CN108550279B - Vehicle drive behavior prediction method based on machine learning - Google Patents
Vehicle drive behavior prediction method based on machine learning Download PDFInfo
- Publication number
- CN108550279B CN108550279B CN201810287172.XA CN201810287172A CN108550279B CN 108550279 B CN108550279 B CN 108550279B CN 201810287172 A CN201810287172 A CN 201810287172A CN 108550279 B CN108550279 B CN 108550279B
- Authority
- CN
- China
- Prior art keywords
- vehicle
- lane
- feature
- crossroad
- pld
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/16—Anti-collision systems
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W40/00—Estimation or calculation of non-directly measurable driving parameters for road vehicle drive control systems not related to the control of a particular sub unit, e.g. by using mathematical models
- B60W40/08—Estimation or calculation of non-directly measurable driving parameters for road vehicle drive control systems not related to the control of a particular sub unit, e.g. by using mathematical models related to drivers or passengers
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Evolutionary Biology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Software Systems (AREA)
- Automation & Control Theory (AREA)
- Transportation (AREA)
- Mechanical Engineering (AREA)
- Traffic Control Systems (AREA)
Abstract
Vehicle drive behavior prediction method based on machine learning, it is related to car networking field, it is intended to using machine learning techniques, the relationship between digging vehicle attribute, road information and running environment information and vehicle drive behavior improves the accuracy of vehicle drive behavior prediction.Specific steps include: step 1, defined feature collection: vehicle characteristics definition, roadway characteristic definition, vehicle running environment definition；Step 2, vehicle movement prediction model: feature extraction and data prediction: vehicle and junction ahead distance feature are extracted, and crossing allows go to action feature extraction, tag extraction；Vehicle movement prediction model: training sample set definition, the training of vehicle movement prediction model；Step 3, vehicle drive behavior prediction model: Gaussian component definition；Step 4, vehicle drive behavior prediction.
Description
Technical field
The present invention relates to car networking fields, and in particular to a kind of vehicle drive behavior prediction method based on machine learning.
Background technique
Vehicle drive behavior prediction serves primarily in the safety-related application of car networking, as crossroad vehicle collision avoidance is supervised
Survey etc..The existing research about vehicle drive behavior prediction, the main historical behavior track for considering vehicle and vehicle driving
Traffic information.These researchs have ignored vehicle self attributes (such as vehicle to modeling idealization of vehicle and place traffic information
Height is wide), vehicle to crossing distance, traffic signals and lane turn to allow to indicate etc. influence vehicle drive behavior it is important because
Element has relatively large deviation to the prediction result of driving behavior and actual conditions so as to cause in real urban road.
The present invention comprehensively considers many factors of above-mentioned influence vehicle drive behavior, utilizes machine learning techniques, wheeled digging machine
Relationship between attribute, road information and running environment information and vehicle drive behavior improves vehicle drive behavior prediction
Accuracy.
Summary of the invention
To vehicle body attribute, road information and running environment and row is driven for existing vehicle drive behavior prediction model
The problem of relation excavation deficiency between, the present invention comprehensively consider the important shadow such as vehicle attribute, road information and running environment
The factor of sound, proposes a kind of vehicle drive behavior prediction method based on machine learning.
Technical solution of the present invention are as follows:
A kind of vehicle drive behavior prediction method based on machine learning, which is characterized in that use full Connection Neural Network
Prediction vehicle movement clusters driving behavior using gauss hybrid models according to the displacement of prediction.Specific method includes such as
Lower step:
Step 1, defined feature collection, including vehicle characteristics definition, roadway characteristic definition, vehicle running environment characterizing definition.
Step 11, vehicle characteristics define
Vehicle characteristics include length of wagon L, body width W, and car speed, acceleration, current driving direction, crossing turns to
Movement, wherein t moment car speed, acceleration are such as respectively labeled as v (t) and a (t), remaining feature is defined respectively as:
Define 1 vehicle heading vDir (t), indicate t moment direction of vehicle movement, between direct north clockwise
Angle indicates, meets:
360 ° of 0≤vDir (t) < (4)
Define 2 vehicle intersections and turn to vMov (t), indicate driving behavior of t moment vehicle when by crossroad, with to
The form of amount characterizes, since crossroad does not allow to reverse end for end, the case where being presently considered vehicle straight trip, turn left, turn right, such as
Formula (2).
3 current vehicle position P (t) are defined, indicate t moment vehicle in CA State Plane III in NAD83 coordinate
Two-dimensional coordinate vector in system, vector items unit are foot (ft).Location information is defined as follows:
P (t)=(x (t), y (t)) (3)
CA State Plane III in NAD83 coordinate system is 1983 North America datum level (NAD) coordinate systems.
To sum up, the feature set feature of t moment vehiclev(t) it is defined as follows:
featurev(t)={ L, W, v (t), a (t), vDir (t), vMov (t), P (t) } (4)
Step 12, roadway characteristic defines
Crossroad or T-shaped road junction are abstracted as quadrangle, with its four apex coordinates since the angle of direction northwest by
According to successively identifying clockwise, mark as follows:
I=(x1, y1, x2, y2, x3, y3, x4, y4) (5)
Defining 4 crossroad set ISet indicates in survey region, the set of all crossroad compositions.
ISet=(I1, I2..., Im...) and (6)
It defines 5 road segment segments and refers to a section of the road between two adjacent crossroads, by road segment segment two sides four crossway
Mouthful identify the road segment segment.The road segment segment set for defining road i is as follows:
RSegSet (i)=(I1I2, I2I3..., IkIm...) and (7)
Wherein Ik(1≤k≤n, n are maximum crossroad number) indicates the crossroad that number is k.
Every road segment segment includes several lanes, defines crossroad Ik, ImBetween road segment segment lane set it is as follows:
IkIm=(lid1, lid2..., lidn) (8)
Wherein lidk(1≤k≤n) indicates lane number.
