CN109727490A - A kind of nearby vehicle behavior adaptive corrective prediction technique based on driving prediction field - Google Patents
A kind of nearby vehicle behavior adaptive corrective prediction technique based on driving prediction field Download PDFInfo
- Publication number
- CN109727490A CN109727490A CN201910071784.XA CN201910071784A CN109727490A CN 109727490 A CN109727490 A CN 109727490A CN 201910071784 A CN201910071784 A CN 201910071784A CN 109727490 A CN109727490 A CN 109727490A
- Authority
- CN
- China
- Prior art keywords
- vehicle
- prediction
- behavior
- driving
- field
- 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
Landscapes
- Traffic Control Systems (AREA)
Abstract
The invention discloses a kind of nearby vehicle behavior adaptive corrective prediction techniques based on driving prediction field, and step 1: nearby vehicle behavior discretization and data set pre-process: by nearby vehicle behavior according to being laterally and longitudinally divided into N number of typical behaviour;Step 2: obtain traffic environment and participate in vehicle time series data: each traffic environment is participated in vehicle and is obtained each moment car's location, speed, acceleration in real time using positioning system;Step 3: establishing driving prediction field: establishing the driving based on safety, efficiency, driver comfort three elements and predict field EP, EP=ES+EE+EC;Step 4: nearby vehicle behavior prediction model is established based on maximum Likelihood;Step 5: nearby vehicle behavior is predicted to correct with model adaptation in real time.The present invention comprehensively considers safety, efficiency and the driver comfort for influencing driver's behavior, establishes driving prediction field and quantification and qualification in the running region of target vehicle, proposes new approaches for nearby vehicle behavior prediction.
Description
Technical field
The invention belongs to intelligent driving technical fields, and in particular to a kind of to predict the nearby vehicle behavior of field certainly based on driving
Adapt to correction prediction technique.
Background technique
Nowadays either advanced driver assistance system or fully automated driving vehicle have all caused each field and have learned
The extensive research interest of person, unquestionably, it is vehicle intellectualized have become the most important trend of development of automobile industry and trend it
One.To find out its cause, intelligent vehicle not only has highly efficient, safer and more clean performance in transportation system, together
When it can also discharge manipulation of the mankind in driving procedure to vehicle.And true traffic environment is often complicated and height is not true
Fixed, the mankind are very outstanding driver in fact in such a case, because people can be by constantly learning have deduction week
Enclose the ability that traffic participant behavior is intended to and predicts their state of motion.Therefore, intelligent vehicle " driving brain " is also only as people
Class equally reaches the distillation to the perception of increasingly traffic environment, can just accomplish that real " intelligence " drives.I.e. from technological layer,
Vehicle should be able to predict the Future movement behavior situation of multiple target in around traffic environment, understand traffic ring to improve it
The ability in border, this is of great significance to rationally efficient trajectory planning and Decision Control is formulated.In recent years, with visual impression
Know the development of technology and the communication technology, laser radar, millimetre-wave radar, camera multi-sensor fusion system enable intelligent vehicle
Enough real-time monitoring surrounding traffic environment, the communication equipments such as car networking, vehicle-to-vehicle (V2V), smart phone can also help vehicle quasi-
Information additional around really is obtained, this provides convenience to the behavior prediction of peripheral object vehicle.
At present, domestic and foreign scholars carry out the driving behavior of this vehicle driver or the identification of driving intention and prediction
A large amount of research work, and achieve brilliant achievement.They are past when carrying out the input quantity data processing of prediction model feature
Toward being by obtaining the vehicle operating parameters of this vehicle (steering wheel angle, longitudinal acceleration and from vehicle and lane center linear distance etc.)
Or driver parameter's (left-hand mirror average fixation number, single Average glance time, single average head level angle etc.)
Come what is realized, if but start with from the detailed data of surrounding vehicles inside and its driver itself, it undoubtedly increases to vehicle mounted communication
The requirement of equipment or telecommunication platform efficiency of transmission, the problem of also jeopardizing information security, therefore this method is not particularly suited for
The behavior prediction of nearby vehicle.Another feasible scheme, is the prototype track by peripheral object vehicle come direct to behavior
It is predicted, can get higher computational efficiency, but such method will be predicted target carriage when carrying out motion prediction as one
A independent individual is studied, and has ignored surrounding traffic environment participant such as other vehicles (including main vehicle) to it
Influence caused by behavior situation, be to be difficult to carry out stable and accurate long-term motion behavior prediction under traffic environment, this be because
Still have the special bus of automatic Pilot ability for the special bus that either people drives, is all excitation of the meeting to ambient enviroment
Make the intelligent body of respective reaction.
Therefore, when predicting nearby vehicle behavior, the long-term influence factor of digging vehicle behavior situation is needed.This hair
It is bright propose it is a kind of based on driving prediction field nearby vehicle behavior adaptive forecasting method, propose consider safety, efficiency,
Field is predicted in the driving of driver comfort Decision Making Effect three elements, and establishes nearby vehicle behavior prediction model based on this, and tie
The Activity recognition result adaptive corrective for closing the vehicle being predicted improves precision of prediction.
Summary of the invention
For the requirement of nearby vehicle behavior prediction long-time stability and reliability, the invention proposes one kind based on driving
Field is predicted in the driving of operator's behaviour decision making elements affect, and proposes a kind of nearby vehicle behavior adaptive prediction side based on this
Method, can the behavior real-time and accurately to peripheral object vehicle make reasonable prediction, mentioned for the decision rule of intelligent vehicle itself
Foundation for reference.The purpose of the present invention can be achieved through the following technical solutions.A kind of periphery vehicle based on driving prediction field
Behavior adaptive forecasting method, specifically includes:
Step1: nearby vehicle behavior discretization and data set pre-process;
Nearby vehicle behavior is divided according to laterally and longitudinally two aspect combinations, discretization is divided into N number of typical behaviour
bi.Denoising is carried out to NGSIM traffic data collection and extracts valid data collection, according to vehicle behavior discretization division methods
It marks each data scaling and corresponds to behavior type.
