CN108470460A - A kind of nearby vehicle Activity recognition method based on smart mobile phone and RNN - Google Patents
A kind of nearby vehicle Activity recognition method based on smart mobile phone and RNN Download PDFInfo
- Publication number
- CN108470460A CN108470460A CN201810320788.2A CN201810320788A CN108470460A CN 108470460 A CN108470460 A CN 108470460A CN 201810320788 A CN201810320788 A CN 201810320788A CN 108470460 A CN108470460 A CN 108470460A
- Authority
- CN
- China
- Prior art keywords
- vehicle
- mobile phone
- rnn
- smart mobile
- behavior
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/09—Arrangements for giving variable traffic instructions
- G08G1/0962—Arrangements for giving variable traffic instructions having an indicator mounted inside the vehicle, e.g. giving voice messages
- G08G1/0967—Systems involving transmission of highway information, e.g. weather, speed limits
- G08G1/096766—Systems involving transmission of highway information, e.g. weather, speed limits where the system is characterised by the origin of the information transmission
- G08G1/096791—Systems involving transmission of highway information, e.g. weather, speed limits where the system is characterised by the origin of the information transmission where the origin of the information is another vehicle
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04M—TELEPHONIC COMMUNICATION
- H04M1/00—Substation equipment, e.g. for use by subscribers
- H04M1/72—Mobile telephones; Cordless telephones, i.e. devices for establishing wireless links to base stations without route selection
- H04M1/724—User interfaces specially adapted for cordless or mobile telephones
- H04M1/72403—User interfaces specially adapted for cordless or mobile telephones with means for local support of applications that increase the functionality
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04M—TELEPHONIC COMMUNICATION
- H04M1/00—Substation equipment, e.g. for use by subscribers
- H04M1/72—Mobile telephones; Cordless telephones, i.e. devices for establishing wireless links to base stations without route selection
- H04M1/724—User interfaces specially adapted for cordless or mobile telephones
- H04M1/72448—User interfaces specially adapted for cordless or mobile telephones with means for adapting the functionality of the device according to specific conditions
- H04M1/72457—User interfaces specially adapted for cordless or mobile telephones with means for adapting the functionality of the device according to specific conditions according to geographic location
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Signal Processing (AREA)
- Computer Networks & Wireless Communication (AREA)
- Human Computer Interaction (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Environmental & Geological Engineering (AREA)
- Atmospheric Sciences (AREA)
- Traffic Control Systems (AREA)
- Navigation (AREA)
Abstract
The nearby vehicle Activity recognition method based on smart mobile phone and RNN that the invention discloses a kind of, belongs to intelligent driving field, including：A. off-line training link：Typical nearby vehicle behavior is concluded and divided, vector coding, the training set as RNN parameter learnings are carried out to the relative characteristic and vehicle behavior that are tracked vehicle and main vehicle using smart mobile phone collected vehicle operation data.B. on-line checking link：Based on real-time traffic scene, main vehicle combines tracked vehicle to form the input that new eigenmatrix is used as trained RNN with the running data from vehicle by 4G communications, distinguishes the affiliated behavior pattern of nearby vehicle.The present invention has the advantage of feasibility and convenience using smart mobile phone as data collection, the hardware of vehicle communication；It is good at handling the characteristic of higher dimensional matrix operation using RNN, while enriching from vehicle and nearby vehicle relative characteristic, promotion discrimination, ensures that Activity recognition has higher real-time.
Description
Technical field
The invention belongs to Vehicular intelligents to drive field, more particularly to a kind of nearby vehicle based on smart mobile phone and RNN
Activity recognition method.
Background technology
In recent years, vehicle behavior identification is turned to from the identification of the monitoring system vehicle behavior of fixed position and is dynamically based on
The nearby vehicle Activity recognition of driving vehicle.The key of Activity recognition is to learn the behavior pattern of vehicle, establishes Activity recognition mould
Then type carries out vehicle behavior identification by trained vehicle behavior identification model, or even can predict vehicle behavior.
In order to give nearby vehicle Activity recognition model to provide training set, first have to collect vehicle using intelligent mobile phone sensor
Running condition information, and establish by 4G networks the communication group of vehicle and vehicle.Currently with smart mobile phone as intelligent automobile shape
State is collected, the research of information transmission is less.Foreign scholar Mucahit Karaduman utilize the GPS of smart mobile phone, acceleration
Meter, the sensor that gyroscope is obtained as vehicle traveling information are classified to vehicle driving trace using HMM, are achieved not
Wrong effect.
