CN116363867A

CN116363867A - Behavior prediction method, device, equipment and medium for intersection object

Info

Publication number: CN116363867A
Application number: CN202211661049.2A
Authority: CN
Inventors: 刘浩泉; 杨天
Original assignee: Uisee Technologies Beijing Co Ltd
Current assignee: Uisee Technologies Beijing Co Ltd
Priority date: 2022-12-23
Filing date: 2022-12-23
Publication date: 2023-06-30

Abstract

The embodiment of the disclosure discloses a behavior prediction method, a device, equipment and a medium of an intersection object, wherein the method comprises the following steps: according to the method, the target area is determined from the intersection departure areas according to the prediction arrival probabilities, the target track is determined according to the target area, the prediction of the target area and the track of the non-motor vehicle object is realized, the behavior prediction of the non-motor vehicle object is realized, the problems of lack of behavior prediction of non-motor vehicle barriers and incapability of predicting the target area are solved, and the problem of poor stability of the neural network prediction result adopted in the prior art is solved.

Description

Behavior prediction method, device, equipment and medium for intersection object

Technical Field

The disclosure relates to the technical field of intelligent driving, in particular to a behavior prediction method, device, equipment and medium for an intersection object.

Background

In recent years, intelligent driving technology of vehicles has become a hot spot in research in the automobile industry. In order for a vehicle to be able to run safely and smoothly, an important and often neglected problem is how to predict the behavior of obstacles around the vehicle for a period of time in the future. Obstacle prediction is a precondition for planning and controlling the next operation of the vehicle.

Currently, a behavior prediction algorithm for an obstacle of a motor vehicle (a trolley, a bus, a truck and the like) directly generates a predicted track (namely, from track to track) in a future period of time through the algorithm according to historical state information of the obstacle; the method mainly uses various deep learning algorithms, the algorithms need to process, label and train a large amount of data, the interpretability is weak, the stability of a predicted result cannot be ensured, and the predicted track is lack of intended semantic information, namely the target area to which an obstacle is going is lacking, so that the method is not beneficial to planning and controlling a vehicle subsequently.

In the process of realizing the invention, the prior art is found to have at least the following technical problems: the stability of the prediction result is poor and the intended semantic information is lacking, so that the follow-up planning control of the vehicle is not facilitated. Moreover, the lack of behavior prediction for non-motor vehicle obstacles, which is an important traffic participant, and the accurate prediction of non-motor vehicle intent and trajectory is critical for safe driving of intelligent driving vehicles.

Disclosure of Invention

In order to solve the above technical problems or at least partially solve the above technical problems, embodiments of the present disclosure provide a method, an apparatus, a device, and a medium for predicting behavior of an intersection object, so as to predict a target area and a track of a non-motor vehicle object, further to predict behavior of the non-motor vehicle object, solve the problem that the prior art lacks behavior prediction of a non-motor vehicle obstacle and cannot predict the target area, and perform behavior prediction in a prediction probability manner, so as to solve the problem that the stability of a neural network prediction result adopted in the prior art is poor.

In a first aspect, an embodiment of the present disclosure provides a behavior prediction method for an intersection object, where the method includes:

acquiring each intersection leaving area corresponding to the intersection to be passed of the current vehicle;

for each object to be predicted in the intersection to be passed, acquiring position relation information between the object to be predicted and each intersection departure area, wherein the object to be predicted is a non-motor vehicle object positioned in the intersection to be passed;

inputting the position relation information into a pre-constructed regional probability prediction model to obtain the predicted forward probability corresponding to the intersection departure region;

And determining a target area corresponding to the object to be predicted in each intersection departure area based on each predicted arrival probability, and determining a target track corresponding to the object to be predicted based on the target area.

In a second aspect, an embodiment of the present disclosure further provides a behavior prediction apparatus for an intersection object, including:

an intersection object behavior prediction apparatus comprising:

the area determining module is used for obtaining each intersection leaving area corresponding to the intersection to be passed of the current vehicle;

the information acquisition module is used for acquiring the position relation information between the to-be-predicted object and each intersection departure area aiming at each to-be-predicted object in the to-be-passed intersections, wherein the to-be-predicted object is a non-motor vehicle object positioned in the to-be-passed intersections;

the probability prediction module is used for inputting the position relation information into a pre-constructed regional probability prediction model to obtain the predicted forward probability corresponding to the intersection departure region;

and the behavior determining module is used for determining a target area corresponding to the object to be predicted in each intersection departure area based on each predicted arrival probability, and determining a target track corresponding to the object to be predicted based on the target area.

In a third aspect, embodiments of the present disclosure further provide an electronic device, including: one or more processors; a storage means for storing one or more programs; the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the behavior prediction method of the intersection object as described above.

In a fourth aspect, the embodiments of the present disclosure also provide a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the behavior prediction method of an intersection object as described above.

According to the behavior prediction method for the intersection object, through obtaining each intersection exit area corresponding to the intersection to be passed, according to the position relation information between each non-motor vehicle to be predicted and each intersection exit area in the intersection to be passed and the area probability prediction model constructed in advance, the predicted forward probability corresponding to each intersection exit area is obtained, then the target area is determined from each intersection exit area according to each predicted forward probability, and the target track is determined according to the target area, prediction of the target area and the track of the non-motor vehicle object is achieved, further, behavior prediction of the non-motor vehicle object is achieved, the problem that the prior art lacks of behavior prediction of non-motor vehicle obstacles and cannot predict the target area is solved, and the problem that the stability of a neural network prediction result is poor is solved through the behavior prediction in a prediction probability mode.

Drawings

The above and other features, advantages, and aspects of embodiments of the present disclosure will become more apparent by reference to the following detailed description when taken in conjunction with the accompanying drawings. The same or similar reference numbers will be used throughout the drawings to refer to the same or like elements. It should be understood that the figures are schematic and that elements and components are not necessarily drawn to scale.

FIG. 1 is a flow chart of a method of behavior prediction of an intersection object in an embodiment of the present disclosure;

fig. 2 is a schematic diagram of each intersection exit area corresponding to an intersection to be passed in an embodiment of the disclosure;

FIG. 3 is a simplified node connection schematic diagram in an embodiment of the present disclosure;

FIG. 4 is a schematic diagram of node connections incorporating time factors in an embodiment of the present disclosure;

FIG. 5 is a schematic diagram of a dynamic Bayesian model in an embodiment of the present disclosure

FIG. 6 is a schematic diagram of an autopilot system in one embodiment of the present disclosure;

fig. 7 is a schematic structural diagram of an intersection object behavior prediction apparatus in an embodiment of the present disclosure;

fig. 8 is a schematic structural diagram of an electronic device in an embodiment of the disclosure.

