CN110532878A - A kind of driving behavior recognition methods based on lightweight convolutional neural networks - Google Patents
A kind of driving behavior recognition methods based on lightweight convolutional neural networks Download PDFInfo
- Publication number
- CN110532878A CN110532878A CN201910681906.7A CN201910681906A CN110532878A CN 110532878 A CN110532878 A CN 110532878A CN 201910681906 A CN201910681906 A CN 201910681906A CN 110532878 A CN110532878 A CN 110532878A
- Authority
- CN
- China
- Prior art keywords
- driving behavior
- convolution
- channel
- convolutional neural
- neural networks
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- G06V20/59—Context or environment of the image inside of a vehicle, e.g. relating to seat occupancy, driver state or inner lighting conditions
- G06V20/597—Recognising the driver's state or behaviour, e.g. attention or drowsiness
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The present invention is the driving behavior recognition methods based on lightweight convolutional neural networks, obtains driving behavior public data collection, obtains the picture of different driving behavior classification；Picture is pre-processed, upsets data set at random and divides training set and test set；Data enhancing is carried out to training set；Lightweight convolutional neural networks are designed, training set input data input network is subjected to feature extraction；The probabilistic forecasting for carrying out each driving behavior classification to the feature vector extracted with classifier instructs network training direction in next step by backpropagation according to training set class label to the probability calculation loss function of prediction；It saves trained driving behavior disaggregated model after the completion of training and saves.The present invention designs lightweight convolutional neural networks by using the information flow of convolution module and enhancing interchannel, the driving behavior disaggregated model that small in size, operation is simple and accuracy rate is high is trained, the identification classification of driving behavior is carried out suitable for vehicle-mounted mobile end.
Description
Technical field
It is specially a kind of based on lightweight convolutional neural networks the present invention relates to computer vision, depth learning technology
Driving behavior recognition methods.
Background technique
Driver distraction, which drives, has become the main reason for causing traffic accident, wherein by driver distraction's driving performance
Driving behavior out has cell phone use while driving (including make a phone call, photos and sending messages, browse web sites, play game etc.), feed and passenger
The behaviors such as interaction.Nowadays vehicle-mounted DAS (Driver Assistant System) is applied on various vehicles more and more widely, efficiently and accurately right
The driving behavior of driver identifies, to prevent from having become auxiliary due to driver distraction's bring traffic accident and driving
An extremely important function in system.For vehicle-mounted DAS (Driver Assistant System), how existing driving behavior number is utilized
According to collection, learn the feature of all kinds of driving behaviors, thus accurately and efficiently to driving behavior carry out Classification and Identification, be reduce due to
The effective means of traffic accident caused by driver distraction's driving.
Existing vehicle-mounted DAS (Driver Assistant System) directly uses generally according to the categorical data of the driving behavior got
The convolutional neural networks such as ResNet, InceptionV3, MobileNetV2, ShuffleNet carry out feature extraction and classification, obtain
Model after to training；Driving behavior picture to be sorted is inputted into convolutional neural networks using the model after the training later
After obtain driving behavior classification prediction result.In the selection of convolutional neural networks, the networks such as ResNet, InceptionV3
The characteristics of be precision compared with high but structure is complicated, model volume is big；The networks such as MobileNetV2, ShuffleNet can reduce operation
Amount, but bring the loss in precision.
As it can be seen that being trained with existing convolutional neural networks to driving behavior data set, it is difficult to guarantee precision
Under conditions of simplify network, obtain model small in size, thus with limited on vehicle-mounted mobile platform.And model it is small in size,
Network is simple, the higher lightweight convolutional neural networks of accuracy rate, can expand the application scenarios of such method significantly, it is low at
Higher accuracy rate is obtained on this all kinds of platforms.
Summary of the invention
To solve technical problem present in the prior art, the present invention proposes the driving based on lightweight convolutional neural networks
Member's Activity recognition method is designed a kind of light-weighted by using the information flow of simple convolution module and enhancing interchannel
It is simply and accurate can to train small in size, operation in conjunction with the existing public data collection of driving behavior for convolutional neural networks structure
The high driving behavior disaggregated model of rate carries out the identification classification of driving behavior suitable for vehicle-mounted mobile end scene.