For nearby vehicle Activity recognition, existing solution either uses radar, camera active obtaining periphery
The running data of vehicle, or data are passively received based on wireless data transmission V2V using between motor vehicles, it is required for additional
Expensive hardware device is built, and smart mobile phone will not give researcher while obtaining abundant information in cost consideration
Too big obstacle is set, while smart mobile phone all carries GPS and inertial sensor, realization positioning, speed measuring function can be facilitated, because
This is using smart mobile phone as the hardware of data acquisition, information transmission, feasibility and convenience with height.
In terms of vehicle behavior identifies modeling, traditional method generally uses HMM, i.e. hidden Markov model, although
In terms of model foundation, HMM can meet vehicle behavior and model requirement for series model well, but the classification capacity of HMM
It is poor while also extremely limited as the vehicle operation characteristic that the observation sequence of mode input can be covered, so the misclassification rate of HMM
It is still higher.RNN, i.e. Recognition with Recurrent Neural Network have just been suggested early in the end of the nineties, but big until nearly 2 years deep learnings
Heat is just widely used, and especially in the fields natural language processing NLP, RNN helps it to achieve rapid progress.It is special
Not in terms of for the Emotion identification of English sentence, frequently with multi input to the RNN models that singly export, i.e., by an English sentence
Each word as multiple input, RNN models export a kind of mood that judgement sentence is intended by pleasure, anger, sorrow, happiness.
The present invention is inspired by the research, is used for reference the method that word is embedded in natural language processing and is adopted to the relative characteristic in vehicle travel process
Encoded with the form of feature vector, using RNN for the support of higher dimensional matrix operation, using Softmax as grader at
Manage more classification problems, can abundant vehicle with respect to travelling characteristic while, effectively improve recognizer accuracy, in real time
Property.
Invention content
The present invention proposes a kind of vehicle behavior recognition methods, accurately can make identification to the behavior of nearby vehicle,
Reference frame is provided for the trajectory planning of intelligent vehicle.
The purpose of the present invention is achieved through the following technical solutions：
A kind of nearby vehicle Activity recognition method based on smart mobile phone and RNN, which is characterized in that including：
Step 1, off-line training link：Typical nearby vehicle behavior is concluded and divided, is utilized based on real-time traffic scene
The running data for collecting nearby vehicle by the smart mobile phone that specific position is put in vehicle, using the form of feature vector to same
One moment main vehicle and the relative characteristic of nearby vehicle are encoded, and a complete vehicle row is indicated using an eigenmatrix
For corresponding behavior carries out handmarking in the form of label vector；The nearby vehicle relative characteristic that will be acquired, marked
Input of the data (including eigenmatrix and its corresponding label vector) as RNN parameter learnings updates model parameter；
Step 2, on-line checking link：Tracked target vehicle passes through smart mobile phone reality by collected from vehicle driving information
When be transferred to smart mobile phone on main vehicle, main vehicle generates new eigenmatrix in conjunction with two vehicle relative characteristics, utilizes trained RNN
It distinguishes and is tracked the affiliated behavior pattern of vehicle.
It is concluded in the step 1 and divides typical nearby vehicle behavior, specially：Typical nearby vehicle behavior is drawn
It is divided into：Front truck behavior：Braking, rear car behavior：With speeding, the behavior of left side vehicle：Lane-change is overtaken other vehicles, doubling, the behavior of right side vehicle：It changes
Road is overtaken other vehicles, doubling.
Periphery vehicle is collected by the smart mobile phone that specific position is put using in vehicle based on real-time traffic scene in step 1
Running data, specially：Smart mobile phone is horizontal positioned, horizontal plane where making smart mobile phone and water where vehicle cross, the longitudinal axis
Plane is parallel, thus resolves vehicular characteristics data using the collected each item data of smart mobile phone, i.e., by the GPS of smart mobile phone
Unified map coordinates system is established for tracked target vehicle and main vehicle, is calculated from vehicle using GPS historical informations by the APP customized
Velocity information, using smart mobile phone gyroscope angle changing of the record vehicle longitudinal axis in map coordinates system and using plus
Speedometer collects vehicle acceleration information.