Detailed Description

Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While certain embodiments of the present disclosure have been shown in the accompanying drawings, it is to be understood that the present disclosure may be embodied in various forms and should not be construed as limited to the embodiments set forth herein, but are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are for illustration purposes only and are not intended to limit the scope of the present disclosure.

It should be noted that the terms "first," "second," and the like in this disclosure are merely used to distinguish between different devices, modules, or units and are not used to define an order or interdependence of functions performed by the devices, modules, or units.

The names of messages or information interacted between the various devices in the embodiments of the present disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

Fig. 1 is a flowchart of a behavior prediction method of an intersection object in an embodiment of the present disclosure. The method can be performed by behavior prediction means of the intersection object, which means can be implemented in software and/or hardware, which means can be arranged in an electronic device. As shown in fig. 1, the method specifically may include the following steps:

s110, acquiring each intersection leaving area corresponding to the intersection to be passed of the current vehicle.

Wherein the current vehicle may be an autonomous vehicle that makes a decision based on obstacle behavior. The intersection to be passed may be an intersection at which the current vehicle is about to arrive. For example, an intersection that appears in the vehicle visual field of the current vehicle may be taken as an intersection to be passed, or an intersection that appears in the vehicle visual field of the vehicle and that is located in the travel route of the current vehicle may be taken as an intersection to be passed.

For the intersections to be passed, the exit areas of the intersections corresponding to the intersections to be passed can be further determined. The intersection departure area may be an area that leaves the intersection to be passed. The intersection exit zone may be defined by a pose (containing position and direction information) and a circle centered on the pose point, the size of the circle representing the size of the intersection exit zone.

Specifically, the intersection departure area can be determined according to the lane topology connection relationship and the lane attribute of the intersection to be passed. For example, the intersection departure area may be defined with the beginning end of the lane leaving the intersection to be passed as the center, and the two ends of the sidewalk surrounding the intersection to be passed as the center.

For example, fig. 2 is a schematic diagram of each intersection exit area corresponding to an intersection to be passed in the embodiment of the present disclosure, referring to fig. 2, an area where two ends of a center line of a pavement are located may be taken as an intersection exit area, or an area where one end of the center line of a motor vehicle lane corresponding to a driving direction is located may be taken as an intersection exit area. The definition of the intersection exit area may be the same for different objects to be predicted in the same intersection to be passed.

In this embodiment, the intersection departure areas corresponding to different intersections to be passed may not be identical. For example, each intersection exit area corresponding to each intersection type may be set in advance, and when each intersection exit area corresponding to the intersection to be passed needs to be determined, the intersection type corresponding to the intersection to be passed is determined first, and then each intersection exit area corresponding to the intersection type is matched.

S120, acquiring position relation information between the object to be predicted and the departure area of each intersection according to each object to be predicted in the intersections to be passed, wherein the object to be predicted is a non-motor vehicle object positioned in the intersections to be passed.

The object to be predicted may be a non-motor vehicle object located in the intersection to be passed, i.e. a non-motor vehicle type obstacle of the current vehicle, such as a pedestrian, a bicycle or a tricycle.

Specifically, all objects of non-motor vehicle types in the intersection to be passed can be used as objects to be predicted so as to predict the behavior of the objects, and further, the automatic driving decision of the current vehicle can be conveniently performed according to the predicted behavior.

In this embodiment, for each object to be predicted, positional relationship information between the object to be predicted and each intersection departure area needs to be acquired. The information of the position relationship between the object to be predicted and the intersection departure area can describe the relationship between the position of the object to be predicted and the position of the intersection departure area. For example, the positional relationship information may be a distance, an angle difference, whether or not the object to be predicted is located in front of the intersection departure area, or the like.

Optionally, obtaining information of a positional relationship between the object to be predicted and the exit area of each intersection includes: for each intersection leaving area, acquiring at least one of the current distance between the object to be predicted and the intersection leaving area, the current angle difference between the object to be predicted and the intersection leaving area and the historical average distance from the object to be predicted to the lane line corresponding to the intersection leaving area; and taking at least one of the obtained current distance, the current angle difference and the historical average distance as the position relation information between the object to be predicted and the intersection departure area.

The current distance may be a euclidean distance between the object to be predicted and the intersection departure area. The current angle difference may be an angle difference between the orientation of the object to be predicted and the intersection exit area. The historical average distance may be an average distance from a historical travel track of the object to be predicted to a corresponding lane line connected to the intersection departure area.

Specifically, at least one of the current distance, the current angle difference, and the historical average distance may be acquired, and the positional relationship information may be determined according to the acquired information. By the method, accurate determination of the position relation information is achieved, euclidean distance, angle difference or historical average distance is used as the position relation information, and prediction accuracy can be further improved.

In the present embodiment, in addition to the euclidean distance, the angle difference, or the history average distance as the positional relationship information, other information may be selected as the positional relationship information, which is not limited in the present embodiment.

Optionally, obtaining information of a positional relationship between the object to be predicted and the exit area of each intersection further includes: judging whether the intersection leaving areas are positioned behind the object to be predicted or not according to each intersection leaving area, if so, determining that a first variable corresponding to the intersection leaving areas is true; judging whether the object to be predicted is positioned in a preset area corresponding to the intersection departure area, if so, determining a second variable corresponding to the intersection departure area as a true value, wherein the preset area is a relevant area of a lane connected to the intersection departure area; and determining the first variable and the second variable corresponding to the intersection departure area as position relation information between the object to be predicted and the intersection departure area.

The preset area may be a vicinity of a lane connected to the intersection exit area, for example, the preset area corresponding to the intersection exit area may be determined according to a preset area size with the lane connected to the intersection exit area as a center.

Specifically, the first variable and the second variable may be boolean variables, i.e. binary scalar values with true or false values. If the intersection departure area is positioned behind the object to be predicted, the first variable takes a true value, and if the object to be predicted is positioned in a preset area corresponding to the intersection departure area, the first variable represents that the object to be predicted is positioned near a lane connected to the intersection departure area, and the second variable takes a true value.

Further, the first truth value and the second truth value can be determined as positional relationship information between the object to be predicted and the intersection departure area. By the method, whether the intersection departure area is positioned behind the object to be predicted or not and whether the object to be predicted is positioned near the lane connected to the intersection departure area or not are determined as the position relation information together, so that the position relation information is enriched, and the prediction accuracy is further improved.

S130, inputting the position relation information into a pre-constructed regional probability prediction model to obtain the predicted forward probability corresponding to each intersection departure region.

In this embodiment, the object to be predicted located in the intersection to be passed needs to go from the current location to a certain intersection exit area to exit the intersection to be passed. Therefore, the region probability prediction model can be adopted to predict the probability that the object to be predicted leaves the intersection to be passed from each intersection leaving region, namely, the probability that the object to be predicted goes to each intersection leaving region from the current position, so as to obtain the predicted going probability corresponding to each intersection leaving region.