Defining 6 track direction lDir (x, y) indicates the direction of travel that lane allows, and wherein x, y indicate the position in lane
Coordinate.Angle defines same vDir.Vehicle is identical as track direction in the driving direction vDir of current lane straight way, even works as front truck
It is in the lane i, then has
VDir (t)=lDiri(x, y), wherein (x, y)=P (t) (9)
Define on the left of 7 lanes can lane change quantity LAL (left available lanes) be under certain lane current location,
The number of lanes that vehicle can change to the left.By the current location (x, y) of current lane i Lane Searches in the same direction to the left, until meeting
Indicate that vehicle can not lane change to the left or until searching leftmost side lane in the same direction to solid line.The number of lanes searched is vehicle
It can lane change quantity LAL on the left of roadi(x, y).
Define on the right side of 8 lanes can lane change amount R AL (right available lanes) be in certain lane current location
Under, number of lanes that vehicle can change to the right.By current location (x, the y) Lane Searches in the same direction to the right of current lane i, until
Encountering solid line indicates that vehicle can not lane change to the right or until searching rightmost side lane in the same direction.The number of lanes searched is
It can lane change amount R AL on the right side of lanei(x, y).
Defining 9 straight trip area lanes allows driving behavior to include straight trip, lane change to the left and to the right lane change.Definition vector sld table
The basic driving behavior for showing that lane allows is as follows:
The driving behavior SLD that straight trip area lane allows has
The driving behavior for defining the permission of 10 crossing area in preparation lanes includes straight trip, is turned left, right-hand bend and u-turn.Define to
It is as follows to measure the basic driving behavior that pld indicates that lane allows:
The driving behavior PLD that crossing area in preparation lane allows can be indicated are as follows:
PLD (x, y)=β1(x, y) pldst+β2pldtl+β3pldtr+β4pldta+β5pldsp (13)
Wherein { β1, β2, β3, β4, β5|βi=0 ∨ βi=1,1≤i≤5, i ∈ N }
βiFor the probability coefficent of certain driving behavior, otherwise it is 0 that choosing, which is 1,.
To sum up, the roadway characteristic collection feature at position (x, y)r(x, y) can be defined as follows:
Step 13, vehicle running environment defines
Define 11 crossing distances, vehicle i and front crossroad ImDistanceIndicate with the front vehicle i along with work as
The distance between crossroad stop line in front of preceding lane.
Define 12t moment, traffic lights TLiSignal, which allows to act, uses vector sigi(t) it indicates.
The crossing for defining 13 vehicle t moments allows go to action to be expressed as IAM (t).This feature is limited by lane permission
Driving behavior PLD and traffic light signals allow to act sigi(t).It is expressed as the Hadamard product of matrix, such as formula (16).
IAM (t)=PLD (P (t)) * sigi(t) (16)
To sum up, t moment vehicle running environment feature set featuree(t) it is defined as follows:
featuree(t)={ VID (t), IAM (t) } (17)
Combining step 11, step 12 and step 13, t moment influence feature set feature (t) definition of vehicle drive behavior
For
Feature (t)=featurev(t)∪featurer(P(t))∪featuree(t) (18)
Step 2, vehicle movement prediction model
Step 21, feature extraction and data prediction
The vehicle characteristics directly acquired include Vehicle length L, vehicle width W, car speed v, vehicle acceleration a, vehicle
Driving direction vDir, vehicle intersection turn to vMov, current vehicle position.
The roadway characteristic directly acquired includes track direction lDir, on the left of lane can lane change quantity LAL, it is variable on the right side of lane
Road amount R AL, the driving behavior SLD that lane allows, the driving behavior PLD that crossing area in preparation lane allows.
Crossroad collection ISet, road segment segment set RSegSet, the driver behavior that traffic light signals allow can directly be obtained
Sig (t), the lane lane where each car and the road segment segment RSeg where each car.
According to the definition of step 1, the feature for needing to extract includes vehicle and front crossroad distance VID, vehicle intersection
Allow go to action IAM, training sample label.
Step 211, vehicle and junction ahead distance feature are extracted
The lane is obtained in front according to track direction lDir where lane lane where vehicle and vehicle on map
The coordinate of two endpoints of stop line AB at crossroad, i.e. A point coordinate (xA, yA) and B point coordinate (xB, yB).From vehicle leading edge
Vertical line is done to straight line AB, acquires length of perpendicular length.Due to road approximation straight way in the data set of research, length can be used
Approximate substitution VID.
AB meets formula (19) in two-dimensional coordinate system.
(yA-yB)·x+(xB-xA)·y+(yB·xA-xB·yA)=0 (19)
Assuming that vehicle location P (t)=(x at this timeC, yC), then the distance length of vehicle to stop line AB meets formula
(20)。
It concentrates the data of each moment point of each car to calculate according to formula (20) data, obtains feature VID.
Step 212, crossing allows go to action feature extraction
The table that crossroad signal lamp changes over time is obtained from data set, i.e., for any crossroad IiIt can be with
Obtain sigi(t).Available lane allows driving behavior PLD from the map lane markings in data set.Then IAM feature
It extracts according to defining 13, sequentially finds the corresponding crossroad in every lane, Hadamard product then is done to sig (t) and PLD
Obtain IAM feature
Step 213, tag extraction
Vehicle movement is denoted as disP, and (Δ x, Δ y) indicate vehicle in (displacement of the t+ Δ t) moment relative to t moment, such as public affairs
Shown in formula (21).
DisP (Δ x, Δ y)=P (x (t+ Δ t), y (t+ Δ t))-P (x (t), y (t)) (21)
Coordinate system and unit are the same as definition 3.
The character representation that will acquire is feature (t) vector of definition, and the training sample displacement tag definition that will acquire is
Label (t), wherein t indicates the time.Feature (t) and label (t) are normalized using min-max method for normalizing
Processing, as shown in formula (22).
Wherein f indicates characteristic point value, fmaxIndicate the maximum value of this feature point value, fminIndicate this feature point value
Minimum value, fNIndicate the characteristic point value after normalization.Feature (t) and label (t) after normalized remember respectively
For featureN(t) and labelN(t)。
Step 22, vehicle movement prediction model
Step 221, training sample set defines
Aforementioned pretreated training sample set is denoted as train, including training sample feature set and training sample tally set.
Training sample label is using displacement label.featureN(t) splice label in sequenceNIt (t) is train samples
Structure.Training sample feature set is denoted as train_x, and training sample tally set is denoted as train_y, is expressed as formula (23)
With formula (24).
Train_y and train_x is corresponded by row.
Step 222, vehicle movement prediction model training
Vehicle movement prediction model training step is as follows:
(4) based on full Connection Neural Network structure (full Connection Neural Network belongs to existing algorithm frame, but input data,
The network number of plies, output layer structure are defined by the present invention), using training sample set train_x as input quantity, calculated using propagated forward
Method calculates (belonging to existing algorithm) excitation value of each layer.
(5) full connection BP neural network is initialized using the resulting network structure of study, is finally adding output layer, output
For vehicle movement disP (Δ x, Δ y).
(6) using mini-batch gradient descent method (belonging to existing algorithm), network is calculated using training tally set train_y
Error, backpropagation calculate loss function to the gradient of weight matrix and bigoted item, finely tune each layer of network parameter, until
Trigger the termination condition of training.(calculation method belongs to existing method)
Step 3, vehicle drive behavior prediction model
Step 31, Gaussian component defines
5 kinds of driving behaviors are set, i.e. vehicle is kept straight on, and is turned left, and is turned right, and u-turn is as you were, respectively corresponds 5 Gausses point
Amount.According to the present invention to the definition of vehicle movement, each Gaussian component is binary Gaussian Profile.
Step 32, gauss hybrid models training
Steps are as follows:
(4) mean value to each Gaussian component and covariance matrix carry out random initializtion, the priori of each Gaussian component
Probability is set as 1/5.
(5) using training sample displacement tally set as input quantity, model is instructed using EM algorithm (belonging to existing algorithm)
Practice.
(6) weight of each Gaussian component, mean value and covariance matrix are obtained.
Step 4, vehicle drive behavior prediction
(1) sample characteristics collection to be predicted is defined first, is denoted as test_x, structure and training sample feature set train_x
It is identical.
(2) displacement prediction model for obtaining sampling feature vectors input step 2 to be predicted, the displacement predicted.
(3) the driving behavior prediction model for predicting that resulting displacement input step 3 obtains, calculating sample belongs to each high
The probability of this distribution, highest probability is driving behavior obtained by the sample predictions.
Beneficial effects of the present invention
The present invention use machine learning techniques, digging vehicle self attributes, road information, running environment and driving behavior it
Between relationship, propose vehicle drive behavior prediction model, achieve the purpose that improve vehicle drive behavior prediction accuracy.
Detailed description of the invention
Fig. 1 road area divides schematic diagram
The full Connection Neural Network structural schematic diagram of Fig. 2
Fig. 3 the method for the present invention flow chart
Specific embodiment
Specific implementation process of the invention is as shown in figure 3, include following 4 aspects:
1. vehicle characteristics, roadway characteristic, vehicle running environment characterizing definition
2. vehicle movement prediction model
3. vehicle drive behavior prediction model
4. vehicle drive behavior prediction method
Step 1, defined feature collection, including vehicle characteristics definition, roadway characteristic definition, vehicle running environment characterizing definition.
Step 11, vehicle characteristics define
Vehicle characteristics include length of wagon L, body width W, and car speed, acceleration, current driving direction, crossing turns to
Movement, wherein t moment car speed, acceleration are such as respectively labeled as v (t) and a (t), remaining feature is defined respectively as:
Define 1 vehicle heading vDir (t), indicate t moment direction of vehicle movement, between direct north clockwise
Angle indicates, meets:
360 ° of 0≤vDir (t) < (7)
Define 2 vehicle intersections and turn to vMov (t), indicate driving behavior of t moment vehicle when by crossroad, with to
The form of amount characterizes, since crossroad does not allow to reverse end for end, the case where being presently considered vehicle straight trip, turn left, turn right, such as
Formula (2).
3 current vehicle position P (t) are defined, indicate t moment vehicle in CA State Plane III in NAD83 coordinate
Two-dimensional coordinate vector in system, vector items unit are foot (ft).Location information is defined as follows:
P (t)=(x (t), y (t)) (3)
CA State Plane III in NAD83 coordinate system is 1983 North America datum level (NAD) coordinate systems.
To sum up, the feature set feature of t moment vehiclev(t) it is defined as follows:
featurev(t)={ L, W, v (t), a (t), vDir (t), vMov (t), P (t) } (4)
Step 12, roadway characteristic defines
It is classified as three parts shown in FIG. 1, including crossing area according to urban road feature, keep straight on area and crossing preparation
Area.With a distance from next crossing farther out, road broken line representation, vehicle can free lane change for road straight trip offset.Crossing prepares
Crossroad or T-shaped road junction, road solid line are closed on by area, and vehicle is unable to changing Lane.Lane has specific crossing to turn
To limitation, vehicle can only be limited as defined in lane in do crossing steering.
Crossroad or T-shaped road junction are abstracted as quadrangle, with its four apex coordinates since the angle of direction northwest by
According to successively identifying clockwise, as shown in Figure 1, label is as follows:
I=(x1, y1, x2, y2, x3, y3, x4, y4) (5)
Defining 4 crossroad set ISet indicates in survey region, the set of all crossroad compositions.
ISet=(I1, I2..., Im...) and (6)
It defines 5 road segment segments and refers to a section of the road between two adjacent crossroads, by road segment segment two sides four crossway
Mouthful identify the road segment segment.The road segment segment set for defining road i is as follows:
RSegSet (i)=(I1I2, I2I3..., IkIm...) and (7)
Wherein Ik(1≤k≤n, n are maximum crossroad number) indicates the crossroad that number is k.
Every road segment segment includes several lanes, defines crossroad Ik, ImBetween road segment segment lane set it is as follows:
IkIm=(lid1, lid2..., lidn) (8)
Wherein lidk(1≤k≤n) indicates lane number.
Defining 6 track direction lDir (x, y) indicates the direction of travel that lane allows, and wherein x, y indicate the position in lane
Coordinate.Angle defines same vDir.Vehicle is identical as track direction in the driving direction vDir of current lane straight way, even works as front truck
It is in the lane i, then has
VDir (t)=lDiri(x, y), wherein (x, y)=P (t) (9)
Define on the left of 7 lanes can lane change quantity LAL (left available lanes) be under certain lane current location,
The number of lanes that vehicle can change to the left.By the current location (x, y) of current lane i Lane Searches in the same direction to the left, until meeting
Indicate that vehicle can not lane change to the left or until searching leftmost side lane in the same direction to solid line.The number of lanes searched is vehicle
It can lane change quantity LAL on the left of roadi(x, y).