Step2: it obtains traffic environment and participates in vehicle time series data;
Each traffic environment participates in vehicle and obtains each moment in real time from vehicle using vehicle-mounted GPS and IMU alignment by union system
Position (x, y), speed (Vx,Vy), acceleration (ax,ay).Main vehicle uses the D2D of LTE module in the V2V communication technology
(Device-To-Device) adjacent communication service (ProSe) obtains the state timing letter of locating traffic environment nearby vehicle in real time
Breath.For the peripheral object vehicle that need to carry out vehicle behavior prediction, the front and back vehicle and adjacent lane of its current lane are taken
The influencer that front and back vehicle occurs as its behavior.
Step3: driving prediction field is established;
It is the intelligent body for making corresponding " going after profits and advoiding disadvantages " reaction to the excitation of ambient enviroment for riding manipulation person, establishes base
Field E is predicted in the driving of safety, efficiency, driver comfort Decision Making Effect three elementsP, EP=ES+EE+EC, wherein safety is pre-
Survey field ES, EFFICIENCY PREDICTION field EE, driving comfort predict field EC.If predetermined period time is Δ T.
(1) safe prediction field is established
Any point position is influenced possessed unit safe gesture value by jth vehicle around in target vehicle driving region
Wherein, (X, Y) is any point position in target vehicle driving region;GSPredict field to permanent for traffic safety
Number;δjThe type of vehicle coefficient of surrounding jth vehicle;MjIt is that the length, width and height of jth vehicle multiply for the equivalent mass ratio of jth vehicle around
Long-pending inverse;(x[j],y[j]) be target vehicle around jth vehicle current time position vector;For target carriage
The velocity vector at surrounding jth vehicle current time;For the acceleration at jth vehicle current time around target vehicle
Spend vector;Δ T is peripheral object vehicle behavior predetermined period time;| | | | 2 be the 2- norm sign of vector.
Then unit safe gesture value possessed by the position of any point in target vehicle driving region
(2) EFFICIENCY PREDICTION field is established
Unit efficiency gesture value possessed by the position of any point in target vehicle driving region
Y is any point lengthwise position in target vehicle driving region;GEFor EFFICIENCY PREDICTION field undetermined constant of driving a vehicle;M0For
The equivalent mass ratio of target vehicle is the inverse of the length, width and height product of target vehicle;y[0]For the longitudinal direction at target vehicle current time
Position;
(3) driving comfort prediction field is established
Unit driving comfort gesture value possessed by the position of any point in target vehicle driving region
(X, Y) is any point position in target vehicle driving region;GCField undetermined constant is predicted for driving driving comfort;
(x[0],y[0]) be target vehicle current time position vector;For the velocity vector at target vehicle current time;
Δ T is peripheral object vehicle behavior predetermined period time;| | | | 2 be the 2- norm sign of vector.
Step4: nearby vehicle behavior prediction model is established;
It is fitted each vehicle behavior biCorresponding similitude track, if the area that target vehicle is run over according to similitude track
Domain isCalculate each behavior of peripheral object vehicle driving prediction field field strength and
Wherein, KSFor the weight coefficient of safe prediction field field strength sum;KEFor the weight coefficient of EFFICIENCY PREDICTION field field strength sum;KS
The weight coefficient of field field strength sum is predicted for driving comfort;
It is the corresponding probability of each vehicle behavior by field strength and normalized
Write out likelihood function L (θ)=Π P_predict(bi), wherein θ={ KS,KE,KC}。
Maximum likelihood is calculated using mature conjugate gradient method based on the NGSIM data set pre-processed in Step1 to estimate
MeteringThat is initial weight COEFFICIENT KS_0,KE_0,KC_0。
Then obtain nearby vehicle behavior prediction pattern function
Step5: nearby vehicle behavior is predicted to correct with model adaptation in real time
Main parking stall is first depending on Step2 in true traffic environment and obtains traffic environment participation vehicle time series data in real time,
Vehicle behavior probability of the peripheral object vehicle in predetermined period time Δ T is predicted in real time according to the prediction model that Step4 is established,
And using the corresponding behavior of maximum probability as the prediction result of nearby vehicle behavior.To further increase precision of prediction, the present invention
Also propose that a kind of weight coefficient returns antidote.Taking the cross of peripheral object vehicle, length travel, speed, acceleration is observation
Variable carries out online recognition to peripheral object vehicle behavior based on HMM model, obtains knowledge of each behavior within predetermined period time
Other probability Precognize(bi)。
Behavior prediction model probability function is written as
Ppredict_k(bi)=fk(KS_k,KE_k,KC_k)
That is Ppredict_k(bi) about KS_k,KE_k,KC_kFunction, wherein KS_kFor k-th of predetermined period time safety field
The weight coefficient of strong sum;KE_kFor the weight coefficient of k-th of predetermined period time efficiency field field strength sum;KC_kFor k-th of prediction week
The weight coefficient of time phase driver training ground field strength sum.
Cost function between structure forecast value and discre value
Wherein N is nearby vehicle behavior type number, N=9.
Construct the rectification functionWherein velocity coefficient is corrected for α.
By the rectification function, predetermined period time Δ T corrects weight coefficient online one by one, realizes adaptive prediction, further
Improve the accuracy rate of nearby vehicle Activity recognition prediction.
Beneficial effects of the present invention:
(1) it participates in influence of the vehicle to target vehicle behavior from surrounding traffic to start with, target carriage work will be only predicted by alleviating
For the defect of single independent individual, the prediction steady in a long-term of nearby vehicle Activity recognition can be realized.
(2) safety, efficiency and the driver comfort for influencing riding manipulation person behavior are comprehensively considered, in target vehicle
Running region establishes driving prediction field and has carried out quantification and qualification, proposes new think of for nearby vehicle behavior prediction
Road.
(3) propose it is a kind of based on driving prediction field nearby vehicle behavior prediction model, have preferable predictablity rate with
Less predicted time.
(4) it proposes that a kind of nearby vehicle behavior prediction returns antidote, is identified by HMM model in predetermined period time
The continuous adaptive corrective prediction model of vehicle behavior result weight coefficient vector, further improve nearby vehicle behavior
Precision of prediction.
Detailed description of the invention
Nearby vehicle behavior adaptive corrective prediction technique block diagram of the Fig. 1 based on driving prediction field;
The discretization of Fig. 2 nearby vehicle behavior divides;
The traffic environment of Fig. 3 target vehicle refers to vehicle;
(a) target vehicle is in middle lane;(b) target vehicle is in right-hand lane;(c) target vehicle is in a left side
Side lane);
The a certain behavior b of Fig. 4 peripheral object vehicleiThe region crossed;
The a certain typical traffic environment H of Fig. 5;
Fig. 6 safe prediction field field strength distribution under H traffic environment;
Fig. 7 EFFICIENCY PREDICTION field field strength distribution under H traffic environment;
Fig. 8 driver comfort under H traffic environment predicts field field strength distribution;
Specific embodiment
The present invention will be further explained below with reference to the attached drawings.
As shown in Figure 1, implementation of the invention includes the following steps:
Step1: nearby vehicle behavior discretization and data set pre-process
According to the characteristics of peripheral object vehicle behavior uncertain factor is more, complicated difficult divides, it would be possible to which behavior is divided into lateral line
To go to be combined division with longitudinal behavior both direction.By in lateral behavior left lane-change (Lane Change to Left),
Keep lane (Lane Keep), right lane-change (Lane Change to Right), the acceleration (Speed in longitudinal behavior
Increase), speed (Speed Keep), deceleration (Speed Decrease) is maintained to be divided into nearby vehicle behavior discretization
N number of typical behaviour bi, N=9, respectively left lane-change deceleration (LCL-SD), Zuo Huandao at the uniform velocity (LCL-SK), left lane-change acceleration
(LCL-SI), it maintains lane to slow down (LK-SD), maintains lane at the uniform velocity (LK-SK), lane is maintained to accelerate (LK-SI), right lane-change subtracts
Fast (LCR-SD), right lane-change are slowed down (LCR-SK), and right lane-change slows down (LCR-SI).Every track corresponds to each as shown in Figure 2
Behavior bi, wherein 1≤i≤9.Computer program is developed using the Pandas data analysis bag of Python, to NGSIM traffic data
Collection carries out denoising and extracts valid data collection, and it is corresponding to mark each data scaling according to vehicle behavior discretization division methods
Behavior type.
Step2: it obtains traffic environment and participates in vehicle time series data
Each traffic environment, which participates in vehicle, has an independent ID, real-time using vehicle-mounted GPS and IMU alignment by union system
Obtain each moment car's location (x, y), speed (Vx,Vy), acceleration (ax,ay).In view of data are transmitted in subsequent step
Real-time and robustness, data acquiring frequency 50Hz, i.e. 0.02s are the duration for taking front and back data break twice.Main vehicle
V2V communication network is accessed by PC5 interface, is used neighbouring with the D2D (Device-To-Device, equipment room) in LTE module
Communication service (ProSe) obtains the state timing information of locating traffic environment nearby vehicle in real time.The V2V communication technology English
Literary full name be Vehicle-To-Vehicle Technology, because its can overcome high-speed mobile cause Doppler effect with
Complicated communication environment is widely applied in intelligent automobile field, and D2D module is not necessarily to that neighbouring vehicle can be established by base station
Being in communication with each other between terminal.Main lock determines a certain vehicle in periphery as behavior prediction target vehicle, according in above-mentioned steps
The traffic environment of the real-time status time series data framework target moving vehicle of acquisition.In true traffic environment, the mankind are driven
Member can by it is preceding to visible area, rearview mirror and rear-camera instrument obtain around traffic information, and intelligent driving system
Same purpose, therefore traffic information pair can be reached by the visual perceptions module such as camera, laser radar and millimetre-wave radar
The influence of target carriage is alternately propagated adjacent to front and back.When constructing the surrounding traffic environment of target vehicle, target vehicle is taken to work as
The influencer that the front and back vehicle in preceding lane and the front and back vehicle of adjacent lane occur as its behavior, if ambient influence vehicle
Number is h.Fig. 3 be respectively in three tunnel one way roads target vehicle be in the mode that different lanes take ambient influence vehicle, a is mesh
Mark vehicle is in middle lane, and h is 6 at this time;B is that target vehicle is in left-hand lane, and h is 4 at this time;C is in for target vehicle
Right-hand lane, h is 4 at this time.
