CN109447171A - A kind of vehicle attitude classification method based on deep learning - Google Patents
A kind of vehicle attitude classification method based on deep learning Download PDFInfo
- Publication number
- CN109447171A CN109447171A CN201811309235.3A CN201811309235A CN109447171A CN 109447171 A CN109447171 A CN 109447171A CN 201811309235 A CN201811309235 A CN 201811309235A CN 109447171 A CN109447171 A CN 109447171A
- Authority
- CN
- China
- Prior art keywords
- vehicle
- deep learning
- method based
- classification method
- training
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/08—Detecting or categorising vehicles
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The vehicle attitude classification method based on deep learning that the present invention provides a kind of.This method comprises: training dataset building, validation data set building, building mobile-net network, network training, prediction.Present invention employs mobile-net models to classify to vehicle attitude, and this improves the speed of the accuracy rate of classification and classification.
Description
Technical field
The present invention relates to technical field of computer vision, are exactly a kind of vehicle attitude classification based on deep learning
Method.
Background technique
With progress of human society, the masses have been owned by the vehicle of oneself, along with produce various car accidents or
Produce traffic congestion.Vehicle attitude classification is the important component of many vehicle application systems, a kind of to cope with various rings
Border, various climate changes, the vehicle attitude classification method for adapting to a variety of models have emphatically for improving vehicle related application system
The meaning wanted.
Classify this problem for vehicle attitude, that is, accurately distinguishes the posture all around of picture vehicle,
The method that can be found in patent system at present mainly has the machine learning method traditional based on SVM, decision tree etc., but these methods
A large amount of priori knowledge and huge calculation amount are needed, it is bad to the vehicle attitude classification robustness under complex scene and weather,
Therefore cause nicety of grading inadequate, cannot be applied in engineering.
Deep learning technology is like a raging fire, produces the sorter networks such as VGG, Inception, Resnet in the recent period.Above-mentioned net
Although network can guarantee nicety of grading, there is a problem of that the training time is long, predicted time is long.Therefore method mentioned above
Using less in engineering.In consideration of it, this method uses mobile-net network, a kind of vehicle appearance based on deep learning is devised
State classification method has good robustness and detection speed, while parameter is few, and model committed memory is small.
Summary of the invention
The shortcomings that in view of passing technology described above, the present invention provides a kind of vehicle attitude classification side based on deep learning
Method, to improve the accuracy and speed of vehicle detection,
To achieve the goals above, the present invention the following steps are included:
Step 1) obtains the vehicle pictures of past few years and existing vehicle pictures on the internet in database, and it is appropriate to carry out
Cut and remove certain background, set mark rule all around, then manually marked, as model training set and
Verifying collection；
Step 2) does data enhancing processing to the picture in all training sets；
Step 3) constructs mobile-net network；
The enhanced training set of data is trained by step 4), and every 1/2 epoch stores a model；
Step 5) training convergence after, verifying collection on analysis there are the problem of, and adjustment hyper parameter, selection verifying collect
The highest model of accuracy rate；
Step 6) is based on the model selected, and to deduction program before building, input tape predicted pictures are predicted.
Above-mentioned steps 1) described in mark rule refer to: according to picture shooting visual angle, it is left that vehicle pictures are divided into front and back
It is right.
Above-mentioned steps 1) described in cutting remove certain background and refer to: many picture shootings are vehicle, are plucked out one
Part constitutes the middle scape figure an of vehicle.
Above-mentioned steps 2) described in data enhancing refer to: to each picture be HSL transformation, low-angle (- 30 ° to+
30 °) enhancing such as rotation transformation plus random noise, random cropping, aspect ratio transformation, it is not possible to it is turning-over changed to do left and right.
Above-mentioned steps 3) described in mobile-net network refer to: width factor 1.0, resolution factor are 1.0
Standard mobile-net network, standard mobile-net network made of being stacked altogether as 22 convolutional layers, 1 full articulamentum.
Above-mentioned steps 4) described in training process refer to: the training dataset after amplification is sent into according to certain batch
Mobile-net network is trained, and then backpropagation updates model parameter again.
Above-mentioned steps 5) described in verifying collection refer to: the verifying collection needs to include all vehicles, each vehicle
There are a certain number of pictures all around.
Above-mentioned steps 6) described in forward direction infer process refer to: building mobile-net network, then parameter using step
It is rapid 5) in selected model parameter, input the picture for 224*224, last output is a number in 0-3, according to
Label mapping when training, corresponds to all around.