Define on the right side of 8 lanes can lane change amount R AL (right available lanes) be in certain lane current location
Under, number of lanes that vehicle can change to the right.By current location (x, the y) Lane Searches in the same direction to the right of current lane i, until
Encountering solid line indicates that vehicle can not lane change to the right or until searching rightmost side lane in the same direction.The number of lanes searched is
It can lane change amount R AL on the right side of lanei(x, y).
Defining 9 straight trip area lanes allows driving behavior to include straight trip, lane change to the left and to the right lane change.Definition vector sld table
The basic driving behavior for showing that lane allows is as follows:
The driving behavior SLD that straight trip area lane allows has
The driving behavior for defining the permission of 10 crossing area in preparation lanes includes straight trip, is turned left, right-hand bend and u-turn.Define to
It is as follows to measure the basic driving behavior that pld indicates that lane allows:
The driving behavior PLD that crossing area in preparation lane allows can be indicated are as follows:
PLD (x, y)=β1(x, y) pldst+β2pldtl+β3pldtr+β4pldta+β5pldsp (13)
Wherein { β1, β2, β3, β4, β5|βi=0V βi=1,1≤i≤5, i ∈ N }
βiFor the probability coefficent of certain driving behavior, otherwise it is 0 that choosing, which is 1,.
To sum up, the roadway characteristic collection feature at position (x, y)r(x, y) can be defined as follows:
Step 13, vehicle running environment defines
Vehicle running environment such as vehicle and junction ahead distance, traffic light signals attribute etc. to driving behavior have directly or
Person influences indirectly.
Define 11 crossing distances, vehicle i and front crossroad ImDistanceIndicate with the front vehicle i along with work as
The distance between crossroad stop line in front of preceding lane.
Define 12t moment, traffic lights TLiSignal, which allows to act, uses vector sigi(t) it indicates.
The crossing for defining 13 vehicle t moments allows go to action to be expressed as IAM (t).This feature is limited by lane permission
Driving behavior PLD and traffic light signals allow to act sigi(t).It is expressed as the Hadamard product of matrix, such as formula (16).
IAM (t)=PLD (P (t)) * sigi(t) (16)
To sum up, t moment vehicle running environment feature set featuree(t) it is defined as follows:
featuree(t)={ VID (t), IAM (t) } (17)
Combining step 11, step 12 and step 13, t moment influence feature set feature (t) definition of vehicle drive behavior
For
Feature (t)=featurev(t)∪featurer(P(t))∪featuree(t) (18)
Step 2, vehicle movement prediction model
Vehicle drive behavior is by vehicle self attributes, travel attribute and running environment attribute etc. in urban road
Many factors influence.The present invention utilizes advantage of the neural network in terms of excavating high dimensional nonlinear data, from a variety of influence vehicles
Vehicle movement prediction network model is trained in the feature of traveling behavior.Model training is divided into three parts: feature extraction, and data are pre-
Processing, displacement prediction network training.
Step 21, feature extraction and data prediction
The vehicle characteristics directly acquired include Vehicle length L, vehicle width W, car speed v, vehicle acceleration a, vehicle
Driving direction vDir, vehicle intersection turn to vMov, current vehicle position.
The roadway characteristic directly acquired includes track direction lDir, on the left of lane can lane change quantity LAL, it is variable on the right side of lane
Road amount R AL, the driving behavior SLD that lane allows, the driving behavior PLD that crossing area in preparation lane allows.
Crossroad collection ISet, road segment segment set RSegSet, the driver behavior that traffic light signals allow can directly be obtained
Sig (t), the lane lane where each car and the road segment segment RSeg where each car.
According to the definition of step 1, the feature for needing to extract includes vehicle and front crossroad distance VID, vehicle intersection
Allow go to action IAM.Since the present invention is when using machine learning techniques training, network is adjusted using supervised learning mode
Parameter, it is also necessary to extract training sample label.
Step 211, vehicle and junction ahead distance feature are extracted
The lane is obtained in front according to track direction lDir where lane lane where vehicle and vehicle on map
The coordinate of two endpoints of stop line AB at crossroad, i.e. A point coordinate (xA, yA) and B point coordinate (xB, yB).From vehicle leading edge
Vertical line is done to straight line AB, acquires length of perpendicular length.Due to road approximation straight way in the data set of research, length can be used
Approximate substitution VID.
AB meets formula (19) in two-dimensional coordinate system.
(yA-yB)·x+(xB-xA)·y+(yB·xA-xB·yA)=0 (19)
Assuming that vehicle location P (t)=(x at this timeC, yC), then the distance length of vehicle to stop line AB meets formula
(20)。
It concentrates the data of each moment point of each car to calculate according to formula (20) data, obtains feature VID.
Step 212, crossing allows go to action feature extraction
The table that crossroad signal lamp changes over time is obtained from data set, i.e., for any crossroad IiIt can be with
Obtain sigi(t).Available lane allows driving behavior PLD from the map lane markings in data set.Then IAM feature
It extracts according to defining 13, sequentially finds the corresponding crossroad in every lane, Hadamard product then is done to sig (t) and PLD
Obtain IAM feature
Step 213, tag extraction
Vehicle movement is denoted as disP, and (Δ x, Δ y) indicate vehicle in (displacement of the t+ Δ t) moment relative to t moment, such as public affairs
Shown in formula (21).
DisP (Δ x, Δ y)=P (x (t+ Δ t), y (t+ Δ t))-P (x (t), y (t)) (21)
Coordinate system and unit are the same as definition 3.
The character representation that will acquire is feature (t) vector of definition, and the training sample displacement tag definition that will acquire is
Label (t), wherein t indicates the time.Feature (t) and label (t) are normalized using min-max method for normalizing
Processing, as shown in formula (22).
Wherein f indicates characteristic point value, fmaxIndicate the maximum value of this feature point value, fminIndicate this feature point value
Minimum value, fNIndicate the characteristic point value after normalization.Feature (t) and label (t) after normalized remember respectively
For featureN(t) and labelN(t)。
Step 22, vehicle movement prediction model
Step 221, training sample set defines
Aforementioned pretreated training sample set is denoted as train, including training sample feature set and training sample tally set.
Training sample label is using displacement label.featureN(t) splice label in sequenceNIt (t) is train samples
Structure.Training sample feature set is denoted as train_x, and training sample tally set is denoted as train_y, is expressed as formula (23)
With formula (24).
Train_y and train_x is corresponded by row.
Step 222, vehicle movement prediction model training
Technical solution of the present invention is using full Connection Neural Network training displacement prediction model, as shown in Fig. 2, input layer receives
Shaped like featureN(t) feature vector, 4-10 layers of the network number of plies, activation primitive uses Relu function, and output layer is one two
Dimensional vector, that is, the vehicle movement predicted define same disP (Δ x, Δ y).Training process includes propagated forward and backpropagation.Before
It is successively trained to first regarding train_x as input quantity when propagating.After current layer training, the hidden layer come will be currently trained
Continue to train as next layer of visible layer.The data of input layer are so transmitted into calculating by the node in hidden layer layer by layer,
Output layer is traveled to always, is made comparisons with final output valve and true value.If the error that propagated forward is finally calculated
Desired value is not achieved, then enters back-propagation process.Backpropagation is based on mini-batch gradient descent method, first with training
Tally set train_y calculates network error, successively finds out error function from back to front by chain rule to the local derviation of each weight
Number, i.e., error function calculates the modification amount of each weight in conjunction with the pace of learning of setting to the gradient of weight.It is primary reversed
After propagation, then by propagated forward calculating error, if error reaches desired value, otherwise deconditioning continues next round
Backpropagation, propagated forward process, iteration continues always, until triggering training termination condition until.
Vehicle movement prediction model training step is as follows:
(7) propagated forward is utilized using training sample set train_x as input quantity based on full Connection Neural Network structure
Algorithm calculates the excitation value of each layer.
(8) full connection BP neural network is initialized using the resulting network structure of study, is finally adding output layer, output
For vehicle movement disP (Δ x, Δ y).
(9) mini-batch gradient descent method is used, calculates network error using training tally set train_y, it is reversed to pass
It broadcasts, calculates loss function to the gradient of weight matrix and bigoted item, the network parameter of each layer of fine tuning, until the end of triggering training
Only condition.
The vehicle movement prediction model that training is completed is denoted as H (x).The same feature of the definition of xNIt (t), is normalized
Feature set afterwards.H (x) is the vehicle movement of prediction, defines same disP (Δ x, Δ y).
Step 3, vehicle drive behavior prediction model
Technical solution of the present invention uses gauss hybrid models, based on the vehicle movement prediction result that step 2 obtains, to vehicle
Driving behavior is clustered.
Step 31, Gaussian component defines
5 kinds of driving behaviors are set, i.e. vehicle is kept straight on, and is turned left, and is turned right, and u-turn is as you were, respectively corresponds 5 Gausses point
Measure N1(μ1, ∑1)~N5(μ5, ∑5).Wherein each Gaussian component is binary Gaussian Profile, and mean value is bivector, association
Variance matrix is the matrix that size is 2x2.
Step 32, gauss hybrid models training
Steps are as follows:
(7) to the mean μ of each Gaussian component1、μ2、μ3、μ4、μ5With covariance matrix ∑1、∑2、∑3、∑4、∑5It carries out
Random initializtion, the prior probability π of each Gaussian componentiIt is set as 1/5, i=1,2,3,4,5.
(8) using training sample displacement tally set as input quantity, model is trained using EM algorithm.
(9) weight of each Gaussian component, mean value and covariance matrix are obtained.
After the completion of model training, calculates sample and belong to some Gaussian Profile NiProbability be
Wherein | ∑i| indicate ∑iDeterminant, Σi -1Indicate ΣiInverse matrix.
Step 4, vehicle drive behavior prediction
(1) sample characteristics collection to be predicted is defined first, is denoted as test_x, structure and training sample feature set train_x
It is identical.
(2) by each of sample characteristics collection to be predicted feature vector featureN(t) displacement that input step 2 obtains
Prediction model, the displacement H (feature predictedN(t))。
(3) the resulting displacement H (feature of predictionN(t)) the driving behavior prediction model that input step 3 obtains calculates
Sample belongs to the probability P of each Gaussian Profilei, highest probability is driving behavior obtained by the sample predictions, i.e. sample predictions
Gained driving behavior
Ct∈ { 1,2,3,4,5 }, respectively corresponds five kinds of driving behaviors.
Innovative point
A kind of vehicle drive behavior prediction method based on machine learning is proposed, reaches and promotes vehicle drive behavior prediction
The purpose of accuracy.For traffic environment complicated in urban road, existing prediction technique based on vehicle historical track, or
It is aided with simple traffic information, prediction result is simultaneously inaccurate.The invention patent has comprehensively considered vehicle self attributes, road information
And running environment information, innovatively full Connection Neural Network and gauss hybrid models are combined, utilize full connection nerve
Network solves the ability of regression problem and the ability of the multi-class cluster of gauss hybrid models, from promotion vehicle drive behavior prediction
Accuracy.
Claims (1)
1. a kind of vehicle drive behavior prediction method based on machine learning, which is characterized in that pre- using full Connection Neural Network
Vehicle movement is surveyed to cluster driving behavior using gauss hybrid models according to the displacement of prediction；
Specific method includes the following steps:
Step 1, defined feature collection, including vehicle characteristics definition, roadway characteristic definition, vehicle running environment characterizing definition；
Step 11, vehicle characteristics define
Vehicle characteristics include length of wagon L, and body width W, car speed, acceleration, current driving direction, crossing turns to dynamic
Make, wherein t moment car speed, acceleration are respectively labeled as v (t) and a (t), remaining feature is defined respectively as:
1 vehicle heading vDir (t) is defined, t moment direction of vehicle movement is indicated, with angle clockwise between direct north
It indicates, meets:
360 ° of 0≤vDir (t) < (1)
It defines 2 vehicle intersections and turns to vMov (t), driving behavior of t moment vehicle when by crossroad is indicated, with vector
Form characterization, since crossroad does not allow to reverse end for end, the case where being presently considered vehicle straight trip, turn left, turn right, such as formula
(2):
3 current vehicle position P (t) are defined, indicate t moment vehicle in CA State Plane III in NAD83 coordinate system
Two-dimensional coordinate vector, vector items unit be foot；Location information is defined as follows:
P (t)=(x (t), y (t))； (3)
CA State Plane III in NAD83 coordinate system is 1983 North America datum level coordinate systems；
To sum up, the feature set feature of t moment vehiclev(t) it is defined as follows:
featurev(t)={ L, W, v (t), a (t), vDir (t), vMov (t), P (t) } (4)
Step 12, roadway characteristic defines
Crossroad or T-shaped road junction are abstracted as quadrangle, with its four apex coordinates according to suitable since the angle of direction northwest
Clockwise successively identifies, and marks as follows:
F=(x1, y1, x2, y2, x3, y3, x4, y4) (5)
Defining 4 crossroad set ISet indicates in survey region, the set of all crossroad compositions；
ISet=(I1, I2..., Im...) and (6)
Define 5 road segment segments and refer to a section of the road between two adjacent crossroads, by road segment segment two sides crossroad Lai
Identify the road segment segment；The road segment segment set for defining road i is as follows:
RSegSet (i)=(I1I2, I2I3..., IkIm...) and (7)
Wherein IkIndicate the crossroad that number is k；
Every road segment segment includes several lanes, defines crossroad Ik, ImBetween road segment segment lane set it is as follows:
IkIm=(lid1, lid2..., lidn) (8)
Wherein lidkIndicate lane number；
Defining 6 track direction lDir (x, y) indicates the direction of travel that lane allows, and wherein x, y indicate the position coordinates in lane；
Angle defines same vDir；Vehicle is identical as track direction in the driving direction vDir of current lane straight way, even at current vehicle
In the lane i, then have
VDir (t)=DDiri(x, y), P (t)=(x (t), y (t)) (9)
Define on the left of 7 lanes can lane change quantity LAL be the number of lanes that vehicle can change to the left under certain lane current location；
By the current location (x, y) of current lane i Lane Searches in the same direction to the left, until encountering solid line indicates that vehicle can not lane change to the left
Or until searching leftmost side lane in the same direction；The number of lanes searched is can lane change quantity LAL on the left of lanei(x, y)；
Define on the right side of 8 lanes can lane change amount R AL be the number of lanes that vehicle can change to the right under certain lane current location；
By current location (x, the y) Lane Searches in the same direction to the right of current lane i, until encountering solid line indicates that vehicle can not lane change to the right
Or until searching rightmost side lane in the same direction；The number of lanes searched is can lane change amount R AL on the right side of lanei(x, y)；
Defining 9 straight trip area lanes allows driving behavior to include straight trip, lane change to the left and to the right lane change；Definition vector sld indicates vehicle
The basic driving behavior that road allows is as follows:
The driving behavior SLD that straight trip area lane allows has:
SLD (x, y)=α1(x, y) sldst+α2(x, y) sldlf+α3(x, y) sldrt (113)
{α1, α2, α3|αi=0V αi=1,1≤i≤3, i ∈ N }
The driving behavior for defining the permission of 10 crossing area in preparation lanes includes straight trip, is turned left, right-hand bend and u-turn；Definition vector
The basic driving behavior that pld indicates that lane allows is as follows:
The driving behavior PLD that crossing area in preparation lane allows can be indicated are as follows:
PLD (x, y)=β1(x, y) pldst+β2pldtl+β3pldtr+β4pldta+β5pldsp (13)
Wherein { β1, β2, β3, β4, β5|βi=0V βi=1,1≤i≤5, i ∈ N }
βiFor the probability coefficent of certain driving behavior, otherwise it is 0 that choosing, which is 1,；
To sum up, the roadway characteristic collection featNre at position (x, y)r(x, y) can be defined as follows:
featurer(x, y)={ lDir (x, y), LAL (x, y), RAL (x, y) } (14)
∪ { SLD (x, y), PLD (x, y) }
Step 13, vehicle running environment defines
Define 11 crossing distances, vehicle i and front crossroad ImDistanceIt indicates with the front vehicle i edge and current lane
The distance between front crossroad stop line；
Define 12t moment, traffic lights TLiSignal, which allows to act, uses vector sigi(t) it indicates；
The crossing for defining 13 vehicle t moments allows go to action to be expressed as IAM (t), which allows go to action to be limited by vehicle
The driving behavior PLD and traffic light signals that road allows allow to act sigi(t), it is expressed as the Hadamard product of matrix, it is such as public
Formula (16),
IAM (t)=PLD (P (t)) * sigi(t) (16)
To sum up, t moment vehicle running environment feature set featuree(t) it is defined as follows:
featuree(t)={ VID (t), IAM (t) } (17)
Combining step 11, step 12 and step 13, the feature set feature (t) that t moment influences vehicle drive behavior are defined as
Feature (t)=featNrev(t)∪featurer(P(t))∪featuree(t) (18)
Step 2, vehicle movement prediction model
Step 21, feature extraction and data prediction
The vehicle characteristics directly acquired include Vehicle length L, vehicle width W, car speed v, vehicle acceleration a, vehicle driving
Direction vDir, vehicle intersection turn to vMov, current vehicle position；
The roadway characteristic directly acquired includes track direction lDir, on the left of lane can lane change quantity LAL, can lane change number on the right side of lane
Measure RAL, the driving behavior SLD that lane allows, the driving behavior PLD that crossing area in preparation lane allows；
Crossroad collection ISet, road segment segment set RSegSet, the driver behavior sig that traffic light signals allow can directly be obtained
(t), the lane lane where each car and road segment segment RSeg where each car；
According to the definition of step 1, the feature for needing to extract includes vehicle and front crossroad distance VID, and vehicle intersection allows
Go to action IAM, training sample label；
Step 211, vehicle and junction ahead distance feature are extracted
The lane is obtained in front cross according to track direction 1Dir where lane lane where vehicle and vehicle on map
The coordinate of two endpoints of stop line AB at crossing, i.e. A point coordinate (xA, yA) and B point coordinate (xB, yB)；From vehicle leading edge to straight
Line AB does vertical line, acquires length of perpendicular length；It, can be approximate with length due to road approximation straight way in the data set of research
Substitute VID；
AB meets formula (19) in two-dimensional coordinate system；
(yA-yB)·x+(xB-xA)·y+(yB·xA-xB·yA)=0 (19)
Assuming that vehicle location P (t)=(x at this timeC, yC), then the distance length of vehicle to stop line AB meets formula (20)；
It concentrates the data of each moment point of each car to calculate according to formula (20) data, obtains feature VID；
Step 212, crossing allows go to action feature extraction
The table that crossroad signal lamp changes over time is obtained from data set, i.e., for any crossroad IiIt is available
sigi(t)；Available lane allows driving behavior PLD from the map lane markings in data set；The then extraction of IAM feature
According to defining 13, the corresponding crossroad in every lane is sequentially found, Hadamard product then is done to sig (t) and PLD and is obtained
IAM feature；
Step 213, tag extraction
Vehicle movement is denoted as disP, and (Δ x, Δ y) indicate vehicle in (displacement of the t+ Δ t) moment relative to t moment, such as formula
(21) shown in；
DisP (Δ x, Δ y)=P (x (t+ Δ t), y (t+ Δ t))-P (x (t), y (t)) (21)
Coordinate system and unit are the same as definition 3；
The character representation that will acquire is feature (t) vector of definition, and the training sample displacement tag definition that will acquire is
Label (t), wherein t indicates the time；Feature (t) and label (t) are normalized using min-max method for normalizing
Processing, as shown in formula (22)；
Wherein f indicates characteristic point value, fmaxIndicate the maximum value of this feature point value, fminIndicate the minimum of this feature point value
Value, fNIndicate the characteristic point value after normalization；Feature (t) and label (t) after normalized are denoted as respectively
featureN(t) and labelN(t)；
Step 22, vehicle movement prediction model
Step 221, training sample set defines
Aforementioned pretreated training sample set is denoted as train, including training sample feature set and training sample tally set；Training
Sample label is using displacement label；featureN(t) splice label in sequenceN(t) be train samples knot
Structure；Training sample feature set is denoted as train_x, and training sample tally set is denoted as train_y, is expressed as formula (23) and public affairs
Formula (24)；
Train_y and train_x is corresponded by row；
Step 222, vehicle movement prediction model training
Vehicle movement prediction model training step is as follows:
(1) propagated forward algorithm meter is utilized using training sample set wain_x as input quantity based on full Connection Neural Network structure
Calculate the excitation value of each layer；
(2) full connection BP neural network is initialized using the resulting network structure of study, is finally adding output layer, is exporting as vehicle
Displacement disP (Δ x, Δ y)；
(3) mini-batch gradient descent method is used, calculates network error, backpropagation, meter using training tally set train_y
Loss function is calculated to the gradient of weight matrix and bigoted item, the network parameter of each layer of fine tuning, until the termination item of triggering training
Part；
Step 3, vehicle drive behavior prediction model
Step 31, Gaussian component defines
5 kinds of driving behaviors are set, i.e. vehicle is kept straight on, and is turned left, and is turned right, and u-turn is as you were, respectively corresponds 5 Gaussian components；Root
Definition according to the present invention to vehicle movement, each Gaussian component are binary Gaussian Profile；
Step 32, gauss hybrid models training
Steps are as follows:
(1) mean value to each Gaussian component and covariance matrix carry out random initializtion, the prior probability of each Gaussian component
It is set as 1/5；
(2) using training sample displacement tally set as input quantity, model is trained using EM algorithm；
(3) weight of each Gaussian component, mean value and covariance matrix are obtained；
Step 4, vehicle drive behavior prediction
(1) sample characteristics collection to be predicted is defined first, is denoted as test_x, and structure is identical as training sample feature set train_x；
(2) displacement prediction model for obtaining sampling feature vectors input step 2 to be predicted, the displacement predicted；
(3) it the driving behavior prediction model for predicting that resulting displacement input step 3 obtains, calculates sample and belongs to each Gauss point
The probability of cloth, highest probability is driving behavior obtained by the sample predictions.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810287172.XA CN108550279B (en) | 2018-04-03 | 2018-04-03 | Vehicle drive behavior prediction method based on machine learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810287172.XA CN108550279B (en) | 2018-04-03 | 2018-04-03 | Vehicle drive behavior prediction method based on machine learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108550279A CN108550279A (en) | 2018-09-18 |
| CN108550279B true CN108550279B (en) | 2019-10-18 |
Family
ID=63513864
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810287172.XA Active CN108550279B (en) | 2018-04-03 | 2018-04-03 | Vehicle drive behavior prediction method based on machine learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108550279B (en) |
Families Citing this family (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109885058B (en) * | 2019-03-12 | 2022-05-20 | 杭州飞步科技有限公司 | Driving track planning method and device, electronic equipment and storage medium |
| JP7086021B2 (en) * | 2019-03-14 | 2022-06-17 | 本田技研工業株式会社 | Behavior predictor |
| CN110020748B (en) * | 2019-03-18 | 2022-02-15 | 杭州飞步科技有限公司 | Trajectory prediction method, apparatus, device and storage medium |
| JP7223629B2 (en) * | 2019-05-13 | 2023-02-16 | 日立Astemo株式会社 | In-vehicle system, external recognition sensor, electronic control unit |
| CN110231820B (en) * | 2019-05-31 | 2022-08-05 | 江苏亿科达科技发展有限公司 | Vehicle running control method based on Internet of things |
| CN110264721B (en) * | 2019-07-01 | 2021-04-20 | 北京理工大学 | Urban intersection surrounding vehicle track prediction method |
| CN110706517B (en) * | 2019-11-07 | 2020-08-28 | 深圳鼎然信息科技有限公司 | Intelligent safety early warning method, device and system for vehicle and storage medium |
| CN111104969B (en) * | 2019-12-04 | 2023-09-19 | 东北大学 | Collision possibility pre-judging method for unmanned vehicle and surrounding vehicles |
| US20220357176A1 (en) * | 2020-01-31 | 2022-11-10 | Grabtaxi Holdings Pte. Ltd. | Methods and data processing systems for predicting road attributes |
| CN111339834B (en) * | 2020-02-04 | 2023-06-02 | 浙江大华技术股份有限公司 | Method for identifying vehicle driving direction, computer device and storage medium |
| CN113835421B (en) * | 2020-06-06 | 2023-12-15 | 华为技术有限公司 | Method and device for training driving behavior decision model |
| CN112308171A (en) * | 2020-11-23 | 2021-02-02 | 浙江天行健智能科技有限公司 | Vehicle position prediction modeling method based on simulated driver |
| WO2022141538A1 (en) * | 2020-12-31 | 2022-07-07 | 华为技术有限公司 | Trajectory prediction method and apparatus |
| CN112699046B (en) * | 2021-01-12 | 2024-03-29 | 抖音视界有限公司 | Application program testing method and device, electronic equipment and storage medium |
| CN113012424B (en) * | 2021-02-22 | 2022-03-08 | 同济大学 | Dynamic evolution prediction method for open type unmanned vehicle group in expressway scene |
| EP4286972A4 (en) * | 2021-02-26 | 2024-03-27 | Huawei Technologies Co., Ltd. | Vehicle driving intention prediction method and apparatus, terminal and storage medium |
| CN113252060B (en) * | 2021-05-31 | 2021-09-21 | 智道网联科技(北京)有限公司 | Vehicle track calculation method and device based on neural network model |
| CN113628441B (en) * | 2021-08-02 | 2022-05-20 | 北京航空航天大学 | Vehicle space-time scheduling method guided by deep-reinforced neural network |
| CN113642644B (en) * | 2021-08-13 | 2024-05-10 | 北京赛目科技有限公司 | Method and device for determining vehicle environment level, electronic equipment and storage medium |
| CN113654569A (en) * | 2021-08-16 | 2021-11-16 | 江铃汽车股份有限公司 | Path planning method, system and storage medium |
| CN113968234B (en) * | 2021-11-29 | 2023-05-02 | 深圳市科莱德电子有限公司 | Vehicle auxiliary driving control method and device and vehicle-mounted controller |
| CN115027497B (en) * | 2022-06-20 | 2024-07-23 | 重庆长安汽车股份有限公司 | Target vehicle cut-in intention prediction method and readable storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101169903A (en) * | 2007-10-19 | 2008-04-30 | 黄辉先 | Intelligent traffic control system for controlling access connection traffic flow |
| CN102176285A (en) * | 2011-02-28 | 2011-09-07 | 江苏怡和科技股份有限公司 | Method for judging behavior patterns of vehicles in video stream |
| CN103021191A (en) * | 2012-11-22 | 2013-04-03 | 浙江理工大学 | Intelligent traffic control device and control method |
| 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 |
| CN107562837A (en) * | 2017-08-24 | 2018-01-09 | 电子科技大学 | A kind of maneuvering Target Tracking Algorithm based on road network |
| CN107609633A (en) * | 2017-05-03 | 2018-01-19 | 同济大学 | The position prediction model construction method of vehicle traveling influence factor based on deep learning in car networking complex network |
-
2018
- 2018-04-03 CN CN201810287172.XA patent/CN108550279B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101169903A (en) * | 2007-10-19 | 2008-04-30 | 黄辉先 | Intelligent traffic control system for controlling access connection traffic flow |
| CN102176285A (en) * | 2011-02-28 | 2011-09-07 | 江苏怡和科技股份有限公司 | Method for judging behavior patterns of vehicles in video stream |
| CN103021191A (en) * | 2012-11-22 | 2013-04-03 | 浙江理工大学 | Intelligent traffic control device and control method |
| CN107609633A (en) * | 2017-05-03 | 2018-01-19 | 同济大学 | The position prediction model construction method of vehicle traveling influence factor based on deep learning in car networking complex 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 |
| CN107562837A (en) * | 2017-08-24 | 2018-01-09 | 电子科技大学 | A kind of maneuvering Target Tracking Algorithm based on road network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108550279A (en) | 2018-09-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108550279B (en) | Vehicle drive behavior prediction method based on machine learning | |
| US11899411B2 (en) | Hybrid reinforcement learning for autonomous driving | |
| US11934962B2 (en) | Object association for autonomous vehicles | |
| CN107609633A (en) | The position prediction model construction method of vehicle traveling influence factor based on deep learning in car networking complex network | |
| Jain et al. | Discrete residual flow for probabilistic pedestrian behavior prediction | |
| CN104914870B (en) | Transfinited the outdoor robot local paths planning method of learning machine based on ridge regression | |
| CN110264721B (en) | Urban intersection surrounding vehicle track prediction method | |
| EP3948656A1 (en) | Pedestrian prediction based on attributes | |
| CN107310550B (en) | Road vehicles travel control method and device | |
| 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 | |
| CN110146910A (en) | A kind of localization method and device merged based on GPS with laser radar data | |
| CN105513354A (en) | Video-based urban road traffic jam detecting system | |
| US11526721B1 (en) | Synthetic scenario generator using distance-biased confidences for sensor data | |
| DE112020002592T5 (en) | SYSTEMS AND METHODS FOR VEHICLE NAVIGATION BASED ON IMAGE ANALYSIS | |
| CN109544909A (en) | Driver's lane-change behavior analysis method based on video frequency vehicle track of taking photo by plane | |
| US20230150549A1 (en) | Hybrid log simulated driving | |
| CN107730889A (en) | A kind of target vehicle search method based on traffic video | |
| CN109035758A (en) | City road network intersection congestion recognition methods based on floating car data mesh mapping | |
| Tran et al. | Traffic flow analysis with multiple adaptive vehicle detectors and velocity estimation with landmark-based scanlines | |
| DE102021132082A1 (en) | END-TO-END SYSTEM TRAINING USING UNITED IMAGES | |
| Ding et al. | A lane detection method based on semantic segmentation | |
| Fazekas et al. | ANN-based classification of urban road environments from traffic sign and crossroad data | |
| Gao et al. | Discretionary cut-in driving behavior risk assessment based on naturalistic driving data | |
| CN116917828A (en) | Instantiating objects in a simulation environment based on log data | |
| CN103279748A (en) | Road detection method based on SIFT-COF feature optical flow |
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 |