CN113561974B - Collision risk prediction method based on coupling of vehicle behavior interaction and road structure - Google Patents

Abstract

The application discloses a collision risk prediction method and a collision risk prediction device based on coupling of vehicle behavior interaction and a road structure, which are based on road structure classification and potential double-vehicle conflict identification and projected to two basic interactive conflict scene models; aiming at vehicle-to-vehicle collision, an intention identification model based on a dynamic Bayesian network is established, and a conditional probability relation between the passing intention of the vehicle and the environment situation and the driving behavior is described; based on the observable information, carrying out probability inference on the environment situation and the behavior semantics; and training and optimizing model parameters of the dynamic Bayesian network based on the natural driving data by adopting an EM algorithm. Based on the passing intention identification result, a Gaussian process regression algorithm is used for predicting the running track of the vehicle and the space-time distribution of the running track of the vehicle, and the collision risk of the two vehicles is estimated. Therefore, under complex traffic scenes such as ramp entry, intersection traffic and the like, collision risk prediction considering multi-vehicle behavior interactive coupling is realized, the method can be widely applied to collision scenes, and the driving safety of intelligent vehicles is improved.

Description

Collision risk prediction method based on coupling of vehicle behavior interaction and road structure

Technical Field

The application relates to the technical field of intelligent driving vehicle environment cognition, in particular to a collision risk prediction method and device based on coupling of vehicle behavior interaction and road structures.

Background

The intelligent driving technology is a basic technology for realizing a safer and more efficient intelligent traffic system in the future, and belongs to the field of hot spot research focused on by various countries. The risk assessment is a key technology of intelligent driving, and the function of the risk assessment is to analyze the driving risk based on the environmental perception information fed back by the sensor so as to provide a decision basis for subsequent driving decisions.

Collision risk is an important component of the risk of driving. In an actual traffic system, a large number of potential conflicts exist among vehicles, for example, in scenes of intersections, remittance and remittance, and the like, behaviors of the conflicting vehicles are interacted with each other, and the behaviors are intended to have the characteristics of time varying, uncertainty and difficulty in direct observation. The existing collision risk prediction method often uses a single vehicle as an analysis object, or can not give an interpretable analysis to a vehicle interaction process, so that the collision risk in a vehicle-to-vehicle collision scene is difficult to effectively predict.

Disclosure of Invention

The present application aims to solve at least one of the technical problems in the related art to some extent.

Therefore, an object of the present application is to provide a collision risk prediction method based on coupling of vehicle behavior interaction and road structure, which can take vehicle history track and road structure information as input, identify potential collision relation of vehicles and predict collision risk thereof, and provide basis for further behavior decision of intelligent vehicles.

Another object of the present application is to propose a collision risk prediction device based on vehicle behavior interactions coupled with road structures.

In order to achieve the above objective, an embodiment of the present application provides a collision risk prediction method based on coupling of vehicle behavior interaction and road structure, the method comprising the following steps:

identifying potential double-car conflicts based on the road structure, and projecting the double-car conflicts to a basic interactive conflict scene model;

establishing an intention recognition model based on a dynamic Bayesian network according to the double-vehicle conflict so as to describe the conditional probability relation between the passing intention of the vehicle and the environment situation and the driving behavior;

respectively establishing a probability map model according to the environmental situation and the semantic behavior so as to respectively utilize observable scene physical information and two-vehicle motion information to carry out probability inference on the environmental situation and the semantic behavior;

Carrying out parameter pre-calibration on the intention recognition model parameters based on experience, and carrying out parameter learning based on an EM algorithm and natural driving data;

and outputting a two-vehicle passing intention recognition result based on the intention recognition model, predicting two-vehicle movement tracks by using a Gaussian process regression algorithm, and outputting a collision risk assessment result based on the Gaussian distribution overlapping degree of the positions of each moment.

To achieve the above object, another embodiment of the present application provides a collision risk prediction apparatus coupled with a road structure based on vehicle behavior interaction, including:

the projection module is used for identifying potential collision of the two vehicles based on the road structure and projecting the collision of the two vehicles to the basic interactive collision scene model;

the modeling module is used for establishing an intention recognition model based on a dynamic Bayesian network according to the double-vehicle conflict so as to describe the conditional probability relation between the passing intention of the vehicle and the environment situation and the driving behavior;

the inference module is used for respectively establishing a probability map model according to the environmental situation and the semantic behaviors so as to respectively utilize observable scene physical information and two-vehicle motion information to carry out probability inference on the environmental situation and the semantic behaviors, taking the environmental situation and the semantic behaviors as observation input and carrying out probability inference on the passing intention of the conflict vehicle based on a dynamic Bayesian network;

The training module is used for carrying out parameter pre-calibration on the intention identification model parameters based on experience, carrying out parameter learning based on an EM algorithm and natural driving data, classifying the natural driving data set based on indexes such as intention and the like, and respectively training a corresponding Gaussian process regression model for subsequent track prediction;

the prediction module is used for outputting a two-vehicle passing intention recognition result based on the intention recognition model, predicting the motion track of the two vehicles by utilizing a Gaussian process regression algorithm, and outputting a collision risk assessment result based on the Gaussian distribution overlapping degree of the positions of the two vehicles at each moment.

The embodiment of the application provides a collision risk prediction method and device based on coupling of vehicle behavior interaction and a road structure, and provides a vehicle intention recognition and track prediction framework which takes the vehicle behavior interaction and the road structure into consideration, and the method and the device are used for quantitatively predicting collision risk under a vehicle collision scene. The framework fuses and considers the influence of the environmental situation and the vehicle behavior on the intention of the driver, and further quantitatively evaluates the collision risk based on the space-time distribution of the predicted two vehicle tracks and the superposition degree thereof, thereby providing a basis for the subsequent decision process of the intelligent vehicle. Based on simulation of human interaction process and training of natural driving data, vehicle motion prediction in complex conflict scenes can be achieved, and through coupling with road structures, description capacity of the model and generalization capacity of different scene migration applications are further enhanced.

Additional aspects and advantages of the application will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the application.

