CN109987102A - A kind of method and apparatus of the High Precision Automatic identification driver's unsafe behaviors based on convolutional neural networks - Google Patents
A kind of method and apparatus of the High Precision Automatic identification driver's unsafe behaviors based on convolutional neural networks Download PDFInfo
- Publication number
- CN109987102A CN109987102A CN201810003621.3A CN201810003621A CN109987102A CN 109987102 A CN109987102 A CN 109987102A CN 201810003621 A CN201810003621 A CN 201810003621A CN 109987102 A CN109987102 A CN 109987102A
- Authority
- CN
- China
- Prior art keywords
- module
- convolutional neural
- neural networks
- network
- driver
- 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
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W40/00—Estimation or calculation of non-directly measurable driving parameters for road vehicle drive control systems not related to the control of a particular sub unit, e.g. by using mathematical models
- B60W40/08—Estimation or calculation of non-directly measurable driving parameters for road vehicle drive control systems not related to the control of a particular sub unit, e.g. by using mathematical models related to drivers or passengers
- B60W40/09—Driving style or behaviour
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W50/00—Details of control systems for road vehicle drive control not related to the control of a particular sub-unit, e.g. process diagnostic or vehicle driver interfaces
- B60W50/08—Interaction between the driver and the control system
- B60W50/14—Means for informing the driver, warning the driver or prompting a driver intervention
Landscapes
- Engineering & Computer Science (AREA)
- Automation & Control Theory (AREA)
- Transportation (AREA)
- Mechanical Engineering (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
Abstract
The method and apparatus of the embodiment of the invention discloses a kind of High Precision Automatic identification driver's unsafe behaviors based on convolutional neural networks, is related to image recognition, pattern-recognition and automatic field.The present invention is transported for Very Important Cargo or the unsafe acts test problems of manned bus man, proposes the solution of DriverBeCog: extracting feature to the realtime graphic of driver using capsule neural network and convolutional neural networks；Two classification are carried out respectively to a variety of behaviors parallel；Picture is passed back by monitoring device and carries out real-time monitoring, unsafe acts is alerted, while relevant information is recorded into database.The present invention uses multiple model concurrent processing；It proposes that a kind of network layer is few, parameter is few, calculation amount is small and is easy to practical convolutional neural networks model；The discrimination of unsafe acts is exceeded and has been fruitful, is conducive to practical；And solve the problems, such as that positive and negative class data volume gap is excessive using weight parameter method of adjustment.
Description
Technical field
The present invention relates to area of pattern recognition and safe driving behavior monitoring fields more particularly to a kind of automatic identification to drive
The method and apparatus of member's unsafe acts.
Background technique
By the end of the year 2016, China's vehicle guaranteeding organic quantity is up to 2.9 hundred million, wherein 1.94 hundred million, automobile, vehicle driver
3.6 hundred million people, and motorist is more than 3.1 hundred million people.The automobile of magnanimity results in a large amount of traffic accident, and these traffic accidents
In, it is much to be caused by uncivil unsafe traffic behavior, such as a hand smokes or makes a phone call, one hand manipulation automobile etc.；Together
When, due to the case where understanding of the driver to safety behavior is inadequate, and there is also a large amount of safety belt buckling frees.For Very Important Cargo transport or
The manned car of person, harm is huge caused by accident.Therefore, society for can automatic identification driver's unlawful practice it is concurrent
The system of alarm has tight demand out.
It is less for the research of driver's unlawful practice both at home and abroad and compare concentration, it is few be capable of it is really commercial,
Associated data set is difficult to obtain, and the data of analysis can substantially be divided into three aspects: direct picture, steering wheel and the more matchmakers of driver
The image of image, side near body area.The content of identification is also to make a phone call, photos and sending messages, operation sound equipment, see the behaviors such as rearward
One or more.
Traditional detection mode is to identify driving behavior by some high-cost sensors, although effect is fine,
It is with high costs, it should not promote.In recent years, some machine learning algorithms are applied to area of pattern recognition, driving behavior is known
It also not joined the machine learning methods such as Adaboost, SVM, presently preferred result mostly comes from neural network.But it is single
Neural network structure can not handle complicated actual state, accuracy rate is barely satisfactory, need one it is more complicated more advanced
Algorithm is completed with model structure.
