based on monocular vision and deep learning,
The road target detection recognition method based on monocular vision and deep learning the following steps are included:
The first step establishes small sample road vehicle data set, collects the road vehicle image under real scene；
Depth convolutional neural networks frame SSD is detected applied to road vehicle, forms new net in conjunction with VGG by second step
Network structure realizes real-time detection end to end, and the training stage uses the weighting parameter of the gradient descent method learning network with momentum,
The loss function that cross entropy loss function and smooth L1 loss function are returned as classifier and position；
Third step, by secondary migration study on VOC data set and small sample road vehicle data set trim network compared with
High-rise weighting parameter；
4th step introduces the characteristic pattern pyramid with lateral connection, is attached to primitive network in a manner of transmitting of jumping, no
Change original network structure, increases a small amount of weighting parameter and calculating operation；
5th step measures the classification accuracy of target using the focal loss function of balance, reduces big in training process
The contribution rate that amount trolley sample learns network weight, promotes training to focus more on difficult example sample.
Further, tricycle, lorry are concentrated in screening public data emphatically in the first step, mark corresponding vehicle mark
Label.
Further, the first step specifically includes:
(1) a certain amount of real roads scene video data are obtained by picture recording equipment, wherein comprising a large amount of all kinds of
The image information of motor vehicle non-motor vehicle and pedestrian；
(2) it writes program to save as the articulating frame in video with formatted image with time interval, in all images
Enterprising pedestrian's work mark, by all images all vehicles and pedestrian be marked and save as location information corresponding
Image tag；It writes program simultaneously to screen VOC data set, obtains the data set comprising all kinds of vehicles and pedestrian
Part.
Further, concrete model is based on SSD target detection model structure, to feature extraction therein in the second step
Network VGG has carried out replacement modification optimization, and in channel dimension equal part, each intermediate features figure obtains matrix1 and matrix2,
And retain the maximum value of two eigenmatrix corresponding points using following operation to obtain new maxout eigenmatrix
Matrix3:matrix3=matrix1-matrix2, the minimum value that matrix3 is truncated make it equal to 0, then matrix3=
matrix3+matrix2；
Training stage uses the weighting parameter of the gradient descent method learning network with momentum, by cross entropy loss function:With smooth L1 loss function: The loss function returned respectively as target category classification and target detection frame position.
Further, using the focal loss function Focal Loss of balance in the 5th step:
Target is measured instead of cross entropy loss function
Classification accuracy, reduce the contribution rate that a large amount of trolley samples learns network weight in training process, promote to train more absorbed
In difficult example sample.
Another object of the present invention is to provide the road targets described in a kind of application based on monocular vision and deep learning
The intelligent traffic administration system platform of detection recognition method.
Another object of the present invention is to provide the road targets described in a kind of application based on monocular vision and deep learning
The smart city of detection recognition method manages platform.
Another object of the present invention is to provide the road targets described in a kind of application based on monocular vision and deep learning
The car networking control platform of detection recognition method.
In conclusion advantages of the present invention and good effect are as follows:
The present invention has abandoned region recommendation step cumbersome in two-period form convolutional neural networks detector, passes through convolution process
Determine that the possible position for surrounding frame and size, design realize the integrated convolutional Neural net based on whole figure candidate and simple regression
Network target detection model, and single convolution kernel is replaced using two-fold product, active coating is substituted using internal integration to realize net
Non-linear, the reduction network parameter amount of network, improves feature to the abstracting power of target, realizes real-time detection end to end.Pre-
The survey stage combines multiple characteristic patterns of different levels to be predicted, increases the semantic meaning representation ability of network model.Final detection knot
The accuracy of fruit and speed are superior to the detection effect of two-period form reference detector Faster R-CNN under same experiment condition.
The mAP of the detection and identification model on training test image data collection can reach 70%, be I7 in CPU
6970K, GPU GTX1080, the detection speed inside saved as on the PC computer of 8G is 23ms/ frame, and about 43 frames/s reach real-time.
It can be seen that the method for the present invention is all bright on detection accuracy measurement mAP and detection speed measurement FPS by the comparison of upper table
Aobvious is more than existing FasterR-CNN method and conventional depth learning method.
Detailed description of the invention
Fig. 1 is the road target detection recognition method stream provided in an embodiment of the present invention based on monocular vision and deep learning
Cheng Tu.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to embodiments, to the present invention
It is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not used to
Limit the present invention.
As shown in Figure 1, provided in an embodiment of the present invention identified based on monocular vision and the detection of the road target of deep learning
Method includes:
S101: establishing small sample road vehicle data set, collects the road vehicle image under real scene；
S102: realize the integrated convolutional neural networks target detection model based on whole figure candidate and simple regression, and by its
It is detected applied to road vehicle, replaces single convolution kernel using two-fold product, substitute active coating using internal integration to realize
Non-linear, the reduction network parameter amount of network improves feature to the abstracting power of target, realizes real-time detection end to end, instruction
Practice the stage using the weighting parameter of the gradient descent method learning network with momentum, cross entropy loss function and smooth L1 are lost
The loss function that function is returned as classifier and position；
S103: higher by secondary migration study trim network on VOC data set and small sample road vehicle data set
The weighting parameter of layer；
S104: introducing the characteristic pattern pyramid with lateral connection, is attached to primitive network in a manner of transmitting of jumping, does not change
Become original network structure, increases a small amount of weighting parameter and calculating operation；
S105: measuring the classification accuracy of target using the focal loss function of balance instead of cross entropy loss function,
The contribution rate that a large amount of trolley samples learn network weight in training process is reduced, training is promoted to focus more on difficult example sample.
In a preferred embodiment of the invention, small sample road vehicle data set is established in step S101, collects true field
Road vehicle image under scape.Specifically, firstly, using with the mobile phone taken pictures with camera function, the photograph of different brands model
Camera and the video camera of different brands model shoot the true street scene of different time diverse geographic location from vehicle
Picture and video；Then unified format analysis processing is done to picture, uses cv::imread () and cv: in opencv:
All pictures from distinct device are unified for jpg format by imwrite () function；Using imagelabel tool to all
All kinds of road targets in picture carry out frame choosing and by its location information and target category information preservation in picture in correspondence
Text file in；The picture handled well and text file random selection are finally divided into two parts, a part includes total picture
The 90% of amount, as the training set of model, another part includes the 10% of total picture amount, the test set as model；
In a preferred embodiment of the invention, it is realized in step S102 using deep learning frame candidate and single based on whole figure
The integrated convolutional neural networks target detection model of recurrence, and it is applied to road vehicle detection, it is replaced using two-fold product
Single convolution kernel substitutes active coating using internal integration to realize the non-linear of network, reduces network parameter amount, improves special
The abstracting power to target is levied, realizes real-time detection end to end.
Concrete model is to have been carried out replacement based on SSD target detection model structure to feature extraction network VGG therein and repaired
Change optimization, each intermediate features figure obtains matrix1 and matrix2 in channel dimension equal part, and retains using following operation
The maximum value of two eigenmatrix corresponding points is to obtain new maxout eigenmatrix matrix3:matrix3=matrix1-
Matrix2, the minimum value that matrix3 is truncated make it equal to 0, then matrix3=matrix3+matrix2.
Training stage uses the weighting parameter of the gradient descent method learning network with momentum, by cross entropy loss function:With smooth L1 loss function: The loss function returned respectively as target category classification and target detection frame position；
In a preferred embodiment of the invention, using the focal loss function Focal Loss of balance in step S105:
Target is measured instead of cross entropy loss function
Classification accuracy, reduce the contribution rate that a large amount of trolley samples learns network weight in training process, promote to train more absorbed
In difficult example sample.
Preferably, screening public data emphatically in S101 and concentrating rare vehicle such as tricycle, lorry, mark pair
The vehicle tag answered, the network model generalization that training obtains on small sample road vehicle data set are stronger；
α=2 is taken by cross validation, and effect is best when γ=2, and the penalty values of final mask are down to 0.31, each sorting room
AP more balance.Detection effect of the heterogeneous networks model on small sample road test collection is compared, the effective of algorithm is demonstrated
Property.The detection speed of final mask reaches 51.10fps, and Average Accuracy 78.6% has reached real-time accurate detection effect, has
There are certain science and practical value.