Drawings

The foregoing and/or additional aspects and advantages of the application will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings, in which:

FIG. 1 is a flow chart of a collision risk prediction method coupled with a road structure based on vehicle behavior interactions, according to one embodiment of the application;

FIG. 2 is a logical block diagram of a collision risk prediction method coupled with a road structure based on vehicle behavior interactions, according to one embodiment of the application;

FIG. 3 is a schematic diagram of a projection process and output information of a real collision scene to two basic collision scenes according to an embodiment of the present application;

FIG. 4 is a schematic diagram of a dynamic Bayesian network for vehicle intent inference in accordance with an embodiment of the present application;

FIG. 5 is a probabilistic graphical illustration for environmental situational inference and behavioral semantic inference in accordance with one embodiment of the present application;

FIG. 6 is a schematic diagram of environmental scenarios involved in two basic conflict scenarios in accordance with one embodiment of the present application;

FIG. 7 is a table of behavior semantics related in two types of basic conflict scenarios in accordance with one embodiment of the present application;

Fig. 8 is a schematic structural diagram of a collision risk prediction apparatus coupled with a road structure based on vehicle behavior interaction according to an embodiment of the present application.

Detailed Description

Embodiments of the present application are described in detail below, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to like or similar elements or elements having like or similar functions throughout. The embodiments described below by referring to the drawings are illustrative and intended to explain the present application and should not be construed as limiting the application.

In the field of intelligent driving, much research has been conducted towards the risk of vehicle-to-vehicle collisions. Modeling for vehicle motion mainly includes two types, fixed motion model-based and behavioral intention-based. The fixed motion model assumes a motion model of a vehicle with a fixed mode such as constant speed, constant acceleration or constant steering angle, longitudinal acceleration and the like, and collision risk prediction indexes such as TTC (Time To Collision), THW (Time Headway) and the like are generated based on the model. Compared with the method, the method based on the behavior intention further combines the road structure in the traffic environment to judge the vehicle behavior intention, further predicts the future movement of the vehicle and analyzes the collision risk. The motion model prediction accuracy based on behavior intention is higher, and the recognition of collision risk by people is more met. However, such methods currently mostly only consider the behavior of the individual vehicles, but lack analysis of the behavioral interaction effects from vehicle to vehicle. From the cause analysis of traffic accidents, when potential conflicts exist between vehicles (such as ramp entry, intersection without signal lights, etc.), one common cause of occurrence of collision accidents is failure of the interaction process (such as no observation of the other party or simultaneous generation of excessive aggressive strategies). Conversely, forming a collaboration based on effective vehicle-to-vehicle interactions can avoid collisions by determining the order of passing the conflict zones. Thus, the behavioral intent-based motion model should further consider the vehicle-to-vehicle interaction process to more effectively predict collision risk. In order to better define the traffic scene studied by the application, referring to related treatises at home and abroad, the application defines an interactive conflict scene as a traffic scene which is formed by the way that two vehicles face potential space-time running conflict and need to coordinate to generate a first-later sequence passing through a conflict point (or region) so as to avoid collision. And in the scenes like rear-end collision, the front and rear vehicles have clear active-passive relation, and an interaction scene is not formed. The uncertainty of the vehicle behavior in the non-interactive scene is smaller, and the existing collision risk method can well realize functions. Therefore, the application is mainly oriented to two interactive scenes of the afflux type conflict and the cross type conflict.

In addition, the evaluation index of collision risk includes safety indexes such as safety distance, potential energy field, collision probability, and the like, in addition to the time indexes represented by TTC and THW. The collision probability index is more fit with uncertainty characteristics of the vehicle intention and uncertainty of a future position prediction algorithm, and is suitable for quantitative description of collision risk.

The following describes a collision risk prediction method and device based on coupling of vehicle behavior interaction and road structure according to an embodiment of the present application with reference to the accompanying drawings.

First, a collision risk prediction method based on coupling of vehicle behavior interaction with a road structure according to an embodiment of the present application will be described with reference to the accompanying drawings.

FIG. 1 is a flow chart of a collision risk prediction method based on coupling of vehicle behavior interactions with road structures, according to one embodiment of the application.

FIG. 2 is a logic block diagram of a collision risk prediction method based on coupling of vehicle behavior interactions with road structures, in accordance with one embodiment of the present application.

As shown in fig. 1 and 2, the collision risk prediction method based on coupling of vehicle behavior interaction with a road structure includes the steps of:

and step S1, identifying potential collision of the two vehicles based on the road structure, and projecting the collision of the two vehicles to the basic interactive collision scene model.

Optionally, in an embodiment of the present application, identifying a potential two-vehicle collision based on the road structure, projecting the two-vehicle collision to the basic interactive collision scene model includes: based on the road structure, taking the lane center line as a reference line to establish projection on a Frenet coordinate system to form an interactive conflict scene model of one of an import conflict scene model and a cross conflict scene model; identifying vehicle pairs that constitute an interactive conflict scene; and initializing an intention identification model according to the confidence level of the prior passing intention of the two conflict vehicles.

Optionally, in an embodiment of the present application, identifying the vehicle pairing that constitutes the interactive conflict scenario includes: and constructing a probability map, taking the states and the environmental information of the target vehicle and the surrounding vehicles as inputs, outputting the collision intensity between the target vehicle and the surrounding vehicles, and forming a double-vehicle collision scene with the vehicle with the highest collision intensity.

In an embodiment of the present application, potential double car conflicts are identified based on road structure and projected into two basic conflict models, namely, an import conflict and a cross conflict, as shown in FIG. 3.

As shown in FIG. 3, the present application is primarily directed to two basic scenarios of afflux conflicts and cross conflicts. Firstly, a Frenet coordinate system is established based on a vehicle reference track defined by a road structure, and an actual conflict scene is projected to two basic conflict models. The Frenet coordinate transformation process can lose geometric information of a real road, and a projection function of a reference track and the road curvature on the reference track is built for saving road curvature characteristics. And then, pairing every two of the vehicle conflict relations existing in the scene, wherein the pairing process takes the conflict strength as a criterion.

Further, in step S1, the two-vehicle collision relationship is modeled in the following manner:

step S11, based on the road structure, taking the lane central line as a reference line to establish projection on the Frenet coordinate system to form a basic conflict model of one of import and intersection. And (3) preserving the curvature along the reference track in the real road and using the curvature in the probability inference of the subsequent environment situation and behavior semantics.

Step S12, identifying the vehicle pairs constituting the interactive conflict scene. Constructing a probability map, taking the states and environmental information of the target vehicle and surrounding vehicles as input, and outputting the collision strength between the target vehicle and the surrounding vehicles and the vehicle structure with the highest collision strengthDouble car conflict scene. Wherein the probability map is at the speed v of the target vehicle-environment vehicle ₁ ,v ₂ Distance l from reference conflict point ₁ ,l ₂ For input and output, the two vehicles have a priority passing intention Pr ₀ Confidence s of (2) ₁ ,s ₂ And s _m ＝P(Pr ₀ |v ₁ ,v ₂ ,l ₁ ,l ₂ ) M=1, 2. The collision strength can be expressed as the intended product c=s of the preferential traffic of two vehicles ₁ ×s ₂ 。

Step S13, based on the identified main conflict object of the target vehicle, outputting the front-rear passing intents S of the two vehicles by using the probability map ₁ ,(1-s ₁ ),s ₂ ,(1-s ₂ ) As an initial value for the intention inference of the following two vehicles.

Step S13, using confidence level S of the priority passing intention of two conflict vehicles ₁ ,s ₂ Initializing a subsequent intent inference network oriented to the interaction process.

And S2, establishing an intention recognition model based on a dynamic Bayesian network according to the double-vehicle conflict so as to describe the conditional probability relation between the vehicle passing intention and the environment situation and the driving behavior.

Optionally, in an embodiment of the present application, establishing an intention recognition model based on a dynamic bayesian network according to a double-vehicle collision to describe a conditional probability relationship between a vehicle passing intention and an environmental situation, and a driving behavior includes: simulating a human behavior interaction process, and constructing a dynamic Bayesian network, wherein the dynamic Bayesian network comprises an environment situation, a driver intention and a vehicle behavior, and the inferred target of the dynamic Bayesian network is the driver intention; establishing a directed connection relation of each variable in the dynamic Bayesian network to form a directed acyclic graph; pre-calibrating dynamic Bayesian network parameters based on experience knowledge; based on all measurable observation information on the time sequence, hidden variable probability inference in the dynamic Bayesian network is carried out, and the priority passing intention confidence of the two conflict vehicles is output.

Specifically, aiming at the studied car-car conflict, an intention identification model based on a dynamic Bayesian network is established. The model deduces the confidence level (namely the passing intention of the two vehicles) that the two vehicles pass through the conflict area preferentially (or lagged) on the two vehicles based on the environmental situation and the behavior semantics in the running process of the vehicles. The dynamic Bayesian network mainly comprises three factors: environmental situation, driver intent, and vehicle behavior, where driver intent is an inferred target for the network.

Influence factors of the passing intention include two types: environmental situation and historical behavior of both parties. The environment situation is coupled with the road structure and the two-vehicle motion state, and is used for describing whether conditions of a certain behavior of the vehicle are provided in the scene, for example, whether the time interval between the incoming vehicle and the rear vehicle meets the incoming requirement or not; the behavior semantics of a vehicle correspond to semantical behaviors with specific meanings, such as "speed-down let-off", etc.

Further, in step S2, the confidence of the two-vehicle first-last traffic intention in the interactive process is estimated in the following manner.

Step S21, simulating a human behavior interaction process, and establishing a dynamic Bayesian network as shown in fig. 4. Implicit variables in a dynamic bayesian network can be divided into three layers: the first layer is an environmental situation and serves as an environmental basis for two vehicle behavior decisions; the second layer is the confidence of the first-last passing intention, and describes the strength of the first-last passing intention of the vehicle in the scene; the third layer is the interactive behavior adopted by the vehicle, comprising two types of behavior (MA) and Request (MR), and can be further divided into two types of transverse behavior and longitudinal behavior. Explicit variables are observable scene physical information such as road structure, vehicle relative motion state, etc. The explicit variables are used for deducing two types of information, namely environmental situation and behavior semantics, and respectively have corresponding deduced models.

Step S22, establishing a directed connection relation of each variable in the dynamic Bayesian network to form a directed acyclic graph (Directed Acyclic Graph, DAG). When variable A points to variable B, the representative A node is the parent node of the B node and has a conditional probability parameter P (B|A).

As shown in FIG. 4, in the directed acyclic graph DAG established by the present application, observation information O is contained ^t Environmental situation P ^t _l (l=1, 2,3, … stands for environmental situation) and behavioral semantics a ^t _l (l＝12,3, … stands for each semantic behavior), and the passing intention of two vehicles. In the present application, the logical relationship between hidden variables can be expressed as: the behavior at the moment is determined by the environmental situation, the traffic intention and the behavior at the last moment, and the traffic intention at the moment is determined by the environmental situation, the behavior at the last moment and the intention at the last moment. The environmental situation at this moment is determined by the environmental situation at the previous moment and the behaviors of both parties.

Step S23, initial calibration is carried out on network parameters based on experience knowledge. The calibration logic is based on causal relation, and the initialization is based on parent node state { f } _n Conditional probability P of inferred child node states (c=c _m |{f _n }). Taking the cut scene as an example, when the distance between two vehicles is large, the cut vehicle has a high possibility of performing the cut (for example, the parameter P (behavior=cut| { f _n })＝0.7)。

And step S24, deducing environmental situation and behavior semantics based on input information at 30 moments, and further accurately deducing the intention of two vehicles. The intent inference process can be expressed as:

s _m ^k ＝P(Pr ₀ |E ^t-29 ～E ^t ,A ^t-29 ～A ^t ),m＝1,2；k＝(t-29)～t.

the inference process adopts a Forward-backward algorithm (Forward-Backward Algorithm) which can integrate Forward and backward probability inferences of reaching the target moment and give accurate inference results of the target moment.

And step S3, respectively establishing a probability map model according to the environmental situation and the semantic behaviors so as to respectively utilize observable scene physical information and two-vehicle motion information to carry out probability inference on the environmental situation and the semantic behaviors.

Optionally, in an embodiment of the present application, a probabilistic graph model is respectively established according to an environmental situation and a semantic behavior, so as to respectively utilize observable scene physical information and two-vehicle motion information to perform probability inference on the environmental situation and the semantic behavior, including: establishing a first probability map model for identifying the environmental situation, wherein the first probability map model takes road structure information, two-vehicle positions and speeds as input and takes whether the vehicle has the environmental situation meeting the behavior condition as output; and establishing a second probability map model for identifying the behavior semantics of the conflict vehicles, wherein the second probability map model takes the motion states of the two vehicles, the road curvature corresponding to the position of the two vehicles and the environmental situation result at the moment as input and takes the behavior semantics identification result as output.

Optionally, in an embodiment of the present application, the probability map model employs a junction tree algorithm for probability inference.

Specifically, an environmental situation and behavior semantic recognition model is constructed. The identification model is constructed based on a probability map model, and the observable scene information and the two-vehicle motion information are respectively utilized to carry out probability inference on the two types of information, namely the environment situation and the semantic behavior. Because the two types of information have corresponding observable information, the parameter calibration of the probability map model can be performed based on the statistical analysis of the observable information.

Further, in step S3, the environmental situation and the semantic behavior are identified in the following manner:

step S31, establishing a probability map model shown in FIG. 5 (a) for identifying the environmental situation P _n . The model takes road structure information, positions and speeds of two vehicles as input and takes whether the vehicles have environmental situations meeting behavior conditions as output. The import-conflict and the cross-conflict have different environmental situations respectively, and the definitions of the situations are shown in the figure, including safe car distance, dynamic gap and the like.

Step S32, establishing a probability map model shown in (b) of FIG. 5 for identifying the semantic behavior A of the vehicle _n . The model takes the motion state of two vehicles, the road curvature corresponding to the position of the two vehicles and the environmental situation result at the moment as input and takes the behavior semantic identification result as output.

In the interaction process, the vehicle behavior semantics can be classified based on the following dimensions: vehicle dimensions to perform actions (e.g., cut-in conflict, cut-in behavior-back behavior); from the direction dimension of the action, the action can be divided into transverse behavior-longitudinal behavior (especially for the convergence conflict, two vehicles do not have fixed conflict points, and therefore, the transverse behavior and the longitudinal behavior of the two vehicles can have specific semantics); the follower dimension includes two types, behavior achievement (MotionAchievement, MA) and behavior Request (MR). Wherein behavior MA means that the host vehicle will take some driving action (switch in from the implementation) without the other party changing the intention of traffic; request MR refers to that the own vehicle requests the other party to change its passing intention by taking some action (e.g. cut-in vehicle requests the vehicle to let go after passing the card position).

In both steps S31 and S32, a junction tree algorithm (Junction Tree Algorithm) is employed to make accurate inferences. The environment situation and behavior semantic inference process can be expressed as:

P(E ^t _l ＝a)＝P(E ^t _l ＝a|O ^t )

P(A ^t _l ＝b)＝P(A ^t _l ＝b|O ^t ,E ^t _l )

the result of the above-mentioned deduction output is the confidence level of various environmental situation and behavior semanteme. The confidence is used as input information of the dynamic Bayesian network for the subsequent intention inference process.

And S4, carrying out parameter pre-calibration on the intention recognition model parameters based on experience, and carrying out parameter learning based on an EM algorithm and natural driving data.

For a dynamic Bayesian network for intent inference, the process comprises two steps, namely, firstly, parameter pre-calibration is carried out based on experience, and parameter learning is carried out based on an EM algorithm and natural driving data.

Specifically, the dynamic bayesian network parameters are trained using the EM algorithm. The EM algorithm has the advantages that under the condition of data missing (part of hidden variables lack of truth labels), continuous iteration of two processes of variable inference and parameter optimization is based to obtain optimal probability network parameters aiming at a data set, and two-vehicle intention effective identification based on semantic observation information is optimally realized,

and S5, outputting a two-vehicle passing intention recognition result based on the intention recognition model, predicting the motion track of the two vehicles by using a Gaussian process regression algorithm, and outputting a collision risk assessment result based on the Gaussian distribution overlapping degree of the positions of the two vehicles at each moment.

Optionally, in an embodiment of the present application, a two-vehicle passing intention recognition result is output based on an intention recognition model, a two-vehicle motion track is predicted by using a gaussian process regression algorithm, and a collision risk assessment result is output based on a gaussian distribution overlapping degree of each time position, including: classifying the vehicle track in the natural driving data based on the intention recognition result to construct a training set; training a Gaussian process regression algorithm model based on training set classification; predicting a future track of the vehicle according to the intention recognition result and the Gaussian process regression algorithm model; and identifying the future collision risk of the two vehicles according to the predicted future track of the vehicles and the overlapping degree of the Gaussian distribution of the vehicle positions.

Specifically, based on the recognized behavior intention, a gaussian regression algorithm (Gaussian Process Regression, GPR) algorithm is used to predict the vehicle running track and its space-time distribution, and finally output the risk of two-vehicle collision.

Specifically, the two-vehicle trajectory and collision risk are predicted in the following manner.

In step S51, in order to implement the trajectory prediction on the condition of intent, the training set needs to be divided first. The training set division basis comprises: final traffic sequence, and interaction intensity classification based on the two-vehicle traffic sequence (weak-medium-strong). The passing sequence can be directly marked based on the final passing sequence of two vehicles in the track data, the interaction strength is classified based on two indexes of situation and intention, and the optimal classification is realized by adopting a K-Means algorithm. In the classified data, the weaker the interaction strength is, the higher the security situation is in the interaction process, and the change of intention of two vehicles is smaller; otherwise, the interaction strength is higher, the security situation in the interaction process is lower, and more intention changes can exist in two vehicles.

Step S52, GPR model training based on training set classification. The GPR model can be expressed as:

f～GP(u,K)

u＝{m(t _i ),t _i ＝1:T}

K＝{k(t _i ,t _j ),i,j＝1:N}

where u is a mean vector (representing the predicted position at each time instant), and K is a covariance matrix describing the distribution of the predicted positions. In the present application, u and K use a polynomial model and a square exponential kernel model, respectively. Based on the training set data of each class, a corresponding GPR model is trained using a maximum likelihood estimation algorithm (Maximum Likelihood Estimate, MLE).

Step S53, inputting a future track in the prediction time domain based on the history track. Firstly, based on input historical data, the probability map model is applied to identify two vehicle behavior intents, and interaction strength attributes are identified based on historical tracks. And selecting a corresponding GPR model to predict the track based on the judging result.

Considering the initial prediction period, the two vehicles have short interaction time and less input information, and the prediction accuracy based on the GPR is lower, so that the method of fusing the GPR prediction result and the constant acceleration model (Constant Acceleration, CA) is adopted in the first 1s of prediction, namely the prediction position takes the average value of the two models.

And S54, predicting the collision risk based on the coincidence ratio of the future tracks of the two vehicles.

First, the potential collision locations of two vehicles are identified. Potential collision points include: front, back, left, right, front left, back left, front right, back right, 8 points. Based on the predicted positions of the two vehicles, taking the closest point as a potential collision point, calculating the Gaussian distribution Overlap ratio (OLR) of the potential collision point positions, and recording the Overlap ratio at the predicted time t as the OLR _t 。

Thereafter, the collision risk prediction result is calculated from the prediction time t (expressed as the predicted number of steps) and the corresponding OLR _t The coupling is given. The smaller the predicted time t is, the position overlap ratio OLR _t The higher the corresponding collision risk is. Thereby, collision risk R _colli Expressed as:

wherein T is the predicted total step length, c _t An attenuation coefficient of less than or equal to 1. c _t The smaller R _colli The more attention is paid to the recent collision risk, and conversely, the pair is increasedConcerns over long-term risk.

By the method, collision risk prediction considering multi-vehicle behavior interactive coupling under complex traffic scenes such as ramp entry and intersection traffic can be realized, the method can be widely applied to various conflict scenes, and the driving safety of intelligent vehicles can be improved.

In order to further analyze the probability of collision of two vehicles in a potential collision and output collision risk, the collision risk prediction method based on coupling of vehicle behavior interaction with a road structure according to the present application is described below with reference to a specific embodiment.

1): potential two-vehicle collisions are identified based on road structure and projected into two basic collision models, the entry collision and the cross collision, as shown in fig. 3.

Specifically, in 1), the two-vehicle collision relationship is modeled in the following manner:

1-1), based on the road structure, taking the lane central line as a reference line to establish the projection of the traffic scene under the Frenet coordinate system, and forming one of two basic conflicts as shown in fig. 3. The projection of the coordinates of the points in the cartesian coordinate system to the Frenet coordinate system comprises two steps: (1) Calculating the shortest distance projection point of the point to be solved on the reference line; (2) Starting from the origin on the reference line, the length of the reference line from the origin to the projection point (longitudinal distance D ₁ ) And the distance from the projected point to the desired point (lateral distance D ₂ ). At this time, the calculated point has a coordinate (D ₁ ，D ₂ )。

After the projection is completed, the curvature information of the original road is saved and recorded as a function C (D ₁ )。

1-2) identifying vehicle pairs that make up an interactive conflict scenario. In the pairing process, a research object can be selected first, and then the vehicle with the biggest collision strength C with the object is searched as a collision object.

In the present application, the collision strength C is defined as the intended product of the collision areas where two vehicles pass preferentially (i.e., the collision strength is greater when both vehicles tend to consider that the own vehicle will pass preferentially over the collision areas compared to each other). Build to targetSpeed v of vehicle-environment vehicle ₁ ,v ₂ Distance l from reference conflict point ₁ ,l ₂ The probability map is input, and two vehicles with priority passing intention Pr are respectively output ₀ Confidence s of (2) ₁ ,s ₂ (i.e. s _m ＝P(Pr ₀ |v ₁ ,v ₂ ,l ₁ ,l ₂ ) M=1, 2). At this time, the collision strength is expressed as c=s ₁ ×s ₂ 。

For calibrating the probability map parameters, a supervised learning method can be adopted, namely, a vehicle which finally and actually forms a conflict is taken as a true value, and a maximum likelihood estimation algorithm (MLE) is used for parameter calibration.

1-3) with confidence s of the priority traffic intention of two conflicting vehicles ₁ ,s ₂ Initializing a subsequent intent inference network oriented to the interaction process.

2): for the studied double-car conflict, a dynamic Bayesian network is established for identifying the confidence that the two-car drivers pass through the conflict area preferentially to themselves respectively. The dynamic Bayesian network mainly comprises three elements: environmental situation, driver intent, and vehicle behavior, where driver intent is an inferred target for the network.

In 2), the confidence of the two-vehicle first-last traffic intention in the interaction process is estimated in the following way:

2-1) simulating human behavior interaction process, and constructing the dynamic Bayesian network shown in fig. 4. In terms of node information definition, the network mainly comprises three layers of nodes: (1) And the environment situation node represents whether the conflict vehicle has the condition of executing the preferential traffic. For example, in a cut scene, a cut-in vehicle can perform a merge only if the cut-in vehicle has a safe vehicle distance or the like. As shown in fig. 6, in the afflux conflict, situation information includes four types: p1, cutting into a longitudinal running space of the vehicle; p2, cutting into a longitudinal running space of the vehicle and the front environment vehicle; p3, the longitudinal running space between the collision vehicles; and P4, dynamically clearance between the straight collision vehicle and the front environment vehicle. In cross-type conflicts, the situation information contains three categories: p1, the time difference from two conflict vehicles to a conflict point; p2, a longitudinal running space of the left side collision vehicle; p3: the lower part collides with the longitudinal running space of the vehicle. The situation information is a discrete hidden variable and comprises two values of "conditional" and "unconditional". (2) The intention nodes represent the confidence that the conflicting vehicles have respectively prioritized passage through the conflicting zones. The intention information is a discrete hidden variable, and comprises two values of 'priority traffic' and 'let traffic'. (3) Behavior nodes describe meanings that conflicting vehicle travel trajectories have. Based on classification dimensions such as conflict vehicles, horizontal-vertical, request-execution, robbery-let-pass, and the like, behavior semantics are shown in the table. Wherein each behavior semantic constitutes a discrete hidden variable and comprises two values of 'having the behavior' and 'not having the behavior'.

In the three-layer node, the environment situation and the behavior semantics can be deduced based on observable physical information such as the vehicle position, the speed and the like, and the observation and inference process will be introduced in step S3. The intention does not directly correspond to observable information, and the inference is performed based on the environment hidden variables and the behavior hidden variables, and the inference result is finally output by the dynamic Bayesian network in the step S2.

2-2) forming a directed acyclic graph based on three layers of nodes. As shown in FIG. 4, the environmental situation at time t is denoted as E ^t _l (l=1, 2,3, … stands for environmental situation), behavior semantics are denoted as a ^t _l (l=1, 2,3, … stands for semantic behavior), all measurable observations are denoted O ^t . When variable A points to variable B, the representative A node is the parent node of the B node and has a conditional probability parameter P (B|A).

2-3) pre-calibrating the dynamic Bayesian network parameters. The calibration process is performed based on experience, and the calibration parameter meaning is based on father node state { f _n Inferred child node state { c=c } _m Conditional probability P (c=c) _m |{f _n }). Taking the cut scene as an example, when the distance between two vehicles is large, it can be determined based on experience that the situation security of the cut vehicle to perform the cut is high (P (situation=security| { f _n -0.8); when the confidence of the first pass of the cut-in vehicle is high, the confidence of the later pass of the cut-in vehicle is high, and the situation security is high, the probability of the host vehicle taking the cut-in behavior is high (P (behavior=cut-in| { f _n })＝0.7))。

2-4) based on all measurable observation information on the time sequence, realizing hidden variable probability inference in the dynamic Bayesian network, and finally outputting the confidence level of the prior passing intention of the two conflict vehicles.

Based on a dynamic Bayesian network, two hidden variables of environmental situation and behavior semantics can be expressed as follows:

P(E ^t _l ＝a)＝P(E ^t _l ＝a|O ^t )

P(A ^t _l ＝b)＝P(A ^t _l ＝b|O ^t ,s ₁ ^t-1 ,s ₂ ^t-1 )

according to the application, under the assumption that observation information of 30 periods is input, the confidence level of the priority passing intention of two vehicles in 30 periods is finally output:

s _m ^k ＝P(Pr ₀ |E ^t-29 ～E ^t ,A ^t-29 ～A ^t ),m＝1,2；k＝(t-29)～t.

in the implementation and test process of the application, an inference algorithm based on a forward-backward algorithm is adopted. Other common dynamic bayesian network inference algorithms can also be used to solve the inference problem of the present application.

3): and respectively establishing observers based on the probability map model aiming at the environmental situation and the vehicle behavior in the step 2). The observer outputs discrete situational assessment and behavioral semantics based on measurable scene physical information (e.g., relative position of two vehicles, vehicle speed, etc.).

In 3), the semantics of the environmental situation and behavior are recognized in the following way:

3-1) establishing a probability map model shown in (a) of fig. 5 for identifying various environmental situations. The input information includes static physical parameters (such as the end position of the ramp junction region, the curvature of the road, etc.) and dynamic physical information (including the positions, speeds, accelerations, etc. of each collision vehicle and each environmental vehicle) of the scene after being projected to the Frenet coordinate system, and the output inferred result is whether the studied environmental situation has the condition that the vehicles execute the preferential pass (such as the situation that the vehicles cut into the scene, whether the collision vehicles in the straight run and the environmental vehicles in front have enough gaps).

In order to calibrate the model parameters of the probability map, the model parameters of the probability map can be optimally designed based on the research on the vehicle-to-vehicle collision process in the real natural driving data, such as statistics of the distribution of information such as relative distance, relative vehicle speed and the like in the process of merging.

3-2) establishing a probability map model as shown in (b) of fig. 5, and identifying the collision vehicle behavior semantics A _n . The model takes the relative motion information of two vehicles and the inferred environmental situation as input to identify corresponding behavior semantics. The behavioral semantic classification is shown in fig. 7.

The classification mode of behavior semantics comprises conflict vehicle, horizontal-vertical, robbing-letting and behavior request-behavior realization. The difference between the behavior Request (MR) and the behavior realization (Motion Achievement, MA) of the application is shown in the following steps: in MR behavior, the vehicle does not have a condition for executing passing intention, which is to request the other party to change intention, and create a condition for own behavior. In MA behavior, the condition is already present, and the vehicle execution behavior does not involve a change in the intention of the other party. For example, in the course of an import-in, assuming that the import-in vehicle is cutting in laterally without a secure import-in, its behavioral semantics should be interpreted as heuristics, requesting that the straight-going vehicle behind it be allowed. Whereas a lateral cut-in behavior with secure entry conditions should be interpreted as its execution of its "priority traffic" intent.

Based on the behavior semantic classification method, a probability map model is established and model parameters are predefined. Further, model parameters may be optimized based on research and statistics of the vehicle-to-vehicle collision process in real natural driving data.

4): the dynamic Bayesian network parameters are calibrated, the process comprises two steps, firstly, parameter pre-calibration is carried out based on experience, and parameter learning is carried out based on an EM algorithm and natural driving data.

Model parameters in a dynamic Bayesian network are trained based on an EM algorithm, and the model parameters are different from the two types of probability map observation models in 3), and the dynamic Bayesian model facing intention inference is difficult to obtain direct observable data statistics, so that direct calibration is difficult. The EM algorithm has the advantage that in the event of data loss (hidden variables have no truth labels), based on continuous iterations of both variable inference and parameter optimization, optimal probability network parameters for the data set are obtained.

5): based on the intent recognition result, a gaussian process regression algorithm (Gaussian Process Regression, GPR) is used to predict the future trajectory of the two vehicles and evaluate their collision risk.

5-1), in order to train the required track prediction model, the vehicle track in the natural driving data needs to be classified first. Based on the output of the intention recognition model, the intention combination of the two conflict vehicles comprises four categories of antecedent-antecedent, let-let. Wherein the probability of each class corresponds to the confidence coefficient product of the corresponding intention of two vehicles, namely P ₁ ＝(s ₁ ×s ₂ )，P ₂ ＝((1-s ₁ )×s ₂ )，P ₃ ＝(s ₁ ×(1-s ₂ ))，P ₄ ＝((1-s ₁ )×(1-s ₂ ) Taking the combination with the highest probability as the classification result at the moment, namely P _m ＝max(P ₁ ,P ₂ ,P ₃ ,P ₄ )。

5-2) training the GPR model based on training set classification. The GPR model can be expressed as:

f～GP(u,K)

u＝{m(t _i ),t _i ＝1:T}

K＝{κ(t _i ,t _j ),i,j＝1:N}

where u is a mean vector (representing the predicted position at each time instant), and K is a covariance matrix describing the distribution of the predicted positions. In the test and verification of the present application, u and K use a five-degree polynomial model and a square exponential kernel model, respectively.

In the training process, the input track is based on P _m Segmentation is performed and used to train the GPR model corresponding to the intent combination, respectively.

5-3) predicting the future track of the vehicle based on the intention recognition result and the GPR model. In the prediction process, a vehicle history track and an intention recognition result are taken as inputs, and track prediction is carried out based on each intention combination and a corresponding GPR model. The output of the prediction model is the position distribution of two vehicles at each time in the future under each intention combination and the probability of each intention combination.

In addition, in consideration of the initial prediction period, the two vehicles are short in interaction time, less in input information and low in prediction accuracy based on GPR, so that a method of fusing a GPR prediction result and a constant acceleration model (Constant Acceleration, CA) is adopted in the first 1s of prediction, namely, the prediction position is the average value of the two models.

5-4) identifying the future collision risk of the two vehicles based on track prediction. Potential collision points include: front, back, left, right, front left, back left, front right, back right, 8 points. Based on the predicted positions of the two vehicles, taking the closest point as a potential collision point, calculating the Gaussian distribution Overlap ratio (OLR) of the potential collision point positions, and recording the Overlap ratio at the predicted time t as the OLR _t 。

Since the predicted output contains trajectory predictions based on four intent combinations, the overlap ratio calculation of the final output is expressed as:

thereafter, the collision risk prediction result is calculated from the prediction time t (expressed as the predicted number of steps) and the corresponding OLR _t The coupling is given. The smaller the predicted time t is, the position overlap ratio OLR _t The higher the corresponding collision risk is. Thereby, collision risk R _colli Expressed as:

wherein T is the predicted total step length, c _t An attenuation coefficient of less than or equal to 1. c _t The smaller R _colli The more the recent collision risk is of concern, and conversely the longer term risk is of increased concern.

According to the collision risk prediction method based on coupling of vehicle behavior interaction and a road structure, a vehicle intention recognition and track prediction framework taking the vehicle behavior interaction and the road structure into consideration is provided for quantitatively predicting collision risk under a vehicle collision scene. The framework fuses and considers the influence of the environmental situation and the vehicle behavior on the intention of the driver, and further quantitatively evaluates the collision risk based on the space-time distribution of the two vehicle tracks and the coincidence degree of the two vehicle tracks, thereby providing a basis for the subsequent decision making process of the intelligent vehicle. The method is based on simulation of human interaction process and training of natural driving data, can realize vehicle motion prediction under complex conflict scenes, and further enhances description capability of a model and generalization capability of different scene migration applications through coupling with a road structure.

Next, a collision risk prediction apparatus based on coupling of vehicle behavior interaction with a road structure according to an embodiment of the present application will be described with reference to the accompanying drawings.

Fig. 8 is a schematic structural diagram of a collision risk prediction apparatus coupled with a road structure based on vehicle behavior interaction according to an embodiment of the present application.

As shown in fig. 8, the collision risk prediction apparatus coupled with a road structure based on vehicle behavior interaction includes: projection module 100, modeling module 200, inference module 300, training module 400, and prediction module 500.

The projection module 100 is configured to identify a potential collision of two vehicles based on the road structure, and project the collision of two vehicles to the basic interactive collision scene model.

The modeling module 200 is configured to establish an intention recognition model based on a dynamic bayesian network according to the collision of two vehicles, so as to describe a conditional probability relationship between the passing intention of the vehicle, an environmental situation and driving behavior.

The inference module 300 is configured to respectively establish a probabilistic graph model according to the environmental situation and the semantic behavior, so as to respectively utilize observable scene physical information and two-vehicle motion information to perform probability inference on the environmental situation and the semantic behavior, and perform probability inference on the passing intention of the collision vehicle based on the dynamic bayesian network by taking the environmental situation and the semantic behavior as observation inputs.

The training module 400 is configured to perform parameter pre-calibration on the intent recognition model parameters based on experience, perform parameter learning based on the EM algorithm and the natural driving data, classify the natural driving data set based on the intent and other indexes, and respectively train the corresponding gaussian process regression model for subsequent trajectory prediction.

The prediction module 500 is configured to output a recognition result of the passing intention of the two vehicles based on the intention recognition model, predict the movement track of the two vehicles by using a gaussian process regression algorithm, and output a collision risk assessment result based on the overlapping degree of gaussian distribution of the positions of the two vehicles at each moment.

Optionally, in an embodiment of the present application, the projection module is further configured to establish a projection on a Frenet coordinate system with a lane center line as a reference line based on the road structure, to form an interactive conflict scene model of one of the afflux conflict scene model and the cross conflict scene model; identifying vehicle pairs that constitute an interactive conflict scene; and initializing an intention identification model according to the confidence level of the prior passing intention of the two conflict vehicles.

Optionally, in one embodiment of the present application, identifying the vehicle pairing that constitutes the interactive conflict scenario includes: and constructing a probability map, taking the states and the environmental information of the target vehicle and the surrounding vehicles as inputs, outputting the collision intensity between the target vehicle and the surrounding vehicles, and forming a double-vehicle collision scene with the vehicle with the highest collision intensity.

It should be noted that the foregoing explanation of the method embodiment is also applicable to the apparatus of this embodiment, and will not be repeated here.

According to the collision risk prediction device based on coupling of vehicle behavior interaction and a road structure, a vehicle intention recognition and track prediction framework taking the vehicle behavior interaction and the road structure into consideration is provided for quantitatively predicting collision risk under a collision scene of a vehicle. The framework fuses and considers the influence of the environmental situation and the vehicle behavior on the intention of the driver, and further quantitatively evaluates the collision risk based on the space-time distribution of the two vehicle tracks and the coincidence degree of the two vehicle tracks, thereby providing a basis for the subsequent decision making process of the intelligent vehicle. Based on simulation of human interaction process and training of natural driving data, vehicle motion prediction in complex conflict scenes can be achieved, and through coupling with road structures, description capacity of the model and generalization capacity of different scene migration applications are further enhanced.

Furthermore, the terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include at least one such feature. In the description of the present application, the meaning of "plurality" means at least two, for example, two, three, etc., unless specifically defined otherwise.

In the description of the present specification, a description referring to terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present application. In this specification, schematic representations of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, the different embodiments or examples described in this specification and the features of the different embodiments or examples may be combined and combined by those skilled in the art without contradiction.

While embodiments of the present application have been shown and described above, it will be understood that the above embodiments are illustrative and not to be construed as limiting the application, and that variations, modifications, alternatives and variations may be made to the above embodiments by one of ordinary skill in the art within the scope of the application.

Claims (6)

1. A collision risk prediction method based on coupling of vehicle behavior interaction and road structure, comprising the steps of:

Identifying potential two-vehicle collisions based on the road structure, projecting the two-vehicle collisions to a basic interactive collision scene model, wherein the identifying potential two-vehicle collisions based on the road structure, projecting the two-vehicle collisions to the basic interactive collision scene model, comprises: based on a road structure, taking a lane center line as a reference line to establish projection on a Frenet coordinate system, forming an interactive conflict scene model of one of an import conflict scene model and a cross conflict scene model, identifying vehicle pairing forming the interactive conflict scene, and initializing an intention identification model by using the confidence of the priority passing intention of the two conflict vehicles;

establishing an intention recognition model based on a dynamic Bayesian network according to the double-vehicle conflict so as to describe the conditional probability relation between the passing intention of the vehicle and the environmental situation and the driving behavior, wherein the establishing the intention recognition model based on the dynamic Bayesian network according to the double-vehicle conflict so as to describe the conditional probability relation between the passing intention of the vehicle and the environmental situation and the driving behavior comprises the following steps: simulating a human behavior interaction process, and constructing a dynamic Bayesian network, wherein parameters of the dynamic Bayesian network comprise an environment situation, driver intention and vehicle behavior, an inferred target of the dynamic Bayesian network is the driver intention, and a directed connection relation of each variable in the dynamic Bayesian network is established to form a directed acyclic graph; pre-calibrating the dynamic Bayesian network parameters based on experience knowledge, training and optimizing the network parameters based on real driving data, deducing hidden variable probability in the dynamic Bayesian network based on all measurable observation information on a time sequence, and outputting the confidence level of the prior passing intention of two conflict vehicles;

Respectively establishing a probability map model according to an environment situation and driving behaviors, respectively utilizing observable scene physical information and two-vehicle movement information, taking the environment situation and the driving behaviors as observation inputs, and carrying out probability inference on the passing intention of the conflict vehicle based on a dynamic Bayesian network, wherein the respectively establishing the probability map model according to the environment situation and the driving behaviors, respectively utilizing observable scene physical information and two-vehicle movement information, taking the environment situation and the driving behaviors as observation inputs, and carrying out probability inference on the passing intention of the conflict vehicle based on the dynamic Bayesian network comprises the following steps: establishing a first probability map model for identifying environmental situations, wherein the first probability map model takes road structure information, positions and speeds of two vehicles as input and takes whether the vehicles have the environmental situations meeting behavior conditions as output; establishing a second probability map model for identifying the behavior semantics of the conflict vehicles, wherein the second probability map model takes the motion state of the two vehicles, the road curvature corresponding to the position of the two vehicles and the environmental situation result at the moment as input and takes the behavior semantics identification result as output;

carrying out parameter pre-calibration on the intention recognition model parameters based on experience, and carrying out parameter learning based on an EM algorithm and natural driving data;

And outputting a two-vehicle passing intention recognition result based on the intention recognition model, predicting the motion track of the two vehicles by using a Gaussian process regression algorithm, and outputting a collision risk assessment result based on the Gaussian distribution overlapping degree of the positions of the two vehicles at each moment.

2. The method of claim 1, wherein the identifying the vehicle pairs that make up the interactive conflict scenario comprises:

and constructing a probability map, taking the states and the environmental information of the target vehicle and the surrounding vehicles as input, outputting the collision intensity of the target vehicle and the surrounding vehicles, and forming a double-vehicle collision scene with the vehicle with the highest collision intensity.

3. The method according to claim 1, wherein the outputting the two-vehicle passing intention recognition result based on the intention recognition model, predicting the two-vehicle motion track by using a gaussian process regression algorithm, and outputting the collision risk assessment result based on the overlapping degree of gaussian distribution of the positions of the two vehicles at each moment, includes:

classifying the vehicle track in the natural driving data based on the intention recognition result to construct a training set;

training a Gaussian process regression algorithm model based on the classified training set;

predicting a future track of the vehicle according to the intention recognition result and the Gaussian process regression algorithm model;

And identifying the future collision risk of the two vehicles according to the predicted future track of the vehicle and the overlapping degree of the Gaussian distribution of the vehicle.

4. The method of claim 1, wherein the probability map model employs a junction tree algorithm for probability inference; the dynamic bayesian network uses forward-backward algorithms for the inference.

5. A collision risk prediction apparatus based on coupling of vehicle behavior interactions with a road structure, comprising:

the projection module is configured to identify a potential collision of two vehicles based on a road structure, and project the collision of two vehicles to a basic interactive collision scene model, where the identifying the potential collision of two vehicles based on the road structure, and projecting the collision of two vehicles to the basic interactive collision scene model, includes: based on a road structure, taking a lane center line as a reference line to establish projection on a Frenet coordinate system, forming an interactive conflict scene model of one of an import conflict scene model and a cross conflict scene model, identifying vehicle pairing forming the interactive conflict scene, and initializing an intention identification model by using the confidence of the priority passing intention of the two conflict vehicles;

the modeling module is configured to establish an intention recognition model based on a dynamic bayesian network according to the double-vehicle conflict to describe a conditional probability relation between a vehicle passing intention and an environmental situation and a driving behavior, wherein the establishing the intention recognition model based on the dynamic bayesian network according to the double-vehicle conflict to describe the conditional probability relation between the vehicle passing intention and the environmental situation and the driving behavior includes: simulating a human behavior interaction process, and constructing a dynamic Bayesian network, wherein parameters of the dynamic Bayesian network comprise an environment situation, driver intention and vehicle behavior, an inferred target of the dynamic Bayesian network is the driver intention, and a directed connection relation of each variable in the dynamic Bayesian network is established to form a directed acyclic graph; pre-calibrating the dynamic Bayesian network parameters based on experience knowledge, training and optimizing the network parameters based on real driving data, deducing hidden variable probability in the dynamic Bayesian network based on all measurable observation information on a time sequence, and outputting the confidence level of the prior passing intention of two conflict vehicles;

The inference module is configured to respectively establish a probability map model according to an environmental situation and a driving behavior, respectively use observable scene physical information and two-vehicle motion information, use the environmental situation and the driving behavior as observation inputs, and perform probability inference on the passing intention of the collision vehicle based on a dynamic bayesian network, where the respectively establish a probability map model according to the environmental situation and the driving behavior, respectively use observable scene physical information and two-vehicle motion information, use the environmental situation and the driving behavior as observation inputs, and perform probability inference on the passing intention of the collision vehicle based on the dynamic bayesian network, and includes: establishing a first probability map model for identifying environmental situations, wherein the first probability map model takes road structure information, positions and speeds of two vehicles as input and takes whether the vehicles have the environmental situations meeting behavior conditions as output; establishing a second probability map model for identifying the behavior semantics of the conflict vehicles, wherein the second probability map model takes the motion state of the two vehicles, the road curvature corresponding to the position of the two vehicles and the environmental situation result at the moment as input and takes the behavior semantics identification result as output;

The training module is used for carrying out parameter pre-calibration on the intention identification model parameters based on experience, carrying out parameter learning based on an EM algorithm and natural driving data, classifying the natural driving data set based on intention indexes, and respectively training a corresponding Gaussian process regression model for subsequent track prediction;

the prediction module is used for outputting a two-vehicle passing intention recognition result based on the intention recognition model, predicting the motion track of the two vehicles by utilizing a Gaussian process regression algorithm, and outputting a collision risk assessment result based on the Gaussian distribution overlapping degree of the positions of the two vehicles at each moment.

6. The apparatus of claim 5, wherein the identifying the vehicle pairs that constitute the interactive conflict scenario comprises:

and constructing a probability map, taking the states and the environmental information of the target vehicle and the surrounding vehicles as input, outputting the collision intensity of the target vehicle and the surrounding vehicles, and forming a double-vehicle collision scene with the vehicle with the highest collision intensity.