Summary of the invention
Based on above discussion, the technical problem to be solved by the present invention is providing a kind of automatic, high precision identification driver
The method and apparatus of unsafe acts can accurately identify the unsafe acts of driver in real time, alarm in time driver,
It reminds it to carry out safety behavior and places on record.
In order to solve the above technical problems, in a first aspect, the embodiment of the invention provides a kind of identifications of automatic, high precision to drive
The method of member's unsafe acts, the method includes following four big steps:
(1) it acquires specific interior realtime graphic and is transferred to local identifying system；
(2) low-level image feature of image is extracted with specific convolutional neural networks；
(3) to a variety of unsafe acts, the high-level characteristic of image is obtained using capsule Processing with Neural Network low-level image feature parallel
And classify；
(4) obtained classification results are passed into alarm module, causes associated alarm, and place on record.
According in a first aspect, in the first possible implementation, the source of interior realtime graphic is from real-time prison
Control equipment.Real-time monitoring equipment is many kinds of, and a critical issue is how to choose a kind of equipment and the identification of rear end is mutually taken
Match.Find a following fact by practice: simple low resolution black and white camera is difficult to acquire enough behavioural informations,
And then cost is high for high-cost infrared binocular camera, it is not necessary that.DriverBeCog uses 360P in whole life cycle
And the triple channel camera of the above resolution ratio.Detection is with identification process, and camera is located at by vehicle room mirror, and acquisition is driven
The positive information for sailing position, the facial information comprising driver, arm position information, direct picture information of the entire upper part of the body etc..
The position of camera and the information of acquisition are one of features of this method.
According in a first aspect, in the second possible implementation, the number of plies of specific convolutional neural networks is no more than 10
Layer, convolution kernel size is between 3*3 to 5*5, for extracting the information of image bottom, side, point including image, shape, color
Deng.The input of this convolutional neural networks is image pixel rgb value, by neuron models in machine learning, deep learning
Correlation technique realize extraction to low-level image feature.The training of convolutional neural networks model, which uses, in DriverBeCog supervision
The mode of study.By the monitor video gathered in advance extraction key frame sample, then manually to the picture frame of sampling into
Rower note (in violation of rules and regulations or not in violation of rules and regulations), finally trains convolutional network model by gradient descent algorithm under conditions of mass data
Parameter.DriverBeCog is in video extraction key frame, using 1 frequency of every 40 frame sampling, according to experiment, in this way
Help to prevent model over-fitting.
According in a first aspect, in the third possible implementation, classifying parallel to a variety of unsafe acts.It is right
The identification of driver's unlawful practice can regard multi-tag classification problem as, i.e. there are a sample (behavior) multiple characteristics (to be
It is no to fasten the safety belt, whether smoke, whether making a phone call), processing such problems generallys use two kinds of ways: single model treatment
With multiple model concurrent processing.The method that DriverBeCog uses multiple model concurrent processing, although because single model is realized
Simply, it is equal to more classification problems (a normal class, the wrong class of multiple intersections), disadvantage is the coupling between different behaviors
It is relatively strong, from the perspective of soft project, it is unfavorable for software development.And multiple models can be used as the judgement of every kind of unlawful practice
One module is easy to Function Extension, and new module is added and is not necessarily to trained module re -training；Due to answering for disparate modules
Miscellaneous degree is different, handles same original image, and each module independence of program is high.When to need to be added new module (such as right for later development
In the identification of other unlawful practices) when, this mode has bigger advantage.
The feature obtained simultaneously using capsule network (CapsuleNet) processing convolutional neural networks, realizes classification.Capsule
Network is the neural network that Hinton is reconfigured in 2017 NIPS meetings, with series of advantages.DriverBeCog
More perfect structure is devised based on this, has used the length of attitude vectors come the reality represented by indicating by a capsule
Probability existing for body；The cosine of an angle between vector has been used to measure the consistency between them；Used length be n to
Amount, rather than has the matrix of n element to indicate a state, so its transformation matrix is with n*n parameter, and more than n
It is a.Capsule neural network has stronger expressive ability, is more suitable the processing as high-level characteristic.