The relative characteristic of the main vehicle of synchronization and nearby vehicle is compiled using the form of feature vector in the step 1
Code, specifically includes following steps：
Step 1.1, it is respectively (xp to define t moment and be tracked vehicle and main wheel paths point coordinatest, ypt)、(xht, yht),
Speed is respectively upt、uht, acceleration is respectively apt、aht, the vehicle longitudinal axis and the angle of map coordinates system Y-axis positive axis are α pt、
αht, two vehicles are horizontal, longitudinally opposed distance is Δ xt=xpt-xht、Δyt=ypt-yht, relative velocity is Δ ut=upt-uht, phase
It is Δ a to accelerationt=apt-aht, vehicle longitudinal axis relative angle is Δ αt=| α pt-αht|, the angular bisector of longitudinal axis angle with
The angle of map coordinates system Y-axis positive axis is βt=α ht+(apt-αht)/2；
Step 1.2, it acquires after a large amount of data by statistical analysis rejecting abnormalities value and to 6 spies described in step 1.1
(laterally opposed distance, longitudinally opposed distance, relative velocity, relative acceleration, vehicle longitudinal axis relative angle, the longitudinal axis are opposite to be pressed from both sides sign
The angular bisector region at angle) division of making interval, keep the region of division more representative by optimizing and revising；
It is vertical to finally obtain laterally opposed distance region, longitudinally opposed distance region, relative velocity region, relative acceleration region, vehicle
Axis relative angle region, longitudinal axis relative angle angular bisector region quantity be respectively n1,n2,n3,n4,n5,n6, then can be with
A use of dimension is N=n1+n2+n3+n4+n5+n6Feature vector x<t>Come indicate a certain moment be tracked vehicle and main vehicle it
Between relative priority is denoted as by 1 when dividing region where feature meets, is otherwise denoted as 0, institute for the element in feature vector
Possible feature vector shares D=n1×n2×n3×n4n5×n6Kind.
A complete vehicle behavior can be indicated in the step 1 using an eigenmatrix, specially：A certain week
The time that side vehicle completes a complete behavior experience is T, then this complete behavior can use the N that T N-dimensional vector is constituted
The eigenmatrix of × T indicates.
The corresponding behavior of its in the step 1 carries out handmarking in the form of label vector, specially：With one 8
The label column vector y of dimension<t>To characterize the affiliated behavior of nearby vehicle, 0~7 element, 8 kinds of periphery vehicles that corresponding front is concluded respectively
Behavior, meet belonging to behavior when rubidium marking be 1, remaining is labeled as 0.
RNN in the step 1 is specially：
RNN, that is, Recognition with Recurrent Neural Network, the present invention is using multi input to the Recognition with Recurrent Neural Network singly exported, structure such as Fig. 2 institutes
Show, the structure simplified is as shown in figure 3, wherein x<t>The feature vector of time step hidden layer, while the hidden layer are corresponded to for input
Also the hidden layer activation value a of previous step can be received<t-1>, wherein a<0>It is directly generally initialized as null vector, finally output prediction knot
FruitIt wherein inputs, activate, output has corresponding weight matrix Wax, Waa,Way, indicate the input value of last time as this
The weight of secondary output valve.
As Fig. 2,3 communication process in, have：
a<t>=g1(Waaa<t-1>+Waxx<t>+ba) (2)
Wherein ba,byIt is two straggling parameters, activation primitive g1Select tan functions, g2Softmax is selected to return, processing is more
Classification problem, wherein activation primitive are used for that non-linear factor is added to model, and the disadvantage of two classification can only be handled by solving linear model
End.
Nearby vehicle relative characteristic data (eigenmatrix and its corresponding label that will be acquired, marked in the step 1
Vector) input as RNN parameter learnings, update model parameter, specially：By the characteristic input initialization in training set
In RNN models afterwards, setup cost function is returned with Softmax, using gradient descent method iteration value at cost is minimized, repeatedly
Final RNN models are obtained after the completion of generation.
The detailed process of the step 2 includes the following steps：
Step 2.1, it is tracked vehicle and main vehicle is obtained from vehicle driving information using intelligent mobile phone sensor in real time；
Step 2.2, tracked target vehicle will give main vehicle by 4G communications from vehicle information real-time Transmission；
Step 2.3, main vehicle is combined from vehicle information and target vehicle information, extracts new eigenmatrix as trained
The input of RNN, RNN export prediction resultDetermine the affiliated behavior type of nearby vehicle.
Beneficial effects of the present invention：
(1) it is obtained using smart mobile phone as running data, the hardware of information transmission, has cheapness advantage；
(2) it uses the form of feature vector to encode the relative priority of driving vehicle, enriches acquired feature letter
Breath；
(3) it is good at handling the characteristic of higher dimensional matrix operation using RNN so that even if still in the case where operation is very complicated
It can guarantee that algorithm has higher real-time；
(4) using Softmax as the grader of RNN models, non-linear factor is added for model, more classification are effectively treated and ask
Topic.
(5) nearby vehicle running data is obtained in a manner of passively receiving information, avoids active probe by traffic, ring
The influence of border factor；
(6) using smart mobile phone as the hardware of data acquisition, information transmission, feasibility and convenience with height；
(7) it utilizes RNN for the support of higher dimensional matrix operation, more classification problems is handled using Softmax as grader,
While abundant vehicle is with respect to travelling characteristic, accuracy, the real-time of recognizer are effectively improved.