The regional probability prediction model may be a probability map model, such as a BN (bayesian Network), a DBN (Dynamic Bayesian Network ), a ridge regression model, a least squares method, or the like.

In this embodiment, the regional probability prediction model may obtain prior probability and conditional probability according to the input positional relationship information, and further determine posterior probability according to the prior probability and the conditional probability, so as to obtain the predicted forward probability corresponding to each intersection departure region.

In a specific embodiment, the regional probability prediction model is a dynamic bayesian model, and the method for obtaining the predicted forward probability corresponding to each intersection departure region by inputting the positional relationship information into the regional probability prediction model constructed in advance may include:

aiming at each intersection leaving area, taking the intersection leaving area as an intention node in a dynamic Bayesian model, and taking at least one of the current distance, the current angle difference and the historical average distance corresponding to the intersection leaving area as each observation discussion node connected with the intention node; for each intention node in the dynamic Bayesian model, determining the prediction forward probability corresponding to the intention node based on each observation argument node connected with the intention node.

Wherein the dynamic Bayesian model can include intent nodes and observation argument nodes; each intended node represents an intersection departure area and one observation-to-discussion node represents a current distance, a current angular difference, or a historical average distance. For example, if the positional relationship information includes a current distance, a current angle difference, and a historical average distance, each intended node (intersection departure area) is connected to three observation-arguments nodes (current distance, current angle difference, historical average distance corresponding to the intersection departure area)

It should be noted that, each intention node is connected to each observation node, which represents that there is a causal relationship between the intention node and the observation node, and a change in the intention node will cause a change in the observation node. Illustratively, fig. 3 is a simple node connection schematic diagram in an embodiment of the disclosure, where a circle in the graph is called a node, m_i represents an intended node, e represents an observation argument node, and an arrow in the graph represents a causal relationship between nodes.

In this embodiment, the predicted arrival probability corresponding to each intention node, that is, the predicted arrival probability corresponding to each intersection departure area may be determined according to each observation and argument node to which each intention node is connected.

In other words, this embodiment realizes that: in case e is observed, the intention node m=m is calculated _i Probability size P (m=m _i I e), which is the predicted forward probability corresponding to the intended node. By the method, the prediction forward probability corresponding to each intersection departure area is predicted based on the dynamic Bayesian model, further the intention prediction of each object to be predicted is realized, compared with the neural network prediction mode, data processing and labeling are not needed, and the prediction result is stable.

Optionally, for each intention node in the dynamic bayesian model, determining, based on each observation argument node connected by the intention node, a predicted forward probability corresponding to the intention node may be: determining, for each intended node, a first conditional probability of each observation argument node under the intended node, a priori probability of the intended node, and an edge probability of each observation argument node to which the intended node is connected; and determining products of the first conditional probabilities and the prior probabilities, and taking the ratio of the products to the edge probabilities as the predicted forward probabilities corresponding to the intention nodes.

Wherein for each observation arguments node to which each intended node is connected, a conditional probability of the observation arguments under that intent may be determined, resulting in a first conditional probability, e.g., P (e| m=m _i ) Representing the intentional node m _i The observations below address a first conditional probability of the node. The prior probability of each intended node may be preset. The edge probability of each observation argument node that the intended node connects can be understood as a normalized variable.

Illustratively, the predicted forward probability corresponding to the intended node may be calculated by the following formula:

wherein P (m=m _i I e) represents the intent node m _i The corresponding predicted forward probability, i.e., posterior probability; p (e|m=m) _i ) To intentionally aim at node m _i The first conditional probabilities of the observation and argument nodes under the condition that the number of the observation and argument nodes connected by the intention node is a plurality of, multiplying all the first conditional probabilities; p (m=m) _i ) Representing intent node m _i Is a priori probability of (2); sigma (sigma) _m P (m, e) represents the edge probability of each observation argument node that the node is intended to connect.

By the method, the prediction forward probability corresponding to each intention node can be calculated, and the probability that the object to be predicted goes to each intersection departure area is accurately predicted.

It should be noted that, for the prior probability of the intention node, the prior probability of the intention node may be determined by further combining the change caused by the time factor in addition to the direct acquisition.

Optionally, determining the prior probability of the intention node includes: obtaining the predicted forward probability corresponding to each intention node at the previous moment and the transition probability of each intention node at the previous moment to the intention node at the current moment; for each intention node at the previous moment, taking the product of the predicted forward probability and the transition probability corresponding to the intention node at the previous moment as the reference probability of the intention node at the previous moment, and determining the sum of the reference probabilities of the intention nodes at the previous moment as the prior probability of the intention node at the current moment.

That is, the prior probability of the intention node at the current time may be determined in combination with the predicted forward probability corresponding to each intention node at the previous time and the transition probability of each intention node at the previous time to transition to the intention node at the current time.

For example, fig. 4 is a schematic node connection diagram combined with a time factor in the embodiment of the disclosure, referring to fig. 4, the calculation of the intention probability at the current time t is affected not only by the observation of the current time t, but also by the calculation result at the previous time t-1 (i.e. the dashed line in fig. 4).

Specifically, for each intention node at the previous moment, taking the product of the corresponding predicted forward probability and the transition probability as the reference probability, and further determining the sum of the reference probabilities of all intention nodes at the previous moment as the prior probability of the intention node at the current moment.

By way of example, the prior probability of the intended node at the current time may be calculated using the following formula:

P(m ^t ＝m _i )＝∑ _j (P(m ^t ＝m _i |m ^t-1 ＝m _j )P(m ^t-1 ＝m _j |e ^t-1 ))；

wherein P (m ^t ＝m _i ) Intent node m representing current time t _i Is a priori probability of P (m ^t ＝m _i |m ^t-1 ＝m _j ) The intended node m representing the last time t-1 _j Intent node m to transition to current time t _i Is a transition probability of P (m) ^t-1 ＝m _j |e ^t-1 ) The intended node m representing the last time t-1 _j The corresponding predicted forward probabilities.

As can be seen from the calculation formula of the prior probability, the predicted forward probability corresponding to the intention node at the final current time can be obtained from the following formula:

wherein P (m ^t ＝m _i |e ^t ) An intention node m representing the current time _i Corresponding predicted probability of going to e ^t For the intention node m _i Corresponding observation arguments node, P (e ^t |m ^t ＝m _i ) Representing a first conditional probability.

In the embodiment, the prior probability of the current moment is determined based on the prediction result of the previous moment, so that the change caused by taking the time factor into consideration in the intention prediction is realized, and the accuracy of the intention prediction is further improved.

Besides the above-mentioned combination of the prediction result of the previous moment to determine the prior probability of the current moment, the reason argument node may be further defined in the dynamic bayesian model, so as to consider the influence of the reason argument node on the prior probability of the current moment.