According to the 4th of first aspect the kind of possible implementation, the alarm module is located locally rather than cloud.When
When driver there are unsafe acts, its record is located locally, but can timing (arrival strong point) biography in the case where there is the shape condition of network
Enter cloud database, in case data are analyzed and strengthened education to driver.
Second aspect, the embodiment of the invention provides a kind of automatic, high precision identification driver's unsafe behaviors device,
Described device four module includes:
(1) it monitoring module: is passed to video flowing in real time；
(2) identification software module；Video flowing is parsed, is classified by neural network；
(3) alarm module: alarming to driver, and records into database；
(4) feedback module: subsequent user feeds back identification situation according to the log in reality and database.
According to second aspect, in the first possible implementation, DriverBeCog is used in whole life cycle
The triple channel camera of 360P and the above resolution ratio.With identification process, camera is located at by vehicle room mirror, adopts for detection
Collect the positive information of operator seat, the facial information comprising driver, arm position information, the direct picture information of the entire upper part of the body
Deng.
According to second aspect, in the second possible implementation, using the image of triple channel as input, pass through convolution
Neural network extracts low-level image feature, then extracts high-level characteristic by capsule network to achieve the purpose that Classification and Identification.Convolution mind
It is preferable in the application effect of image recognition through this depth feedforward neural network of network.DriverBeCog training neural network
Characteristic frame of the data from camera in car acquisition video extraction, data are flowed by pretreatment, convolutional neural networks outflow, are passed through
Capsule network is exported.It is a big innovation of DriverBeCog device using convolutional neural networks+capsule network method
Point.
According to second aspect, in the third possible implementation, alarm module not only provides the warning function of sound,
And can be by logout into database, each record includes and is not limited to: time of origin, behavior, the duration, instantaneous
Image etc..When driver improves behavior, alarm module makes corresponding reaction (stop alarm).
According to second aspect, in the fourth possible implementation, on backstage (or in cloud service), feedback module
Timing demands user (driver and syndic) carries out manual evaluation to the record of alarm generation and the sampling of nonevent record.Instead
Feedback module is used to check the inaccurate place of software identification module, in the product of a new generation in software identification module
Convolutional neural networks and capsule neural network are adjusted training (fine tuning).
The third aspect, the embodiment of the invention provides a kind of automatic, high precision identification driver's unsafe behaviors device,
Including identifying system described in any possible implementation of second aspect or second aspect.
Fourth aspect, the embodiment of the invention provides a kind of precision improvement sides of automatic identification driver's unsafe behaviors
Method, which is characterized in that the driver's unsafe behaviors automatic recognition system using first aspect or first aspect is any can
Method described in the implementation of energy is identified.
Detailed description of the invention
Fig. 1 be the DriverBeCog of an embodiment of the present invention make a phone call/smoke/safety belt buckling free identification device frame
Composition.
Fig. 2 is the acquisition image schematic diagram of the DriverBeCog method of an embodiment of the present invention.
Fig. 3 is the structure chart of the DriverBeCog method convolutional neural networks of an embodiment of the present invention.
Fig. 4 is the DriverBeCog method Capsule schematic network structure of an embodiment of the present invention.
Fig. 5 is the DriverBeCog method work flow diagram of an embodiment of the present invention.
Specific embodiment
Below according to drawings and examples, specific embodiments of the present invention will be described in further detail.Implement below
Example is not intended to limit the scope of the invention for illustrating the present invention.
As shown in Fig. 1 device architecture figure, the embodiment of the invention provides a kind of automatic, high precision, identification driver is dangerous
The device of behavior, unsafe acts herein include but is not limited to such as three kinds of figure, which includes:
Monitoring module extracts live video stream, is transferred to identification module, identification module (makes a phone call to three behaviors, is not safety
Band, smoking) determined parallel, once there is a kind of behavior, then alarm module can be triggered, if three behaviors do not occur,
Then continue to identify.When alarm module is triggered, driver will receive corresponding alarm, and is apprised of and stops the dangerous row immediately
For, while the relevant informations such as unsafe acts, involved party, time of origin, representative grabgraf being placed on record.Database receives this
A little information records, and when there is network automatic synchronization to cloud.User fixes a time cycle (such as one month) to database
Middle carry out review, collect user feedback in feedback module, then re -training or continues training and promotes identification module
Performance.
