CN109858553A - Monitoring model update method, updating device and the storage medium of driving condition - Google Patents
Monitoring model update method, updating device and the storage medium of driving condition Download PDFInfo
- Publication number
- CN109858553A CN109858553A CN201910103594.1A CN201910103594A CN109858553A CN 109858553 A CN109858553 A CN 109858553A CN 201910103594 A CN201910103594 A CN 201910103594A CN 109858553 A CN109858553 A CN 109858553A
- Authority
- CN
- China
- Prior art keywords
- monitoring model
- node
- convolutional neural
- neural networks
- probability value
- 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
- 238000012544 monitoring process Methods 0.000 title claims abstract description 230
- 238000000034 method Methods 0.000 title claims abstract description 81
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 157
- 238000010606 normalization Methods 0.000 claims abstract description 48
- 238000012545 processing Methods 0.000 claims abstract description 20
- 238000012549 training Methods 0.000 claims description 18
- 238000013528 artificial neural network Methods 0.000 claims description 13
- 230000007423 decrease Effects 0.000 claims description 8
- 230000001537 neural effect Effects 0.000 claims description 8
- 230000006870 function Effects 0.000 claims description 7
- 241001269238 Data Species 0.000 claims description 5
- 230000000750 progressive effect Effects 0.000 claims description 4
- 230000002159 abnormal effect Effects 0.000 abstract description 25
- 238000010586 diagram Methods 0.000 description 18
- 230000005856 abnormality Effects 0.000 description 9
- 230000006399 behavior Effects 0.000 description 7
- 206010048232 Yawning Diseases 0.000 description 6
- 238000004891 communication Methods 0.000 description 6
- 238000004364 calculation method Methods 0.000 description 5
- 244000269722 Thea sinensis Species 0.000 description 4
- 239000011159 matrix material Substances 0.000 description 4
- 230000036651 mood Effects 0.000 description 4
- 230000001133 acceleration Effects 0.000 description 3
- 240000002853 Nelumbo nucifera Species 0.000 description 2
- 235000006508 Nelumbo nucifera Nutrition 0.000 description 2
- 235000006510 Nelumbo pentapetala Nutrition 0.000 description 2
- 206010039203 Road traffic accident Diseases 0.000 description 2
- 230000019771 cognition Effects 0.000 description 2
- 238000012937 correction Methods 0.000 description 2
- 239000006185 dispersion Substances 0.000 description 2
- 230000008921 facial expression Effects 0.000 description 2
- 230000005484 gravity Effects 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- 241001282135 Poromitra oscitans Species 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000005314 correlation function Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 239000011521 glass Substances 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000004904 shortening Methods 0.000 description 1
- 238000010181 skin prick test Methods 0.000 description 1
Abstract
The invention discloses a kind of monitoring model update methods of driving condition, this method comprises: reading the newly-increased sample data of the driving condition of driving condition collector acquisition and the first category of the affiliated driving condition of newly-increased sample data;The critical data in newly-increased sample data is obtained, processing is carried out to the critical data and forms normalization characteristic vector;The normalization characteristic vector sum first category is obtained into newly-increased sample data in the output probability value of the node of initial tree convolutional neural networks monitoring model as the input of initial tree convolutional neural networks monitoring model;The first category is added in the node of initial tree convolutional neural networks monitoring model according to the output probability value, initial tree convolutional neural networks monitoring model is updated.The invention also discloses updating devices and storage medium.The present invention can be realized through incremental data dynamic update abnormal driving condition supervision model, and can guarantee the low redundancy of model in model dynamic updating process, improve abnormal driving state recognition precision and model robustness.
Description
Technical field
The present invention relates to the monitoring model update method of field of communication technology more particularly to driving condition, updating device and
Storage medium.
Background technique
Traffic accident caused by fatigue driving in recent years, dispersion attention, changeable in mood driving, sudden illness etc. is also gradually
Increase.In order to inherently reduce the generation of traffic accident situation, driving condition supervision model is established, driver can be monitored
Driving behavior and give alarm to abnormal driving behavior, it is negative to the driving for improving the driving ability of driver and reduce driver
Lotus, and the relationship between driver and vehicle and traffic environment is coordinated, it is of great significance.
But traditional driving condition supervision model only passes through acquisition various kinds of sensors information, to determine whether driver is in
Fatigue driving state, identification abnormal driving state is single, and the factor for influencing to drive shape often has a variety of aspects, such as recognizes negative
Lotus is overweight, external interference, vehicle running state etc..Only judge that the single abnormal driving status monitoring of fatigue driving can not
Meet the requirement of high-precision, strong robust, and since monitoring model is to utilize existing nominal data train classification models, and utilize
The disaggregated model carries out identification classification to driving condition.When this class model models the driving condition of user, often to user
Existing driving condition modeling, for reflecting its current driving mode.Over time, have very much can for the driving condition of user
It can change, when the new abnormal driving behavior of one kind occurs in user, monitoring model can not correctly know new behavior
Do not classify.When there is new abnormal driving state, the mode of re -training can only be taken, in this processing mode, is needed
Exponentially type increases with the increase of sample size training time, and model redundancy is higher, is unable to satisfy wanting for incremental learning
It asks.
Summary of the invention
It is a primary object of the present invention to propose monitoring model update method, updating device and the storage of a kind of driving condition
Medium, it is intended to realize through incremental data dynamic update abnormal driving condition supervision model, and can be updated in model dynamic
The low redundancy for guaranteeing model in journey, improves abnormal driving state recognition precision and model robustness.
To achieve the above object, the present invention provides a kind of monitoring model update method of driving condition, the method includes
Following steps:
It is driven belonging to the newly-increased sample data and newly-increased sample data for reading the driving condition of driving condition collector acquisition
Sail the first category of state;
Obtain the critical data in newly-increased sample data, to the critical data carry out processing formed normalization characteristic to
Amount;
By the normalization characteristic vector sum first category, as the input of initial tree convolutional neural networks monitoring model,
Newly-increased sample data is obtained in the output probability value of the node of initial tree convolutional neural networks monitoring model;
The first category is added to the node of initial tree convolutional neural networks monitoring model according to the output probability value
In, initial tree convolutional neural networks monitoring model is updated.
Preferably, the critical data obtained in newly-increased sample data carries out processing formation to the critical data and returns
Before the step of one change feature vector, the method also includes:
The newly-increased sample data is pre-processed.
Preferably, described by the normalization characteristic vector sum first category, it is monitored as initial tree convolutional neural networks
The input of model obtains newly-increased sample data in the step of the output probability value of the node of initial tree convolutional neural networks monitoring model
Suddenly include:
By the normalization characteristic vector sum first category, the first order of initial tree convolutional neural networks monitoring model is inputted
Branch node exports newly-increased sample data in the sample probability value of first order branch node;
According to newly-increased sample data in the sample probability value of first order branch node, newly-increased sample data is calculated in initial tree
The output probability value of the first order branch node of convolutional neural networks monitoring model;
The node that the first category is added to initial tree convolutional neural networks monitoring model according to output probability value
In, the step of being updated to initial tree convolutional neural networks monitoring model includes:
According to the output probability value of first order branch node, initial tree convolutional neural networks prison is added in the first category
It surveys in the node of model, initial tree convolutional neural networks monitoring model is updated.
Preferably, the basis increases sample data newly in the sample probability value of first order branch node, calculates newly-increased sample
Data include: in the step of output probability value of the first order branch node of initial tree convolutional neural networks monitoring model
Take the average value of the sample probability value of more parts of newly-increased sample datas of the same first order branch node as this
The first node probability value of level-one branch node;
The processing of softmax function is carried out to the first node probability value, obtains newly-increased sample data in initial tree convolution
The output probability value of each first order branch node of neural network monitoring model.
Preferably, initial tree volume is added in the first category by the output probability value according to first order branch node
In the node of product neural network monitoring model, the step of being updated to initial tree convolutional neural networks monitoring model, includes:
When the output probability value of each first order branch node is respectively less than preset threshold, using the first category as new
First order branch node is added in initial tree convolutional neural networks monitoring model;
In the output probability value of each first order branch node, there is the output probability value of multiple first order branch nodes to be greater than
When preset threshold, new first order branch node is generated in initial convolutional neural networks monitoring model, by the first category
It is added in the child node of the new first order branch node generated;
Initial number convolutional neural networks monitoring model is updated.