The present invention is based on the driving behavior recognition methods of lightweight convolutional neural networks, comprising the following steps:
Driving behavior data set disclosed in S1, acquisition obtains corresponding to a series of pictures under different driving behavior classification；
S2, picture is pre-processed, then upsets data set at random, the data set after upsetting is divided, is instructed
Practice collection and test set；
S3, data enhancing processing is carried out to training set, to increase the diversity of training sample；
Training set data input lightweight convolutional neural networks are carried out feature by S4, design lightweight convolutional neural networks
It extracts；
S5, the probabilistic forecasting for carrying out each driving behavior classification to the feature vector extracted with classifier, according to instruction
Practice collection class label to the probability calculation loss function of prediction, and is instructed under lightweight convolutional neural networks by backpropagation
One step trains direction；
Trained driving behavior disaggregated model is saved after the completion of S6, training.
In a preferred embodiment, the lightweight convolutional neural networks include at least one convolution module, multiple bottlenecks
Module, the down-sampling bottleneck module that multiple step-lengths are 2, global average pond layer and full articulamentum；Multiple bottleneck modules with it is multiple
Step-length be 2 down-sampling bottleneck module connection type are as follows: after 1-3 bottleneck block coupled in series with the bottleneck module of a down-sampling
Connection forms a submodule, then multiple submodule is together in series and is connected between convolution module and global average pond layer；
The average pond layer of the overall situation is connect with full articulamentum.
Compared with the prior art, the invention has the following advantages and beneficial effects:
1, designed convolutional neural networks structure replaces the convolution module of 3*3 with the convolution module of 1*3 and 3*1, with
And reduce operand, thus the simple light weight of driving behavior disaggregated model, only 11MB or so with the depth convolution module of 1*1
Size, platform applicatory is extensive.
2, in recognition methods of the present invention, data are enhanced, expansion processing, increases the diversity of training sample；Fortune
The information flow promoted between each channel is shuffled with channel；The precision of neural network is improved using hard-swish activation primitive,
Accuracy rate is high, can reach 97.33% accuracy rate after training 50 times with 32 batches, can effectively predict driving behavior
Classification.
Detailed description of the invention
Fig. 1 is overall flow figure of the invention；
Fig. 2 is the structural schematic diagram of convolution module；
Fig. 3 is the process flow diagram of bottleneck module；
Fig. 4 is the process flow diagram for the down-sampling bottleneck module that step-length is 2.
Specific embodiment
Present invention will now be described in further detail with reference to the embodiments and the accompanying drawings, but embodiments of the present invention are unlimited
In this.
Embodiment
The data set that this method uses includes at least ten kinds to the classification of driving behavior, is respectively as follows: normal driving, the right hand
Mobile phone, right hand making and receiving calls, left hand is played to play mobile phone, left hand making and receiving calls, adjust in-vehicle device, feed, take thing from heel row, see mirror
Son arranges dress hair and passenger speaks.As shown in Figure 1, the present invention carries out knowledge method for distinguishing to driving behavior, including walks as follows
It is rapid:
Driving behavior data set disclosed in S1, acquisition obtains corresponding to a series of pictures under different driving behavior classification；Its
In, picture is successively read in by behavior class label.
S2, picture is pre-processed, then upsets data set at random, in training set: test set=8:2 ratio is to upsetting
Data set afterwards is divided, and training set and test set are obtained；
Preprocessing process are as follows: obtain driving behavior picture from data set, linear normalization processing first is carried out to picture:
F (x)=(x-min)/(max-min), wherein x is that a channel value of the pixel of image in RGB color space is (total
Have tri- channels R, G, B), in RGB color space, minimum value min=0, maximum value max=255；To all pixels of image
After three channels of point all carry out linear normalization processing, with area interpolation algorithm, lightweight convolution is scaled the images to
Neural network sets the size of clocking requirement, i.e. 224*224*3.
Data set partition process are as follows: it is advanced to the picture successively read in by behavior class label it is row stochastic upset, according to
Training set: test set=8:2 ratio divides the data set after upsetting.Wherein, training set is used to train driver's row
For the model of classification, test set is used to the classification performance of test model.
S3, data enhancing processing is carried out to training set, to increase the diversity of training sample；
Since the driving behavior that camera takes will receive the factors such as angle, distance, illumination and driver's colour of skin
Influence, in order to increase the diversity of training sample, so that neural network be made to have a better robustness, the present invention is from training data
In randomly select some pictures, based on these pictures carry out data enhancement operations.Data enhancing includes: rotation processing, by image
It rotates a certain angle；Translation processing, translates certain distance for image along the horizontal or vertical direction；Random cropping, random cropping
The image block of size is specified in original image；Color jitter changes the saturation degree, lightness and/or tone of image.