As shown in Fig. 2 Image Acquisition schematic diagram, the embodiment of the invention provides a kind of automatic, high precision identification drivers not
The Image Acquisition mode of safety behavior.This mode describes optimal acquisition angles and mode in practice, being capable of substantially pole
High recognition accuracy.
As shown in Fig. 3 convolutional neural networks schematic diagram, the embodiment of the invention provides a kind of identifications of automatic, high precision to drive
The convolutional neural networks technology model of member's unsafe acts.One of network structure in method is illustrated only in this figure, wherein
Conv2D layers represent two-dimensional convolution, and Maxpooling represents maximum down-sampling layer, and Dense represents full articulamentum,
Activation is activation primitive.
As shown in figure 4, outputting and inputting for capsule structure is vector, rather than the mark that neuron is handled in neural network
Amount.A large amount of Capsule structure composition Capsule network, for handling the characteristic value of convolutional neural networks output.
It will be understood by those skilled in the art that the serial number size of each step is not in the method for various embodiments of the present invention
Mean the successive of execution sequence, the execution sequence of each step should be determined by its function and internal logic, without coping with the present invention
The implementation process of specific embodiment constitutes any restriction.
Various embodiments of the present invention further explained below:
As shown in the method work flow diagram of Fig. 5, specific implementation step are as follows:
S1: a large amount of related interior video datas are collected, in case supervised learning.
S2: key images are removed with the mode of extraction key frame, then are pre-processed with cutting, deformation, rotation, change color etc.
Mode is processed into standard picture, and is manually marked and (manually determine whether these images belong to unsafe acts).
S3: different network structure and training hyper parameter (parameter of each behavior parallelism recognition is different) are set, to S2
In data be trained.
S4: being repeated S3, compares training set and verifying collection accuracy rate, to obtain a suitable high-precision model.
S5: it is formal to start to work, live video stream, the mould for being pre-processed, and being obtained using S4 are transmitted with acquisition equipment
Type is determined.
S6: decide whether the alarm module that sets out according to judgement result, if it is determined that being unsafe acts, then reported to driver
It is alert, if safety behavior, then return to S5.
S7: alarm system is to relevant informations such as data-base recording unsafe acts, involved party, time of origin, real-time video frames.
S8: collect user feedback, then re -training or continue training promoted identification module performance.
Those of ordinary skill in the art will appreciate that realizing all or part of the process in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, the program can be stored in a computer-readable storage and be situated between
In matter, the program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein, the storage medium can be
Magnetic disk, CD, read-only memory (Read-Only Memory, ROM) or random access memory (Random Access
Memory, RAM) etc..
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
In the technical scope disclosed by the present invention, any changes or substitutions that can be easily thought of by those familiar with the art, all answers
It is included within the scope of the present invention.Therefore, protection scope of the present invention should be subject to the protection scope in claims.
Claims (11)
1. a kind of method of the High Precision Automatic identification driver's unsafe behaviors based on convolutional neural networks, which is characterized in that
The method includes the steps:
(1) it acquires specific interior realtime graphic and is transferred to local identifying system；
(2) low-level image feature of image is extracted with specific convolutional neural networks；
(3) to a variety of unsafe acts, the high-level characteristic of image is obtained using capsule Processing with Neural Network low-level image feature parallel
And classify；Imbalanced training sets problem is handled using the method for adjusting weight when to network training；
(4) obtained classification results are passed into alarm module, causes associated alarm, and place on record.
2. the method according to claim 1, wherein the Image Acquisition source of the automatic identifying method is real
When monitoring device, mutually arrange in pairs or groups to choose a kind of equipment and the identification process of rear end in miscellaneous real-time monitoring equipment,
Carried out a large amount of practices and found a following fact: simple low resolution black and white camera is difficult to acquire enough behavior letters
Breath；And then cost is high for high-cost infrared binocular camera, it is not necessary that, therefore DriverBeCog is in whole life cycle
Use 360P and the triple channel camera of the above resolution ratio；
With identification process, camera is located at by vehicle room mirror for detection, the positive information of operator seat is acquired, comprising driving
Facial information, arm position information, the direct picture information of the entire upper part of the body etc. of member；The position of camera and the information of acquisition
It is one of the feature of this method.