Preferably, initial tree volume is added in the first category by the output probability value according to first order branch node
In the node of product neural network monitoring model, the step of being updated to initial tree convolutional neural networks monitoring model, includes:
In the output probability value of each first order branch node, the output probability value of only one first order branch node is big
When preset threshold, the output probability value of the second level branch node in the first order branch node is calculated;
According to the output probability value of second level branch node, initial tree convolutional neural networks prison is added in the first category
It surveys in the progressive child node of second level branch node described in model or second level branch node, to initial tree convolutional neural networks
Monitoring model is updated.
Preferably, by the normalization characteristic vector sum first category, initial tree convolutional neural networks monitoring model is inputted
First order branch node, exporting newly-increased sample data in the step of sample probability value of first order branch node includes:
By the normalization characteristic vector sum first category, each the first of initial tree convolutional neural networks monitoring model is inputted
Grade branch node;
Obtain the weighted value of each first order branch node of initial tree convolutional neural networks model;
Newly-increased sample data is exported each according to the weighted value of each first order branch node of the normalization characteristic vector sum
The sample probability value of first order branch node.
Preferably, described that the first category is added by the monitoring of initial tree convolutional neural networks according to the output probability value
In the node of model, the step of being updated to initial tree convolutional neural networks monitoring model, includes:
The first category is added to the node of initial tree convolutional neural networks monitoring model according to the output probability value
In;
The training of gradient decline is carried out to the node, and updates the node weight of initial convolutional neural networks monitoring model
Weight completes the update to initial tree convolutional neural networks monitoring model.
In addition, to achieve the above object, the present invention also provides a kind of updating device, the updating device include: memory,
Processor and the monitoring model more new procedures for being stored in the driving condition that can be run on the memory and on the processor,
The monitoring model of the driving condition more new procedures realize the monitoring of driving condition as described above when being executed by the processor
The step of model update method.
In addition, to achieve the above object, the present invention also provides a kind of storage medium, driving is stored on the storage medium
The monitoring model of state more new procedures, the monitoring model of the driving condition more new procedures realize institute as above when being executed by processor
The step of monitoring model update method for the driving condition stated.
The present invention reads the newly-increased sample data and newly-increased sample data of the driving condition of driving condition collector acquisition
The first category of affiliated driving condition;The critical data in newly-increased sample data is obtained, processing shape is carried out to the critical data
At normalization characteristic vector;By the normalization characteristic vector sum first category, mould is monitored as initial tree convolutional neural networks
The input of type obtains newly-increased sample data in the output probability value of the node of initial tree convolutional neural networks monitoring model;According to
The output probability value first category is added in the node of initial tree convolutional neural networks monitoring model, rolls up to initial tree
Product neural network monitoring model is updated.
By the above-mentioned means, the present invention can be by handling newly-increased sample data, the normalizing that will be formed after processing
The first category for changing feature vector and the affiliated driving condition of newly-increased sample data inputs initial tree convolutional neural networks monitoring model
In, the output probability value of node is obtained, the classification of affiliated driving condition is added by initial monitor model according to output probability value
In node, initial tree convolutional neural networks monitoring model is updated, realizes and is supervised by incremental data dynamic update abnormal driving condition
Model is surveyed, and can guarantee the low redundancy of model in model dynamic updating process, improves abnormal driving state recognition precision
With model robustness, effectively solve the problems, such as that batch learning training time index increases, when greatly shortening the training of monitoring model
Between
Detailed description of the invention
Fig. 1 is the terminal structure schematic diagram for the hardware running environment that the embodiment of the present invention is related to;
Fig. 2 is the flow diagram of the monitoring model update method first embodiment of driving condition of the present invention;
Fig. 3 is the flow diagram of the monitoring model update method second embodiment of driving condition of the present invention;
Fig. 4 is the flow diagram of the monitoring model update method 3rd embodiment of driving condition of the present invention;
Fig. 5 is the flow diagram of the monitoring model update method fourth embodiment of driving condition of the present invention;
Fig. 6 is the flow diagram of the 5th embodiment of monitoring model update method of driving condition of the present invention;
Fig. 7 is the flow diagram of the monitoring model update method sixth embodiment of driving condition of the present invention;
Fig. 8 is the flow diagram of the 7th embodiment of monitoring model update method of driving condition of the present invention;
Fig. 9 is the flow diagram of the 8th embodiment of monitoring model update method of driving condition of the present invention.
The embodiments will be further described with reference to the accompanying drawings for the realization, the function and the advantages of the object of the present invention.
Specific embodiment
It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not intended to limit the present invention.
The primary solutions of the embodiment of the present invention are: reading the newly-increased sample of the driving condition of driving condition collector acquisition
The first category of notebook data and the affiliated driving condition of newly-increased sample data;The critical data in newly-increased sample data is obtained, it is right
The critical data carries out processing and forms normalization characteristic vector;By the normalization characteristic vector sum first category, as first
The input of beginning tree convolutional neural networks monitoring model obtains newly-increased sample data in initial tree convolutional neural networks monitoring model
The output probability value of node;Initial tree convolutional neural networks are added in the first category according to the output probability value and monitor mould
In the node of type, initial tree convolutional neural networks monitoring model is updated.
Increasing is fixed and are unable to satisfy to the single abnormality of the existing identification of existing driving condition supervision model, identification model
The problem of requirement that amount study updates.
The present invention realizes through incremental data dynamic update abnormal driving condition supervision model, and can model dynamic more
The low redundancy for guaranteeing model during new, improves abnormal driving state recognition precision and model robustness.
As shown in Figure 1, Fig. 1 is the terminal structure schematic diagram for the hardware running environment that the embodiment of the present invention is related to.
The terminal of that embodiment of the invention can be PC, be also possible to smart phone, tablet computer, MP4 (Moving Picture
Experts Group Audio Layer IV, dynamic image expert's compression standard audio level 3) player, portable computer
Etc. packaged type terminal device having a display function.
As shown in Figure 1, the terminal may include: processor 1001, such as CPU, network interface 1004, user interface
1003, memory 1005, communication bus 1002.Wherein, communication bus 1002 is for realizing the connection communication between these components.
User interface 1003 may include display screen (Display), input unit such as keyboard (Keyboard), optional user interface
1003 can also include standard wireline interface and wireless interface.Network interface 1004 optionally may include that the wired of standard connects
Mouth, wireless interface (such as WI-FI interface).Memory 1005 can be high speed RAM memory, be also possible to stable memory
(non-volatile memory), such as magnetic disk storage.Memory 1005 optionally can also be independently of aforementioned processor
1001 storage device.
Preferably, terminal can also include camera, RF (Radio Frequency, radio frequency) circuit, sensor, audio
Circuit, WiFi module etc..Wherein, sensor such as optical sensor, motion sensor and other sensors.Specifically, light
Sensor may include ambient light sensor and proximity sensor, wherein ambient light sensor can according to the light and shade of ambient light come
The brightness of display screen is adjusted, proximity sensor can close display screen and/or backlight when mobile terminal is moved in one's ear.As
One kind of motion sensor, gravity accelerometer can detect the size of (generally three axis) acceleration in all directions, quiet
Size and the direction that can detect that gravity when only, the application that can be used to identify mobile terminal posture are (such as horizontal/vertical screen switching, related
Game, magnetometer pose calibrating), Vibration identification correlation function (such as pedometer, tap) etc.;Certainly, mobile terminal can also match
The other sensors such as gyroscope, barometer, hygrometer, thermometer, infrared sensor are set, details are not described herein.
It will be understood by those skilled in the art that the restriction of the not structure paired terminal of terminal structure shown in Fig. 1, can wrap
It includes than illustrating more or fewer components, perhaps combines certain components or different component layouts.
As shown in Figure 1, as may include that operating system, network are logical in a kind of memory 1005 of computer storage medium
Believe the monitoring model more new procedures of module, Subscriber Interface Module SIM and driving condition.