Description of the drawings
Fig. 1 is the flow chart of the present invention；
Fig. 2 is the structure chart of RNN in the present invention；
Fig. 3 is the simple structure figure of RNN in the present invention.
Specific implementation mode
The present invention is described in detail with specific embodiment below in conjunction with the accompanying drawings.The present embodiment is with technical solution of the present invention
Premised on implemented.Detailed real-time mode and specific operation process are given, protection scope of the present invention is not limited to following
Embodiment.
It sets all vehicle individuals for participating in vehicle behavior identification and is well placed smart mobile phone all in accordance with specific position, intelligence
Mobile phone has the function of each item data needed for acquisition, can form information mutual communication between smart mobile phone, can inform vehicle identification each other
And exchange data；Each vehicle can not only be used as and be tracked vehicle, but also can be used as main vehicle；Once having set main vehicle, the Chinese herbaceous peony
Vehicle adjacent around left and right is set to tracked target vehicle afterwards；Since the distance of Adjacent vehicles is closer, set maximum logical
Communication distance is 250m.
As shown in Figure 1, a kind of nearby vehicle Activity recognition method based on smart mobile phone and RNN, including：
Step 1, off-line training link：Typical nearby vehicle behavior is concluded and divided, is utilized based on real-time traffic scene
The running data for collecting nearby vehicle by the smart mobile phone that specific position is put in vehicle, using the form of feature vector to same
One moment main vehicle and the relative characteristic of nearby vehicle are encoded, it is possible thereby to indicate one completely using an eigenmatrix
Vehicle behavior, corresponding behavior carries out handmarking in the form of label vector；The nearby vehicle that will be acquired, marked
Input of the relative characteristic data (eigenmatrix and its corresponding label vector) as RNN parameter learnings updates model parameter；
Step 2, on-line checking link：Tracked target vehicle passes through smart mobile phone reality by collected from vehicle driving information
When be transferred to smart mobile phone on main vehicle, main vehicle generates new eigenmatrix in conjunction with two vehicle relative characteristics, utilizes trained RNN
It distinguishes and is tracked the affiliated behavior pattern of vehicle.
It is concluded in the step 1 and divides typical nearby vehicle behavior and be specially：Typical nearby vehicle behavior is drawn
It is divided into：Front truck behavior：Braking, rear car behavior：With speeding, the behavior of arranged on left and right sides vehicle：Lane-change is overtaken other vehicles, doubling.
Periphery vehicle is collected by the smart mobile phone that specific position is put using in vehicle based on real-time traffic scene in step 1
Running data be specially：Smart mobile phone is parallel with vehicle transverse and longitudinal axis, thus utilizes the collected each item data of smart mobile phone
It resolves vehicular characteristics data, i.e., is that tracked target vehicle and main vehicle establish unified map reference by the GPS of smart mobile phone
System utilizes velocity information of the GPS historical informations measuring and calculating from vehicle by the APP customized, the gyroscope of smart mobile phone is utilized to record vehicle
Angle changing and utilization accelerometer of the longitudinal axis in map coordinates system collect vehicle acceleration information.
The relative characteristic of the main vehicle of synchronization and nearby vehicle is compiled using the form of feature vector in the step 1
Code, specifically includes following steps：
Step 1.1, it is respectively (xp to define t moment and be tracked vehicle and main wheel paths point coordinatest, ypt)、(xht, yht),
Speed is respectively upt、uht, acceleration is respectively apt、aht, the vehicle longitudinal axis and the angle of map coordinates system Y-axis positive axis are α pt、
αht, two vehicles are horizontal, longitudinally opposed distance is Δ xt=xpt-xht、Δyt=ypt-yht, relative velocity is Δ ut=upt-uht, phase
It is Δ a to accelerationt=apt-aht, vehicle longitudinal axis relative angle is Δ αt|αpt-αht|, angular bisector and the ground of longitudinal axis angle
The angle of figure coordinate system Y-axis positive axis is βt=α ht(αpt-αht)/2；
Step 1.2, it acquires after a large amount of data by statistical analysis rejecting abnormalities value and to 6 spies described in step 1.1
Sign makes the division of interval, keeps the region of division more representative by optimizing and revising；Finally obtain it is laterally opposed away from