3. according to the method described in claim 1, it is characterized by:
The number of plies of specific convolutional neural networks is no more than 10 layers, and convolution kernel size is between 3*3 to 5*5, for extracting image bottom
The information of layer, side, point, shape, color including image etc.；The input of this convolutional neural networks is image pixel rgb value, is passed through
Correlation technique in neuron models, deep learning in machine learning realizes the extraction to low-level image feature；
The training of convolutional neural networks model is by the way of supervised learning in DriverBeCog, the prison that will be gathered in advance
Control video extraction key frame is sampled, and (in violation of rules and regulations or not in violation of rules and regulations) is then manually labeled to the picture frame of sampling, is finally existed
The parameter of convolutional network model is trained under conditions of mass data by gradient descent algorithm；
DriverBeCog is helped using 1 frequency of every 40 frame sampling according to experiment in this way in video extraction key frame
In preventing model over-fitting.
4. according to the method described in claim 1, it is characterized by:
Classify parallel to a variety of unsafe acts, because can regard multi-tag point as to the identification of driver's unlawful practice
Class problem, i.e. a sample (behavior) have multiple characteristics (whether fasten the safety belt, whether smoke, whether making a phone call), so
Processing such problems generallys use two kinds of ways: single model treatment and multiple model concurrent processing, DriverBeCog are used
The method of multiple model concurrent processing, although because single model realize it is simple, be equal to more classification problems (a normal class,
The wrong class of multiple intersections), disadvantage is that the coupling between different behaviors is stronger, from the perspective of soft project, is unfavorable for
Software development, and multiple models can be used as a module to the judgement of every kind of unlawful practice, be easy to Function Extension, and new module is added
Without to trained module re -training；
Since the complexity of disparate modules is different, same original image is handled, each module independence of program is high, when later development needs
When new module (such as the identification of other unlawful practices) is added, this mode has bigger advantage；
Training when, due to sample be it is unbalanced, that is, in the image acquired safety with uneasiness full images data volumes seriously lose
Weighing apparatus, during specific training pattern, this method, which uses, to be increased positive class (in violation of rules and regulations with uneasy universal class) and weighs in loss contribution
The way of weight parameter, solves the problems, such as its imbalanced training sets.
5. the method according to claim 1, wherein handling convolutional Neural net with capsule network (CapsuleNet)
The feature that network obtains realizes classification；Capsule network is the neural network that Hinton is reconfigured in 2017NIPS meeting, tool
There is series of advantages.DriverBeCog devises more perfect structure based on this, and the length of attitude vectors has been used to indicate
The probability by existing for the entity represented by a capsule；The cosine of an angle between vector has been used to measure between them
Consistency；Having used length is the vector of n, rather than has the matrix of n element to indicate a state, so its transformation matrix
With n*n parameter, and more than n is a.Capsule neural network has stronger expressive ability, is more suitable as high-level characteristic
Processing.
6. the method according to claim 1, wherein the alarm module is located locally rather than cloud, when driving
When the person of sailing there are unsafe acts, its record is located locally, but it is incoming periodically (to reach strong point) in the case where there is the shape condition of network
Cloud database, in case data are analyzed and strengthened education to driver.
7. a kind of device of the High Precision Automatic identification driver's unsafe behaviors based on convolutional neural networks, which is characterized in that
Described device includes:
(1) it monitoring module: is passed to video flowing in real time；
(2) identification software module；Video flowing is parsed, is classified by neural network；
(3) alarm module: alarming to driver, and records into database；
(4) feedback module: subsequent user feeds back identification situation according to the log in reality and database.
8. device according to claim 7, which is characterized in that the monitoring module:
DriverBeCog uses 360P and the triple channel camera of the above resolution ratio in whole life cycle；It detects and identified
Cheng Zhong, camera are located at by vehicle room mirror, acquire the positive information of operator seat, facial information, hand comprising driver
Arm location information, direct picture information of the entire upper part of the body etc..
9. device according to claim 7, which is characterized in that the identification software module:
Using the image of triple channel as input, low-level image feature is extracted by convolutional neural networks, is then extracted by capsule network
High-level characteristic is to achieve the purpose that Classification and Identification；This depth feedforward neural network of convolutional neural networks, in answering for image recognition
It is preferable with effect.Characteristic frame of the data from camera in car acquisition video extraction of DriverBeCog training neural network, number
It is exported according to by pretreatment inflow, convolutional neural networks outflow, by capsule network；Use convolutional neural networks+capsule net
The method of network is a big innovative point of DriverBeCog device.
10. device according to claim 7, which is characterized in that the alarm module:
Alarm module not only provides the warning function of sound, and can be by logout into database, and each record includes
And be not limited to: time of origin, behavior, duration, instantaneous picture etc., when driver improves behavior, alarm module is made accordingly
Reaction (stop alarm).
11. device according to claim 7, which is characterized in that the feedback module:
On backstage (or in cloud service), note of the feedback module timing demands user (driver and syndic) to alarm generation
Record and the sampling of nonevent record carry out manual evaluation；Feedback module is used to check the inaccurate place of software identification module,
With a new generation product in in software identification module convolutional neural networks and capsule neural network be adjusted training
(fine tuning)。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810003621.3A CN109987102A (en) | 2018-01-03 | 2018-01-03 | A kind of method and apparatus of the High Precision Automatic identification driver's unsafe behaviors based on convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810003621.3A CN109987102A (en) | 2018-01-03 | 2018-01-03 | A kind of method and apparatus of the High Precision Automatic identification driver's unsafe behaviors based on convolutional neural networks |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109987102A true CN109987102A (en) | 2019-07-09 |
Family
ID=67128906
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810003621.3A Pending CN109987102A (en) | 2018-01-03 | 2018-01-03 | A kind of method and apparatus of the High Precision Automatic identification driver's unsafe behaviors based on convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109987102A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110422178A (en) * | 2019-08-14 | 2019-11-08 | 李斯特技术中心（上海）有限公司 | Driving information acquisition methods, device and Information Acquisition System |
| CN110472549A (en) * | 2019-08-09 | 2019-11-19 | 紫荆智维智能科技研究院(重庆)有限公司 | Based on the vehicle-mounted GPU train driver movement real-time identifying system accelerated and method |
| CN111368789A (en) * | 2020-03-18 | 2020-07-03 | 腾讯科技（深圳）有限公司 | Image recognition method and device, computer equipment and storage medium |
| CN111461063A (en) * | 2020-04-24 | 2020-07-28 | 武汉大学 | Behavior identification method based on graph convolution and capsule neural network |
| CN111563494A (en) * | 2020-07-16 | 2020-08-21 | 平安国际智慧城市科技股份有限公司 | Behavior identification method and device based on target detection and computer equipment |
| CN111874000A (en) * | 2020-07-22 | 2020-11-03 | 重庆长安新能源汽车科技有限公司 | Method for judging safety level of driving behavior and storage medium |
| CN112528952A (en) * | 2020-12-25 | 2021-03-19 | 合肥诚记信息科技有限公司 | Working state intelligent recognition system for electric power business hall personnel |
| CN115071725A (en) * | 2022-08-02 | 2022-09-20 | 广东车卫士信息科技有限公司 | Driving behavior analysis method and device |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102263937A (en) * | 2011-07-26 | 2011-11-30 | 华南理工大学 | Driver's driving behavior monitoring device and monitoring method based on video detection |
| CN106096607A (en) * | 2016-06-12 | 2016-11-09 | 湘潭大学 | A kind of licence plate recognition method |
| CN106485224A (en) * | 2016-10-13 | 2017-03-08 | 北京智芯原动科技有限公司 | A kind of seatbelt wearing recognition methodss and device |
-
2018
- 2018-01-03 CN CN201810003621.3A patent/CN109987102A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102263937A (en) * | 2011-07-26 | 2011-11-30 | 华南理工大学 | Driver's driving behavior monitoring device and monitoring method based on video detection |
| CN106096607A (en) * | 2016-06-12 | 2016-11-09 | 湘潭大学 | A kind of licence plate recognition method |
| CN106485224A (en) * | 2016-10-13 | 2017-03-08 | 北京智芯原动科技有限公司 | A kind of seatbelt wearing recognition methodss and device |
Non-Patent Citations (2)
| Title |
|---|
| SARA SABOUR等: "Dynamic Routing Between Capsules", 《31ST CONFERENCE ON NEURAL INFORMATION PROCESSING SYSTEM》 * |
| 陈倬: "基于随机Dropout卷积神经网络的人体行为识别", 《科技资讯》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110472549A (en) * | 2019-08-09 | 2019-11-19 | 紫荆智维智能科技研究院(重庆)有限公司 | Based on the vehicle-mounted GPU train driver movement real-time identifying system accelerated and method |
| CN110422178A (en) * | 2019-08-14 | 2019-11-08 | 李斯特技术中心（上海）有限公司 | Driving information acquisition methods, device and Information Acquisition System |
| CN111368789A (en) * | 2020-03-18 | 2020-07-03 | 腾讯科技（深圳）有限公司 | Image recognition method and device, computer equipment and storage medium |
| CN111368789B (en) * | 2020-03-18 | 2023-05-26 | 腾讯科技（深圳）有限公司 | Image recognition method, device, computer equipment and storage medium |
| CN111461063A (en) * | 2020-04-24 | 2020-07-28 | 武汉大学 | Behavior identification method based on graph convolution and capsule neural network |
| CN111461063B (en) * | 2020-04-24 | 2022-05-17 | 武汉大学 | Behavior identification method based on graph convolution and capsule neural network |
| CN111563494A (en) * | 2020-07-16 | 2020-08-21 | 平安国际智慧城市科技股份有限公司 | Behavior identification method and device based on target detection and computer equipment |
| CN111874000A (en) * | 2020-07-22 | 2020-11-03 | 重庆长安新能源汽车科技有限公司 | Method for judging safety level of driving behavior and storage medium |
| CN112528952A (en) * | 2020-12-25 | 2021-03-19 | 合肥诚记信息科技有限公司 | Working state intelligent recognition system for electric power business hall personnel |
| CN112528952B (en) * | 2020-12-25 | 2022-02-11 | 合肥诚记信息科技有限公司 | Working state intelligent recognition system for electric power business hall personnel |
| CN115071725A (en) * | 2022-08-02 | 2022-09-20 | 广东车卫士信息科技有限公司 | Driving behavior analysis method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109987102A (en) | A kind of method and apparatus of the High Precision Automatic identification driver's unsafe behaviors based on convolutional neural networks | |
| US11783601B2 (en) | Driver fatigue detection method and system based on combining a pseudo-3D convolutional neural network and an attention mechanism | |
| CN104637246B (en) | Driver multi-behavior early warning system and danger evaluation method | |
| CN105488453B (en) | A kind of driver based on image procossing does not fasten the safety belt detection recognition method | |
| CN103824420B (en) | Fatigue driving identification system based on heart rate variability non-contact measurement | |
| CN103049740B (en) | Fatigue state detection method based on video image and device | |
| CN105769120A (en) | Fatigue driving detection method and device | |
| CN105956548A (en) | Driver fatigue state detection method and device | |
| Hicham et al. | Vehicle type classification using convolutional neural network | |
| CN101950355A (en) | Method for detecting fatigue state of driver based on digital video | |
| CN109948616A (en) | Image detecting method, device, electronic equipment and computer readable storage medium | |
| Ragab et al. | A visual-based driver distraction recognition and detection using random forest | |
| CN105354985A (en) | Fatigue driving monitoring device and method | |
| CN110472511A (en) | A kind of driver status monitoring device based on computer vision | |
| CN109977771A (en) | Verification method, device, equipment and the computer readable storage medium of driver identification | |
| Yan et al. | Recognizing driver inattention by convolutional neural networks | |
| CN116612335B (en) | Few-sample fine-granularity image classification method based on contrast learning | |
| Xing et al. | Application of face recognition based on CNN in fatigue driving detection | |
| CN112614102A (en) | Vehicle detection method, terminal and computer readable storage medium thereof | |
| CN112052829B (en) | Pilot behavior monitoring method based on deep learning | |
| CN113642646A (en) | Image threat article classification and positioning method based on multiple attention and semantics | |
| Zheng et al. | Rapid screening of children with autism spectrum disorders through face image classification | |
| CN116434203A (en) | Anger driving state identification method considering language factors of driver | |
| Umut et al. | Detection of driver sleepiness and warning the driver in real-time using image processing and machine learning techniques | |
| CN110516538A (en) | The double violation assessment method of leaving the post in prison based on deep learning target detection |
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: 20190709 |