In terminal shown in Fig. 1, network interface 1004 is mainly used for connecting background server, carries out with background server
Data communication;User interface 1003 is mainly used for connecting client (user terminal), carries out data communication with client;And processor
1001 can be used for calling the monitoring model more new procedures of the driving condition stored in memory 1005, and execute following operation:
It is driven belonging to the newly-increased sample data and newly-increased sample data for reading the driving condition of driving condition collector acquisition
Sail the first category of state;
Obtain the critical data in newly-increased sample data, to the critical data carry out processing formed normalization characteristic to
Amount;
By the normalization characteristic vector sum first category, as the input of initial tree convolutional neural networks monitoring model,
Newly-increased sample data is obtained in the output probability value of the node of initial tree convolutional neural networks monitoring model;
The first category is added to the node of initial tree convolutional neural networks monitoring model according to the output probability value
In, initial tree convolutional neural networks monitoring model is updated.
Further, processor 1001 can call the monitoring model of the driving condition stored in memory 1005 to update journey
Sequence also executes following operation: the critical data obtained in newly-increased sample data carries out processing to the critical data and is formed
Before the step of normalization characteristic vector, the method also includes:
The newly-increased sample data is pre-processed.
Further, processor 1001 can call the monitoring model of the driving condition stored in memory 1005 to update journey
Sequence also executes following operation: it is described by the normalization characteristic vector sum first category, it is supervised as initial tree convolutional neural networks
The input of model is surveyed, the output probability value for increasing sample data newly in the node of initial tree convolutional neural networks monitoring model is obtained
Step includes:
By the normalization characteristic vector sum first category, the first order of initial tree convolutional neural networks monitoring model is inputted
Branch node exports newly-increased sample data in the sample probability value of first order branch node;
According to newly-increased sample data in the sample probability value of first order branch node, newly-increased sample data is calculated in initial tree
The output probability value of the first order branch node of convolutional neural networks monitoring model;
The node that the first category is added to initial tree convolutional neural networks monitoring model according to output probability value
In, the step of being updated to initial tree convolutional neural networks monitoring model includes:
According to the output probability value of first order branch node, initial tree convolutional neural networks prison is added in the first category
It surveys in the node of model, initial tree convolutional neural networks monitoring model is updated.
Further, processor 1001 can call the monitoring model of the driving condition stored in memory 1005 to update journey
Sequence, also execute following operation: the basis increases sample data newly in the sample probability value of first order branch node, calculates newly-increased sample
Notebook data includes: in the step of output probability value of the first order branch node of initial tree convolutional neural networks monitoring model
Take the average value of the sample probability value of more parts of newly-increased sample datas of the same first order branch node as this
The first node probability value of level-one branch node;
The processing of softmax function is carried out to the first node probability value, obtains newly-increased sample data in initial tree convolution
The output probability value of each first order branch node of neural network monitoring model.
Further, processor 1001 can call the monitoring model of the driving condition stored in memory 1005 to update journey
Sequence, also executes following operation: the output probability value according to first order branch node, and initial tree is added in the first category
In the node of convolutional neural networks monitoring model, the step of being updated to initial tree convolutional neural networks monitoring model, includes:
When the output probability value of each first order branch node is respectively less than preset threshold, using the first category as new
First order branch node is added in initial tree convolutional neural networks monitoring model;
In the output probability value of each first order branch node, there is the output probability value of multiple first order branch nodes to be greater than
When preset threshold, new first order branch node is generated in initial convolutional neural networks monitoring model, by the first category
It is added in the child node of the new first order branch node generated;
Initial number convolutional neural networks monitoring model is updated.
Further, processor 1001 can call the monitoring model of the driving condition stored in memory 1005 to update journey
Sequence, also executes following operation: the output probability value according to first order branch node, and initial tree is added in the first category
In the node of convolutional neural networks monitoring model, the step of being updated to initial tree convolutional neural networks monitoring model, includes:
In the output probability value of each first order branch node, the output probability value of only one first order branch node is big
When preset threshold, the output probability value of the second level branch node in the first order branch node is calculated;
According to the output probability value of second level branch node, initial tree convolutional neural networks prison is added in the first category
It surveys in the progressive child node of second level branch node described in model or second level branch node, to initial tree convolutional neural networks
Monitoring model is updated.
Further, processor 1001 can call the monitoring model of the driving condition stored in memory 1005 to update journey
Sequence also executes following operation: by the normalization characteristic vector sum first category, inputting initial tree convolutional neural networks and monitors mould
The first order branch node of type, exporting newly-increased sample data in the step of sample probability value of first order branch node includes:
By the normalization characteristic vector sum first category, each the first of initial tree convolutional neural networks monitoring model is inputted
Grade branch node;
Obtain the weighted value of each first order branch node of initial tree convolutional neural networks model;
Newly-increased sample data is exported each according to the weighted value of each first order branch node of the normalization characteristic vector sum
The sample probability value of first order branch node.
Further, processor 1001 can call the monitoring model of the driving condition stored in memory 1005 to update journey
Sequence also executes following operation: described that initial tree convolutional neural networks are added in the first category according to the output probability value
In the node of monitoring model, the step of being updated to initial tree convolutional neural networks monitoring model, includes:
The first category is added to the node of initial tree convolutional neural networks monitoring model according to the output probability value
In;
The training of gradient decline is carried out to the node, and updates the node weight of initial convolutional neural networks monitoring model
Weight completes the update to initial tree convolutional neural networks monitoring model.
Based on above-mentioned hardware configuration, embodiment of the present invention method is proposed.
It is the flow diagram of the monitoring model update method first embodiment of driving condition of the present invention referring to Fig. 2, Fig. 2,
The monitoring model update method of the driving condition includes:
Step S10, the newly-increased sample data of the driving condition of reading driving condition collector acquisition and newly-increased sample number
According to the first category of affiliated driving condition;
The present invention is applied to the monitoring system of driving condition, and the monitoring system of driving condition includes initial tree convolutional Neural net
The building module and monitoring model dynamic update module of network monitoring model, when system acquisition is not belonging to predefined driving condition classification
First category newly-increased sample data when, monitoring model dynamic update module read driving condition collector collect driving shape
The newly-increased sample data of state and the classification of the affiliated driving condition of newly-increased sample data.
Step S20 obtains the critical data in newly-increased sample data, carries out processing to the critical data and forms normalization
Feature vector;
In embodiment, the critical data in newly-increased sample data is obtained, such as: driving condition collector includes camera,
Camera takes in video and increases sample data newly, critical data be the driver in video image face and hand etc. with it is affiliated
The relevant data of driving condition classification can carry out feature coding, Chi Hua, normalization etc. to the critical data of newly-increased sample data
Reason forms normalization characteristic vector, and such as newly-increased sample data is that video increases sample data newly, video can be increased newly sample data into
The processing such as row feature coding, pond and normalization, form the normalization characteristic vector of description video, that is, are formed and describe newly-increased sample
Newly-increased sample data normalization characteristic vector, it is assumed that blink is increased into initial monitor model, the blink of reading is new
Increasing sample data, the critical data for increasing sample data newly is the eye and facial expression image data (i.e. RGB data) of driver,
Eye and facial expression image data are handled, the normalization characteristic vector of description video is formed.
Step S30, by the normalization characteristic vector sum first category, as initial tree convolutional neural networks monitoring model
Input, obtain newly-increased sample data in the output probability value of the node of initial tree convolutional neural networks monitoring model;
Initial tree convolutional neural networks monitoring model is that Tree-CNN monitoring model is driven for abnormal driving status monitoring
The state for the person of sailing can be there are four basic dimensions, the reality of load, external interference and vehicle on the load respectively recognized, body
Border travel situations, the load of cognition, such as driver's spirit is judged by the heart rate of driver, skin pricktest, myoelectricity physiological data
The case where equal loads, the load on body, as by camera acquisition glasses direction of gaze, yawn, eyes blink
Situation judges the body burdens situation such as driver drowsy, fatigue.External interference, such as shooting driver by camera is
It is phoning with mobile telephone, or is having tea and the external interferences situations such as the interference of passenger when driving.The actual travel situation of vehicle,
Vehicle actual travel situation can be judged by vehicle traveling information that vehicle-mounted CAN bus obtains.