From region, longitudinally opposed distance region, relative velocity region, relative acceleration region, vehicle longitudinal axis relative angle region, the longitudinal axis
The angular bisector region quantity of relative angle is respectively n1,n2,n3,n4,n5,n6, then it is N=n that can use a dimension1
+n2+n3+n4+n5+n6Feature vector x<t>Indicate that a certain moment is tracked relative priority between vehicle and main vehicle, for spy
Element in sign vector is denoted as 1 when dividing region where feature meets, and is otherwise denoted as 0, all possible feature vector is shared
D=n1×n2×n3×n4×n5×n6Kind, in the present embodiment, laterally opposed distance [- 6,6] is pressed to 0.5 spacing average mark
It is additional (- ∞, -6) at 24 regions, totally 26 regions (6 ,+∞), unit m；Longitudinally opposed distance is divided into (- ∞ ,-
50),[-50,-20),[-20,-10),[-10,-5),[-5,-4),[-4,-3),[-3,-2),[-2,-1),[-1,-0),[0,
1), [1,2), [2,3), [4,5), [5,10), [10,20), [20,50), [50 ,+∞) totally 18 regions, unit m；It will be opposite
Speed is divided into (- ∞, -20), and [- 20, -10), [- 10, -5), [- 5,0), [0,5), [5,10), [10,20), [20 ,+∞) altogether
8 regions, unit km/h；Relative acceleration is divided into (- ∞, -6), [- 6, -3), [- 3,0), [0,3), [3,6), [6 ,+
∞) totally 6 regions, unit m/s2；Vehicle longitudinal axis relative angle [0,90] is divided into 9 regions by 10 spacing, it is single
Position is degree；Semiaxis is born to Y-axis be divided into 18 regions as longitudinal axis folder by 10 spacing from map coordinates system Y-axis positive axis
The partitioning standards of the angular bisector at angle.
A complete vehicle behavior, specially a certain week can be indicated in the step 1 using an eigenmatrix
The time that side vehicle completes a complete behavior experience is T, then this complete behavior can use the N that T N-dimensional vector is constituted
The eigenmatrix of × T indicates.
It is specially to be tieed up with one 8 that the corresponding behavior of its in the step 1 carries out handmarking in the form of label vector
Label column vector y<t>To characterize the affiliated behavior of nearby vehicle, 0~7 element, 8 kinds of vehicle-surroundings that corresponding front is concluded respectively
Vehicle behavior, rubidium marking is 1 when meeting affiliated behavior, remaining is labeled as 0.
RNN in the step 1 is specially：
RNN, that is, Recognition with Recurrent Neural Network, a kind of multi input is to the Recognition with Recurrent Neural Network structure singly exported as shown in Fig. 2, it is simple
Single structure is as shown in figure 3, wherein x<t>The feature vector of time step hidden layer is corresponded to for input, while the hidden layer can also receive
The hidden layer activation value a of previous step<t-1>, wherein a<0>It is directly generally initialized as null vector, finally exports prediction result
It wherein inputs, activate, output has corresponding weight matrix Wax, Waa,Way, indicate that last input value is used as and this time export
The weight of value.
As Fig. 2,3 communication process in, have：
a<t>=g1(Waaa<t-1>+Waxx<t>+ba) (2)
Wherein ba,byIt is two straggling parameters, activation primitive g1Select tanh, g2It selects Softmax to return, handles more points
Class problem, wherein activation primitive are used for that non-linear factor is added to model, solve the drawbacks of linear model can only handle two classification.
Nearby vehicle relative characteristic data (eigenmatrix and its corresponding label that will be acquired, marked in the step 1
Vector) input as RNN parameter learnings, updating model parameter is specially：By the characteristic input initialization in training set
In RNN models afterwards, setup cost function is returned with Softmax, using gradient descent method iteration value at cost is minimized, repeatedly
Final RNN models are obtained after the completion of generation.
The detailed process of the step 2 includes the following steps：
Step 2.1, it is tracked vehicle and main vehicle is obtained from vehicle driving information using intelligent mobile phone sensor in real time；
Step 2.2, tracked target vehicle will give main vehicle by 4G communications from vehicle information real-time Transmission；
Step 2.3, main vehicle is combined from vehicle information and target vehicle information, extracts new eigenmatrix as trained
The input of RNN, RNN export prediction resultDetermine the affiliated behavior type of nearby vehicle.
The series of detailed descriptions listed above only for the present invention feasible embodiment specifically
Bright, they are all without departing from equivalent implementations made by technical spirit of the present invention not to limit the scope of the invention
Or change should all be included in the protection scope of the present invention.