Detailed description of the invention
Fig. 1 is the input example of the vehicle attitude classification method of the invention based on deep learning
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.
Referring to Fig. 1, the vehicle attitude classification method of the invention based on deep learning includes training dataset building, tests
Demonstrate,prove data set building, building mobile-net network, network training, prediction.
The following steps are included:
Step 1) obtains the vehicle pictures of past few years and existing vehicle pictures on the internet in database, and it is appropriate to carry out
Cut and remove certain background, set mark rule all around, then manually marked, as model training set and
Verifying collection；
Step 2) does data enhancing processing to the picture in all training sets；
Step 3) constructs mobile-net network；
The enhanced training set of data is trained by step 4), and every 1/2 epoch stores a model；
Step 5) training convergence after, verifying collection on analysis there are the problem of, and adjustment hyper parameter, selection verifying collect
The highest model of accuracy rate；
Step 6) is based on the model selected, and to deduction program before building, input tape predicted pictures are predicted.
Above-mentioned steps 1) described in mark rule refer to: according to picture shooting visual angle, it is left that vehicle pictures are divided into front and back
It is right.
Above-mentioned steps 1) described in cutting remove certain background and refer to: many picture shootings are vehicle, are plucked out one
Part constitutes the middle scape figure an of vehicle.
Above-mentioned steps 2) described in data enhancing refer to: to each picture be HSL transformation, low-angle (- 30 ° to+
30 °) enhancing such as rotation transformation plus random noise, random cropping, aspect ratio transformation, it is not possible to it is turning-over changed to do left and right.
Above-mentioned steps 3) described in mobile-net network refer to: width factor 1.0, resolution factor are 1.0
Standard mobile-net network, standard mobile-net network made of being stacked altogether as 22 convolutional layers, 1 full articulamentum.
Above-mentioned steps 4) described in training process refer to: the training dataset after amplification is sent into according to certain batch
Mobile-net network is trained, and then backpropagation updates model parameter again.
Above-mentioned steps 5) described in verifying collection refer to: the verifying collection needs to include all vehicles, each vehicle
There are a certain number of pictures all around.
Above-mentioned steps 6) described in forward direction infer process refer to: building mobile-net network, then parameter using step
It is rapid 5) in selected model parameter, input the picture for 224*224, last output is a number in 0-3, according to
Label mapping when training, corresponds to all around.
Vehicle attitude classification method based on deep learning of the invention has fully considered each vehicle, all angles, each
Kind scene can be quickly obtained recognition result using mobile-net network, and keep higher accuracy rate, have very
Good robustness.
Method provided by the present invention is described in detail above, specific case used herein is to of the invention
Principle and embodiment is expounded, method and its core of the invention that the above embodiments are only used to help understand
Thought；At the same time, for those skilled in the art, according to the thought of the present invention, in specific embodiment and application range
Upper there will be changes, in conclusion the contents of this specification are not to be construed as limiting the invention.
Claims (8)
1. a kind of vehicle attitude classification method based on deep learning, which comprises the following steps:
Step 1) obtains the vehicle pictures of past few years and existing vehicle pictures on the internet in database, is suitably cut
Certain background is removed, mark rule all around is set, is then manually marked, training set and verifying as model
Collection；
Step 2) does data enhancing processing to the picture in all training sets；
Step 3) constructs mobile-net network；
The enhanced training set of data is trained by step 4), and every 1/2 epoch stores a model；
After step 5) training convergence, analysis on verifying collection there are the problem of, and adjustment hyper parameter, selection be accurate in verifying collection
The highest model of rate；
Step 6) is based on the model selected, and to deduction program before building, input tape predicted pictures are predicted.
2. a kind of vehicle attitude classification method based on deep learning according to claim 1, characterized in that in step 1)
Described in mark rule refer to: according to picture shooting visual angle, vehicle pictures are divided into all around.
3. a kind of vehicle attitude classification method based on deep learning according to claim 1, characterized in that in step 1)
Described in cutting remove certain background and refer to: many picture shootings are vehicle, are plucked out a part, constitute a vehicle
Middle scape figure.
4. a kind of vehicle attitude classification method based on deep learning according to claim 1, characterized in that in step 2)
Described in data enhancing refer to: to each picture do HSL transformation, low-angle (- 30 ° to+30 °) rotation transformation, add at random
The enhancing such as noise, random cropping, aspect ratio transformation, it is not possible to it is turning-over changed to do left and right.
5. a kind of vehicle attitude classification method based on deep learning according to claim 1, characterized in that in step 3)
Described in mobile-net network refer to: width factor 1.0, resolution factor be 1.0 standard mobile-net network,
Standard mobile-net network made of being stacked altogether as 22 convolutional layers, 1 full articulamentum.
6. a kind of vehicle attitude classification method based on deep learning according to claim 1, characterized in that in step 4)
Described in training process refer to: by the training dataset after amplification, be sent into mobile-net network according to certain batch and carry out
Training, then backpropagation updates model parameter again.
7. a kind of vehicle attitude classification method based on deep learning according to claim 1, characterized in that in step 5)
Described in verifying collection refer to: the verifying collection needs to include all vehicles, each vehicle has a certain number of front and backs
The picture of left and right.
8. a kind of vehicle attitude classification method based on deep learning according to claim 1, characterized in that in step 6)
Described in forward direction infer that process refers to: building mobile-net network, then parameter is joined using model selected in step 5)
Number, inputs the picture for 224*224, last output is a number in 0-3, label mapping when according to training, right
It should arrive all around.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811309235.3A CN109447171A (en) | 2018-11-05 | 2018-11-05 | A kind of vehicle attitude classification method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811309235.3A CN109447171A (en) | 2018-11-05 | 2018-11-05 | A kind of vehicle attitude classification method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109447171A true CN109447171A (en) | 2019-03-08 |
Family
ID=65551910
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811309235.3A Pending CN109447171A (en) | 2018-11-05 | 2018-11-05 | A kind of vehicle attitude classification method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109447171A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109910891A (en) * | 2019-03-20 | 2019-06-21 | 奇瑞汽车股份有限公司 | Control method for vehicle and device |
| CN110458225A (en) * | 2019-08-08 | 2019-11-15 | 北京深醒科技有限公司 | A kind of vehicle detection and posture are classified joint recognition methods |
| CN110751197A (en) * | 2019-10-14 | 2020-02-04 | 上海眼控科技股份有限公司 | Picture classification method, picture model training method and equipment |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9063930B2 (en) * | 2011-09-24 | 2015-06-23 | Z Advanced Computing, Inc. | Method and system for analyzing or resolving ambiguities in image recognition for gesture, emotion, or expression recognition for a human |
| CN105740906A (en) * | 2016-01-29 | 2016-07-06 | 中国科学院重庆绿色智能技术研究院 | Depth learning based vehicle multi-attribute federation analysis method |
| CN105975941A (en) * | 2016-05-31 | 2016-09-28 | 电子科技大学 | Multidirectional vehicle model detection recognition system based on deep learning |
| CN106250838A (en) * | 2016-07-27 | 2016-12-21 | 乐视控股（北京）有限公司 | vehicle identification method and system |
| CN107133570A (en) * | 2017-04-07 | 2017-09-05 | 武汉睿智视讯科技有限公司 | A kind of vehicle/pedestrian detection method and system |
| CN108010030A (en) * | 2018-01-24 | 2018-05-08 | 福州大学 | A kind of Aerial Images insulator real-time detection method based on deep learning |
| CN108509954A (en) * | 2018-04-23 | 2018-09-07 | 合肥湛达智能科技有限公司 | A kind of more car plate dynamic identifying methods of real-time traffic scene |
| 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 |
| CN108632530A (en) * | 2018-05-08 | 2018-10-09 | 阿里巴巴集团控股有限公司 | A kind of data processing method of car damage identification, device, processing equipment and client |
| CN108682010A (en) * | 2018-05-08 | 2018-10-19 | 阿里巴巴集团控股有限公司 | Processing method, processing equipment, client and the server of vehicle damage identification |
-
2018
- 2018-11-05 CN CN201811309235.3A patent/CN109447171A/en active Pending
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9063930B2 (en) * | 2011-09-24 | 2015-06-23 | Z Advanced Computing, Inc. | Method and system for analyzing or resolving ambiguities in image recognition for gesture, emotion, or expression recognition for a human |
| CN105740906A (en) * | 2016-01-29 | 2016-07-06 | 中国科学院重庆绿色智能技术研究院 | Depth learning based vehicle multi-attribute federation analysis method |
| CN105975941A (en) * | 2016-05-31 | 2016-09-28 | 电子科技大学 | Multidirectional vehicle model detection recognition system based on deep learning |
| CN106250838A (en) * | 2016-07-27 | 2016-12-21 | 乐视控股（北京）有限公司 | vehicle identification method and system |
| CN107133570A (en) * | 2017-04-07 | 2017-09-05 | 武汉睿智视讯科技有限公司 | A kind of vehicle/pedestrian detection method and system |
| CN108010030A (en) * | 2018-01-24 | 2018-05-08 | 福州大学 | A kind of Aerial Images insulator real-time detection method based on deep learning |
| 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 |
| CN108509954A (en) * | 2018-04-23 | 2018-09-07 | 合肥湛达智能科技有限公司 | A kind of more car plate dynamic identifying methods of real-time traffic scene |
| CN108632530A (en) * | 2018-05-08 | 2018-10-09 | 阿里巴巴集团控股有限公司 | A kind of data processing method of car damage identification, device, processing equipment and client |
| CN108682010A (en) * | 2018-05-08 | 2018-10-19 | 阿里巴巴集团控股有限公司 | Processing method, processing equipment, client and the server of vehicle damage identification |
Non-Patent Citations (3)
| Title |
|---|
| ANDREW G. HOWARD等: "MobileNets: Efficient Convolutional Neural Networks for Mobile VisionApplications", 《ARXIV:1704.04861V1 [CS.CV]》 * |
| BAEK等: "Real-time Detection, Tracking, and Classification of Moving and Stationary Objects using Multiple Fisheye Images", 《2018 IEEE INTELLIGENT VEHICLES SYMPOSIUM (IV)》 * |
| 卞聪超: "基于车辆行驶姿态的主动安全系统设计与实现", 《中国优秀硕士学位论文全文数据库·工程科技Ⅱ辑》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109910891A (en) * | 2019-03-20 | 2019-06-21 | 奇瑞汽车股份有限公司 | Control method for vehicle and device |
| CN110458225A (en) * | 2019-08-08 | 2019-11-15 | 北京深醒科技有限公司 | A kind of vehicle detection and posture are classified joint recognition methods |
| CN110751197A (en) * | 2019-10-14 | 2020-02-04 | 上海眼控科技股份有限公司 | Picture classification method, picture model training method and equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US9129191B2 (en) | Semantic object selection | |
| US9129192B2 (en) | Semantic object proposal generation and validation | |
| CN106920229B (en) | Automatic detection method and system for image fuzzy area | |
| AU2019213369B2 (en) | Non-local memory network for semi-supervised video object segmentation | |
| WO2020169043A1 (en) | Dense crowd counting method, apparatus and device, and storage medium | |
| CN104063876B (en) | Interactive image segmentation method | |
| CN108140032B (en) | Apparatus and method for automatic video summarization | |
| CN108564129B (en) | Trajectory data classification method based on generation countermeasure network | |
| WO2019114523A1 (en) | Classification training method, server and storage medium | |
| CN113674140B (en) | Physical countermeasure sample generation method and system | |
| CN106294344B (en) | Video retrieval method and device | |
| CN106960195A (en) | A kind of people counting method and device based on deep learning | |
| CN107862270A (en) | Face classification device training method, method for detecting human face and device, electronic equipment | |
| CN109447171A (en) | A kind of vehicle attitude classification method based on deep learning | |
| CN109492596B (en) | Pedestrian detection method and system based on K-means clustering and regional recommendation network | |
| CN105808610A (en) | Internet picture filtering method and device | |
| CN110390302A (en) | A kind of objective detection method | |
| CN109033144B (en) | Three-dimensional model retrieval method based on sketch | |
| CN107730553B (en) | Weak supervision object detection method based on false-true value search method | |
| CN110020650B (en) | Inclined license plate recognition method and device based on deep learning recognition model | |
| CN106203296B (en) | The video actions recognition methods of one attribute auxiliary | |
| US8204889B2 (en) | System, method, and computer-readable medium for seeking representative images in image set | |
| CN111783779A (en) | Image processing method, apparatus and computer-readable storage medium | |
| CN111652141B (en) | Question segmentation method, device, equipment and medium based on question numbers and text lines | |
| CN111079507A (en) | Behavior recognition method and device, computer device and readable storage medium |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190308 |