Optionally, after taking the intersection exit area as the intended node in the dynamic bayesian model for each intersection exit area, the method further includes: the first variable and the second variable corresponding to the intersection departure area are used as the reason arguments of the connection intention node;

accordingly, determining the prior probability of the intended node further includes: determining second conditional probabilities of the intent node under the causal argument nodes connecting the intent node; taking the sum of the second conditional probabilities as a first value, taking the sum of the reference probabilities of the intention nodes at the previous moment as a second value, and taking the product of the first value and the second value as the prior probability of the intention node at the current moment.

The first variable and the second variable corresponding to the intersection departure area can be used as the reason statement nodes corresponding to the intersection departure area, are connected with the intention nodes corresponding to the intersection departure area, form causal relation with the intention nodes, and cause the change of the intention nodes when the reason statement nodes change.

Fig. 5 is a schematic structural diagram of a dynamic bayesian model according to an embodiment of the present disclosure, where REV and IN are causal arguments, respectively represent a first variable and a second variable, a_1, a_2, … … a_m represent each intention node, and d_to_area, theta_diff, avg_d_to_lane are observation arguments, respectively represent a current distance, a current angle difference, and a historical average distance; the broken line connected to each intended node represents the predicted result of the previous time.

Specifically, the prior probability of the intention node at the current moment is determined, and the prior probability of each intention node at the previous moment can be further determined by combining each second conditional probability of the intention node and the reference probability of each intention node at the previous moment. Wherein the second conditional probability represents an intended conditional probability under a specific causal argument.

In the present embodiment, the sum of the two second conditional probabilities is taken as the first numerical value; taking the sum of the reference probabilities of all the intention nodes at the previous moment as a second numerical value; the product of the first value and the second value is the prior probability of the intended node at the current time.

Illustratively, the prior probability may be calculated by the following formula:

wherein P (m ^t ＝a ₁ ) Node a representing the intention of the current time ₁ C represents two causal arguments nodes that affect the intent node: the first and second variables may be regarded as 1 in terms of their prior probability P (c) since they can give objective results before computation. Finally, the prior probability of the intended node at the current time may be expressed by the following formula:

further, the predicted forward probability corresponding to the intended node at the current time may be calculated by the following formula:

wherein, the liquid crystal display device comprises a liquid crystal display device,

Representing a specific intention a ₁ Observation theory below->

Conditional probability, i.e.)>

Is a first conditional probability of P (m ^t ＝a ₁ |m ^t-1 ＝m _j ) The intended node m representing the last time t-1 _j Intent node a transitioning to current time t ₁ Is a transition probability of P (m) ^t ＝a ₁ And c) represents the probability of intent condition under the specific reason argument c, i.e. a ₁ Is a second conditional probability of (2).

Representing intent node a ₁ The observations of the connection concern the edge probability of the node.

In the embodiment, the reason argument node is introduced into the dynamic Bayesian model to consider the influence of the reason argument node on the prior probability of the current moment, so that the accuracy of the prior probability is further improved, and the accuracy of the prediction result of the current moment is further improved.

The first conditional probability, the second conditional probability, and the transition probability may be set manually or may be obtained through training.

In a specific embodiment, before inputting the positional relationship information into the pre-constructed regional probability prediction model to obtain the predicted forward probability corresponding to each intersection departure region, the method further comprises: acquiring a training data set, wherein the training data set comprises sample position relation information between a sample object and each sample leaving area and each area forward probability label corresponding to each sample position relation information; and training the regional probability prediction model based on the training data set to obtain a first conditional probability table, a second conditional probability table and a transition probability table, or to obtain the first conditional probability model, the second conditional probability model and the transition probability model.

Wherein a first conditional probability table or a first conditional probability model is used for determining each first conditional probability, a second conditional probability table or a second conditional probability model is used for determining each second conditional probability, and a transition probability table or a transition probability model is used for determining each transition probability.

Specifically, the training data set may be a small sample data set, including sample position relation information corresponding to each sample departure area, and each area forward probability label corresponding to each sample position relation information.

The training data set may be input to the region probability prediction model, so that the region probability prediction model obtains a first conditional probability table, a second conditional probability table, and a transition probability table according to the forward probability labels of the regions, or obtains the first conditional probability model, the second conditional probability model, and the transition probability model.

After the probability table or the probability model is obtained, the first conditional probability of the observation statement node under a certain intention node can be queried through the first conditional probability table or the first conditional probability model, the second conditional probability of the intention node under a certain reason statement node can be queried through the second conditional probability table or the second conditional probability model, and the transition probability among the intention nodes can be queried through the transition probability table or the transition probability model. According to the queried probability, the edge probability can be further calculated, the posterior probability is further calculated, and the predicted forward probability corresponding to the intersection departure area is obtained.

By means of the training data set training mode, the first conditional probability, the second conditional probability and the transition probability can be accurately determined, manual setting is not needed, accuracy of all probabilities is guaranteed, and meanwhile prediction efficiency is improved.

The above-described method for calculating the predicted forward probability from the first conditional probability, the second conditional probability, and the transition probability may be understood as: the probability of going to each region under the known observation theory is predicted from the known probabilities of going to various observation theory (position, angle difference, etc.) under each specific region.

And S140, determining a target area corresponding to the object to be predicted in each intersection departure area based on each prediction arrival probability, and determining a target track corresponding to the object to be predicted based on the target area.

Specifically, after the predicted arrival probabilities of the object to be predicted going to the intersection departure areas are obtained, the target area may be determined from the intersection departure areas according to all the predicted arrival probabilities. Wherein the number of target areas may be one or more.

For example, all the predicted arrival probabilities may be ranked in order of probability from high to low, and the intersection departure region corresponding to the first N predicted arrival probabilities may be used as the target region, or the intersection departure region corresponding to the highest predicted arrival probability may be used as the target region.

After the target areas are obtained, a target track of the object to be predicted to the target area can be predicted for each target area. The target trajectory is generated, for example, by means of a curve model or by means of a deep learning network. In order to further improve the reliability of the target track, the embodiment can also predict the target track by combining the environment information.

In a specific embodiment, determining, based on the target area, a target track corresponding to the object to be predicted may include the following steps:

step 11, acquiring state information and environment information of an object to be predicted at the current time, wherein the state information comprises first position information and first speed information of the object to be predicted, and the environment information comprises second position information and second speed information of other objects to be predicted in a crossing to be passed;

step 12, determining speed control information of the object to be predicted at the current time according to the environmental information and the state information at the current time, and determining the state information of the object to be predicted at the next time based on the state information and the speed control information of the object to be predicted at the current time;

step 13, re-acquiring the environmental information at the current time by taking the next time as the current time, determining the speed control information according to the environmental information and the state information at the current time until the state information of the object to be predicted at the last time is determined, and determining the target track corresponding to the object to be predicted according to the state information at each time; the first position information of the object to be predicted at the last moment is the target area.

For the above step 11, considering that there are a plurality of objects to be predicted in the intersection to be passed, the positions and speeds of the objects to be predicted will affect each other, so when predicting the target track of the current object to be predicted, not only the state information of the object to be predicted at the current time, that is, the first position information and the first speed information of the object to be predicted, but also the environmental information of the object to be predicted at the current time, that is, the second position information and the second speed information of other objects to be predicted, may be obtained.

For the above step 12, according to the environmental information and the state information at the current time, the optimal control amount, i.e. the speed control information, of the object to be predicted at the current time may be solved. Further, the state information of the next moment is determined according to the speed control information at the current moment. Wherein the speed control information may be information in the form of a speed vector.

Further, the next time is taken as the current time, the environmental information under the current time is obtained again, and the execution step 12 is returned until the state information of the object to be predicted under the last time is predicted, wherein the first position information under the last time corresponds to the target area, namely, the object to be predicted under the last time reaches the target area. Finally, the target track can be determined according to the state information at all times.

Through the steps 11-13, the state information under each time is predicted, the environment information under each time is considered, and the prediction accuracy of the track is ensured. In addition, compared with the mode of deep learning or curve model, after the state information of each moment is determined, the environment information can be updated in the steps, so that the state information of the next moment can be solved on the basis of the updated environment information, and the like.

For the step 12, optionally, determining the speed control information of the object to be predicted at the current time according to the environmental information and the state information at the current time includes: constructing a distance cost function, wherein the distance cost function is used for describing the relation between the object distance and the cost between the object to be predicted and other objects to be predicted; and determining the speed control information corresponding to the minimum distance cost according to the environmental information and the state information at the current moment by taking the calculation result of the distance cost function as a target.

Specifically, the distance cost function may be used to calculate a corresponding distance cost according to the object distance between the object to be predicted and other objects to be predicted. The objective of constructing the distance cost function is to: interactions of the object to be predicted with other objects to be predicted in the environment are considered; in order to avoid collisions between the object to be predicted and other objects to be predicted, the minimum distance between the object to be predicted and other objects to be predicted should be as large as possible, and thus the distance cost should be as small as possible.

In this embodiment, the speed control information corresponding to the minimum distance cost may be obtained with the result of the calculation of the distance cost function being minimized as a target, that is, with the distance cost being minimized as a target. And the optimal speed control information is obtained by taking the minimum calculation result of the distance cost function as a target, so that the speed control information can meet the requirement that the distance between objects is as large as possible, and further collision between the objects is avoided.

For the construction process of the distance cost function, in a specific implementation manner, the construction process of the distance cost function includes: constructing a relation between the minimum object distance between the object to be predicted and other objects to be predicted and speed control information according to square distance change between the object to be predicted and other objects to be predicted; and constructing a distance cost function according to the relation between the minimum object distance and the speed control information and a preset first super parameter.

Wherein the square distance change is used to describe the amount of distance change between objects at each time instant. In this embodiment, assuming that the other objects to be predicted all keep the current speed to move at a constant speed, the square distance between the objects to be predicted and the other objects to be predicted may be expressed by the following formula:

d _mn (t,u _m )＝‖p _m +tu _m -p _n -v _n ‖ ² ；

wherein t represents the current time, u _m Representing speed control information, p _m Representing object o to be predicted _m P _n Representing other objects to be predicted o _n V of the second position information of (v) _n Representing other objects to be predicted o _n Is set, and the second speed information of (1).

Further, for the formula of square distance change, the time t can be derived, and the minimum distance between the object to be predicted and other objects to be predicted is obtained at the moment:

further, the formula obtained after deriving the time t is carried into the formula of square distance change, so that the relation between the minimum object distance and the speed control information can be obtained:

wherein k=p _m -p _n ，q＝u _m -v _n . Further, the first super parameter may be presetAnd constructing a distance cost function from the relationship, the distance cost function being expressed by the following formula:

wherein sigma _d For presetting a first super parameter, controlling other objects o to be predicted _n Is the influence range of sigma _d The larger the object o to be predicted _m Tends to be farther away from other objects o to be predicted _n 。

Specifically, the distance cost function can be solved by a gradient descent method, so that speed control information is obtained. By the method, the distance cost function is accurately constructed, and the accuracy of speed control information solved based on the distance cost function is further guaranteed.

Considering that a plurality of other objects to be predicted may exist in the environment, when the plurality of other objects to be predicted all affect the objects to be predicted, a final distance cost function may also be constructed, where the final distance cost function is a weighted sum of costs of the other objects to be predicted.

Optionally, after constructing the distance cost function according to the relationship between the minimum object distance and the speed control information and the preset first super parameter, the method further includes: if the number of the other objects to be predicted is multiple, determining cost weights corresponding to the other objects to be predicted according to the angle difference and the distance between the objects to be predicted and the other objects to be predicted at the current time, and determining distance cost components corresponding to the other objects to be predicted according to the distance cost function and the cost weights corresponding to the other objects to be predicted; and constructing a final distance cost function according to the sum of the distance cost components corresponding to the other objects to be predicted.

The angle difference may be an angle that the current direction of the object to be predicted passes when rotating to other objects to be predicted, and the distance may be a distance between two objects at the current time.

For example, the cost weights corresponding to other objects to be predicted may be calculated according to the following formula:

wherein w is _mn Representing other objects to be predicted o _n The corresponding cost weight, phi is the object o to be predicted at the current time _m With other objects to be predicted o _n The definition of the angle difference, k, is as described above, and beta, sigma _w For the super-parameters, beta determines the maximum value of the distance cost term, sigma _w Action and sigma _d Similarly, the scope of action of other objects to be predicted is affected.

After the cost weights corresponding to the other objects to be predicted are obtained, the product of the cost weights corresponding to the other objects to be predicted and the distance cost function can be used as the distance cost component corresponding to the other objects to be predicted, and then the sum of the distance cost components corresponding to all the other objects to be predicted is used as the final distance cost function.

Illustratively, at t _i Time of day, object to be predicted o _m The distance cost function from the environment is defined as:

I _m (u _m )＝∑ _n≠m w _mn ·E _mn (u _m )；

wherein I is _m (u _m ) Is the final distance cost function. Specifically, the final distance cost function can be solved by a gradient descent method, so that speed control information is obtained. By the method, the construction of the distance cost function under a plurality of other objects to be predicted is realized, so that the track prediction considering that the objects to be predicted can avoid collision with other objects is realized, and the track prediction precision is further improved.

In addition to solving the optimal control amount according to the distance cost function, in this embodiment, the optimal control amount may be solved in combination with the speed cost function.

For example, optionally, determining the speed control information of the object to be predicted at the current time according to the environmental information and the state information at the current time further includes: constructing a speed cost function, wherein the speed cost function is used for describing the relation between the speed difference and the cost between the speed control information and the expected speed of the object to be predicted; and constructing a total cost function according to the speed cost function and the distance cost function, and determining speed control information corresponding to the minimum total cost according to the environmental information and the state information at the current moment by taking the calculation result of the total cost function as a target.

Specifically, the speed cost function may be used to calculate a corresponding speed cost according to a speed difference between the speed control information of the object to be predicted and the desired speed. The purpose of constructing the velocity cost function is to: so that the speed control information is as close as possible to the desired speed. The smaller the speed difference between the speed control information and the desired speed, the smaller the speed cost calculated by the speed cost function.

For example, the constructed speed cost function may be expressed by the following formula:

wherein S is _m (u _m ) The cost of the speed is represented by the velocity,

the expected speed is indicated, and the expected speed corresponding to each object type can be preset, so that the expected speed corresponding to the object to be predicted can be matched.

Further, the sum of the speed cost function and the distance cost function can be used as a total cost function; or constructing a total cost function according to the speed cost function, the weight corresponding to the speed cost function, the distance cost function and the weight corresponding to the distance cost function. Specifically, the total cost function can be solved by a gradient descent method, so that speed control information is obtained.

Further, the optimal control quantity is solved by taking the total cost calculated by the total cost function as a target, and the speed control information is obtained. By the method, control quantity solution of combining the speed cost and the distance cost is realized, the fact that the speed of the object to be predicted at each moment is close to the expected speed in theory is considered, collision with other objects can be avoided in theory, and accuracy of track prediction is further improved.

In this embodiment, the direction cost function may be combined in addition to the speed cost function.

Optionally, before constructing the total cost function according to the speed cost function and the distance cost function, the method further includes: constructing a direction cost function, wherein the direction cost function is used for describing the relationship between the direction difference and the cost between the speed control information of the object to be predicted and the target area; correspondingly, constructing a total cost function according to the speed cost function and the distance cost function, including: and constructing a total cost function according to the speed cost function, the direction cost function and the distance cost function.

The direction cost function may be used to calculate a corresponding direction cost according to a direction difference between the speed control information and the target area. The objective of constructing the directional cost function is to: the speed direction in the speed control information is as close as possible to the direction in which the target area is located. The smaller the direction difference between the speed control information and the target area, the smaller the direction cost calculated by the direction cost function.

For example, the constructed directional cost function may be expressed by the following formula:

wherein z is _m Representing coordinate points in the target region, e.g. the region edge point closest to the object to be predicted in the target region, or the region center point of the target region, D _m (u _m ) Representing the directional cost.

Further, a total cost function may be constructed from the speed cost function, the direction cost function, and the distance cost function. Specifically, the sum of the speed cost function, the direction cost function and the distance cost function can be directly determined as the total cost function; or, the total cost function can be constructed according to the speed cost function, the weight corresponding to the speed cost function, the direction cost function, the weight corresponding to the direction cost function, and the distance cost function.

Illustratively, the total cost function may be expressed by the following formula:

E _m (u _m )＝I _m (u _m )+λ ₁ S _m (u _m )+λ ₂ D _m (u _m )；

wherein E is _m (u _m ) Representing the total cost, lambda ₁ And lambda (lambda) ₂ Is a super parameter used to adjust the impact of speed cost and direction cost. Specifically, the formula can be solved by a gradient descent method to obtain speed control information. By the method, control quantity solving combining the speed cost, the direction cost and the distance cost is realized, the fact that the speed of the object to be predicted at each moment is close to the expected speed in theory is considered, collision with other objects is avoided, the direction of the speed approaches to the direction of the target area, and the accuracy of track prediction is further improved.

In this embodiment, the speed control information at each time is obtained by solving at each time, and then the state information at the next time is obtained based on the speed control information at that time, and so on, the state information at a total of N times can be obtained.

For the above step 13, in a specific embodiment, determining the state information of the object to be predicted at the next time based on the state information and the speed control information of the object to be predicted at the current time includes: determining first speed information of the object to be predicted at the next moment according to a preset second super parameter, the first speed information of the object to be predicted at the current moment and the speed control information; and determining the first position information of the object to be predicted at the next time according to the first position information of the object to be predicted at the current time, the first speed information at the next time and the time difference between the current time and the next time.

Specifically, the product between the first speed information and the preset second super parameter can be determined, the product between the result of subtracting the preset second super parameter from the set value and the speed control information can be determined, and the sum of the two products is taken as the first speed information at the next moment. Wherein the set value may be 1.

Further, a product between the first speed information at the next time and the time difference may be determined, and a sum of the product and the first position information at the current time may be used as the first position information at the next time.

For example, see the following formula:

wherein, the state information of the object to be predicted at the current time is defined as

The first position information representing the current time point can be two-dimensional coordinates; />

The first speed information representing the current time instant may be a speed vector. />

First location information indicative of the next moment, < > in->

The first speed information at the next moment is represented, alpha represents a preset second super parameter, and the value is between 0 and 1.

In the above embodiment, the state information of the next time is determined based on the speed control information of the current time, so that the state information of the next time can be determined according to the state information and the speed control information of the previous time, further, the state information of each time is sequentially determined, the relevance among the state information of each time is ensured, and the track prediction accuracy is further improved.

It should be noted that, in the formulas provided in this embodiment, the respective super parameters α, β, σ involved _d ,σ _w ,λ ₁ ,λ ₂ The method can collect a large amount of track data of the non-motor vehicles in the intersection area in advance, and each super-parameter can be estimated and obtained through a maximum likelihood method.

The method provided by the embodiment can realize the prediction of the target area of each object to be predicted and the prediction of the target track going to each target area, further realize the behavior prediction of each object to be predicted, and automatically predict the intention and the track of the non-motor vehicle by the method provided by the embodiment when the automatic driving vehicle passes through the intersection, so that the automatic driving vehicle makes more reasonable decision and planning, and achieves safer and stable driving effect.

For example, the method provided by the present embodiment may be performed by a behavior prediction module of an autonomous vehicle. Referring to fig. 6, fig. 6 is a schematic diagram of an automatic driving system in an embodiment of the disclosure, where a behavior prediction module may perform intent prediction and trajectory prediction on each object to be predicted according to information output by a received sensing module, a map module, and a positioning module, so that a planning decision module makes a reasonable path decision plan according to a prediction result, and finally controls a vehicle action through a control module.

According to the behavior prediction method for the intersection object, the intersection exit areas corresponding to the intersections to be passed are obtained, the predicted arrival probability corresponding to the intersection exit areas is obtained according to the position relation information between the objects to be predicted and the intersection exit areas and the area probability prediction model constructed in advance for each non-motor vehicle to be predicted object in the intersections to be passed, the target area is determined from the intersection exit areas according to the predicted arrival probability, the target track is determined according to the target area, the prediction of the target area and the track of the non-motor vehicle object is achieved, the behavior prediction of the non-motor vehicle object is achieved, the problem that the behavior prediction of the non-motor vehicle obstacle is lacked and the target area cannot be predicted in the prior art is solved, and the problem that the stability of the prediction result of the neural network adopted in the prior art is poor is solved by performing the behavior prediction in the prediction probability mode.

Fig. 7 is a schematic structural diagram of an intersection object behavior prediction apparatus in an embodiment of the present disclosure. As shown in fig. 7: the device comprises: a region determination module 710, an information acquisition module 720, a probability prediction module 730, and a behavior determination module 740.

The area determining module 710 is configured to obtain each intersection departure area corresponding to an intersection to be passed of the current vehicle;

the information obtaining module 720 is configured to obtain, for each object to be predicted in the intersection to be passed, positional relationship information between the object to be predicted and each intersection exit area, where the object to be predicted is a non-motor vehicle object located in the intersection to be passed;

the probability prediction module 730 is configured to input each piece of positional relationship information into a pre-constructed regional probability prediction model, so as to obtain a predicted forward probability corresponding to each intersection departure region;

the behavior determining module 740 is configured to determine a target area corresponding to the object to be predicted in each intersection departure area based on each predicted arrival probability, and determine a target track corresponding to the object to be predicted based on the target area.

The behavior prediction device for the intersection object provided by the embodiment of the present disclosure may execute steps in the behavior prediction method for the intersection object provided by the embodiment of the present disclosure, and the execution steps and the beneficial effects are not described herein.

Fig. 8 is a schematic structural diagram of an electronic device in an embodiment of the disclosure. Referring now in particular to fig. 8, a schematic diagram of an electronic device 500 suitable for use in implementing embodiments of the present disclosure is shown. The electronic device shown in fig. 8 is merely an example and should not be construed to limit the functionality and scope of use of the disclosed embodiments.

As shown in fig. 8, an electronic device 500 may include a processing means (e.g., a central processor, a graphics processor, etc.) 501 that may perform various suitable actions and processes to implement the methods of embodiments as described in the present disclosure according to a program stored in a Read Only Memory (ROM) 502 or a program loaded from a storage means 508 into a Random Access Memory (RAM) 503. In the RAM 503, various programs and data required for the operation of the electronic apparatus 500 are also stored. The processing device 501, the ROM 502, and the RAM 503 are connected to each other via a bus 504. An input/output (I/O) interface 505 is also connected to bus 504.

In particular, according to embodiments of the present disclosure, the processes described above with reference to flowcharts may be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product comprising a computer program embodied on a non-transitory computer readable medium, the computer program comprising program code for performing the method shown in the flowchart, thereby implementing the intersection object behavior prediction method as described above. In such an embodiment, the computer program may be downloaded and installed from a network via the communication means 509, or from the storage means 508, or from the ROM 502. The above-described functions defined in the methods of the embodiments of the present disclosure are performed when the computer program is executed by the processing device 501.

It should be noted that the computer readable medium described in the present disclosure may be a computer readable signal medium or a computer readable storage medium, or any combination of the two. The computer readable storage medium can be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the foregoing. More specific examples of the computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the context of this disclosure, a computer-readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. In the present disclosure, however, the computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, with the computer-readable program code embodied therein. Such a propagated data signal may take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination of the foregoing. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to: electrical wires, fiber optic cables, RF (radio frequency), and the like, or any suitable combination of the foregoing.

The computer readable medium may be contained in the electronic device; or may exist alone without being incorporated into the electronic device. The computer readable medium carries one or more programs which, when executed by the electronic device, cause the electronic device to:

Alternatively, the electronic device may perform other steps described in the above embodiments when the above one or more programs are executed by the electronic device.

In the context of this disclosure, a machine-readable medium may be a tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. The machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

Scheme 1, a behavior prediction method of an intersection object, the method comprising:

The method according to claim 2, wherein the obtaining the information of the positional relationship between the object to be predicted and each intersection departure area includes:

for each intersection leaving area, acquiring at least one of a current distance between the object to be predicted and the intersection leaving area, a current angle difference between the object to be predicted and the intersection leaving area and a historical average distance from the object to be predicted to a lane line corresponding to the intersection leaving area;

and taking the obtained at least one of the current distance, the current angle difference and the historical average distance as the position relation information between the object to be predicted and the intersection departure area.

The method according to the scheme 3, wherein the regional probability prediction model is a dynamic bayesian model, the inputting the positional relationship information into the regional probability prediction model constructed in advance, to obtain the predicted arrival probability corresponding to the intersection departure region, includes:

For each intersection leaving area, taking the intersection leaving area as an intention node in the dynamic Bayesian model, and taking at least one of the current distance, the current angle difference and the historical average distance corresponding to the intersection leaving area as each observation discussion node connected with the intention node;

for each of the intent nodes in the dynamic bayesian model, determining a predicted forward probability corresponding to the intent node based on each of the observation and argument nodes connected by the intent node.

Solution 4, the method according to claim 3, wherein for each of the intent nodes in the dynamic bayesian model, determining, based on each of the observation argument nodes connected by the intent node, a predicted arrival probability corresponding to the intent node includes:

determining, for each of the intended nodes, a first conditional probability of each of the observation argument nodes under the intended node, an a priori probability of the intended node, and an edge probability of each of the observation argument nodes to which the intended node is connected;

and determining products of the first conditional probabilities and the prior probabilities, and taking the ratio of the products to the edge probabilities as a predicted forward probability corresponding to the intention node.

Scheme 5, the method according to scheme 4, determining the prior probability of the intended node, comprising:

obtaining the predicted forward probability corresponding to each intention node at the previous moment and the transition probability of each intention node at the previous moment to the intention node at the current moment;

for each intention node at the previous moment, taking the product of the predicted forward probability and the transition probability corresponding to the intention node at the previous moment as the reference probability of the intention node at the previous moment, and determining the sum of the reference probabilities of the intention nodes at the previous moment as the prior probability of the intention node at the current moment.

The method according to claim 6, wherein the obtaining the information of the positional relationship between the object to be predicted and each intersection departure area further includes:

judging whether the intersection leaving areas are positioned behind the object to be predicted or not according to each intersection leaving area, if yes, determining that a first variable corresponding to the intersection leaving areas is true;

judging whether the object to be predicted is positioned in a preset area corresponding to the intersection departure area, if so, determining a second variable corresponding to the intersection departure area as a true value, wherein the preset area is a relevant area of a lane connected to the intersection departure area;

And determining the first variable and the second variable corresponding to the intersection departure area as position relation information between the object to be predicted and the intersection departure area.

The method according to claim 7, wherein after said taking the intersection exit area as the intention node in the dynamic bayesian model for each of the intersection exit areas, further comprises:

the first variable and the second variable corresponding to the intersection departure area are used as the reason arguments nodes connected with the intention nodes;

the determining the prior probability of the intended node further includes:

determining second conditional probabilities of the intent node under the causal argument nodes connecting the intent node;

taking the sum of the second conditional probabilities as a first value, taking the sum of the reference probabilities of the intention nodes at the previous moment as a second value, and taking the product of the first value and the second value as the prior probability of the intention node at the current moment.

In an aspect 8, according to the method of the aspect 7, before the inputting the positional relationship information into the pre-constructed area probability prediction model to obtain the predicted arrival probability corresponding to the intersection departure area, the method further includes:

Acquiring a training data set, wherein the training data set comprises sample position relation information between a sample object and each sample leaving area and each area forward probability label corresponding to each sample position relation information;

training the regional probability prediction model based on the training data set to obtain a first conditional probability table, a second conditional probability table and a transition probability table, or to obtain a first conditional probability model, a second conditional probability model and a transition probability model;

wherein the first conditional probability table or the first conditional probability model is used for determining each of the first conditional probabilities, the second conditional probability table or the second conditional probability model is used for determining each of the second conditional probabilities, and the transition probability table or the transition probability model is used for determining each of the transition probabilities.

The method according to claim 9, according to claim 1, wherein the determining, based on the target area, the target track corresponding to the object to be predicted includes:

acquiring state information and environment information of the object to be predicted at the current moment, wherein the state information comprises first position information and first speed information of the object to be predicted, and the environment information comprises second position information and second speed information of other objects to be predicted in the intersection to be passed;

Determining speed control information of the object to be predicted at the current time according to the environment information and the state information at the current time, and determining the state information of the object to be predicted at the next time based on the state information and the speed control information of the object to be predicted at the current time;

the next moment is taken as the current moment, the environmental information under the current moment is acquired again, the speed control information is determined according to the environmental information and the state information under the current moment until the state information of the object to be predicted under the last moment is determined, and the target track corresponding to the object to be predicted is determined according to the state information under each moment;

and the first position information of the object to be predicted at the last moment is the target area.

The method according to claim 10, according to claim 9, wherein the determining the speed control information of the object to be predicted at the current time according to the environmental information and the state information at the current time includes:

constructing a distance cost function, wherein the distance cost function is used for describing the relation between the object distance and cost between the object to be predicted and other objects to be predicted;

And determining speed control information corresponding to the minimum distance cost according to the environmental information and the state information at the current moment by taking the calculation result of the distance cost function as a target.

Solution 11, the method according to solution 10, the constructing a distance cost function, includes:

constructing a relation between the minimum object distance between the object to be predicted and other objects to be predicted and speed control information according to square distance change between the object to be predicted and other objects to be predicted;

and constructing a distance cost function according to the relation between the minimum object distance and the speed control information and a preset first super parameter.

The method according to claim 12, wherein after the constructing the distance cost function according to the relationship between the minimum object distance and the speed control information and the preset first super parameter, the method further includes:

if the number of the other objects to be predicted is multiple, determining cost weights corresponding to the other objects to be predicted according to the angle difference and the distance between the objects to be predicted and the other objects to be predicted at the current time, and determining distance cost components corresponding to the other objects to be predicted according to the distance cost function and the cost weights corresponding to the other objects to be predicted;

And constructing a final distance cost function according to the sum of the distance cost components corresponding to the other objects to be predicted.

The method according to claim 13, wherein the determining the speed control information of the object to be predicted at the current time according to the environmental information and the state information at the current time further includes:

constructing a speed cost function, wherein the speed cost function is used for describing the relation between the speed difference and the cost between the speed control information of the object to be predicted and the expected speed;

and constructing a total cost function according to the speed cost function and the distance cost function, and determining speed control information corresponding to the minimum total cost according to the environmental information and the state information at the current moment by taking the calculation result of the total cost function as a target.

Scheme 14, the method of scheme 13, before said constructing a total cost function from said velocity cost function and said distance cost function, further comprising:

constructing a direction cost function, wherein the direction cost function is used for describing the relationship between the direction difference and the cost between the speed control information of the object to be predicted and the target area;

Correspondingly, the constructing a total cost function according to the speed cost function and the distance cost function includes:

and constructing a total cost function according to the speed cost function, the direction cost function and the distance cost function.

The method according to claim 15, according to claim 9, wherein the determining the state information of the object to be predicted at the next time based on the state information and the speed control information of the object to be predicted at the current time includes:

determining first speed information of the object to be predicted at the next moment according to a preset second super parameter, the first speed information of the object to be predicted at the current moment and the speed control information;

and determining the first position information of the object to be predicted at the next moment according to the first position information of the object to be predicted at the current moment, the first speed information at the next moment and the time difference between the current moment and the next moment.

Scheme 16, a behavior prediction apparatus for an intersection object, comprising:

Scheme 17, an electronic device, the electronic device comprising:

one or more processors;

a storage means for storing one or more programs;

the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the methods of any of aspects 1-15.

Scheme 18, a computer readable storage medium having stored thereon a computer program which when executed by a processor implements the method according to any of the claims 1-15.

The foregoing description is only of the preferred embodiments of the present disclosure and description of the principles of the technology being employed. It will be appreciated by persons skilled in the art that the scope of the disclosure referred to in this disclosure is not limited to the specific combinations of features described above, but also covers other embodiments which may be formed by any combination of features described above or equivalents thereof without departing from the spirit of the disclosure. Such as those described above, are mutually substituted with the technical features having similar functions disclosed in the present disclosure (but not limited thereto).

Claims

1. A method for predicting behavior of an intersection object, the method comprising:

2. The method according to claim 1, wherein the obtaining positional relationship information between the object to be predicted and each of the intersection exit areas includes:

3. The method according to claim 2, wherein the regional probability prediction model is a dynamic bayesian model, the inputting each piece of positional relationship information into a pre-constructed regional probability prediction model to obtain a predicted arrival probability corresponding to each intersection departure region, and the method includes:

4. A method according to claim 3, wherein said determining, for each of said intended nodes in said dynamic bayesian model, a predicted forward probability corresponding to said intended node based on each of said observation-based nodes to which said intended node is connected, comprises:

5. The method of claim 4, wherein determining the prior probability of the intended node comprises:

6. The method of claim 5, wherein the obtaining positional relationship information between the object to be predicted and each of the intersection exit areas further comprises:

7. The method of claim 6, further comprising, after said treating the intersection exit zone as an intended node in the dynamic bayesian model for each of the intersection exit zones:

the determining the prior probability of the intended node further includes:

8. An intersection object behavior prediction apparatus, comprising:

9. An electronic device, the electronic device comprising:

one or more processors;

a storage means for storing one or more programs;

the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the method of any of claims 1-7.

10. A computer readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the method according to any of claims 1-7.