The building module of initial tree convolutional neural networks monitoring model (Tree-CNN monitoring model), first by abnormal data point
At several dimensions, such as 4 above-mentioned dimensions, then each dimension is successively divided again, constantly opens branch just as tree
Leaf is dissipated, the classification that final leaf node obtains is exactly the final classification to be identified, Tree-CNN monitoring model includes root node, divides
Zhi Jiedian, leaf node, leaf node are the finish node in Tree-CNN monitoring model, abnormal driving Tree-CNN monitoring model
In, abnormal driving can be divided into 4 dimensions, and 4 dimensions continue to divide, such as: the external interference dimension of 4 dimensions continues to be divided into
Phone with mobile telephone, have tea, passenger interfere 3 classifications, phone with mobile telephone, have tea, passenger interference be final node, do not continue to divide, then
Classification, classification of having tea, the passenger's interference classification of phoning with mobile telephone are that leaf node is divided into and multiplies if passenger interferes classification to can also continue to divide
Visitor beats driver, passenger robs two classifications of steering wheel, and passenger beats driver's node, passenger robs steering wheel node as finish node,
It does not continue to divide, then it is leaf node that passenger, which beats driver's node, and passenger's interfering nodes are the root section that passenger beats driver's node
Point, passenger beat the branch node that driver's node is passenger's interference.
The initial monitor model building method of the monitoring system of driving condition includes: for example, driving condition collector is vehicle
When carrying camera,
1) the abnormal driving status data and its classification of vehicle-mounted camera acquisition are read;
2) crucial pretreatment, including illumination correction method, Connected component Fast Labeling, direction are carried out to driver's video
Projection etc.;
3) sequence of operations such as feature coding, pond and normalization are passed through to key messages such as driver face and hands,
Ultimately form the normalization characteristic vector of description video;
4) combine the normalization characteristic vector of extraction with abnormal class building Tree-CNN monitoring model.
When driving condition collector further includes acceleration gyro sensor, the acceleration for the new category demarcated is read
Gyroscope sensor data equally carries out 2) handling with 3) item, and the comprehensive data with vehicle-mounted camera acquisition, which construct, is based on convolution
The Tree-CNN monitoring model of neural network, i.e. building form initial tree convolutional neural networks monitoring model.
Tree-CNN monitoring model has used for reference layer classifier, tree convolutional neural networks be made of node and data structure in
Tree as, each node has the ID of oneself, father (Parent) and child (Children), and (Net handles image to net
Convolutional neural networks), (" Labels Transform ", label corresponding to each node come root node and minor matters point to LT
It says, can be a kind of division to final classification classification, be exactly final class categories for leaf node.), wherein most
Top is the root node of tree.For a sample, it can be sent to root node network first, classification is gone to obtain " super-
Then classes " send sample according to the classification that root classifier obtains then according to " super-classes " recognized
Enter corresponding node to go further to classify, obtains the classification of one more " specific ", recurrence is successively carried out, until sorting out me
Final desired classification.For the fine grit classification of abnormality, such as there is following a pile label: fatigue driving, attention
Dispersion, sudden illness, changeable in mood driving.If identifying changeable in mood driving, we may undergo such process, first exist
The abnormality of cognition level is found in this pile, then finds changeable in mood driving again.
The first category that the normalization characteristic vector sum is increased newly to the affiliated driving condition of sample data is rolled up as initial tree
The input of product neural network monitoring model obtains the node for increasing sample data newly in initial tree convolutional neural networks monitoring model
Output probability value, such as: need it is newly-increased yawn, both abnormal driving behaviors of closing one's eyes, these two types are collected in advance more
A image increases sample data newly and obtains the matrix (such as 244*244*3) of feature vector by pretreatment, is then input to initial
In the first order branch node of the network of Tree-CNN monitoring model, every layer multiplied by initial model training when obtained weight, most
The sample probability value O (k, m, i) of first order branch node is obtained eventually, and O (k, m, i) indicates that i-th part of newly-increased sample data belongs to m
The sample probability value of k-th of node when class, wherein [1, K] k ∈, m ∈ [1, M], i ∈ [1, I], K initial number convolutional neural networks
First order branch node number in monitoring model, M indicate that the classification number of affiliated driving condition in newly-increased sample data, I indicate each
The sample number of a newly-increased classification.
Take the average value of the sample probability value of more parts of newly-increased sample datas of the same first order branch node as this
The first node probability value of level-one branch node, calculation formula are as follows:
Wherein, K indicates that first order branch node number in initial number convolutional neural networks monitoring model, M indicate newly-increased sample
The classification number of affiliated driving condition in data, I indicate the sample number of each newly-increased classification;O (k, m, i) indicates that i-th part increases newly
The sample probability value of k-th of node when sample data belongs to m class, Oavg (k, m) indicate the newly-increased sample data of m class,
The first node probability value of k-th of first order branch node;
The processing of softmax function is carried out to first node probability value, obtains newly-increased sample data in initial tree convolutional Neural
The output probability value of each first order branch node of Network Monitoring Model, calculation formula are as follows:
Wherein, L (k, m) indicates the newly-increased sample data of m class, in the output probability value of k-th of first order branch node.
If in the output probability value of first order branch node, the output probability value of only one first order branch node is greater than
When preset threshold in initial Tree-CNN monitoring model, continue according to above-mentioned calculation method, calculate be greater than preset threshold should
The output probability value of branch node under first order branch node, i.e. the output probability value of second level branch node, if the second level
When the output probability value of only one second level branch node is greater than preset threshold in the output probability value of branch node, with such
Push away, continue calculate third level branch node output probability value, until by first category be added Tree-CNN monitoring model at
For leaf node.
The first category is added initial tree convolutional neural networks according to the output probability value and monitors mould by step S40
In the node of type, initial tree convolutional neural networks monitoring model is updated.
The first category is added to the node of initial tree convolutional neural networks monitoring model according to the output probability value
In, it is possible to understand that m class is increased newly the first category of the affiliated driving condition of sample data by ground, and initial tree convolutional Neural net is added
In the node of network monitoring model,
According to formula (3), the output probability value of first order branch node is arranged according to the case where probability distribution, is obtained
To set S.
S=[lm1, lm2, lm3..., lmM], maxk(lm1(k))≥maxk(lm2(k))≥…≥maxk(lmM(k)) (3)
It is raw in initial convolutional neural networks monitoring model if there is multiple probability values to be greater than preset threshold in output set S
The first order branch node of Cheng Xin the first category is added in the child node of the new first order branch node generated.Example
Such as newly-increased classification is vehicle abnormality (brake failure, tire problem), then says that the vehicle abnormality is set as the first new fraction
Brake failure, tire problem are added the child node of vehicle abnormality, form leaf node by Zhi Jiedian with fatigue driving peer.
If all output probability values are both less than preset threshold in output set S, using the first category as new
First order branch node be added initial tree convolutional neural networks monitoring model in, first order branch node be leaf node.
If maximum output probability value l in output set Sm1When greater than preset threshold, as shown in formula (4), i.e. the first order
In the output probability value of branch node, when the output probability value of only one first order branch node is greater than preset threshold, it is known that
The relationship of first category and the first order branch node is larger, continues to calculate under the first order branch node greater than preset threshold
Branch node output probability value, i.e. the output probability value of second level branch node, if the output of second level branch node is general
When the output probability value of only one second level branch node is greater than preset threshold in rate value, and so on, continue to calculate third
The output probability value of grade branch node becomes leaf node until first category to be added in Tree-CNN monitoring model.
By the above-mentioned means, realizing the study to new category, as incremental learning dynamic updates.Such as newly-increased classification does not belong to
In any existing major class, then it is individually juxtaposed to leaf node with fatigue driving major class, occurs similar feelings again later
Condition, then directly output is not as a result, continue to divide.
After first order branch node M class completes aforesaid operations, the second level branch node rank of tree will be moved into.Second
Fraction detail point calculating method is identical with the mode that model is added in first category, when new classification will be added in model.When into
When entering next rank, only show that those distribute to the sample image of the newly-increased classification of specific child node.Generally, for one
A rank, the sample image of all M classes are all transmitted by different child nodes.For example, being added to two newly to a child node
Class.If child node is a leaf node, it will be changed to a branch node, and the branch node is now with 3 son sections
Point.If child node is a branch node, the process of likelihood matrix is computed repeatedly, and determines how and adds the two new classes
Into its output.Semi-supervised on how to increase the decision of tree: under given restraint condition algorithm itself determine how
Increase tree.The value of parameter, such as maximum child node, the depth capacity of tree of node can be limited according to demand.It is distributed in tree
After new class, the training based on supervision gradient decline is carried out to modified/new node.In this way, whole network need not be then modified, and
And only impacted subnetwork needs re -training/fine tuning.Change it is original when there is new abnormal driving state, can only
The mode for taking re -training, by incremental data dynamic update abnormal driving condition supervision model, can model dynamic more
The low redundancy for guaranteeing model during new, improves abnormal driving state recognition precision and model robustness.