Claims (10)
1. a kind of nearby vehicle Activity recognition method based on smart mobile phone and RNN, which is characterized in that include the following steps：
Step 1, off-line training link：Typical nearby vehicle behavior is concluded and divided, is utilized in vehicle based on real-time traffic scene
The running data for collecting nearby vehicle in by the smart mobile phone that specific position is put, using the form of feature vector to same a period of time
The relative characteristic for carving main vehicle and nearby vehicle is encoded, and a complete vehicle behavior is indicated using an eigenmatrix,
Its corresponding behavior carries out handmarking in the form of label vector；The nearby vehicle relative characteristic number that will be acquired, marked
Input according to (including eigenmatrix and its corresponding label vector) as RNN parameter learnings updates model parameter；
Step 2, on-line checking link：Tracked target vehicle is passed from vehicle driving information by smart mobile phone in real time by collected
The smart mobile phone being defeated by main vehicle, main vehicle generate new eigenmatrix in conjunction with two vehicle relative characteristics, are distinguished using trained RNN
It is tracked the affiliated behavior pattern of vehicle.
2. a kind of nearby vehicle Activity recognition method based on smart mobile phone and RNN according to claim 1, feature exist
In being concluded in the step 1 and divide typical nearby vehicle behavior, specially：Typical nearby vehicle behavior is divided into：
Front truck behavior：Braking, rear car behavior：With speeding, the behavior of left and right sides vehicle：Lane-change is overtaken other vehicles, doubling.
3. a kind of nearby vehicle Activity recognition method based on smart mobile phone and RNN according to claim 1, feature exist
In collecting periphery vehicle using the smart mobile phone put by specific position in vehicle based on real-time traffic scene in the step 1
Running data, specially：Smart mobile phone is parallel with vehicle transverse and longitudinal axis, and vehicle is resolved using the collected data of smart mobile phone
Characteristic is that tracked target vehicle and main vehicle establish unified map coordinates system by the GPS of smart mobile phone, by APP
Using the measuring and calculating of GPS historical informations from the velocity information of vehicle, the vehicle longitudinal axis is recorded in map reference using the gyroscope of smart mobile phone
Angle changing and utilization accelerometer in system collect vehicle acceleration information.
4. a kind of nearby vehicle Activity recognition method based on smart mobile phone and RNN according to claim 1, feature exist
In being encoded, had to the relative characteristic of the main vehicle of synchronization and nearby vehicle using the form of feature vector in the step 1
Body includes the following steps：
Step 1.1, it is respectively (xp to define t moment and be tracked vehicle and main wheel paths point coordinatest, ypt)、(xht, yht), speed point
It Wei not upt、uht, acceleration is respectively apt、aht, the vehicle longitudinal axis and the angle of map coordinates system Y-axis positive axis are α pt、αht, two
Che Heng, longitudinally opposed distance are respectively Δ xt=xpt-xht、Δyt=ypt-yht, relative velocity is Δ ut=upt-uht, relatively
Acceleration is Δ at=apt-aht, vehicle longitudinal axis relative angle is Δ αt=| α pt-αht|, angular bisector and the ground of longitudinal axis angle
The angle of figure coordinate system Y-axis positive axis is βt=α ht+(αpt-αht)/2；
Step 1.2, it is made by statistical analysis rejecting abnormalities value and to 6 features in step 1.1 after acquiring a large amount of data
The division of interval keeps the region of division more representative by optimizing and revising；Finally obtain laterally opposed distance region,
Longitudinally opposed distance region, relative velocity region, relative acceleration region, vehicle longitudinal axis relative angle region, the opposite folder of the longitudinal axis
The angular bisector region quantity at angle is respectively n1,n2,n3,n4,n5,n6, then it is N=n that can use a dimension1+n2+n3+
n4+n5+n6Feature vector x<t>Indicate that a certain moment is tracked relative priority between vehicle and main vehicle, for feature vector
In element, when feature meet where divide region when be denoted as 1, be otherwise denoted as 0, all possible feature vector shares D=n1
×n2×n3×n4×n5×n6Kind.
5. a kind of nearby vehicle Activity recognition method based on smart mobile phone and RNN according to claim 1, feature exist
In indicating a complete vehicle behavior using an eigenmatrix in the step 1, specially：A certain nearby vehicle is complete
Time at a complete behavior experience is T, then this complete behavior can use the spy of N × T of T N-dimensional vector composition
Matrix is levied to indicate.
6. a kind of nearby vehicle Activity recognition method based on smart mobile phone and RNN according to claim 5, feature exist
In the corresponding behavior of its in the step 1 carries out handmarking in the form of label vector, specially：The mark tieed up with one 8
Sign column vector y<t>To characterize the affiliated behavior of nearby vehicle, 0~7 element, 8 kinds of vehicle-surroundings vehicle rows that correspondence is concluded respectively
For rubidium marking is 1 when meeting affiliated behavior, remaining is labeled as 0.