Application principle of the invention is further described combined with specific embodiments below.
Road target detection recognition method provided in an embodiment of the present invention based on monocular vision and deep learning is specifically wrapped
It includes:
(1) small sample road vehicle data set is established, the road vehicle image under real scene is collected, and is screened emphatically public
Vehicle rare in data set such as tricycle, lorry are opened, corresponding vehicle tag is marked, on small sample road vehicle data set
The network model generalization that training obtains is stronger；Specific data processing method is as follows:
A. by road monitoring camera, automobile data recorder, mobile phone camera, a variety of picture recordings such as hand-held camera are set
It is standby to obtain a certain amount of real roads scene video data, wherein the figure comprising a large amount of all kinds of motor vehicle non-motor vehicles and pedestrian
As information.
B. the image that the articulating frame in video is saved as certain format by program at a time interval is write, in all figures
As enterprising pedestrian's work marks, by all images all vehicles and pedestrian be marked and location information saved as into correspondence
Image tag；It writes program simultaneously to screen VOC data set, obtains the data comprising all kinds of vehicles and pedestrian
Collect part.(VOC:Pascal VOC data set, a standardized target detection public data collection, provides 20 class target images,
Including all kinds of vehicles and pedestrian)
(2) the integrated convolutional neural networks target detection model based on whole figure candidate and simple regression is realized in design, and is made
Single convolution kernel is replaced with two-fold product, active coating is substituted using internal integration to realize the non-linear of network, reduces network
Parameter amount improves feature to the abstracting power of target, realizes real-time detection end to end.Training stage uses the gradient with momentum
The weighting parameter of descent method learning network is returned using cross entropy loss function and smooth L1 loss function as classifier and position
The loss function returned；
(3) pass through secondary migration study trim network higher level on VOC data set and small sample road vehicle data set
Weighting parameter, enhance the character representation ability of model, Average Accuracy is compared to improving 10 percentage points in (2)；
(4) it is directed to Small object vehicle missing inspection problem, the characteristic pattern pyramid with lateral connection is introduced, increases to a certain extent
Contextual information in character representation, the pyramid newly increased are attached to primitive network in a manner of transmitting of jumping, do not change
Become original network structure, only increase a small amount of weighting parameter and calculating operation, not to the detection speed of algorithm entirety
It makes a significant impact；
(5) it is unevenly distributed unbalance problem between vehicle sample class caused by weighing apparatus for a variety of models under road camera, adopts
Cross entropy loss function is replaced with the focal loss function of balance to measure the classification accuracy of target, is reduced big in training process
The contribution rate that amount trolley sample learns network weight, promotes training to focus more on difficult example sample.(5) and the combination of (4) is certain
Intensive vehicle detection accuracy is improved in degree.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Made any modification within mind and principle, equivalent replacement and improvement etc., should all be included in the protection scope of the present invention.
Claims (8)
1. a kind of road target detection recognition method based on monocular vision and deep learning, which is characterized in that described based on single
Visually feel with the road target detection recognition method of deep learning the following steps are included:
The first step establishes small sample road vehicle data set, collects the road vehicle image under real scene；
Depth convolutional neural networks frame SSD is detected applied to road vehicle, forms new network knot in conjunction with VGG by second step
Structure realizes real-time detection end to end, and the training stage uses the weighting parameter of the gradient descent method learning network with momentum, will hand over
Pitch the loss function that entropy loss function and smooth L1 loss function are returned as classifier and position；
Third step passes through secondary migration study trim network higher level on VOC data set and small sample road vehicle data set
Weighting parameter；
4th step introduces the characteristic pattern pyramid with lateral connection, is attached to primitive network in a manner of transmitting of jumping, does not change
Network structure originally increases a small amount of weighting parameter and calculating operation；
5th step measures the classification accuracy of target using the focal loss function of balance, reduces a large amount of small in training process
The contribution rate that vehicle sample learns network weight promotes training to focus more on difficult example sample.
2. the road target detection recognition method based on monocular vision and deep learning as described in claim 1, which is characterized in that
Tricycle, lorry are concentrated in screening public data emphatically in the first step, mark corresponding vehicle tag.
3. the road target detection recognition method based on monocular vision and deep learning as described in claim 1, which is characterized in that
The first step specifically includes:
(1) a certain amount of real roads scene video data are obtained by picture recording equipment, wherein comprising a large amount of all kinds of motor-driven
The image information of Chefei's motor vehicle and pedestrian；
(2) program is write to save as the articulating frame in video with formatted image with time interval, it is enterprising in all images
Pedestrian's work mark, by all images all vehicles and pedestrian be marked and location information saved as into corresponding image
Label；It writes program simultaneously to screen VOC data set, obtains the data set portion comprising all kinds of vehicles and pedestrian
Point.
4. the road target detection recognition method based on monocular vision and deep learning as described in claim 1, which is characterized in that
Concrete model is to be replaced based on SSD target detection model structure to feature extraction network VGG therein in the second step
Repair changes optimization, and in channel dimension equal part, each intermediate features figure obtains matrix1 and matrix2, and uses following operation
Retain the maximum value of two eigenmatrix corresponding points to obtain new maxout eigenmatrix matrix3:matrix3=
Matrix1-matrix2, the minimum value that matrix3 is truncated make it equal to 0, then matrix3=matrix3+matrix2；
Training stage uses the weighting parameter of the gradient descent method learning network with momentum, by cross entropy loss function:With smooth L1 loss function: The loss function returned respectively as target category classification and target detection frame position.
5. the road target detection recognition method based on monocular vision and deep learning as described in claim 1, which is characterized in that
Using the focal loss function FocalLoss of balance in 5th step:
Point of target is measured instead of cross entropy loss function
Class accuracy reduces the contribution rate that a large amount of trolley samples learn network weight in training process, training is promoted to focus more on difficulty
Example sample.
6. a kind of known using described in Claims 1 to 5 any one based on monocular vision and the detection of the road target of deep learning
The intelligent traffic administration system platform of other method.
7. a kind of known using described in Claims 1 to 5 any one based on monocular vision and the detection of the road target of deep learning
The smart city of other method manages platform.
8. a kind of known using described in Claims 1 to 5 any one based on monocular vision and the detection of the road target of deep learning
The car networking control platform of other method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811547843.8A CN109766769A (en) | 2018-12-18 | 2018-12-18 | A kind of road target detection recognition method based on monocular vision and deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811547843.8A CN109766769A (en) | 2018-12-18 | 2018-12-18 | A kind of road target detection recognition method based on monocular vision and deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109766769A true CN109766769A (en) | 2019-05-17 |
Family
ID=66451497
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811547843.8A Pending CN109766769A (en) | 2018-12-18 | 2018-12-18 | A kind of road target detection recognition method based on monocular vision and deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109766769A (en) |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110276268A (en) * | 2019-05-28 | 2019-09-24 | 江苏金海星导航科技有限公司 | Distortion driving video vehicle detection method based on metric learning and SSD network |
| CN110334646A (en) * | 2019-07-02 | 2019-10-15 | 朱恺晗 | A kind of detection method of single classification obstacle recognition based on SSD |
| CN110472503A (en) * | 2019-07-11 | 2019-11-19 | 桂林电子科技大学 | A kind of road target detection method, device and storage medium |
| CN110490188A (en) * | 2019-07-02 | 2019-11-22 | 天津大学 | A kind of target object rapid detection method based on SSD network improvement type |
| CN110503154A (en) * | 2019-08-27 | 2019-11-26 | 携程计算机技术（上海）有限公司 | Method, system, electronic equipment and the storage medium of image classification |
| CN111310592A (en) * | 2020-01-20 | 2020-06-19 | 杭州视在科技有限公司 | Detection method based on scene analysis and deep learning |
| CN111522946A (en) * | 2020-04-22 | 2020-08-11 | 成都中科云集信息技术有限公司 | Paper quality evaluation method based on attention long-short term memory recurrent neural network |
| CN111578940A (en) * | 2020-04-24 | 2020-08-25 | 哈尔滨工业大学 | Indoor monocular navigation method and system based on cross-sensor transfer learning |
| CN112270286A (en) * | 2020-11-09 | 2021-01-26 | 北京机电工程研究所 | Shadow interference resistant monochrome video target tracking method |
| CN112489138A (en) * | 2020-12-02 | 2021-03-12 | 中国船舶重工集团公司第七一六研究所 | Target situation information intelligent acquisition system based on wearable equipment |
| CN112507840A (en) * | 2020-12-02 | 2021-03-16 | 中国船舶重工集团公司第七一六研究所 | Man-machine hybrid enhanced small target detection and tracking method and system |
| CN112926495A (en) * | 2021-03-19 | 2021-06-08 | 高新兴科技集团股份有限公司 | Vehicle detection method based on multistage convolution characteristic cascade |
| CN113159116A (en) * | 2021-03-10 | 2021-07-23 | 中国科学院大学 | Small sample image target detection method based on class interval balance |
| CN113269118A (en) * | 2021-06-07 | 2021-08-17 | 重庆大学 | Monocular vision forward vehicle distance detection method based on depth estimation |
| CN113537244A (en) * | 2021-07-23 | 2021-10-22 | 深圳职业技术学院 | Livestock image target detection method and device based on light-weight YOLOv4 |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107833236A (en) * | 2017-10-31 | 2018-03-23 | 中国科学院电子学研究所 | Semantic vision positioning system and method are combined under a kind of dynamic environment |
| CN108304787A (en) * | 2018-01-17 | 2018-07-20 | 河南工业大学 | Road target detection method based on convolutional neural networks |
| CN108596053A (en) * | 2018-04-09 | 2018-09-28 | 华中科技大学 | A kind of vehicle checking method and system based on SSD and vehicle attitude classification |
-
2018
- 2018-12-18 CN CN201811547843.8A patent/CN109766769A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107833236A (en) * | 2017-10-31 | 2018-03-23 | 中国科学院电子学研究所 | Semantic vision positioning system and method are combined under a kind of dynamic environment |
| CN108304787A (en) * | 2018-01-17 | 2018-07-20 | 河南工业大学 | Road target detection method based on convolutional neural networks |
| CN108596053A (en) * | 2018-04-09 | 2018-09-28 | 华中科技大学 | A kind of vehicle checking method and system based on SSD and vehicle attitude classification |
Non-Patent Citations (1)
| Title |
|---|
| 张泽轩等: "基于深度学习的道路实景行人车辆检测", 《图形图像》 * |
Cited By (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110276268A (en) * | 2019-05-28 | 2019-09-24 | 江苏金海星导航科技有限公司 | Distortion driving video vehicle detection method based on metric learning and SSD network |
| CN110334646A (en) * | 2019-07-02 | 2019-10-15 | 朱恺晗 | A kind of detection method of single classification obstacle recognition based on SSD |
| CN110490188A (en) * | 2019-07-02 | 2019-11-22 | 天津大学 | A kind of target object rapid detection method based on SSD network improvement type |
| CN110472503A (en) * | 2019-07-11 | 2019-11-19 | 桂林电子科技大学 | A kind of road target detection method, device and storage medium |
| CN110503154A (en) * | 2019-08-27 | 2019-11-26 | 携程计算机技术（上海）有限公司 | Method, system, electronic equipment and the storage medium of image classification |
| CN111310592B (en) * | 2020-01-20 | 2023-06-16 | 杭州视在科技有限公司 | Detection method based on scene analysis and deep learning |
| CN111310592A (en) * | 2020-01-20 | 2020-06-19 | 杭州视在科技有限公司 | Detection method based on scene analysis and deep learning |
| CN111522946A (en) * | 2020-04-22 | 2020-08-11 | 成都中科云集信息技术有限公司 | Paper quality evaluation method based on attention long-short term memory recurrent neural network |
| CN111578940A (en) * | 2020-04-24 | 2020-08-25 | 哈尔滨工业大学 | Indoor monocular navigation method and system based on cross-sensor transfer learning |
| CN111578940B (en) * | 2020-04-24 | 2021-05-11 | 哈尔滨工业大学 | Indoor monocular navigation method and system based on cross-sensor transfer learning |
| CN112270286A (en) * | 2020-11-09 | 2021-01-26 | 北京机电工程研究所 | Shadow interference resistant monochrome video target tracking method |
| CN112270286B (en) * | 2020-11-09 | 2024-05-10 | 北京机电工程研究所 | Shadow interference resistant monochromatic video target tracking method |
| CN112507840A (en) * | 2020-12-02 | 2021-03-16 | 中国船舶重工集团公司第七一六研究所 | Man-machine hybrid enhanced small target detection and tracking method and system |
| CN112489138B (en) * | 2020-12-02 | 2024-02-20 | 中国船舶集团有限公司第七一六研究所 | Target situation information intelligent acquisition system based on wearable equipment |
| CN112489138A (en) * | 2020-12-02 | 2021-03-12 | 中国船舶重工集团公司第七一六研究所 | Target situation information intelligent acquisition system based on wearable equipment |
| CN112507840B (en) * | 2020-12-02 | 2024-06-18 | 中国船舶集团有限公司第七一六研究所 | Man-machine hybrid enhanced small target detection and tracking method and system |
| CN113159116A (en) * | 2021-03-10 | 2021-07-23 | 中国科学院大学 | Small sample image target detection method based on class interval balance |
| CN112926495A (en) * | 2021-03-19 | 2021-06-08 | 高新兴科技集团股份有限公司 | Vehicle detection method based on multistage convolution characteristic cascade |
| CN113269118A (en) * | 2021-06-07 | 2021-08-17 | 重庆大学 | Monocular vision forward vehicle distance detection method based on depth estimation |
| CN113537244A (en) * | 2021-07-23 | 2021-10-22 | 深圳职业技术学院 | Livestock image target detection method and device based on light-weight YOLOv4 |
| CN113537244B (en) * | 2021-07-23 | 2024-03-15 | 深圳职业技术学院 | Livestock image target detection method and device based on lightweight YOLOv4 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109766769A (en) | A kind of road target detection recognition method based on monocular vision and deep learning | |
| CN106448184B (en) | Vehicle identification method and vehicle appearance recognition methods | |
| CN110660222B (en) | Intelligent environment-friendly electronic snapshot system for black-smoke road vehicle | |
| CN110245644A (en) | A kind of unmanned plane image transmission tower lodging knowledge method for distinguishing based on deep learning | |
| CN103902976B (en) | A kind of pedestrian detection method based on infrared image | |
| CN107085696A (en) | A kind of vehicle location and type identifier method based on bayonet socket image | |
| CN108074244B (en) | Safe city traffic flow statistical method integrating deep learning and background difference method | |
| CN111368687A (en) | Sidewalk vehicle illegal parking detection method based on target detection and semantic segmentation | |
| CN109165682A (en) | A kind of remote sensing images scene classification method merging depth characteristic and significant characteristics | |
| CN103679214B (en) | Vehicle checking method based on online Class area estimation and multiple features Decision fusion | |
| CN109740676A (en) | Object detection moving method based on similar purpose | |
| CN105608431A (en) | Vehicle number and traffic flow speed based highway congestion detection method | |
| CN110119726A (en) | A kind of vehicle brand multi-angle recognition methods based on YOLOv3 model | |
| CN109902629A (en) | A kind of real-time vehicle target detection model under vehicles in complex traffic scene | |
| CN112668375B (en) | Tourist distribution analysis system and method in scenic spot | |
| CN107273832A (en) | Licence plate recognition method and system based on integrating channel feature and convolutional neural networks | |
| CN113205107A (en) | Vehicle type recognition method based on improved high-efficiency network | |
| CN114267082B (en) | Bridge side falling behavior identification method based on depth understanding | |
| CN109086803A (en) | A kind of haze visibility detection system and method based on deep learning and the personalized factor | |
| CN108648210B (en) | Rapid multi-target detection method and device under static complex scene | |
| CN113450573A (en) | Traffic monitoring method and traffic monitoring system based on unmanned aerial vehicle image recognition | |
| CN110674887A (en) | End-to-end road congestion detection algorithm based on video classification | |
| CN111414861A (en) | Method for realizing detection processing of pedestrians and non-motor vehicles based on deep learning | |
| CN111540203B (en) | Method for adjusting green light passing time based on fast-RCNN | |
| CN114596316A (en) | Road image detail capturing method based on semantic segmentation |
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: 20190517 |
|
| RJ01 | Rejection of invention patent application after publication |