It further, is the stream of the monitoring model update method second embodiment of driving condition of the present invention referring to Fig. 3, Fig. 3
Journey schematic diagram.Based on above-mentioned embodiment shown in Fig. 2, can also include: before step S20
Step S50 pre-processes the newly-increased sample data;
Newly-increased sample data is pre-processed, such as: driver's video of vehicle-mounted camera acquisition increases sample data newly
When, crucial pretreatment, including illumination correction method, Connected component Fast Labeling, direction projection etc. are carried out to driver's video;
Critical data processing is formed normalization characteristic vector by the critical data for obtaining pretreated newly-increased sample data.Increase new
Increase the accuracy of sample data.
It further, is the stream of the monitoring model update method 3rd embodiment of driving condition of the present invention referring to Fig. 4, Fig. 4
Journey schematic diagram.Based on the above embodiments, step S30 can also include:
The normalization characteristic vector sum first category is inputted initial tree convolutional neural networks monitoring model by step S31
First order branch node, export newly-increased sample data in the sample probability value of first order branch node;
The first category that the normalization characteristic vector sum is increased newly to the affiliated driving condition of sample data is rolled up as initial tree
The input of product neural network monitoring model obtains the node for increasing sample data newly in initial tree convolutional neural networks monitoring model
Output probability value, such as: need it is newly-increased yawn, both abnormal driving behaviors of closing one's eyes, these two types are collected in advance more
A image increases sample data newly and obtains the matrix (such as 244*244*3) of feature vector by pretreatment, is then input to initial
In the first order branch node of the network of Tree-CNN monitoring model, every layer multiplied by initial model training when obtained weight, most
The sample probability value O (k, m, i) of first order branch node is obtained eventually, and O (k, m, i) indicates that i-th part of newly-increased sample data belongs to m
The sample probability value of k-th of node when class, wherein [1, K] k ∈, m ∈ [1, M], i ∈ [1, I], K initial number convolutional neural networks
First order branch node number in monitoring model, M indicate that the classification number of affiliated driving condition in newly-increased sample data, I indicate each
The sample number of a newly-increased classification.
Step S32 calculates newly-increased sample data according to newly-increased sample data in the sample probability value of first order branch node
In the output probability value of the first order branch node of initial tree convolutional neural networks monitoring model;
It further, is the stream of the monitoring model update method fourth embodiment of driving condition of the present invention referring to Fig. 5, Fig. 5
Journey schematic diagram.Based on the above embodiments, step S32 can also include:
Step S321 takes the average value of the sample probability value of more parts of newly-increased sample datas of the same first order branch node
First node probability value as the first order branch node;
First node probability value calculation formula is as follows:
Wherein, K indicates that first order branch node number in initial number convolutional neural networks monitoring model, M indicate newly-increased sample
The classification number of affiliated driving condition in data, I indicate the sample number of each newly-increased classification;O (k, m, i) indicates that i-th part increases newly
The sample probability value of k-th of node when sample data belongs to m class, Oavg (k, m) indicate the newly-increased sample data of m class,
The first node probability value of k-th of first order branch node;
Step S322 carries out the processing of softmax function to the first node probability value, obtains newly-increased sample data first
The output probability value of each first order branch node of beginning tree convolutional neural networks monitoring model.
The output probability value calculation formula of first order branch node is as follows:
Wherein, L (k, m) indicates the newly-increased sample data of m class, in the output probability value of k-th of first order branch node.
By the classification of the affiliated driving condition of newly-increased sample data first order branch node sample probability value, according to step
S321 and step S322 calculate newly-increased sample data, in the first order branch node of initial tree convolutional neural networks monitoring model
First node probability value carries out the output probability value that processing obtains first order branch node to first node probability value.
Step S40 can also include:
According to the output probability value of the first order branch node initial tree convolution mind is added in the first category by step S41
In node through Network Monitoring Model, initial tree convolutional neural networks monitoring model is updated.
According to the size of first order branch node output probability value and preset threshold, initial tree convolution is added in first category
In the node of neural network monitoring model, initial tree convolutional neural networks monitoring model is updated.
It further, is the stream of the 5th embodiment of monitoring model update method of driving condition of the present invention referring to Fig. 6, Fig. 6
Journey schematic diagram.Based on the above embodiments, step S41 can also include:
Step S411, when the output probability value of each first order branch node is respectively less than preset threshold, by the first kind
It is not added in initial tree convolutional neural networks monitoring model as new first order branch node;
When the output probability value of each first order branch node is respectively less than preset threshold, using the first category as new the
Level-one branch node is added in initial tree convolutional neural networks monitoring model, and first order branch node is leaf node.
Step S412 in the output probability value of each first order branch node, there is the output of multiple first order branch nodes
When probability value is greater than preset threshold, new first order branch node is generated in initial convolutional neural networks monitoring model, by institute
First category is stated to be added in the child node of the new first order branch node generated;
In the output probability value of first order branch node, there is the output probability value of multiple first order branch nodes to be greater than default
When threshold value, new first order branch node is generated in initial convolutional neural networks monitoring model, the first category is added
In the child node of the new first order branch node generated.Such as newly-increased classification is vehicle abnormality (brake failure, tire problem
Deng), then say that the vehicle abnormality is set as new first order branch node, it is at the same level with fatigue driving, by brake failure, tire problem
The child node of vehicle abnormality is added, forms leaf node.
Step S413 is updated initial number convolutional neural networks monitoring model.
When first category being added in the node of initial tree convolutional neural networks monitoring model, to modified/new node
The training based on supervision gradient decline is carried out, and updates the node weights of initial convolutional neural networks monitoring model, is completed to first
The update of beginning tree convolutional neural networks monitoring model.
It further, is the stream of the monitoring model update method sixth embodiment of driving condition of the present invention referring to Fig. 7, Fig. 7
Journey schematic diagram.Based on the above embodiments, step S41 can also include:
Step S414, in the output probability value of each first order branch node, only one first order branch node it is defeated
When probability value is greater than preset threshold out, the output probability value of the second level branch node in the first order branch node is calculated;
In the output probability value of first order branch node, the output probability value of only one first order branch node is greater than pre-
If when threshold value, it is known that the relationship of first category and the first order branch node is larger, continue to calculate be greater than preset threshold this
The output probability value of branch node under level-one branch node, i.e. the output probability value of second level branch node.
According to the output probability value of the second level branch node initial tree convolution is added in the first category by step S415
In the progressive child node of second level branch node described in neural network monitoring model or second level branch node, initial tree is rolled up
Product neural network monitoring model is updated.
If in the output probability value of second level branch node, thering is the output probability value of multiple second level branch nodes to be greater than pre-
If when threshold value, the first category is added in the child node of the new second level branch node generated.
When the output probability value of second level branch node is both less than preset threshold, using the first category as new second
Grade branch node is added in initial tree convolutional neural networks monitoring model, and second level branch node is leaf node.
If the output probability value of only one second level branch node is greater than in the output probability value of second level branch node
When preset threshold, and so on, continue the output probability value for calculating third level branch node, is added until by first category
Become leaf node in Tree-CNN monitoring model.
Initial tree convolutional neural networks prison is added in the classification of the affiliated driving condition according to the output probability value of node
It surveys in the node of model, the training based on supervision gradient decline is carried out to modified/new node, and update initial convolutional Neural
The node weights of Network Monitoring Model complete the update to initial tree convolutional neural networks monitoring model, realize to new category
Study, as incremental learning dynamic update.
It further, is the stream of the 7th embodiment of monitoring model update method of driving condition of the present invention referring to Fig. 8, Fig. 8
Journey schematic diagram.Based on the above embodiments, step S31 can also include:
The normalization characteristic vector sum first category is inputted initial tree convolutional neural networks and monitors mould by step S311
Each first order branch node of type;
The first category that the normalization characteristic vector sum is increased newly to the affiliated driving condition of sample data is rolled up as initial tree
The input of product neural network monitoring model obtains the node for increasing sample data newly in initial tree convolutional neural networks monitoring model
Output probability value, such as: need it is newly-increased yawn, both abnormal driving behaviors of closing one's eyes, these two types are collected in advance more
A image increases sample data newly and obtains the matrix (such as 244*244*3) of feature vector by pretreatment, is then input to initial
In the first order branch node of the network of Tree-CNN monitoring model.
Step S312 obtains the weighted value of each first order branch node of initial tree convolutional neural networks model;
The weighted value of initial number convolutional neural networks model branch nodes at different levels is formed in training, obtains each first order
The weighted value of branch node.
Step S313 exports newly-increased sample according to the weighted value of each first order branch node of the normalization characteristic vector sum
Sample probability value of the data in each first order branch node.
Normalization characteristic vector is exported first order branch of the newly-increased sample data in monitoring model multiplied by weighted value to save
The sample probability value O (k, m, i) of point, O (k, m, i) indicate the sample of k-th of node when i-th part of newly-increased sample data belongs to m class
This probability value, wherein [1, K] k ∈, m ∈ [1, M], i ∈ [1, I], the first fraction in K initial number convolutional neural networks monitoring model
Branch number of nodes, M indicate that the classification number of affiliated driving condition in newly-increased sample data, I indicate the sample number of each newly-increased classification.
It further, is the stream of the 8th embodiment of monitoring model update method of driving condition of the present invention referring to Fig. 9, Fig. 9
Journey schematic diagram.Based on the above embodiments, step S40 can also include:
The first category is added initial tree convolutional neural networks according to the output probability value and monitors mould by step S42
In the node of type;
According to the output probability value of the node of monitoring model, initial tree convolutional Neural is added in the classification of the driving condition
In node in network model, such as: it is added to new category of yawning to fatigue driving, then is equally become with initial bow etc.
The branch node of fatigue driving yawns, bows as the leaf node of monitoring model.
Step S43, the training of gradient decline is carried out to the node, and updates initial convolutional neural networks monitoring model
Node weights complete the update to initial tree convolutional neural networks monitoring model.
After new classification is added, after new class is distributed in tree, modified/new node is carried out based on supervision gradient decline
Training, update the weight of initial model, complete the update to initial tree convolutional neural networks monitoring model.It need not then modify whole
A network, and only impacted subnetwork needs re -training/fine tuning.
The monitoring model updating device of driving condition of the present invention includes: memory, processor and is stored in the memory
The monitoring model more new procedures of driving condition that are upper and can running on the processor, the monitoring model of the driving condition is more
New procedures realize the step of monitoring model update method of driving condition as described above when being executed by the processor.
Wherein, the monitoring model of the driving condition run on the processor more new procedures are performed realized side
Method can refer to each embodiment of monitoring model update method of driving condition of the present invention, and details are not described herein again.
The present invention also provides a kind of storage mediums.
The monitoring model more new procedures of driving condition, the monitoring mould of the driving condition are stored on storage medium of the present invention
Type more new procedures realize the step of monitoring model update method of driving condition as described above when being executed by processor.
Wherein, the monitoring model of the driving condition run on the processor more new procedures are performed realized side
Method can refer to each embodiment of monitoring model update method of driving condition of the present invention, and details are not described herein again.
It should be noted that, in this document, the terms "include", "comprise" or its any other variant are intended to non-row
His property includes, so that the process, method, article or the system that include a series of elements not only include those elements, and
And further include other elements that are not explicitly listed, or further include for this process, method, article or system institute it is intrinsic
Element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including being somebody's turn to do
There is also other identical elements in the process, method of element, article or system.
The serial number of the above embodiments of the invention is only for description, does not represent the advantages or disadvantages of the embodiments.
Through the above description of the embodiments, those skilled in the art can be understood that above-described embodiment side
Method can be realized by means of software and necessary general hardware platform, naturally it is also possible to by hardware, but in many cases
The former is more preferably embodiment.Based on this understanding, technical solution of the present invention substantially in other words does the prior art
The part contributed out can be embodied in the form of software products, which is stored in one as described above
In storage medium (such as ROM/RAM, magnetic disk, CD), including some instructions are used so that terminal device (it can be mobile phone,
Computer, server, air conditioner or network equipment etc.) execute method described in each embodiment of the present invention.
The above is only a preferred embodiment of the present invention, is not intended to limit the scope of the invention, all to utilize this hair
Equivalent structure or equivalent flow shift made by bright specification and accompanying drawing content is applied directly or indirectly in other relevant skills
Art field, is included within the scope of the present invention.
Claims (10)
1. a kind of monitoring model update method of driving condition, which is characterized in that described method includes following steps:
Shape is driven belonging to the newly-increased sample data and newly-increased sample data for reading the driving condition of driving condition collector acquisition
The first category of state;
The critical data in newly-increased sample data is obtained, processing is carried out to the critical data and forms normalization characteristic vector;
The normalization characteristic vector sum first category is obtained as the input of initial tree convolutional neural networks monitoring model
Output probability value of the newly-increased sample data in the node of initial tree convolutional neural networks monitoring model;
The first category is added in the node of initial tree convolutional neural networks monitoring model according to the output probability value, it is right
Initial tree convolutional neural networks monitoring model is updated.
2. the monitoring model update method of driving condition as described in claim 1, which is characterized in that described to obtain newly-increased sample
Critical data in data, before carrying out the step of processing forms normalization characteristic vector to the critical data, the method
Further include:
The newly-increased sample data is pre-processed.
3. the monitoring model update method of driving condition as claimed in claim 2, which is characterized in that described by the normalization
Feature vector and first category obtain newly-increased sample data first as the input of initial tree convolutional neural networks monitoring model
Begin tree convolutional neural networks monitoring model node output probability value the step of include:
By the normalization characteristic vector sum first category, the first order branch of initial tree convolutional neural networks monitoring model is inputted
Node exports newly-increased sample data in the sample probability value of first order branch node;
According to newly-increased sample data in the sample probability value of first order branch node, newly-increased sample data is calculated in initial tree convolution
The output probability value of the first order branch node of neural network monitoring model;
It is described that the first category is added in the node of initial tree convolutional neural networks monitoring model according to output probability value, it is right
The step of initial tree convolutional neural networks monitoring model is updated include:
According to the output probability value of first order branch node, initial tree convolutional neural networks are added in the first category and monitor mould
In the node of type, initial tree convolutional neural networks monitoring model is updated.
4. the monitoring model update method of driving condition as claimed any one in claims 1 to 3, which is characterized in that described
According to newly-increased sample data in the sample probability value of first order branch node, newly-increased sample data is calculated in initial tree convolutional Neural
The step of output probability value of the first order branch node of Network Monitoring Model includes:
Take the average value of the sample probability value of more parts of newly-increased sample datas of the same first order branch node as the first order
The first node probability value of branch node;
The processing of softmax function is carried out to the first node probability value, obtains newly-increased sample data in initial tree convolutional Neural
The output probability value of each first order branch node of Network Monitoring Model.
5. the monitoring model update method of driving condition as claimed in claim 4, which is characterized in that described according to the first fraction
The output probability value of Zhi Jiedian the first category is added in the node of initial tree convolutional neural networks monitoring model, to first
Begin tree convolutional neural networks monitoring model the step of being updated includes:
When the output probability value of each first order branch node is respectively less than preset threshold, using the first category as new first
Grade branch node is added in initial tree convolutional neural networks monitoring model;
In the output probability value of each first order branch node, there is the output probability value of multiple first order branch nodes to be greater than default
When threshold value, new first order branch node is generated in initial convolutional neural networks monitoring model, the first category is added
In the child node of the new first order branch node generated;
Initial number convolutional neural networks monitoring model is updated.
6. the monitoring model update method of driving condition as claimed in claim 5, which is characterized in that described according to the first fraction
The output probability value of Zhi Jiedian the first category is added in the node of initial tree convolutional neural networks monitoring model, to first
Begin tree convolutional neural networks monitoring model the step of being updated includes:
In the output probability value of each first order branch node, the output probability value of only one first order branch node is greater than pre-
If when threshold value, calculating the output probability value of the second level branch node in the first order branch node;
According to the output probability value of second level branch node, initial tree convolutional neural networks are added in the first category and monitor mould
In the progressive child node of second level branch node described in type or second level branch node, initial tree convolutional neural networks are monitored
Model is updated.
7. the monitoring model update method of driving condition as claimed in claim 6, which is characterized in that
By the normalization characteristic vector sum first category, the first order branch of initial tree convolutional neural networks monitoring model is inputted
Node, exporting newly-increased sample data in the step of sample probability value of first order branch node includes:
By the normalization characteristic vector sum first category, each first fraction of initial tree convolutional neural networks monitoring model is inputted
Zhi Jiedian;
Obtain the weighted value of each first order branch node of initial tree convolutional neural networks model;
Newly-increased sample data is exported each first according to the weighted value of each first order branch node of the normalization characteristic vector sum
The sample probability value of grade branch node.
8. the monitoring model update method of driving condition as claimed in claim 7, which is characterized in that described according to the output
Probability value the first category is added in the node of initial tree convolutional neural networks monitoring model, to initial tree convolutional Neural net
The step of network monitoring model is updated include:
The first category is added in the node of initial tree convolutional neural networks monitoring model according to the output probability value;
The training of gradient decline is carried out to the node, and updates the node weights of initial convolutional neural networks monitoring model, it is complete
The update of pairs of initial tree convolutional neural networks monitoring model.
9. a kind of updating device, which is characterized in that the updating device includes: memory, processor and is stored in the storage
On device and the monitoring model more new procedures of driving condition that can run on the processor, the monitoring model of the driving condition
The monitoring model such as driving condition described in any item of the claim 1 to 8 is realized when more new procedures are executed by the processor
The step of update method.
10. a kind of storage medium, which is characterized in that the monitoring model for being stored with driving condition on the storage medium updates journey
Sequence is realized as described in any one of claims 1 to 8 when the monitoring model of the driving condition more new procedures are executed by processor
Driving condition monitoring model update method the step of.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910103594.1A CN109858553B (en) | 2019-01-31 | 2019-01-31 | Method, device and storage medium for updating driving state monitoring model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910103594.1A CN109858553B (en) | 2019-01-31 | 2019-01-31 | Method, device and storage medium for updating driving state monitoring model |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109858553A true CN109858553A (en) | 2019-06-07 |
CN109858553B CN109858553B (en) | 2023-12-12 |
Family
ID=66897446
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910103594.1A Active CN109858553B (en) | 2019-01-31 | 2019-01-31 | Method, device and storage medium for updating driving state monitoring model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109858553B (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110693508A (en) * | 2019-09-02 | 2020-01-17 | 中国航天员科研训练中心 | Multi-channel cooperative psychophysiological active sensing method and service robot |
CN112580795A (en) * | 2019-09-29 | 2021-03-30 | 华为技术有限公司 | Neural network acquisition method and related equipment |
CN113837834A (en) * | 2021-09-22 | 2021-12-24 | 东北大学 | Shopping basket recommendation method based on deep learning |
WO2021258847A1 (en) * | 2020-06-23 | 2021-12-30 | 华为技术有限公司 | Driving decision-making method, device, and chip |
CN114355184A (en) * | 2022-01-05 | 2022-04-15 | 国网江苏省电力有限公司宿迁供电分公司 | High-voltage circuit breaker state monitoring and early warning system and method based on online learning |
CN114565964A (en) * | 2022-03-03 | 2022-05-31 | 网易(杭州)网络有限公司 | Emotion recognition model generation method, recognition method, device, medium and equipment |
Citations (26)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2007122362A (en) * | 2005-10-27 | 2007-05-17 | Toyota Motor Corp | State estimation method using neural network and state estimation apparatus using neural network |
CN101038623A (en) * | 2006-03-13 | 2007-09-19 | 欧姆龙株式会社 | Feature point detecting device, feature point detecting method, and feature point detecting program |
JP2007257043A (en) * | 2006-03-20 | 2007-10-04 | Nissan Motor Co Ltd | Occupant state estimating device and occupant state estimating method |
US20100158316A1 (en) * | 2008-12-24 | 2010-06-24 | Denso Corporation | Action estimating apparatus, method for updating estimation model, and program |
CN102750827A (en) * | 2012-06-26 | 2012-10-24 | 浙江大学 | System for sampling and identifying data of driver response behaviors under group guidance information |
CN104036299A (en) * | 2014-06-10 | 2014-09-10 | 电子科技大学 | Human eye contour tracking method based on local texture AAM |
US20140358840A1 (en) * | 2012-01-13 | 2014-12-04 | Pulse Function F6 Ltd | Apparatus, system and method for risk indicator calculation for driving behaviour and for reconstructing a vehicle trajectory |
US20150092056A1 (en) * | 2013-09-30 | 2015-04-02 | Sackett Solutions & Innovations | Driving assistance systems and methods |
CN104504400A (en) * | 2015-01-08 | 2015-04-08 | 中国科学院重庆绿色智能技术研究院 | Detection method of driver's abnormal behavior modeled based on online behavior |
US20160104486A1 (en) * | 2011-04-22 | 2016-04-14 | Angel A. Penilla | Methods and Systems for Communicating Content to Connected Vehicle Users Based Detected Tone/Mood in Voice Input |
CN106909879A (en) * | 2017-01-11 | 2017-06-30 | 开易(北京)科技有限公司 | A kind of method for detecting fatigue driving and system |
CN106934368A (en) * | 2017-03-13 | 2017-07-07 | 长安大学 | A kind of driving fatigue detecting system and recognition methods based on the dynamic achievement data of eye |
CN107316457A (en) * | 2016-04-27 | 2017-11-03 | 上海炬宏信息技术有限公司 | Judge whether road traffic condition meets the method for automatic driving |
CN107437147A (en) * | 2017-08-02 | 2017-12-05 | 辽宁友邦网络科技有限公司 | Reduce the vehicle travel risk dynamic assessment method and its system of freight logistics scene |
US20180025289A1 (en) * | 2016-07-20 | 2018-01-25 | Qualcomm Incorporated | Performance Provisioning Using Machine Learning Based Automated Workload Classification |
US20180053102A1 (en) * | 2016-08-16 | 2018-02-22 | Toyota Jidosha Kabushiki Kaisha | Individualized Adaptation of Driver Action Prediction Models |
CN107907886A (en) * | 2017-11-07 | 2018-04-13 | 广东欧珀移动通信有限公司 | Travel conditions recognition methods, device, storage medium and terminal device |
CN108248608A (en) * | 2017-12-28 | 2018-07-06 | 北京百度网讯科技有限公司 | For assessing the method and apparatus of the comfort level of control loop |
CN108363968A (en) * | 2018-01-31 | 2018-08-03 | 上海瀚所信息技术有限公司 | A kind of tired driver driving monitoring system and method based on key point extraction |
CN108423006A (en) * | 2018-02-02 | 2018-08-21 | 辽宁友邦网络科技有限公司 | A kind of auxiliary driving warning method and system |
CN108520268A (en) * | 2018-03-09 | 2018-09-11 | 浙江工业大学 | The black box antagonism attack defense method evolved based on samples selection and model |
CN108596087A (en) * | 2018-04-23 | 2018-09-28 | 合肥湛达智能科技有限公司 | A kind of driving fatigue degree detecting regression model based on dual network result |
CN108688675A (en) * | 2017-03-29 | 2018-10-23 | 马自达汽车株式会社 | Vehicle drive support system |
CN108806667A (en) * | 2018-05-29 | 2018-11-13 | 重庆大学 | The method for synchronously recognizing of voice and mood based on neural network |
US20180348785A1 (en) * | 2017-06-06 | 2018-12-06 | PlusAI Corp | Method and system for integrated global and distributed learning in autonomous driving vehicles |
CN109284661A (en) * | 2017-07-21 | 2019-01-29 | 松下电器(美国)知识产权公司 | Learning data generating means, learning data generation method and program |
-
2019
- 2019-01-31 CN CN201910103594.1A patent/CN109858553B/en active Active
Patent Citations (26)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2007122362A (en) * | 2005-10-27 | 2007-05-17 | Toyota Motor Corp | State estimation method using neural network and state estimation apparatus using neural network |
CN101038623A (en) * | 2006-03-13 | 2007-09-19 | 欧姆龙株式会社 | Feature point detecting device, feature point detecting method, and feature point detecting program |
JP2007257043A (en) * | 2006-03-20 | 2007-10-04 | Nissan Motor Co Ltd | Occupant state estimating device and occupant state estimating method |
US20100158316A1 (en) * | 2008-12-24 | 2010-06-24 | Denso Corporation | Action estimating apparatus, method for updating estimation model, and program |
US20160104486A1 (en) * | 2011-04-22 | 2016-04-14 | Angel A. Penilla | Methods and Systems for Communicating Content to Connected Vehicle Users Based Detected Tone/Mood in Voice Input |
US20140358840A1 (en) * | 2012-01-13 | 2014-12-04 | Pulse Function F6 Ltd | Apparatus, system and method for risk indicator calculation for driving behaviour and for reconstructing a vehicle trajectory |
CN102750827A (en) * | 2012-06-26 | 2012-10-24 | 浙江大学 | System for sampling and identifying data of driver response behaviors under group guidance information |
US20150092056A1 (en) * | 2013-09-30 | 2015-04-02 | Sackett Solutions & Innovations | Driving assistance systems and methods |
CN104036299A (en) * | 2014-06-10 | 2014-09-10 | 电子科技大学 | Human eye contour tracking method based on local texture AAM |
CN104504400A (en) * | 2015-01-08 | 2015-04-08 | 中国科学院重庆绿色智能技术研究院 | Detection method of driver's abnormal behavior modeled based on online behavior |
CN107316457A (en) * | 2016-04-27 | 2017-11-03 | 上海炬宏信息技术有限公司 | Judge whether road traffic condition meets the method for automatic driving |
US20180025289A1 (en) * | 2016-07-20 | 2018-01-25 | Qualcomm Incorporated | Performance Provisioning Using Machine Learning Based Automated Workload Classification |
US20180053102A1 (en) * | 2016-08-16 | 2018-02-22 | Toyota Jidosha Kabushiki Kaisha | Individualized Adaptation of Driver Action Prediction Models |
CN106909879A (en) * | 2017-01-11 | 2017-06-30 | 开易(北京)科技有限公司 | A kind of method for detecting fatigue driving and system |
CN106934368A (en) * | 2017-03-13 | 2017-07-07 | 长安大学 | A kind of driving fatigue detecting system and recognition methods based on the dynamic achievement data of eye |
CN108688675A (en) * | 2017-03-29 | 2018-10-23 | 马自达汽车株式会社 | Vehicle drive support system |
US20180348785A1 (en) * | 2017-06-06 | 2018-12-06 | PlusAI Corp | Method and system for integrated global and distributed learning in autonomous driving vehicles |
CN109284661A (en) * | 2017-07-21 | 2019-01-29 | 松下电器(美国)知识产权公司 | Learning data generating means, learning data generation method and program |
CN107437147A (en) * | 2017-08-02 | 2017-12-05 | 辽宁友邦网络科技有限公司 | Reduce the vehicle travel risk dynamic assessment method and its system of freight logistics scene |
CN107907886A (en) * | 2017-11-07 | 2018-04-13 | 广东欧珀移动通信有限公司 | Travel conditions recognition methods, device, storage medium and terminal device |
CN108248608A (en) * | 2017-12-28 | 2018-07-06 | 北京百度网讯科技有限公司 | For assessing the method and apparatus of the comfort level of control loop |
CN108363968A (en) * | 2018-01-31 | 2018-08-03 | 上海瀚所信息技术有限公司 | A kind of tired driver driving monitoring system and method based on key point extraction |
CN108423006A (en) * | 2018-02-02 | 2018-08-21 | 辽宁友邦网络科技有限公司 | A kind of auxiliary driving warning method and system |
CN108520268A (en) * | 2018-03-09 | 2018-09-11 | 浙江工业大学 | The black box antagonism attack defense method evolved based on samples selection and model |
CN108596087A (en) * | 2018-04-23 | 2018-09-28 | 合肥湛达智能科技有限公司 | A kind of driving fatigue degree detecting regression model based on dual network result |
CN108806667A (en) * | 2018-05-29 | 2018-11-13 | 重庆大学 | The method for synchronously recognizing of voice and mood based on neural network |
Non-Patent Citations (3)
Title |
---|
张旭东;严宏斌;张忠伟;周亚同;: "一种考虑驾驶员姿态及光照因素的疲劳驾驶稳健检测算法", 电子技术, no. 12, pages 26 - 30 * |
杨格兰;邓晓军;刘琮;: "基于深度时空域卷积神经网络的表情识别模型", 中南大学学报(自然科学版), no. 07, pages 2311 - 2319 * |
郭克友, 储江伟, 王荣本: "驾驶员眼部状态识别方法的研究", 系统工程与电子技术, no. 10, pages 1186 - 1188 * |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110693508A (en) * | 2019-09-02 | 2020-01-17 | 中国航天员科研训练中心 | Multi-channel cooperative psychophysiological active sensing method and service robot |
CN112580795A (en) * | 2019-09-29 | 2021-03-30 | 华为技术有限公司 | Neural network acquisition method and related equipment |
WO2021258847A1 (en) * | 2020-06-23 | 2021-12-30 | 华为技术有限公司 | Driving decision-making method, device, and chip |
CN113837834A (en) * | 2021-09-22 | 2021-12-24 | 东北大学 | Shopping basket recommendation method based on deep learning |
CN113837834B (en) * | 2021-09-22 | 2023-09-01 | 东北大学 | Shopping basket recommendation method based on deep learning |
CN114355184A (en) * | 2022-01-05 | 2022-04-15 | 国网江苏省电力有限公司宿迁供电分公司 | High-voltage circuit breaker state monitoring and early warning system and method based on online learning |
CN114355184B (en) * | 2022-01-05 | 2023-09-26 | 国网江苏省电力有限公司宿迁供电分公司 | Online learning-based high-voltage circuit breaker state monitoring and early warning method |
CN114565964A (en) * | 2022-03-03 | 2022-05-31 | 网易(杭州)网络有限公司 | Emotion recognition model generation method, recognition method, device, medium and equipment |
Also Published As
Publication number | Publication date |
---|---|
CN109858553B (en) | 2023-12-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109858553A (en) | Monitoring model update method, updating device and the storage medium of driving condition | |
Huang et al. | HCF: A hybrid CNN framework for behavior detection of distracted drivers | |
Abreu et al. | Evaluating sign language recognition using the myo armband | |
CN110309861B (en) | Multi-modal human activity recognition method based on generation of confrontation network | |
KR20190094133A (en) | An artificial intelligence apparatus for recognizing object and method for the same | |
CN105354986B (en) | Driver's driving condition supervision system and method | |
CN109726771A (en) | Abnormal driving detection model method for building up, device and storage medium | |
Fekete et al. | Towards a computational theory of experience | |
CN109949438A (en) | Abnormal driving monitoring model method for building up, device and storage medium | |
CN103745235A (en) | Human face identification method, device and terminal device | |
CN109522945A (en) | One kind of groups emotion identification method, device, smart machine and storage medium | |
CN111797854B (en) | Scene model building method and device, storage medium and electronic equipment | |
CN109260566A (en) | Enhance sleep technology using shadow casting technique | |
CN110516113B (en) | Video classification method, video classification model training method and device | |
CN113255617B (en) | Image recognition method and device, electronic equipment and computer-readable storage medium | |
CN110135497A (en) | Method, the method and device of Facial action unit intensity estimation of model training | |
CN111753683A (en) | Human body posture identification method based on multi-expert convolutional neural network | |
CN111798978A (en) | User health assessment method and device, storage medium and electronic equipment | |
JP2022120775A (en) | On-device activity recognition | |
CN113435335B (en) | Microscopic expression recognition method and device, electronic equipment and storage medium | |
CN116935188B (en) | Model training method, image recognition method, device, equipment and medium | |
CN113269010B (en) | Training method and related device for human face living body detection model | |
CN114444705A (en) | Model updating method and device | |
CN113762585B (en) | Data processing method, account type identification method and device | |
CN109976726A (en) | Vehicle-mounted Edge intelligence computing architecture, method, system and 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 | ||
CB02 | Change of applicant information | ||
CB02 | Change of applicant information |
Address after: 518000 706, building 3b, hongrongyuan Shangjun phase II, Longping community, Dalang street, Longhua District, Shenzhen City, Guangdong Province Applicant after: Jintu computing technology (Shenzhen) Co.,Ltd. Address before: 518000 area a, 18 / F, building 1, creative park, broadcasting group, No.1 Qingqing Road, Qinghu community, Longhua sub district office, Longhua District, Shenzhen City, Guangdong Province Applicant before: SHENZHEN SEMISKY TECHNOLOGY Co.,Ltd. |
|
GR01 | Patent grant | ||
GR01 | Patent grant |