7. a kind of nearby vehicle Activity recognition method based on smart mobile phone and RNN according to claim 1, feature exist
In the RNN in the step 1 is designed specifically to：
RNN using multi input to the Recognition with Recurrent Neural Network that singly exports, by x<t>As the corresponding time step hidden layer of input feature to
Amount, while the hidden layer can also receive the hidden layer activation value a of previous step<t-1>, wherein a<0>It directly is initialized as null vector, most
After export prediction resultIt wherein inputs, activate, output has corresponding weight matrix Wax, Waa,Way, indicate the defeated of last time
Enter weight of the value as this time output valve；
Wherein communication process is as follows：
a<t>=g1(Waaa<t-1>+Waxx<t>+ba)
Wherein ba,byIt is straggling parameter, activation primitive g respectively1Select tanh, g2Softmax is selected to return, wherein activation primitive is used
That non-linear factor is added to model.
8. a kind of nearby vehicle Activity recognition method based on smart mobile phone and RNN according to claim 1, feature exist
In by the nearby vehicle relative characteristic data for acquiring, having marked (eigenmatrix and its corresponding label vector) in the step 1
As the input of RNN parameter learnings, model parameter is updated, specially：After the characteristic input initialization in training set
In RNN models, setup cost function is returned with Softmax, using gradient descent method iteration value at cost is minimized, iteration is complete
Final RNN models are obtained after.
9. a kind of nearby vehicle Activity recognition method based on smart mobile phone and RNN according to claim 1, feature exist
In the detailed process of the step 2 includes the following steps：
Step 2.1, it is tracked vehicle and main vehicle is obtained from vehicle driving information using intelligent mobile phone sensor in real time；
Step 2.2, tracked target vehicle will give main vehicle by wireless transmission from vehicle information real-time Transmission；
Step 2.3, main vehicle is combined from vehicle information and target vehicle information, extracts new eigenmatrix as trained RNN's
Input, RNN export prediction resultDetermine the affiliated behavior type of nearby vehicle.
10. a kind of nearby vehicle Activity recognition method based on smart mobile phone and RNN according to claim 9, feature
It is, wireless transmission is using 4G networks in the step 2.2.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810320788.2A CN108470460B (en) | 2018-04-11 | 2018-04-11 | Peripheral vehicle behavior identification method based on smart phone and RNN |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810320788.2A CN108470460B (en) | 2018-04-11 | 2018-04-11 | Peripheral vehicle behavior identification method based on smart phone and RNN |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108470460A true CN108470460A (en) | 2018-08-31 |
| CN108470460B CN108470460B (en) | 2020-08-28 |
Family
ID=63263155
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810320788.2A Active CN108470460B (en) | 2018-04-11 | 2018-04-11 | Peripheral vehicle behavior identification method based on smart phone and RNN |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108470460B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109284701A (en) * | 2018-09-06 | 2019-01-29 | 南京威尔思汽车部件科技有限公司 | A kind of driving recognition methods based on regional correlation |
| CN109727490A (en) * | 2019-01-25 | 2019-05-07 | 江苏大学 | A kind of nearby vehicle behavior adaptive corrective prediction technique based on driving prediction field |
| CN109948654A (en) * | 2019-02-15 | 2019-06-28 | 山东师范大学 | Motoring condition recognition methods, system and equipment based on user behavior data |
| WO2020151059A1 (en) * | 2019-01-22 | 2020-07-30 | 江苏大学 | Method for recognizing surrounding vehicle behaviors in complex road conditions based on hmm-svm two-layer improved model |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100128975A1 (en) * | 2008-11-24 | 2010-05-27 | Toyota Motor Engineering & Manufacturing North America, Inc. | Method and system for object recognition based on a trainable dynamic system |
| CN105719313A (en) * | 2016-01-18 | 2016-06-29 | 中国石油大学(华东) | Moving object tracking method based on intelligent real-time video cloud |
| CN106990714A (en) * | 2017-06-05 | 2017-07-28 | 李德毅 | Adaptive Control Method and device based on deep learning |
| CN107133974A (en) * | 2017-06-02 | 2017-09-05 | 南京大学 | The vehicle type classification method that Gaussian Background modeling is combined with Recognition with Recurrent Neural Network |
| CN107492251A (en) * | 2017-08-23 | 2017-12-19 | 武汉大学 | It is a kind of to be identified and driving condition supervision method based on the driver identity of machine learning and deep learning |
| CN107563332A (en) * | 2017-09-05 | 2018-01-09 | 百度在线网络技术（北京）有限公司 | For the method and apparatus for the driving behavior for determining unmanned vehicle |
-
2018
- 2018-04-11 CN CN201810320788.2A patent/CN108470460B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100128975A1 (en) * | 2008-11-24 | 2010-05-27 | Toyota Motor Engineering & Manufacturing North America, Inc. | Method and system for object recognition based on a trainable dynamic system |
| CN105719313A (en) * | 2016-01-18 | 2016-06-29 | 中国石油大学(华东) | Moving object tracking method based on intelligent real-time video cloud |
| CN107133974A (en) * | 2017-06-02 | 2017-09-05 | 南京大学 | The vehicle type classification method that Gaussian Background modeling is combined with Recognition with Recurrent Neural Network |
| CN106990714A (en) * | 2017-06-05 | 2017-07-28 | 李德毅 | Adaptive Control Method and device based on deep learning |
| CN107492251A (en) * | 2017-08-23 | 2017-12-19 | 武汉大学 | It is a kind of to be identified and driving condition supervision method based on the driver identity of machine learning and deep learning |
| CN107563332A (en) * | 2017-09-05 | 2018-01-09 | 百度在线网络技术（北京）有限公司 | For the method and apparatus for the driving behavior for determining unmanned vehicle |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109284701A (en) * | 2018-09-06 | 2019-01-29 | 南京威尔思汽车部件科技有限公司 | A kind of driving recognition methods based on regional correlation |
| WO2020151059A1 (en) * | 2019-01-22 | 2020-07-30 | 江苏大学 | Method for recognizing surrounding vehicle behaviors in complex road conditions based on hmm-svm two-layer improved model |
| CN109727490A (en) * | 2019-01-25 | 2019-05-07 | 江苏大学 | A kind of nearby vehicle behavior adaptive corrective prediction technique based on driving prediction field |
| CN109727490B (en) * | 2019-01-25 | 2021-10-12 | 江苏大学 | Peripheral vehicle behavior self-adaptive correction prediction method based on driving prediction field |
| CN109948654A (en) * | 2019-02-15 | 2019-06-28 | 山东师范大学 | Motoring condition recognition methods, system and equipment based on user behavior data |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108470460B (en) | 2020-08-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114384920B (en) | Dynamic obstacle avoidance method based on real-time construction of local grid map | |
| CN108470460A (en) | A kind of nearby vehicle Activity recognition method based on smart mobile phone and RNN | |
| CN113954864B (en) | Intelligent automobile track prediction system and method integrating peripheral automobile interaction information | |
| CN112347993B (en) | Expressway vehicle behavior and track prediction method based on vehicle-unmanned aerial vehicle cooperation | |
| CN105956560B (en) | A kind of model recognizing method based on the multiple dimensioned depth convolution feature of pondization | |
| CN110796856A (en) | Vehicle lane change intention prediction method and training method of lane change intention prediction network | |
| CN107492251A (en) | It is a kind of to be identified and driving condition supervision method based on the driver identity of machine learning and deep learning | |
| CN106599827A (en) | Small target rapid detection method based on deep convolution neural network | |
| CN109285348A (en) | A kind of vehicle behavior recognition methods and system based on two-way length memory network in short-term | |
| CN109727490A (en) | A kind of nearby vehicle behavior adaptive corrective prediction technique based on driving prediction field | |
| CN113044064B (en) | Vehicle self-adaptive automatic driving decision method and system based on meta reinforcement learning | |
| CN107967486A (en) | A kind of nearby vehicle Activity recognition method based on V2V communications with HMM-GBDT mixed models | |
| CN110288835A (en) | A kind of nearby vehicle behavior real-time identification method based on kinematics predictive compensation mechanism | |
| Hu et al. | Learning a deep cascaded neural network for multiple motion commands prediction in autonomous driving | |
| CN105989062A (en) | Defining method based on electric vehicle travelling track characteristics and data mining technology | |
| CN113238970B (en) | Training method, evaluation method, control method and device of automatic driving model | |
| Wang et al. | Multi-information-based convolutional neural network with attention mechanism for pedestrian trajectory prediction | |
| CN110866477A (en) | Track motion mode identification method and device | |
| CN109389246B (en) | Neural network-based vehicle destination area range prediction method | |
| CN111474934A (en) | Personalized cruise control system based on high-precision map | |
| CN116202533B (en) | Vehicle track navigation planning method and device based on terrain reconstruction | |
| CN109508686A (en) | A kind of Human bodys' response method based on the study of stratification proper subspace | |
| Wang et al. | End-to-end self-driving approach independent of irrelevant roadside objects with auto-encoder | |
| CN115062202A (en) | Method, device, equipment and storage medium for predicting driving behavior intention and track | |
| CN114516336A (en) | Vehicle track prediction method considering road constraint conditions |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |