CN115743179A - Vehicle probability multi-mode expected trajectory prediction method - Google Patents
Vehicle probability multi-mode expected trajectory prediction method Download PDFInfo
- Publication number
- CN115743179A CN115743179A CN202211498750.7A CN202211498750A CN115743179A CN 115743179 A CN115743179 A CN 115743179A CN 202211498750 A CN202211498750 A CN 202211498750A CN 115743179 A CN115743179 A CN 115743179A
- Authority
- CN
- China
- Prior art keywords
- vehicle
- lane
- information
- predicted
- characteristic information
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Traffic Control Systems (AREA)
Abstract
The invention belongs to the technical field of intelligent driving, and particularly relates to a vehicle probability multi-mode expected trajectory prediction method, which comprises the following steps of 1: in order to reduce the complexity of the neural network, extracting the characteristic information of the predicted vehicle and the peripheral vehicles as integral input information; carrying out data processing on speed, acceleration, coordinates and behavior intention labels contained in the traffic vehicle; step 2: preprocessing data and generating a data set for training, testing and verifying a neural network model; and step 3: acquiring characteristic information of a predicted vehicle and surrounding traffic vehicles through a one-way LSTM encoder, transmitting the encoded characteristic information into a behavior intention recognition module, training the behavior intention recognition module, and recognizing the probability of each behavior intention through a softmax function; step four: outputting a future multi-modal trajectory of the vehicle through the bidirectional LSTM and the mixed density network; the intelligent driving safety prediction method is reasonable in structure, practical significance is achieved in predicting multi-modal tracks and improving intelligent driving safety decisions, and multiple tracks which may occur in the future of the vehicle can be effectively predicted through historical track information.
Description
Technical Field
The invention relates to the technical field of intelligent driving, in particular to a vehicle probability multi-mode expected trajectory prediction method.
Background
In order to safely and effectively travel in complex traffic scenarios, an intelligent drive automobile needs to have the ability to predict the intent and future trajectory of the surrounding vehicle. Trajectory prediction also constitutes an important part of an intelligent driving Local Dynamic Driving Map (LDDM). The LDDM comprises the static situation of the environment at the current moment and the dynamic situation cognition at the future moment, and is a compact road environment modeling method for situation cognition of a complex dynamic driving traffic environment. The good track prediction capability not only helps the vehicle make a decision in advance, thereby reducing the incidence rate of traffic accidents, but also can better improve the safety, stability and efficiency of autonomous driving vehicle running.
Different autonomous vehicles behave differently under the same scenario, i.e., there are multiple possible outcomes in the future, because there is an inherent uncertainty in predicting the future. The uncertainty problem in predicting the future leads to the multi-modal nature of trajectory prediction, making trajectory prediction a challenging problem.
Multi-modal expected trajectory prediction aims to generate multiple possible and safe trajectories for traffic vehicles in a complex highly dynamic environment. The difficulty of multi-modal prediction is that there are many possibilities for future behavior or trajectory corresponding to the same input, and the data obtained in the real world, i.e. the true value of the prediction task, is sampled from these many possibilities once, and others may not be known. Multi-modal trajectory prediction involves two major tasks: (1) how to represent the multi-modal nature of the prediction: different targets may have different future trajectories for the same historical trajectory. (2) how to model the interaction between objects: the behavior between objects is influenced not only by its own intention but also by other objects in the surroundings.
In recent years, many researchers have studied future trajectory information of a vehicle, and trajectory prediction models are roughly classified into a prediction model based on physical constraints and a prediction model based on data driving. The prediction model based on physical constraints mainly considers state quantities such as vehicle motion states, road environment factors and automobile characteristics and adopts a dynamics and kinematics model to predict the future motion trend of a vehicle, but the model is too dependent on the certainty of the current state of the vehicle and the integrity of model input, and the uncertainty of control input variables is generally modeled by a Kalman filtering method and a Monte Carlo method to improve the prediction precision of motion tracks.
In order to solve the problem of low long-term prediction accuracy in a dynamic environment, more and more learners pay attention to the track prediction based on deep learning. The method can predict the distribution condition of the running track of the future traffic vehicle based on the LSTM network model of the convolution social pool, but the method uses the convolution network to extract the environmental features, increases the complexity of the environment modeling of the predicted vehicle, lacks the interpretability of track prediction and is difficult to meet the real-time requirement of intelligent driving of the vehicle. The method comprises the steps of constructing a vehicle behavior intention model and a track prediction model of the expressway based on UB-LSTM, predicting a future single-mode track of the vehicle, effectively identifying the future behavior intention of the vehicle and fitting the vehicle track in an optimized mode, wherein the intention probability output by the intention recognition model is not applied to the track prediction model, so that the track output by the track prediction model has a large error with a real track.
The LSTM has good performance in processing the problem of the long time domain track sequence in the dynamic environment, makes up the problem of vehicle model prediction, but has a defect in predicting the specific driving behavior intention of the intelligent driving traffic vehicle. In order to confirm the complex intelligent driving traffic environment, the intelligent driving traffic vehicle and the driving scene thereof need to be subjected to specific prediction. The intelligent driving traffic vehicle can generate various driving behaviors such as straight line driving, left lane changing, right lane changing, acceleration, deceleration and the like in the driving process, and in the driving process of the intelligent driving traffic vehicle, the specific driving behaviors of the predicted vehicle influence the cognition of the vehicle on the complex traffic driving environment, and influence modules such as decision planning and the like to make the most reasonable decision and control.
In order to solve the defects, a probabilistic multi-modal expected trajectory prediction model is provided by deeply extracting environmental characteristic information. The model generates a variety of likely future trajectory information directly from historical environmental information. The model is built based on a Seq2Seq encoder-decoder architecture, representing the overall environment of the vehicle by inputting both the predicted vehicle and the surrounding vehicle characteristic information into the RNN. Meanwhile, end-to-end training can be efficiently carried out on the traffic vehicles in the scene.
Disclosure of Invention
This section is for the purpose of summarizing some aspects of embodiments of the invention and to briefly introduce some preferred embodiments. In this section, as well as in the abstract and the title of the invention of this application, simplifications or omissions may be made to avoid obscuring the purpose of the section, the abstract and the title, and such simplifications or omissions are not intended to limit the scope of the invention.
The present invention has been made in view of the problems occurring in the prior art.
Therefore, the invention aims to provide a vehicle probability multi-mode expected trajectory prediction method, which has practical significance in predicting multi-mode trajectories and improving intelligent driving safety decisions and can effectively predict multiple possible future trajectories of a vehicle through historical trajectory information.
To solve the above technical problem, according to an aspect of the present invention, the present invention provides the following technical solutions:
a vehicle probability multi-modal expected trajectory prediction method comprises the following steps:
wherein the content of the first and second substances,
step 1: in order to reduce the complexity of the neural network, extracting the characteristic information of the predicted vehicle and the peripheral vehicles as integral input information; carrying out data processing on speed, acceleration, coordinates and behavior intention labels contained in the traffic vehicle;
step 2: preprocessing data and generating a data set for training, testing and verifying a neural network model;
and step 3: acquiring characteristic information of a predicted vehicle and surrounding traffic vehicles through a one-way LSTM encoder, transmitting the encoded characteristic information into a behavior intention recognition module, training the behavior intention recognition module, and recognizing the probability of each behavior intention through a softmax function;
and 4, step 4: acquiring characteristic information of a predicted vehicle and surrounding traffic vehicles through a bidirectional LSTM encoder;
and 5: combining the intention probabilities of the behavior intention recognition module and the coding characteristic information of the track prediction module into a middle vector;
step 6: inputting the complete intermediate vector into an LSTM decoder and a mixed density network to perform multi-modal prediction;
and 7: and training a multi-mode vehicle track prediction module, taking a maximum likelihood function loss function of minimized negative logarithms as an optimization target, and outputting tracks with higher probability.
As a preferable aspect of the vehicle probability multi-modal expected trajectory prediction method according to the present invention, wherein: the input information comprises predicted vehicle historical characteristic information and environmental characteristic information, and the input information is as follows:
whereinx t ,y t Respectively the transverse and longitudinal coordinate information of the predicted vehicle,respectively predicted vehicle lateral longitudinal speed information,respectively, the predicted lateral and longitudinal acceleration information of the vehicle, d hw ,t hw ,t tc The vehicle headway, the time interval and the collision time of the predicted vehicle and the front vehicle are respectively.
As an optimization method of the vehicle probability multi-modal expected trajectory prediction methodA table, wherein: the environmental information is characterized as ps = (LF, LA, LB, MF, MB, RF, RA, RB) is the state feature information of the traffic vehicles in the vicinity of the predicted vehicle.Wherein, the first and the second end of the pipe are connected with each other,is the interaction information of the predicted vehicle and the surrounding traffic vehicles, is the interaction information between the surrounding traffic vehicles, reflects the absolute distance between the vehicles on the left lane and the right lane of the predicted vehicle,
as a preferable aspect of the vehicle probability multi-modal expected trajectory prediction method according to the present invention, wherein: the classification steps of the track changing tracks of the vehicle are as follows:
(1) Extracting intersection points of the track and the lane lines, and recording the time at the moment;
(2) Calculating a yaw angle theta;
(3) Forward and backward deducing a yaw angle of the sampling point;
(4)|θ|<θ b (course angle threshold of starting point of lane change), defined as straight-line driving and reverse drivingThe definition of it is lane change;
(5) Determining a lane change starting point and a lane change end point;
in the formula, x t ,y t Is the abscissa, x, of the vehicle at time t t+1 ,y t+1 Is the abscissa and ordinate of the vehicle at time t + 1.
As a preferable aspect of the vehicle probability multi-modal expected trajectory prediction method according to the present invention, wherein: the intelligent driving traffic vehicle driving behavior intention recognition module outputs five driving behavior probabilities of a vehicle at the current moment, such as lane keeping, lane changing on the left, lane changing on the right, lane changing on the left accelerating, lane changing on the right accelerating, and the like, through a normalized exponential function (Softmax function) based on historical state information of the traffic vehicle. Setting historical state characteristic information M of the overall environment of the intelligent driving traffic vehicle at the current moment as an input vector of a vehicle motion prediction model, and C = (C) 1 ,c 2 ,c 3 ,c 4 ,c 5 ) An intention category vector output for a driving behavior intention recognition module, c 1 ,c 2 ,c 3 ,c 4 ,c 5 Respectively represents 5 driving intention categories of lane keeping, left lane changing, right lane changing, left accelerating lane changing and right accelerating lane changing.
The vehicle behavior intent recognition module loss function is:
in the formula, L c Identifying a module penalty function for the behavioral intent; y is n Is a one-hot encoding of the nth sample label; p is a radical of n The class probability for the nth prediction sample.
Wherein ω is 1 ,ω 2 ,ω 3 ,ω 4 ,ω 5 Respectively representing 5 driving intention category probabilities of lane keeping, left lane changing, right lane changing, left accelerating lane changing and right accelerating lane changing.
As a preferable aspect of the vehicle probability multi-modal expected trajectory prediction method according to the present invention, wherein: and predicting the probability distribution of the coordinate information of the future track through the MDN network, and outputting various behaviors and tracks which may occur in the future of the vehicle. The invention adopts the combination of n Gaussian functions as the kernel function of the MDN, and the trace distribution probability output by the MDN layer is as follows:
in the formula, x is an input characteristic parameter; o is the position of the vehicle at a certain moment; n is the number of the mixing kernel functions; alpha is alpha i (x) Is a model weight coefficient; mu.s i (x) Is the center of the ith kernel function.
As a preferable aspect of the vehicle probability multi-modal expected trajectory prediction method according to the present invention, wherein: ensuring that the sum of the model weight coefficients is 1 and each item is greater than 0, and simultaneously ensuring sigma by exponential operation i Is positive.
As a preferable aspect of the vehicle probability multi-modal expected trajectory prediction method according to the present invention, wherein: and training a network model, and using a maximum likelihood function loss function of minimized negative logarithms as an optimization target.
In the formula, X obs For a sequence of historical trajectories of the predicted vehicle, C k Driving behaviour predicted for the driving behaviour prediction phase, G for the trajectory prediction phaseOf future trajectory of the vehicle.
Compared with the prior art, the invention has the beneficial effects that: the method has practical significance in predicting the multi-modal track and improving the intelligent driving safety decision, and can effectively predict a plurality of tracks which may occur in the future of the vehicle through the historical track information. The invention provides a multi-modal characteristic of a track prediction model expressing a prediction result through an interactive relation between deep modeling targets. The driving behavior intention recognition model recognizes various behavior intention probabilities of vehicle lane keeping, left lane changing, right lane changing, left acceleration lane changing and right acceleration lane changing by extracting state characteristic information and environment interaction characteristic information between a predicted vehicle and a surrounding traffic vehicle in a complex dynamic traffic environment, and the track prediction model predicts multi-mode track information of the vehicle in the future 6s according to state characteristic information of history 3 s. The multi-mode track prediction model has good performances in driving behavior intention recognition and multi-mode tracks based on the state information and interaction information of the predicted vehicle and surrounding traffic vehicles, makes up the defects of specific behavior prediction and environment interaction characteristic information of the expected track of the intelligent driving system, and provides prior information for the dynamic situation cognition of the subsequent working traffic environment.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the present invention will be described in detail with reference to the accompanying drawings and detailed embodiments, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise. Wherein:
FIG. 1 is a schematic structural view of the present invention;
FIG. 2 is a schematic view of the model structure of the present invention;
FIG. 3 is a diagram illustrating the effect of multi-modal trajectory prediction according to the present invention.
Detailed Description
In order to make the aforementioned objects, features and advantages of the present invention more comprehensible, embodiments accompanying figures are described in detail below.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, but the present invention may be practiced in other ways than those specifically described herein, and it will be apparent to those of ordinary skill in the art that the present invention may be practiced without departing from the spirit and scope of the present invention, and therefore the present invention is not limited to the specific embodiments disclosed below.
To make the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention will be described in detail with reference to the accompanying drawings.
The invention provides the following technical scheme: a vehicle probability multi-mode expected trajectory prediction method has practical significance in predicting multi-mode trajectories and improving intelligent driving safety decisions, and can effectively predict multiple possible future trajectories of a vehicle through historical trajectory information;
example 1
In a complex dynamic traffic environment, the intelligent driving transportation vehicle motion prediction not only considers the motion state of the predicted vehicle, but also considers the environmental information of the predicted vehicle, namely the characteristic information of the surrounding transportation vehicles and the interaction characteristic information between the predicted vehicle and the surrounding transportation vehicles. In order to enable the motion prediction model to understand interactive behaviors among vehicles, input information comprises predicted vehicle historical characteristic information and environmental characteristic information, and the input information is as follows:
in the formula, M t For the input amount of the intelligent driving traffic vehicle motion prediction model,as characteristic information of the predicted vehicle, E t T is the characteristic information and the interaction characteristic information of the traffic vehicles around the predicted vehicle, and the time length of the historical track of the vehicle. Wherein x t ,y t Respectively the transverse and longitudinal coordinate information of the predicted vehicle,respectively predicted vehicle transverse and longitudinal speed information,respectively, the predicted vehicle transverse and longitudinal acceleration information, d hw ,t hw ,t tc The vehicle headway, the time interval and the collision time of the predicted vehicle and the front vehicle are respectively.
The classification steps of the track changing tracks of the vehicle are as follows:
extract the intersection of the trajectory and the lane line and record the time at that moment
Calculate the yaw angle θ
Forward and reverse deduction of the yaw angle of the sample point
·|θ|<θ b (course angle threshold of starting point of lane change), defined as straight line driving, otherwise defined as lane change
Determining a lane change start and end
In the formula, x t ,y t Is the abscissa, x, of the vehicle at time t t+1 ,y t+1 Is the abscissa and ordinate of the vehicle at time t + 1.
The environment information is represented by eight directions of the predicted vehicle, including Left Front (LF), right Left (LA), left rear (LB), right front (MF), right rear (MB), right Front (RF), right front (RA), and right Rear (RB). The environmental information is characterized as ps = (LF, LA, LB, MF, MB, RF, RA, RB) is the state feature information of the traffic vehicles around the predicted vehicle.Wherein the content of the first and second substances,is the interaction information of the predicted vehicle and the surrounding traffic vehicles, is the interaction information among the surrounding traffic vehicles, reflects the absolute distance between the vehicles in the left lane and the right lane of the predicted vehicle,
the probability multi-modal expected trajectory prediction model provided by the invention consists of a driving behavior intention recognition module and an expected trajectory prediction module, and is shown in FIG. 2. The intelligent driving traffic vehicle driving behavior intention recognition module outputs five driving behavior probabilities of a vehicle at the current moment, such as lane keeping, lane changing on the left, lane changing on the right, lane changing on the left accelerating, lane changing on the right accelerating, and the like, through a normalized exponential function (Softmax function) based on historical state information of the traffic vehicle. Setting historical state characteristic information M of the overall environment of the intelligent driving traffic vehicle at the current moment as an input vector of a vehicle motion prediction model, and C = (C) 1 ,c 2 ,c 3 ,c 4 ,c 5 ) An intention category vector output for a driving behavior intention recognition module, c 1 ,c 2 ,c 3 ,c 4 ,c 5 Respectively representing 5 driving intention categories of lane keeping, left lane changing, right lane changing, left accelerating lane changing and right accelerating lane changing. Omega is a probability vector composed of driving behavior intents, where omega is 1 ,ω 2 ,ω 3 ,ω 4 ,ω 5 Respectively representing 5 driving intention category probabilities of lane keeping, left lane changing, right lane changing, left accelerating lane changing and right accelerating lane changing. The driving intention category probability is:
the vehicle behavior intent recognition module loss function is:
in the formula, L c Identifying a module penalty function for the behavioral intent; y is n Is a one-hot encoding of the nth sample label; p is a radical of n The class probability for the nth prediction sample.
The expected trajectory prediction module outputs a probability distribution of future time domain vehicle trajectories based on the historical encoded information and the driving behavior intent probability. The traffic vehicle expected trajectory prediction module is composed of a full connection layer, an encoder, a decoder, an MLP and an MDN network, wherein the full connection layer extracts characteristic information of a traffic vehicle historical state as input of the encoder, and the encoder encodes the input characteristic information into a context vector by adopting a recurrent neural network LSTM for improving the context of the current state. The context vector generated by the encoder and the behavior recognition vector output by the driving behavior intention recognition module generate a vehicle track information intermediate vector and serve as the input of the decoder. The MLP network and the MDN network input the decoder's output vector, allowing the model to predict the probability distribution of future traces based on intent recognition. In order to enable the traffic vehicle to better accord with the diversity of driving behaviors and reflect the uncertainty of the track, the probability distribution of the coordinate information of the future track is predicted through the MDN network, and various behaviors and tracks which may occur in the future of the vehicle are output. The invention adopts the combination of n Gaussian functions as the kernel function of the MDN, and the trace distribution probability output by the MDN layer is as follows:
in the formula, x is an input characteristic parameter; o is the position of the vehicle at a certain moment; n is the number of the mixing kernel functions; alpha (alpha) ("alpha") i (x) Is a model weight coefficient; mu.s i (x) Is the center of the ith kernel function.
Ensuring that the sum of the model weight coefficients is 1 and each item is greater than 0, and simultaneously ensuring sigma through exponential operation i Is positive.
And training a network model, and using a maximum likelihood function loss function of minimized negative logarithms as an optimization target.
In the formula, X obs For a sequence of historical trajectories of the predicted vehicle, C k And G is Gaussian distribution of the future track of the vehicle predicted in the track prediction stage.
Although the present invention has been described above with reference to the embodiments, various modifications may be made thereto without departing from the scope of the invention. In particular, the various features of the disclosed embodiments of the invention may be used in any combination, provided that no structural conflict exists, and the combinations are not exhaustively described in this specification merely for the sake of brevity and resource conservation. Therefore, it is intended that the invention not be limited to the particular embodiments disclosed, but that the invention will include all embodiments falling within the scope of the appended claims.
Claims (8)
1. A vehicle probability multi-mode expected trajectory prediction method is characterized by comprising the following steps: the method comprises the following steps:
wherein the content of the first and second substances,
step 1: in order to reduce the complexity of a neural network, extracting characteristic information of a predicted vehicle and surrounding vehicles as integral input information; carrying out data processing on speed, acceleration, coordinates and behavior intention labels contained in the traffic vehicle;
step 2: preprocessing data and generating a data set for training, testing and verifying a neural network model;
and step 3: acquiring characteristic information of a predicted vehicle and surrounding traffic vehicles through a one-way LSTM encoder, transmitting the encoded characteristic information into a behavior intention recognition module, training the behavior intention recognition module, and recognizing the probability of each behavior intention through a softmax function;
and 4, step 4: acquiring characteristic information of a predicted vehicle and surrounding traffic vehicles through a bidirectional LSTM encoder;
and 5: combining the intention probabilities of the behavior intention recognition module and the coding characteristic information of the track prediction module into a middle vector;
step 6: inputting the complete intermediate vector into an LSTM decoder and a mixed density network to perform multi-modal prediction;
and 7: and training a multi-mode vehicle track prediction module, taking a maximum likelihood function loss function of minimized negative logarithms as an optimization target, and outputting a track with higher probability.
2. The probabilistic multi-modal vehicle prospective trajectory prediction method according to claim 1, wherein: the input information comprises the historical characteristic information of the predicted vehicle and the environmental characteristic information, and the input information comprises the following components:
whereinx t ,y t Respectively are the transverse and longitudinal coordinate information of the predicted vehicle,respectively predicted vehicle lateral longitudinal speed information,respectively, the predicted lateral and longitudinal acceleration information of the vehicle, d hw ,t hw ,t tc The vehicle headway, the time interval and the collision time of the predicted vehicle and the front vehicle are respectively.
3. The probabilistic multi-modal vehicle prospective trajectory prediction method according to claim 1, wherein: the environmental information is characterized as Is the state characteristic information of the traffic vehicles around the predicted vehicle, where ps = (LF, LA, LB, MF, MB, RF, RA, RB) is the vehicle bearing information of the predicted left front, right left, left rear, right front, right middle, and right rear, respectively.Wherein the content of the first and second substances,is the interaction information of the predicted vehicle and the surrounding traffic vehicles, is the interaction information among the surrounding traffic vehicles, reflects the absolute distance between the vehicles in the left lane and the right lane of the predicted vehicle,
4. the vehicle probabilistic multi-modal expected trajectory prediction method according to claim 1, wherein: the classification steps of the track changing tracks of the vehicle are as follows:
(1) Extracting intersection points of the track and the lane lines, and recording the time at the moment;
(2) Calculating a yaw angle theta;
(3) Forward and backward deducing a yaw angle of the sampling point;
(4)|θ|<θ b (a course angle threshold value of a lane change starting point) is defined as straight line driving, otherwise, the lane change is defined as lane change;
(5) Determining a lane change starting point and a lane change end point;
in the formula, x t ,y t Is the abscissa, x, of the vehicle at time t t+1 ,y t+1 Is the abscissa and ordinate of the vehicle at time t + 1.
5. The vehicle probabilistic multi-modal expected trajectory prediction method according to claim 1, wherein: the intelligent driving traffic vehicle driving behavior intention recognition module outputs current through a normalized exponential function (Softmax function) based on historical state information of the traffic vehicleAt the moment, the vehicle keeps on the lane, changes the lane on the left, changes the lane on the right, changes the lane with the left acceleration, changes the lane with the right acceleration and so on five kinds of driving behavior probabilities. Setting historical state characteristic information M of the overall environment of the intelligent driving traffic vehicle at the current moment as an input vector of a vehicle motion prediction model, and C = (C) 1 ,c 2 ,c 3 ,c 4 ,c 5 ) An intention category vector output for a driving behavior intention recognition module, c 1 ,c 2 ,c 3 ,c 4 ,c 5 Respectively representing 5 driving intention categories of lane keeping, left lane changing, right lane changing, left accelerating lane changing and right accelerating lane changing.
The vehicle behavior intent recognition module loss function is:
in the formula, L c Identifying a module penalty function for the behavioral intent; y is n Is a one-hot encoding of the nth sample label; p is a radical of n The class probability for the nth prediction sample.
Wherein omega 1 ,ω 2 ,ω 3 ,ω 4 ,ω 5 Respectively representing 5 driving intention category probabilities of lane keeping, left lane changing, right lane changing, left accelerating lane changing and right accelerating lane changing.
6. The vehicle probabilistic multi-modal expected trajectory prediction method according to claim 1, wherein: and predicting the probability distribution of the future track coordinate information through the MDN network, and outputting various behaviors and tracks which are possible to occur in the future of the vehicle. The invention adopts the combination of n Gaussian functions as the kernel function of the MDN, and the trace distribution probability output by the MDN layer is as follows:
in the formula, x is an input characteristic parameter; o is the position of the vehicle at a certain moment; n is the number of the mixing kernel functions; alpha is alpha i (x) Is a model weight coefficient; mu.s i (x) Is the center of the ith kernel function.
8. The vehicle probabilistic multi-modal expected trajectory prediction method according to claim 1, wherein: and training a network model, and using a maximum likelihood function loss function of minimized negative logarithms as an optimization target.
In the formula, X obs For a sequence of historical trajectories of the predicted vehicle, C k And G is Gaussian distribution of the future track of the vehicle predicted in the track prediction stage.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211498750.7A CN115743179A (en) | 2022-11-28 | 2022-11-28 | Vehicle probability multi-mode expected trajectory prediction method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202211498750.7A CN115743179A (en) | 2022-11-28 | 2022-11-28 | Vehicle probability multi-mode expected trajectory prediction method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN115743179A true CN115743179A (en) | 2023-03-07 |
Family
ID=85339480
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202211498750.7A Pending CN115743179A (en) | 2022-11-28 | 2022-11-28 | Vehicle probability multi-mode expected trajectory prediction method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115743179A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117077042A (en) * | 2023-10-17 | 2023-11-17 | 北京鑫贝诚科技有限公司 | Rural level crossing safety early warning method and system |
CN117333847A (en) * | 2023-12-01 | 2024-01-02 | 山东科技大学 | Track prediction method and system based on vehicle behavior recognition |
CN117523821A (en) * | 2023-10-09 | 2024-02-06 | 苏州大学 | System and method for predicting vehicle multi-mode driving behavior track based on GAT-CS-LSTM |
-
2022
- 2022-11-28 CN CN202211498750.7A patent/CN115743179A/en active Pending
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117523821A (en) * | 2023-10-09 | 2024-02-06 | 苏州大学 | System and method for predicting vehicle multi-mode driving behavior track based on GAT-CS-LSTM |
CN117077042A (en) * | 2023-10-17 | 2023-11-17 | 北京鑫贝诚科技有限公司 | Rural level crossing safety early warning method and system |
CN117077042B (en) * | 2023-10-17 | 2024-01-09 | 北京鑫贝诚科技有限公司 | Rural level crossing safety early warning method and system |
CN117333847A (en) * | 2023-12-01 | 2024-01-02 | 山东科技大学 | Track prediction method and system based on vehicle behavior recognition |
CN117333847B (en) * | 2023-12-01 | 2024-03-15 | 山东科技大学 | Track prediction method and system based on vehicle behavior recognition |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11899411B2 (en) | Hybrid reinforcement learning for autonomous driving | |
CN106874597B (en) | highway overtaking behavior decision method applied to automatic driving vehicle | |
CN115743179A (en) | Vehicle probability multi-mode expected trajectory prediction method | |
CN111746559B (en) | Method and system for predicting lane changing intention of front vehicle | |
CN111930110A (en) | Intent track prediction method for generating confrontation network by combining society | |
CN113044064B (en) | Vehicle self-adaptive automatic driving decision method and system based on meta reinforcement learning | |
CN112734808B (en) | Trajectory prediction method for vulnerable road users in vehicle driving environment | |
CN115683145B (en) | Automatic driving safety obstacle avoidance method based on track prediction | |
CN115053237A (en) | Vehicle intent prediction neural network | |
Zhu et al. | Interaction-aware cut-in trajectory prediction and risk assessment in mixed traffic | |
CN114212110A (en) | Obstacle trajectory prediction method, obstacle trajectory prediction device, electronic device, and storage medium | |
Yu et al. | Machine Learning-Based Vehicle Intention Trajectory Recognition and Prediction for Autonomous Driving | |
CN114030485A (en) | Automatic driving automobile man lane change decision planning method considering attachment coefficient | |
CN115062202A (en) | Method, device, equipment and storage medium for predicting driving behavior intention and track | |
Lu et al. | A sharing deep reinforcement learning method for efficient vehicle platooning control | |
Ren et al. | Self-learned intelligence for integrated decision and control of automated vehicles at signalized intersections | |
Atakishiyev et al. | Explaining autonomous driving actions with visual question answering | |
Gao et al. | Deep learning‐based hybrid model for the behaviour prediction of surrounding vehicles over long‐time periods | |
CN114889608A (en) | Attention mechanism-based vehicle lane change prediction method | |
CN113799799A (en) | Security compensation method and device, storage medium and electronic equipment | |
Pavelko et al. | Modification and Experimental Validation of a Logistic Regression Vehicle-Pedestrian Model | |
Yang et al. | Deep Reinforcement Learning Lane-Changing Decision Algorithm for Intelligent Vehicles Combining LSTM Trajectory Prediction | |
Xing et al. | A personalized deep learning approach for trajectory prediction of connected vehicles | |
Tan et al. | RCP‐RF: A comprehensive road‐car‐pedestrian risk management framework based on driving risk potential field | |
Adadi et al. | Explainable Artificial Intelligence for Intelligent Transportation Systems: Are We There Yet? |
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 |