S4, design lightweight convolutional neural networks carry out training set input data input lightweight convolutional neural networks
Feature extraction；
Lightweight convolutional neural networks include at least one convolution module, multiple bottleneck modules, multiple step-lengths be 2 under adopt
Sample bottle eck mould block, global average pond layer and full articulamentum；The down-sampling bottleneck module that multiple bottleneck modules and multiple step-lengths are 2
Connection type are as follows: connect to form a submodule with the bottleneck module of a down-sampling after 1-3 bottleneck block coupled in series, then
Multiple submodule, which is together in series, to be connected between convolution module and global average pond layer；And global average pond layer with connect entirely
Layer connection.The network mainly passes through the down-sampling ring mould that at least one convolution module, multiple bottleneck modules and multiple step-lengths are 2
Block extracts data characteristics, obtains classification results finally by the average pondization of the overall situation and full articulamentum；In the present embodiment, the network
Specific structure parameter is as shown in table 1, and the function of full articulamentum is realized by two 1*1 convolution after global average pond layer.
1 network architecture parameters table of table
The structure of convolution module is as shown in Fig. 2, include sequentially connected convolutional layer, batch regularization and activation primitive.Bottleneck
The process flow for the down-sampling bottleneck module that module and step-length are 2, it is as shown in Figure 3,4 respectively.It is 2 in bottleneck module and step-length
In the process flow of down-sampling bottleneck module, each section is specific as follows:
Depth convolution: separately carrying out feature extraction for the region of image and channel, first individually rolls up to each channel of input
Product obtains characteristic pattern, then point-by-point across the channel convolution of 1*1 is done to obtained characteristic pattern.This operation can be big compared to traditional convolution
It is big to reduce operand.And depth convolution operation of step-length when being 2, it is special that higher is extracted while reducing photo resolution
Sign avoided directly through pondization down-sampled the problem of causing a part of characteristic information to be lost.
Channel separation: being divided into two parts for input channel, carries out next step operation respectively.
It expands convolution: injecting cavity in convolution kernel and carry out convolution operation, empty number is determined by extension convolution rate.Expansion
Convolution can increase receptive field in the case where not doing pond, while remain more information.And by the expansion of 3*3 in this method
It opens convolution and is divided into 1*3 expansion convolution sum 3*1 expansion two step of convolution progress, further reduce the operand of convolution operation.
Channel connection: it two parts will be divided into before carries out the characteristic pattern after expansion convolution operation and connected by channel.
It shuffles in channel: since the depth convolution done before does not have the circulation of characteristic information in interchannel, output can be caused special
The reduction of ability to express is levied, this method is added channel shuffle operation, upsets putting in order for original channel, increase the letter of interchannel
Breath circulation, promotes the ability to express of network model.
1*1 convolution: for depth convolution, dimensionality reduction is carried out to input feature value, reduces parameter.
It is added: in conjunction with the thought of Remanent Model, allowing feature cross-layer to transmit, prevent network caused by deepening due to bottleneck layer from moving back
Change.It, directly will be after input picture and above-mentioned 1*1 convolution again through batch regularization, activation primitive treated figure in bottleneck module
Piece is added；And in the down-sampling bottleneck module that step-length is 2, input picture is subjected to 3*3 and is averaged behind pond and above-mentioned 1*1 convolution
Through batch regularization, activation primitive, treated that picture is added again afterwards.
Wherein, activation primitive:
" RE " expression uses Relu6 as nonlinear activation function；
F (x) ≈ min (max (0, x+N (0,1)), 6)
" HS " expression uses hard-swish as nonlinear activation function, which can be efficiently modified network essence
Degree, and very big calculation amount will not be brought；
K indicates driving behavior classification number.
S5, the probabilistic forecasting for carrying out each driving behavior classification to the feature vector extracted with softmax classifier,
According to training set class label to the probability calculation cross entropy loss function of prediction, and lightweight convolution is instructed by backpropagation
The next step training direction of neural network；
Calculate the process of loss function and backpropagation are as follows: the driving behavior picture of input passes through lightweight convolutional Neural
After network progress feature extraction obtains feature vector, is classified using softmax function (classifier) to feature vector, compareed
The driving behavior class label of the driving behavior picture of input calculates loss function using cross entropy；According to loss function and
The optimizer of stochastic gradient descent carries out backpropagation, to start training next time.
It saves trained driving behavior disaggregated model after the completion of S6, training and saves；The model is small in size, only
The size of 11MB or so.
S7, input test picture to lightweight convolutional neural networks, using trained driving behavior disaggregated model with
And classifier immediately arrives at driving behavior class prediction result.
The test data separated from data set is input to lightweight convolutional neural networks, using trained before
Driving behavior disaggregated model and softmax classifier, can immediately arrive at the class label of driving behavior, to obtain
Driving behavior recognition result.
The above embodiment is a preferred embodiment of the present invention, but embodiments of the present invention are not by above-described embodiment
Limitation, other any changes, modifications, substitutions, combinations, simplifications made without departing from the spirit and principles of the present invention,
It should be equivalent substitute mode, be included within the scope of the present invention.
Claims (10)
1. a kind of driving behavior recognition methods based on lightweight convolutional neural networks, which comprises the following steps:
Driving behavior data set disclosed in S1, acquisition obtains corresponding to a series of pictures under different driving behavior classification；
S2, picture is pre-processed, then upsets data set at random, the data set after upsetting is divided, training set is obtained
And test set；
S3, data enhancing processing is carried out to training set, to increase the diversity of training sample；
Training set input data input lightweight convolutional neural networks are carried out feature by S4, design lightweight convolutional neural networks
It extracts；
S5, the probabilistic forecasting for carrying out each driving behavior classification to the feature vector extracted with classifier, according to training set
Class label instructs by backpropagation the next step of lightweight convolutional neural networks to the probability calculation loss function of prediction
Training direction；
Trained driving behavior disaggregated model is saved after the completion of S6, training.
2. driving behavior recognition methods according to claim 1, which is characterized in that the lightweight convolutional neural networks
Including at least one convolution module, multiple bottleneck modules, the down-sampling bottleneck module that multiple step-lengths are 2, global average pond layer
With full articulamentum；The connection type for the down-sampling bottleneck module that multiple bottleneck modules and multiple step-lengths are 2 are as follows: 1-3 ring mould
It connect to form a submodule with the bottleneck module of a down-sampling after block series connection, then multiple submodule is together in series and is connected to
Between convolution module and global average pond layer；The average pond layer of the overall situation is connect with full articulamentum.
3. driving behavior recognition methods according to claim 2, which is characterized in that the convolution module includes successively connecting
Convolutional layer, batch regularization and the activation primitive connect.
4. driving behavior recognition methods according to claim 2, which is characterized in that the process flow of the bottleneck module
Include:
Depth convolution: separately carrying out feature extraction for the region of image and channel, first obtains to the independent convolution in each channel of input
Across channel convolution point by point is done to characteristic pattern, then to obtained characteristic pattern；
Channel separation: being divided into two parts for input channel, carries out next step operation respectively；
It expands convolution: injecting cavity in convolution kernel and carry out convolution operation, empty number is determined by extension convolution rate；
Channel connection: it two parts will be divided into before carries out the characteristic pattern after expansion convolution operation and connected by channel；
It shuffles in channel: upsetting putting in order for original channel, increase the information flow of interchannel, promote the expression energy of network model
Power；
1*1 convolution: dimensionality reduction is carried out to input feature value, reduces parameter；
Be added: will input picture with criticized regularization again after above-mentioned 1*1 convolution, treated that picture is added for activation primitive.
5. driving behavior recognition methods according to claim 2, which is characterized in that the down-sampling bottle that the step-length is 2
The process flow of eck mould block includes:
Depth convolution: separately carrying out feature extraction for the region of image and channel, first obtains to the independent convolution in each channel of input
Across channel convolution point by point is done to characteristic pattern, then to obtained characteristic pattern；
Channel separation: being divided into two parts for input channel, carries out next step operation respectively；
It expands convolution: injecting cavity in convolution kernel and carry out convolution operation, empty number is determined by extension convolution rate；
Channel connection: it two parts will be divided into before carries out the characteristic pattern after expansion convolution operation and connected by channel；
It shuffles in channel: upsetting putting in order for original channel, increase the information flow of interchannel, promote the expression energy of network model
Power；
1*1 convolution: dimensionality reduction is carried out to input feature value, reduces parameter；
Be added: will input picture carry out behind average pond with criticized that regularization, treated for activation primitive after above-mentioned 1*1 convolution again
Picture is added.
6. driving behavior recognition methods according to claim 4 or 5, which is characterized in that the expansion convolution is by 3*3's
Expansion convolution is divided into 1*3 expansion convolution sum 3*1 expansion two step of convolution and carries out.
7. driving behavior recognition methods according to claim 1, which is characterized in that the processing of data enhancing described in step S3
It include: rotation processing, by the certain angle of image rotation；Image, is translated a spacing by translation processing along the horizontal or vertical direction
From；The image block of size is specified in random cropping original image；Color jitter changes the saturation degree, lightness and/or color of image
It adjusts.
8. driving behavior recognition methods according to claim 1, which is characterized in that according to training set in step S2: surveying
Examination collection=8:2 ratio divides the data set after upsetting.
9. driving behavior recognition methods according to claim 1, which is characterized in that preprocessing process in step S2 are as follows:
Driving behavior picture is obtained from data set, linear normalization processing first is carried out to picture, then uses area interpolation algorithm,
Picture is zoomed into the size that lightweight convolutional neural networks set clocking requirement.
10. driving behavior recognition methods according to claim 1, which is characterized in that the driving behavior identification side
Method further comprises the steps of:
S7, input test picture to lightweight convolutional neural networks using trained driving behavior disaggregated model and divide
Class device immediately arrives at driving behavior class prediction result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910681906.7A CN110532878B (en) | 2019-07-26 | 2019-07-26 | Driver behavior identification method based on lightweight convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910681906.7A CN110532878B (en) | 2019-07-26 | 2019-07-26 | Driver behavior identification method based on lightweight convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110532878A true CN110532878A (en) | 2019-12-03 |
| CN110532878B CN110532878B (en) | 2022-11-29 |
Family
ID=68661827
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910681906.7A Active CN110532878B (en) | 2019-07-26 | 2019-07-26 | Driver behavior identification method based on lightweight convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110532878B (en) |
Cited By (26)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111062399A (en) * | 2019-12-12 | 2020-04-24 | 易诚高科（大连）科技有限公司 | Monitoring video face recognition method based on color dithering and image mixing |
| CN111209972A (en) * | 2020-01-09 | 2020-05-29 | 中国科学院计算技术研究所 | Image classification method and system based on hybrid connectivity deep convolution neural network |
| CN111553190A (en) * | 2020-03-30 | 2020-08-18 | 浙江工业大学 | Image-based driver attention detection method |
| CN111582101A (en) * | 2020-04-28 | 2020-08-25 | 中国科学院空天信息创新研究院 | Remote sensing image detection method and system |
| CN111626186A (en) * | 2020-05-25 | 2020-09-04 | 宁波大学 | Driver distraction detection method |
| CN111832446A (en) * | 2020-06-30 | 2020-10-27 | 东南大学 | Driver posture identification method based on double-view-angle video data |
| CN111860316A (en) * | 2020-07-20 | 2020-10-30 | 上海汽车集团股份有限公司 | Driving behavior recognition method and device and storage medium |
| CN111860427A (en) * | 2020-07-30 | 2020-10-30 | 重庆邮电大学 | Driving distraction identification method based on lightweight class eight-dimensional convolutional neural network |
| CN112053354A (en) * | 2020-09-15 | 2020-12-08 | 上海应用技术大学 | Track slab crack detection method |
| CN112487996A (en) * | 2020-12-02 | 2021-03-12 | 重庆邮电大学 | Driving behavior recognition method based on DenseNet121 network |
| CN112732797A (en) * | 2021-01-26 | 2021-04-30 | 武汉理工大学 | Fuel cell automobile driving behavior analysis method, device and storage medium |
| CN112784723A (en) * | 2021-01-14 | 2021-05-11 | 金陵科技学院 | Road traffic safety protection model based on IFast-RCNN algorithm |