Step3: driving prediction field is established
Either it is the special bus or the special bus for having automatic Pilot ability that people drives, is all a meeting to vehicle around
The intelligent body of corresponding " going after profits and advoiding disadvantages " reaction is made in the excitation of road collaboration traffic environment, and nearby vehicle driving behavior has very
Big uncertainty, changeability, and influenced respectively by the containing from the various expected revenuses of vehicle.Therefore, below in peripheral object
A kind of driving prediction field E based on Decision Making Effect factor is established in vehicle driving regionP, to quantify nearby vehicle behavior generation
Influence factor.According to riding manipulation person behavior three elements (safety, efficiency, driver comfort) is influenced, driving prediction field is divided
For three son prediction fields, respectively safe prediction field ES, EFFICIENCY PREDICTION field EE, driving comfort predict field EC.Wherein, EP=ES+EE+
EC.For convenience of description, a certain typical traffic environment H shown in Fig. 5 is taken.
(1) safe prediction field is established
Safe prediction field characterizes influence of the safety to peripheral object vehicle driving maneuvers person.With the front and back of target vehicle to
H periphery vehicular traffic is " charge " for generating safe field potential, and front and back to the position of h periphery vehicular traffic, speed and is added
Speed is as the primary variables for influencing safe gesture value.
Writing out in target vehicle driving region any point position is influenced possessed unit safe by jth vehicle around
Gesture value
Wherein, (X, Y) is any point position in target vehicle driving region;GSPredict field to permanent for traffic safety
Number;δjThe type of vehicle coefficient of surrounding jth vehicle;MjIt is that the length, width and height of jth vehicle multiply for the equivalent mass ratio of jth vehicle around
Long-pending inverse;(x[j],y[j]) be target vehicle around jth vehicle current time position vector;For target carriage
The velocity vector at surrounding jth vehicle current time;For the acceleration at jth vehicle current time around target vehicle
Spend vector;Δ T is peripheral object vehicle behavior predetermined period time;|| ||2For the 2- norm sign of vector.
Then unit safe gesture value possessed by the position of any point in target vehicle driving region
Such as Fig. 6, emulation obtains safe prediction field field strength distribution under H traffic environment.
(2) EFFICIENCY PREDICTION field is established
EFFICIENCY PREDICTION field characterizes influence of the safety to peripheral object vehicle driving maneuvers person.It is to generate effect with target vehicle
" charge " of rate field potential, using the lengthwise position of target vehicle as the primary variables for influencing efficiency gesture value.
Write out unit efficiency gesture value possessed by the position of any point in target vehicle driving region
Y is any point lengthwise position in target vehicle driving region;GEFor EFFICIENCY PREDICTION field undetermined constant of driving a vehicle;M0For
The equivalent mass ratio of target vehicle is the inverse of the length, width and height product of target vehicle;y[0]For the longitudinal direction at target vehicle current time
Position;
Such as Fig. 7, emulation obtains EFFICIENCY PREDICTION field field strength distribution under H traffic environment.
(3) driving comfort prediction field is established
Driving comfort predicts influence of the field characterization driver comfort to peripheral object vehicle driving maneuvers person.With target vehicle
For " charge " for generating driving comfort field potential, go to the transverse and longitudinal acceleration of a certain position of running region as shadow target vehicle
Ring the primary variables of driving comfort gesture value.
Write out unit driving comfort gesture value possessed by the position of any point in target vehicle driving region
(X, Y) is any point position in target vehicle driving region;GCField undetermined constant is predicted for driving driving comfort;
(x[0],y[0]) be target vehicle current time position vector;For the velocity vector at target vehicle current time;
Δ T is peripheral object vehicle behavior predetermined period time;|| ||2For the 2- norm sign of vector.
Such as Fig. 8, emulation obtains driving comfort under H traffic environment and predicts field field strength distribution.
Step4: nearby vehicle behavior prediction model is established
By taking target vehicle is in middle lane as an example, safety zone in travelable region is divided into 9 according to behavior type
A behavior hot-zone.The decision-making level of simulation peripheral object vehicle generates and executes a certain vehicle behavior biCan blur estimation respectively drive a vehicle son
It predicts that the containing of field field strength sum influences, according to similarity principle, the central point of each behavior hot-zone is taken to terminate as each behavior type
The position of the target vehicle at moment.It is fitted each vehicle behavior biCorresponding similitude track, pick-up is inswept under the track
Area be integral domain
By driving prediction field field strength and it is normalized, i.e., field field strength is predicted into the driving of each vehicle behavior and turns
For the corresponding prediction probability of the vehicle behavior,
Write out likelihood function L (θ)=Π Ppredict(bi), wherein θ={ KS,KE,KC}。
Maximum is calculated using mature conjugate gradient method as sample set using the NGSIM data set pre-processed in Step1
Possibility predication amountThat is initial weight COEFFICIENT KS,0,KE,0,KC,0。
Then obtain nearby vehicle behavior prediction pattern function
Step5: nearby vehicle behavior is predicted to correct with model adaptation in real time
Operate nearby vehicle human driver or intelligent driving system in periphery traffic environment also by itself decision machine
The influence of system, pursuing the expectation of safety, efficiency, driver comfort has distinct individual character, and it is expected to pursue and drive
It is stable in the time domain of behavior long-term forecast.Main parking stall is first depending on Step2 and obtains friendship in real time in true traffic environment
Logical environment participates in vehicle time series data, predicts peripheral object vehicle in predetermined period in real time according to the prediction model that Step4 is established
Between vehicle typical behaviour probability P in Δ Tpredict(bi), exporting the corresponding vehicle behavior type of maximum behavior probability is prediction knot
Fruit.
To further increase precision of prediction, the present invention also proposes that a kind of weight coefficient returns antidote, true to reduce
Weight coefficient deviation gives behavior prediction result bring error rate in scene.When a predicted time period Δ T is ended, utilize
Nearby vehicle typical behaviour type in the Δ T of hidden Markov model (HMM) identification prediction cycle time.
If hidden Markov model is a five-tuple (Q, V, A, B, π).Observable state is expressed as V={ V1,V2,…,
VM, M is the number of observation state;Hidden state is expressed as Q={ Q1,Q2,…,QN, N is the number of hidden state.I is length
For the status switch of T, O is corresponding observation sequence, I={ I1,I2,…,IT, O={ O1,O2,…,OT}。
A=[aij]N×NFor hidden state transition probability matrix, in element representation HMM model between each hidden state
Transition probability.Wherein,
aij=P (It+1=Qj∣It=Qi), i=1,2 ..., N;J=1,2 ..., N
It is in t moment, hidden state QiUnder conditions of, it is Q in t+1 moment hidden statejProbability.
B=[bj(k)]N×MFor confusion matrix, in element representation HMM model between each hidden state and observation state
Transition probability.Wherein,
bj(k)=P (Ot=Vk∣It=Qj), k=1,2 ..., M;J=1,2 ..., N
It indicates in t moment, hidden state QjUnder the conditions of being, observation state OtProbability.
π=(πi) it is initial state probabilities matrix, wherein πi=P (I1=Qi), i=1,2 ..., N is initial time t=1
Each hidden state QiProbability.
Two stages can be divided by carrying out nearby vehicle Activity recognition with HMM model:
A. model training learns: initializing to each vehicle behavior identification model, obtains initial parameter N, M, A, B, π;
Extract the characteristic O that the NGSIM traffic data that Step1 is handled well concentrates each acquisition time in Δ Tt(lateral displacement,
Speed, acceleration and length travel, speed, acceleration), each time point characteristic OtThe observation sequence O=of composition
O1O2O3……O△T.It is used using observation sequence as the input of HMM parameter learning according to the initial parameter after model initialization
Baum-Welch iterative algorithm adjusts model λ=(A, B, π) parameter, maximizes probability function, progressive updating model parameter,
Finally obtain the optimal HMM model of each behavior type;
B. on-line testing identifies: and then trained vehicle behavior HMM identification model is utilized, by peripheral object to be identified
Vehicle forms observation sequence as mode input, using forwards algorithms calculating observation sequence each after feature extraction, coding
Probability P under HMM modelrecognize_k(bi)。
Behavior prediction model probability function is written as
Ppredict_k(bi)=fk(KS_k,KE_k,KC_k)
That is Ppredict_k(bi) about KS_k,KE_k,KC_kFunction, k indicate k-th of predetermined period time Δ T.
Write out the residual sum of squares (RSS) function of prediction probability value Yu identification probability value, i.e. cost function
Construct the rectification functionWherein velocity coefficient is corrected for α.
Then by KS_k+1,KE_k+1,KC_k+1Substitute into update prediction model parameters, come predict it is next pass through predetermined period when
Between target vehicle behavior in Δ T.By the rectification function, predetermined period time corrects weight coefficient online one by one, adaptive to realize
It should predict, further increase the accuracy rate of nearby vehicle Activity recognition prediction.
The series of detailed descriptions listed above only for feasible embodiment of the invention specifically
Protection scope bright, that they are not intended to limit the invention, it is all without departing from equivalent implementations made by technical spirit of the present invention
Or change should all be included in the protection scope of the present invention.
Claims (8)
1. a kind of nearby vehicle behavior adaptive corrective prediction technique based on driving prediction field, which is characterized in that including as follows
Step:
Step 1: nearby vehicle behavior discretization and data set pre-process;
Step 2: obtaining traffic environment and participate in vehicle time series data;
Step 3: establishing driving prediction field, including safe prediction field, EFFICIENCY PREDICTION field, driving comfort prediction field;
Step 4: establishing nearby vehicle behavior prediction model
Step 5: nearby vehicle behavior is predicted to correct with model adaptation in real time.
2. a kind of nearby vehicle behavior adaptive corrective prediction technique based on driving prediction field according to claim 1,
It is characterized in that, the detailed process of the step 1 includes:
Nearby vehicle behavior is divided according to laterally and longitudinally two aspect combinations, discretization is divided into N number of typical behaviour bi, right
NGSIM traffic data collection carries out denoising and extracts valid data collection, is marked according to vehicle behavior discretization division methods
Each data scaling corresponds to behavior type.
3. a kind of nearby vehicle behavior adaptive corrective prediction technique based on driving prediction field according to claim 2,
It is characterized in that, N number of typical behaviour specifically:
Left lane-change slows down, Zuo Huandao at the uniform velocity, left lane-change accelerate, lane maintained to slow down, maintain lane at the uniform velocity, maintain lane to accelerate,
Right lane-change is slowed down, right lane-change is slowed down, right lane-change is slowed down.
4. a kind of nearby vehicle behavior adaptive corrective prediction technique based on driving prediction field according to claim 2,
It is characterized in that, the specific implementation of the step 2 includes:
Each traffic environment participates in vehicle and obtains each moment car's location in real time using vehicle GPS and IMU alignment by union system
(x, y), speed (Vx, Vy), acceleration (ax, ay);Main vehicle uses the D2D adjacent communication service of LTE module in the V2V communication technology
The state timing information of locating traffic environment nearby vehicle is obtained in real time.
5. a kind of nearby vehicle behavior adaptive corrective prediction technique based on driving prediction field according to claim 1,
It is characterized in that, the specific implementation of the step 3 includes:
The method for building up of the safe prediction field:
To h periphery vehicular traffic it is " charge " for generating safe field potential with the front and back of target vehicle, front and back is handed over to h periphery
The position that is open to traffic, velocity and acceleration are as the primary variables for influencing safe gesture value;
Writing out in target vehicle driving region any point position is influenced possessed unit safe gesture value by jth vehicle around
Wherein, (X, Y) is any point position in target vehicle driving region;GsField undetermined constant is predicted for traffic safety;δjWeek
Enclose the type of vehicle coefficient of jth vehicle;MjIt is falling for the length, width and height product of jth vehicle for the equivalent mass ratio of jth vehicle around
Number;(x[j], y[j]) be target vehicle around jth vehicle current time position vector;It is around target vehicle
The velocity vector at j vehicle current time;For the vector acceleration at jth vehicle current time around target vehicle;
Δ T is peripheral object vehicle behavior predetermined period time;||||2For the 2- norm sign of vector;
Then unit safe gesture value possessed by the position of any point in target vehicle driving region
The method for building up of the EFFICIENCY PREDICTION field:
Using target vehicle as " charge " of generation efficiency field potential, using the lengthwise position of target vehicle as the master for influencing efficiency gesture value
Want variable;
Write out unit efficiency gesture value possessed by the position of any point in target vehicle driving region
Y is any point lengthwise position in target vehicle driving region;GEFor EFFICIENCY PREDICTION field undetermined constant of driving a vehicle;M0For target
The equivalent mass ratio of vehicle is the inverse of the length, width and height product of target vehicle;y[0]For longitudinal position at target vehicle current time
It sets;
The method for building up of driving comfort prediction field:
It is " charge " for generating driving comfort field potential with target vehicle, target vehicle is gone to the transverse and longitudinal of a certain position of running region
To acceleration as the primary variables for influencing driving comfort gesture value;
Write out unit driving comfort gesture value possessed by the position of any point in target vehicle driving region
(X, Y) is any point position in target vehicle driving region;GCField undetermined constant is predicted for driving driving comfort;(x[0],
y[0]) be target vehicle current time position vector;For the velocity vector at target vehicle current time;Δ T is
Peripheral object vehicle behavior predetermined period time;||||2For the 2- norm sign of vector.
6. a kind of nearby vehicle behavior adaptive corrective prediction technique based on driving prediction field according to claim 1,
It is characterized in that, the specific implementation of the step 4 includes:
By taking target vehicle is in middle lane as an example, safety zone in travelable region is divided into 9 rows according to behavior type
For hot-zone;The decision-making level of simulation peripheral object vehicle generates and executes a certain vehicle behavior biCan blur estimation respectively drive a vehicle son predict
The containing of field field strength sum influences, and according to similarity principle, takes the central point of each behavior hot-zone as each behavior type finish time
Target vehicle position;It is fitted each vehicle behavior biCorresponding similitude track, a face inswept under the track of picking up the car
Product is integral domain
Wherein, KSFor the weight coefficient of safe prediction field field strength sum;KEFor the weight coefficient of EFFICIENCY PREDICTION field field strength sum;KSTo drive
Sail the weight coefficient of comfortable prediction field field strength sum;
By driving prediction field field strength and it is normalized, i.e., field field strength is predicted into the driving of each vehicle behavior and switchs to this
Vehicle joins the corresponding prediction probability of behavior,
Write out likelihood function L (θ)=Π Ppredict(bi), wherein θ={ KS, KE, KC};
Maximum likelihood is calculated using mature conjugate gradient method as sample set using the NGSIM data set pre-processed in step 1
EstimatorThat is initial weight COEFFICIENT KS, 0, KE, 0, KC, 0。
Obtain nearby vehicle behavior prediction pattern function
7. a kind of nearby vehicle behavior adaptive corrective prediction technique based on driving prediction field according to claim 1,
It is characterized in that, the specific implementation of the step 5 includes:
It is first depending on step 2 and obtains traffic environment participation vehicle time series data in real time, the prediction model established according to step 4 is pre- in real time
Survey vehicle typical behaviour probability P of the peripheral object vehicle in predetermined period time Δ Tpredict(bi), export maximum behavior probability
Corresponding vehicle behavior type is prediction result.
8. a kind of nearby vehicle behavior adaptive corrective prediction technique based on driving prediction field according to claim 7,
It is characterized by further comprising: returning antidote using a kind of weight coefficient for prediction result, the antidote is specific
Are as follows:
At the end of a predicted time period Δ T, the week in the Δ T of hidden Markov model identification prediction cycle time is utilized
Side vehicle typical behaviour type;
If hidden Markov model is a five-tuple (Q, V, A, B, π), observation state is expressed as V={ V1, V2..., VM, M
For the number of observation state;Hidden state is expressed as Q={ Q1, Q2..., QN, N is the number of hidden state, and I is that length is T
Status switch, O are corresponding observation sequence, I={ I1, I2..., IT, O={ O1, O2..., OT};
A=[aij]N×NTransfer for hidden state transition probability matrix, in element representation HMM model between each hidden state
Probability;Wherein,
aij=P (It+1=Qj|It=Qi), i=1,2 ..., N;J=1,2 ..., N
It is in t moment, hidden state QiUnder conditions of, it is Q in t+1 moment hidden statejProbability;
B=[bj(k)]N×MFor confusion matrix, turn between each hidden state and observation state in element representation HMM model
Move probability;Wherein,
bj(k)=P (Ot=Vk|It=Qj), k=1,2 ..., M;J=1,2 ..., N
It indicates in t moment, hidden state QjUnder the conditions of being, observation state OtProbability;
π=(πi) it is initial state probabilities matrix, wherein πi=P (I1=Qi), i=1,2 ..., N are initial time t=1 each hidden
Q containing stateiProbability;
Nearby vehicle Activity recognition, which is carried out, with HMM model is divided into two stages:
A. model training learns: initializing to each vehicle behavior prediction model, obtains initial parameter N, M, A, B, π;It extracts
The NGSIM traffic data that step 1 is handled well concentrates the characteristic O of each acquisition time in Δ Tt, i.e. lateral displacement, speed
Degree, acceleration and length travel, speed, acceleration, each time point characteristic OtThe observation sequence O=of composition
O1O2O3......OΔT;It is adopted using observation sequence as the input of HMM parameter learning according to the initial parameter after model initialization
Model λ=(A, B, π) parameter is adjusted with Baum-Welch iterative algorithm, maximizes probability function, progressive updating model ginseng
Number, finally obtains the optimal HMM model of each behavior type;
B. on-line testing identifies: utilizing trained vehicle behavior HMM model, peripheral object vehicle to be identified is mentioned through feature
Formation observation sequence is general under each HMM model using forwards algorithms calculating observation sequence as mode input after taking, encoding
Rate Precognize_k(bi);
Behavior prediction model probability function is written as
Ppredict_k(bi)=fk(KS_k, KE_K, KC_k)
That is Ppredict_k(bi) about KS_k, KE_k, KC_kFunction, k indicate k-th of predetermined period time Δ T;
Write out the residual sum of squares (RSS) function of prediction probability value Yu identification probability value, i.e. cost function
Construct the rectification functionWherein velocity coefficient is corrected for α;
Then by KS_k+1, KE_k+1, KC_k+1It substitutes into and updates prediction model parameters, it is next by predetermined period time Δ T to predict
Interior target vehicle behavior, by the rectification function, predetermined period time corrects weight coefficient online one by one, realizes adaptive prediction,
Improve the accuracy rate of nearby vehicle Activity recognition prediction.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910071784.XA CN109727490B (en) | 2019-01-25 | 2019-01-25 | Peripheral vehicle behavior self-adaptive correction prediction method based on driving prediction field |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910071784.XA CN109727490B (en) | 2019-01-25 | 2019-01-25 | Peripheral vehicle behavior self-adaptive correction prediction method based on driving prediction field |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109727490A true CN109727490A (en) | 2019-05-07 |
CN109727490B CN109727490B (en) | 2021-10-12 |
Family
ID=66300095
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910071784.XA Active CN109727490B (en) | 2019-01-25 | 2019-01-25 | Peripheral vehicle behavior self-adaptive correction prediction method based on driving prediction field |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109727490B (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110288835A (en) * | 2019-06-28 | 2019-09-27 | 江苏大学 | A kind of nearby vehicle behavior real-time identification method based on kinematics predictive compensation mechanism |
CN110851948A (en) * | 2019-08-27 | 2020-02-28 | 清华大学 | Driving environment situation assessment method and device under unstructured road condition |
CN110920631A (en) * | 2019-11-27 | 2020-03-27 | 北京三快在线科技有限公司 | Method and device for controlling vehicle, electronic equipment and readable storage medium |
CN111338335A (en) * | 2019-12-31 | 2020-06-26 | 清华大学 | Vehicle local track planning method under structured road scene |
CN113177508A (en) * | 2021-05-18 | 2021-07-27 | 中移(上海)信息通信科技有限公司 | Method, device and equipment for processing driving information |
WO2021196879A1 (en) * | 2020-03-31 | 2021-10-07 | 华为技术有限公司 | Method and device for recognizing driving behavior of vehicle |
CN113561996A (en) * | 2021-09-23 | 2021-10-29 | 国汽智控(北京)科技有限公司 | Trajectory planning method and device |
CN114852099A (en) * | 2021-02-03 | 2022-08-05 | 郑州宇通客车股份有限公司 | Method for predicting lane changing behavior of motor vehicle |
CN117922576A (en) * | 2024-03-22 | 2024-04-26 | 山东科技大学 | Automatic driving vehicle lane change decision method based on data and knowledge double driving |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20090140882A1 (en) * | 2007-12-03 | 2009-06-04 | Robert Thomas Arenburg | Dynamic speed limit system |
CN101529486A (en) * | 2006-11-01 | 2009-09-09 | 丰田自动车株式会社 | Cruise control plan evaluation device and method |
CN106408981A (en) * | 2016-10-21 | 2017-02-15 | 江苏大学 | Unmanned vehicle information optimization system and method based on vehicle information sharing |
CN106950956A (en) * | 2017-03-22 | 2017-07-14 | 合肥工业大学 | The wheelpath forecasting system of fusional movement model and behavior cognitive model |
CN107967486A (en) * | 2017-11-17 | 2018-04-27 | 江苏大学 | A kind of nearby vehicle Activity recognition method based on V2V communications with HMM-GBDT mixed models |
CN108387242A (en) * | 2018-02-07 | 2018-08-10 | 西南交通大学 | Automatic Pilot lane-change prepares and executes integrated method for planning track |
CN108470460A (en) * | 2018-04-11 | 2018-08-31 | 江苏大学 | A kind of nearby vehicle Activity recognition method based on smart mobile phone and RNN |
-
2019
- 2019-01-25 CN CN201910071784.XA patent/CN109727490B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101529486A (en) * | 2006-11-01 | 2009-09-09 | 丰田自动车株式会社 | Cruise control plan evaluation device and method |
US20090140882A1 (en) * | 2007-12-03 | 2009-06-04 | Robert Thomas Arenburg | Dynamic speed limit system |
CN106408981A (en) * | 2016-10-21 | 2017-02-15 | 江苏大学 | Unmanned vehicle information optimization system and method based on vehicle information sharing |
CN106950956A (en) * | 2017-03-22 | 2017-07-14 | 合肥工业大学 | The wheelpath forecasting system of fusional movement model and behavior cognitive model |
CN107967486A (en) * | 2017-11-17 | 2018-04-27 | 江苏大学 | A kind of nearby vehicle Activity recognition method based on V2V communications with HMM-GBDT mixed models |
CN108387242A (en) * | 2018-02-07 | 2018-08-10 | 西南交通大学 | Automatic Pilot lane-change prepares and executes integrated method for planning track |
CN108470460A (en) * | 2018-04-11 | 2018-08-31 | 江苏大学 | A kind of nearby vehicle Activity recognition method based on smart mobile phone and RNN |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110288835A (en) * | 2019-06-28 | 2019-09-27 | 江苏大学 | A kind of nearby vehicle behavior real-time identification method based on kinematics predictive compensation mechanism |
CN110288835B (en) * | 2019-06-28 | 2021-08-17 | 江苏大学 | Surrounding vehicle behavior real-time identification method based on kinematic prediction compensation mechanism |
CN110851948A (en) * | 2019-08-27 | 2020-02-28 | 清华大学 | Driving environment situation assessment method and device under unstructured road condition |
CN110920631B (en) * | 2019-11-27 | 2021-02-12 | 北京三快在线科技有限公司 | Method and device for controlling vehicle, electronic equipment and readable storage medium |
WO2021103841A1 (en) * | 2019-11-27 | 2021-06-03 | 北京三快在线科技有限公司 | Control vehicle |
CN110920631A (en) * | 2019-11-27 | 2020-03-27 | 北京三快在线科技有限公司 | Method and device for controlling vehicle, electronic equipment and readable storage medium |
CN111338335A (en) * | 2019-12-31 | 2020-06-26 | 清华大学 | Vehicle local track planning method under structured road scene |
CN111338335B (en) * | 2019-12-31 | 2021-02-26 | 清华大学 | Vehicle local track planning method under structured road scene |
WO2021196879A1 (en) * | 2020-03-31 | 2021-10-07 | 华为技术有限公司 | Method and device for recognizing driving behavior of vehicle |
CN114852099A (en) * | 2021-02-03 | 2022-08-05 | 郑州宇通客车股份有限公司 | Method for predicting lane changing behavior of motor vehicle |
CN113177508A (en) * | 2021-05-18 | 2021-07-27 | 中移(上海)信息通信科技有限公司 | Method, device and equipment for processing driving information |
CN113177508B (en) * | 2021-05-18 | 2022-04-08 | 中移(上海)信息通信科技有限公司 | Method, device and equipment for processing driving information |
CN113561996A (en) * | 2021-09-23 | 2021-10-29 | 国汽智控(北京)科技有限公司 | Trajectory planning method and device |
CN113561996B (en) * | 2021-09-23 | 2021-12-21 | 国汽智控(北京)科技有限公司 | Trajectory planning method and device |
CN117922576A (en) * | 2024-03-22 | 2024-04-26 | 山东科技大学 | Automatic driving vehicle lane change decision method based on data and knowledge double driving |
CN117922576B (en) * | 2024-03-22 | 2024-05-17 | 山东科技大学 | Automatic driving vehicle lane change decision method based on data and knowledge double driving |
Also Published As
Publication number | Publication date |
---|---|
CN109727490B (en) | 2021-10-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109727490A (en) | A kind of nearby vehicle behavior adaptive corrective prediction technique based on driving prediction field | |
Zhang et al. | Human-like autonomous vehicle speed control by deep reinforcement learning with double Q-learning | |
US11726477B2 (en) | Methods and systems for trajectory forecasting with recurrent neural networks using inertial behavioral rollout | |
CN110796856B (en) | Vehicle lane change intention prediction method and training method of lane change intention prediction network | |
CN110007675B (en) | Vehicle automatic driving decision-making system based on driving situation map and training set preparation method based on unmanned aerial vehicle | |
CN110264721B (en) | Urban intersection surrounding vehicle track prediction method | |
CN111145552B (en) | Planning method for vehicle dynamic lane changing track based on 5G network | |
CN109726804B (en) | Intelligent vehicle driving behavior personification decision-making method based on driving prediction field and BP neural network | |
CN112799386B (en) | Robot path planning method based on artificial potential field and reinforcement learning | |
CN109460023A (en) | Driver's lane-changing intention recognition methods based on Hidden Markov Model | |
CN107886750B (en) | Unmanned automobile control method and system based on beyond-visual-range cooperative cognition | |
CN113044064B (en) | Vehicle self-adaptive automatic driving decision method and system based on meta reinforcement learning | |
CN110304074A (en) | A kind of hybrid type driving method based on stratification state machine | |
CN112257850A (en) | Vehicle track prediction method based on generation countermeasure network | |
CN110288835B (en) | Surrounding vehicle behavior real-time identification method based on kinematic prediction compensation mechanism | |
CN112162555A (en) | Vehicle control method based on reinforcement learning control strategy in hybrid vehicle fleet | |
CN112356830A (en) | Intelligent parking method based on model reinforcement learning | |
CN103914985A (en) | Method for predicting future speed trajectory of hybrid power bus | |
CN111625989A (en) | Intelligent vehicle influx method and system based on A3C-SRU | |
CN111401458A (en) | Multi-model target state prediction method and system based on deep reinforcement learning | |
CN115662166B (en) | Automatic driving data processing method and automatic driving traffic system | |
CN115257745A (en) | Automatic driving lane change decision control method based on rule fusion reinforcement learning | |
CN111580526B (en) | Cooperative driving method for fixed vehicle formation scene | |
Jiang et al. | Implementation of human-like driver model based on recurrent neural networks | |
CN108470460B (en) | Peripheral vehicle behavior identification method based on smart phone and RNN |
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 |