| CN112907600A (en) * | 2021-03-10 | 2021-06-04 | 江苏禹空间科技有限公司 | Optimization method and system of target detection model |
| CN113052189A (en) * | 2021-03-30 | 2021-06-29 | 电子科技大学 | Improved MobileNet V3 feature extraction network |
| CN113111938A (en) * | 2021-04-09 | 2021-07-13 | 中国工程物理研究院电子工程研究所 | Terrain classification method based on digital elevation model data |
| CN113128353A (en) * | 2021-03-26 | 2021-07-16 | 安徽大学 | Emotion sensing method and system for natural human-computer interaction |
| CN113378653A (en) * | 2021-05-19 | 2021-09-10 | 海南师范大学 | Method and device for detecting in-vehicle behavior of driver |
| CN113449656A (en) * | 2021-07-01 | 2021-09-28 | 淮阴工学院 | Driver state identification method based on improved convolutional neural network |
| CN113505628A (en) * | 2021-04-02 | 2021-10-15 | 上海师范大学 | Target identification method based on lightweight neural network and application thereof |
| CN113536919A (en) * | 2021-06-10 | 2021-10-22 | 重庆邮电大学 | Signal modulation recognition algorithm based on data enhancement and convolutional neural network |
| CN113780385A (en) * | 2021-08-30 | 2021-12-10 | 武汉理工大学 | Driving risk monitoring method based on attention mechanism |
| CN114187581A (en) * | 2021-12-14 | 2022-03-15 | 安徽大学 | Driver distraction fine-grained detection method based on unsupervised learning |
| CN114266934A (en) * | 2021-12-10 | 2022-04-01 | 上海应用技术大学 | Dangerous action detection method based on cloud storage data |
| CN114782931A (en) * | 2022-04-22 | 2022-07-22 | 电子科技大学 | Driving behavior classification method for improved MobileNetv2 network |
| CN115082698A (en) * | 2022-06-28 | 2022-09-20 | 华南理工大学 | Distracted driving behavior detection method based on multi-scale attention module |
| CN115186770A (en) * | 2022-09-08 | 2022-10-14 | 北京邮电大学 | Driver identity recognition method and electronic equipment |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107633220A (en) * | 2017-09-13 | 2018-01-26 | 吉林大学 | A kind of vehicle front target identification method based on convolutional neural networks |
| CN108491858A (en) * | 2018-02-11 | 2018-09-04 | 南京邮电大学 | Method for detecting fatigue driving based on convolutional neural networks and system |
| EP3379460A1 (en) * | 2017-03-21 | 2018-09-26 | Axis AB | Quality measurement weighting of image objects |
| CN108875674A (en) * | 2018-06-29 | 2018-11-23 | 东南大学 | A kind of driving behavior recognition methods based on multiple row fusion convolutional neural networks |
| CN108921030A (en) * | 2018-06-04 | 2018-11-30 | 浙江大学 | A kind of SAR automatic target recognition method of Fast Learning |
| CN109446922A (en) * | 2018-10-10 | 2019-03-08 | 中山大学 | A kind of method for detecting human face of real-time robust |
| CN109784298A (en) * | 2019-01-28 | 2019-05-21 | 南京航空航天大学 | A kind of outdoor on-fixed scene weather recognition methods based on deep learning |
| US20190180418A1 (en) * | 2017-06-05 | 2019-06-13 | Adasky, Ltd. | Scene-based nonuniformity correction using a convolutional recurrent neural network |
-
2019
- 2019-07-26 CN CN201910681906.7A patent/CN110532878B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3379460A1 (en) * | 2017-03-21 | 2018-09-26 | Axis AB | Quality measurement weighting of image objects |
| US20190180418A1 (en) * | 2017-06-05 | 2019-06-13 | Adasky, Ltd. | Scene-based nonuniformity correction using a convolutional recurrent neural network |
| CN107633220A (en) * | 2017-09-13 | 2018-01-26 | 吉林大学 | A kind of vehicle front target identification method based on convolutional neural networks |
| CN108491858A (en) * | 2018-02-11 | 2018-09-04 | 南京邮电大学 | Method for detecting fatigue driving based on convolutional neural networks and system |
| CN108921030A (en) * | 2018-06-04 | 2018-11-30 | 浙江大学 | A kind of SAR automatic target recognition method of Fast Learning |
| CN108875674A (en) * | 2018-06-29 | 2018-11-23 | 东南大学 | A kind of driving behavior recognition methods based on multiple row fusion convolutional neural networks |
| CN109446922A (en) * | 2018-10-10 | 2019-03-08 | 中山大学 | A kind of method for detecting human face of real-time robust |
| CN109784298A (en) * | 2019-01-28 | 2019-05-21 | 南京航空航天大学 | A kind of outdoor on-fixed scene weather recognition methods based on deep learning |
Non-Patent Citations (2)
| Title |
|---|
| ZHANG QIANG ET AL.: ""Vehicle color recognition using multiple-layer feature representations of lightweight convolutional neural network"", 《SIGNAL PROCESSING》 * |
| 黄良辉 等: ""基于轻量级卷积神经网络的人脸识别方法"", 《湖南工业大学学报》 * |
Cited By (38)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111062399A (en) * | 2019-12-12 | 2020-04-24 | 易诚高科（大连）科技有限公司 | Monitoring video face recognition method based on color dithering and image mixing |
| CN111062399B (en) * | 2019-12-12 | 2023-04-25 | 易诚高科（大连）科技有限公司 | Monitoring video face recognition method based on color dithering and image mixing |
| CN111209972A (en) * | 2020-01-09 | 2020-05-29 | 中国科学院计算技术研究所 | Image classification method and system based on hybrid connectivity deep convolution neural network |
| CN111553190A (en) * | 2020-03-30 | 2020-08-18 | 浙江工业大学 | Image-based driver attention detection method |
| CN111582101A (en) * | 2020-04-28 | 2020-08-25 | 中国科学院空天信息创新研究院 | Remote sensing image detection method and system |
| CN111582101B (en) * | 2020-04-28 | 2021-10-01 | 中国科学院空天信息创新研究院 | Remote sensing image target detection method and system based on lightweight distillation network |
| CN111626186A (en) * | 2020-05-25 | 2020-09-04 | 宁波大学 | Driver distraction detection method |
| CN111832446A (en) * | 2020-06-30 | 2020-10-27 | 东南大学 | Driver posture identification method based on double-view-angle video data |
| CN111860316A (en) * | 2020-07-20 | 2020-10-30 | 上海汽车集团股份有限公司 | Driving behavior recognition method and device and storage medium |
| CN111860316B (en) * | 2020-07-20 | 2024-03-19 | 上海汽车集团股份有限公司 | Driving behavior recognition method, device and storage medium |
| CN111860427A (en) * | 2020-07-30 | 2020-10-30 | 重庆邮电大学 | Driving distraction identification method based on lightweight class eight-dimensional convolutional neural network |
| CN111860427B (en) * | 2020-07-30 | 2022-07-01 | 重庆邮电大学 | Driving distraction identification method based on lightweight class eight-dimensional convolutional neural network |
| CN112053354A (en) * | 2020-09-15 | 2020-12-08 | 上海应用技术大学 | Track slab crack detection method |
| CN112053354B (en) * | 2020-09-15 | 2024-01-30 | 上海应用技术大学 | Rail plate crack detection method |
| CN112487996A (en) * | 2020-12-02 | 2021-03-12 | 重庆邮电大学 | Driving behavior recognition method based on DenseNet121 network |
| CN112487996B (en) * | 2020-12-02 | 2023-07-28 | 重庆邮电大学 | Driving behavior recognition method based on DenseNet121 network |
| CN112784723A (en) * | 2021-01-14 | 2021-05-11 | 金陵科技学院 | Road traffic safety protection model based on IFast-RCNN algorithm |
| CN112732797A (en) * | 2021-01-26 | 2021-04-30 | 武汉理工大学 | Fuel cell automobile driving behavior analysis method, device and storage medium |
| CN112907600A (en) * | 2021-03-10 | 2021-06-04 | 江苏禹空间科技有限公司 | Optimization method and system of target detection model |
| CN113128353A (en) * | 2021-03-26 | 2021-07-16 | 安徽大学 | Emotion sensing method and system for natural human-computer interaction |
| CN113128353B (en) * | 2021-03-26 | 2023-10-24 | 安徽大学 | Emotion perception method and system oriented to natural man-machine interaction |
| CN113052189A (en) * | 2021-03-30 | 2021-06-29 | 电子科技大学 | Improved MobileNet V3 feature extraction network |
| CN113052189B (en) * | 2021-03-30 | 2022-04-29 | 电子科技大学 | Improved MobileNet V3 feature extraction network |
| CN113505628A (en) * | 2021-04-02 | 2021-10-15 | 上海师范大学 | Target identification method based on lightweight neural network and application thereof |
| CN113111938A (en) * | 2021-04-09 | 2021-07-13 | 中国工程物理研究院电子工程研究所 | Terrain classification method based on digital elevation model data |
| CN113378653A (en) * | 2021-05-19 | 2021-09-10 | 海南师范大学 | Method and device for detecting in-vehicle behavior of driver |
| CN113536919A (en) * | 2021-06-10 | 2021-10-22 | 重庆邮电大学 | Signal modulation recognition algorithm based on data enhancement and convolutional neural network |
| CN113449656A (en) * | 2021-07-01 | 2021-09-28 | 淮阴工学院 | Driver state identification method based on improved convolutional neural network |
| CN113449656B (en) * | 2021-07-01 | 2022-08-02 | 淮阴工学院 | Driver state identification method based on improved convolutional neural network |
| CN113780385A (en) * | 2021-08-30 | 2021-12-10 | 武汉理工大学 | Driving risk monitoring method based on attention mechanism |
| CN114266934A (en) * | 2021-12-10 | 2022-04-01 | 上海应用技术大学 | Dangerous action detection method based on cloud storage data |
| CN114187581A (en) * | 2021-12-14 | 2022-03-15 | 安徽大学 | Driver distraction fine-grained detection method based on unsupervised learning |
| CN114187581B (en) * | 2021-12-14 | 2024-04-09 | 安徽大学 | Driver distraction fine granularity detection method based on unsupervised learning |
| CN114782931A (en) * | 2022-04-22 | 2022-07-22 | 电子科技大学 | Driving behavior classification method for improved MobileNetv2 network |
| CN114782931B (en) * | 2022-04-22 | 2023-09-29 | 电子科技大学 | Driving behavior classification method for improving mobilenet v2 network |
| CN115082698A (en) * | 2022-06-28 | 2022-09-20 | 华南理工大学 | Distracted driving behavior detection method based on multi-scale attention module |
| CN115082698B (en) * | 2022-06-28 | 2024-04-16 | 华南理工大学 | Distraction driving behavior detection method based on multi-scale attention module |
| CN115186770A (en) * | 2022-09-08 | 2022-10-14 | 北京邮电大学 | Driver identity recognition method and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110532878B (en) | 2022-11-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110532878A (en) | A kind of driving behavior recognition methods based on lightweight convolutional neural networks | |
| CN108520535B (en) | Object classification method based on depth recovery information | |
| CN110929602B (en) | Foundation cloud picture cloud identification method based on convolutional neural network | |
| US20200151497A1 (en) | Semantic segmentation with soft cross-entropy loss | |
| CN109657595B (en) | Key feature region matching face recognition method based on stacked hourglass network | |
| CN112084866A (en) | Target detection method based on improved YOLO v4 algorithm | |
| CN109886225A (en) | A kind of image gesture motion on-line checking and recognition methods based on deep learning | |
| CN108830252A (en) | A kind of convolutional neural networks human motion recognition method of amalgamation of global space-time characteristic | |
| CN112163449B (en) | Lightweight multi-branch feature cross-layer fusion image semantic segmentation method | |
| CN109101914A (en) | It is a kind of based on multiple dimensioned pedestrian detection method and device | |
| CN110046575A (en) | Based on the remote sensing images scene classification method for improving residual error network | |
| CN112801040B (en) | Lightweight unconstrained facial expression recognition method and system embedded with high-order information | |
| CN109741331A (en) | A kind of display foreground method for segmenting objects | |
| CN111401117B (en) | Neonate pain expression recognition method based on double-current convolutional neural network | |
| CN113627376B (en) | Facial expression recognition method based on multi-scale dense connection depth separable network | |
| CN110610124B (en) | Image generation method based on generation countermeasure network | |
| Guo et al. | Dense123'color enhancement dehazing network | |
| CN115457498A (en) | Urban road semantic segmentation method based on double attention and dense connection | |
| CN110991317B (en) | Crowd counting method based on multi-scale perspective sensing network | |
| CN109685713A (en) | Makeup analog control method, device, computer equipment and storage medium | |
| CN110532925A (en) | Driver Fatigue Detection based on space-time diagram convolutional network | |
| CN115082698B (en) | Distraction driving behavior detection method based on multi-scale attention module | |
| CN110378234A (en) | Convolutional neural networks thermal imagery face identification method and system based on TensorFlow building | |
| CN114882537B (en) | Finger new visual angle image generation method based on nerve radiation field | |
| CN115731597A (en) | Automatic segmentation and restoration management platform and method for mask image of face mask |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |