Vehicle trajectory prediction method and device, computer equipment and storage medium

Abstract

The invention provides a vehicle track prediction method and device, computer equipment and a storage medium, and belongs to the technical field of automatic driving. According to the vehicle trajectory prediction method, a short-time domain trajectory of a vehicle in a second time period is predicted according to motion data of the vehicle in a first time period. According to the historical lane changing data of the road where the vehicle is located, multiple long-time-domain tracks of the vehicle in the third time period are predicted, then according to the short-time-domain tracks, the track of the vehicle is screened out from the multiple long-time-domain tracks, and it is achieved that the track of the vehicle is predicted by combining long and short time domains through a mathematical method; compared with a method for predicting the trajectory through a neural network model, the method has the advantages that the calculation amount consumed by prediction is reduced, and the trajectory prediction cost is reduced.

Description

Technical Field

[0001] The present application relates to the technical field of autonomous driving, and particularly to a vehicle trajectory prediction method, apparatus, computer device, and storage medium. Background Art

[0002] With the development of autonomous driving technology, computer devices such as in-vehicle terminals can obtain the motion data of vehicles around the host vehicle through sensors mounted on the host vehicle, and then predict the trajectories of the surrounding vehicles based on the motion data to assist autonomous driving.

[0003] Currently, a vehicle trajectory prediction method is to process the motion data of surrounding vehicles through a trajectory prediction model to obtain the trajectories of the surrounding vehicles, so as to achieve the prediction of the trajectories of the surrounding vehicles.

[0004] In the application of the above method, a large amount of historical data is used to train the trajectory prediction model, and the model training process and subsequent maintenance of the trajectory prediction model consume a large amount of computing resources, resulting in a high cost of trajectory prediction through the above method. Summary of the Invention

[0005] Embodiments of the present application provide a vehicle trajectory prediction method, apparatus, computer device, and storage medium for reducing the cost of trajectory prediction. The technical solutions are as follows:

[0006] In a first aspect, a vehicle trajectory prediction method is provided, and the method includes:

[0007] According to the motion data of a vehicle in a first time period, through Gaussian process regression, predict the short-time domain trajectory of the vehicle in a second time period. The first time period is any time period before the current moment, the second time period is any time period after the current moment, the motion data includes the lateral displacement and longitudinal displacement corresponding to multiple time points of the vehicle in the first time period, and the short-time domain trajectory includes the lateral displacement and longitudinal displacement corresponding to multiple time points of the vehicle in the second time period;

[0008] According to the short-time domain trajectory of the vehicle and the historical lane-changing data of the road where the vehicle is located, combined with multiple sets of fitting parameters, predict multiple long-time domain trajectories of the vehicle in a third time period. Each long-time domain trajectory corresponds to a set of fitting parameters. The historical lane-changing data includes the historical driving records of multiple vehicles passing on the road. The third time period is any time period after the current moment, the third time period includes the second time period, and the long-time domain trajectory includes the lateral displacement and longitudinal displacement corresponding to multiple time points of the vehicle in the third time period;

[0009] Based on the short-time domain trajectory of the vehicle, screen out the trajectory of the vehicle from multiple long-time domain trajectories of the vehicle.

[0010] The vehicle trajectory prediction method provided by this application predicts the short-term trajectory of a vehicle in a second time period based on the movement data of the vehicle in a first time period. It also predicts multiple long-term trajectories of the vehicle in a third time period according to the historical lane-changing data of the road where the vehicle is located. Then, according to the short-term trajectory, it screens out the trajectory of the vehicle from multiple long-term trajectories, realizing the combination of short-term and long-term predictions of the vehicle's trajectory through mathematical methods. Compared with predicting the trajectory through a neural network model, it reduces the computational amount consumed by the prediction and lowers the cost of trajectory prediction.

[0011] In some embodiments, predicting the short-term trajectory of a vehicle in a second time period based on the movement data of the vehicle in a first time period through Gaussian process regression includes:

[0012] According to the movement data of the vehicle in the first time period, through Gaussian process regression, obtain the prior distribution that the lateral displacement of the vehicle in the first time period follows;

[0013] According to the prior distribution that the lateral displacement of the vehicle in the first time period follows and the longitudinal displacement corresponding to each time point in the second time period, obtain the posterior distribution that the lateral displacement corresponding to each time point in the second time period follows;

[0014] According to the posterior distribution that the lateral displacement corresponding to each time point in the second time period follows, predict the lateral displacements corresponding to multiple time points of the vehicle in the second time period to obtain the short-term trajectory of the vehicle in the second time period.

[0015] In some embodiments, the prior distribution that the lateral displacement of the vehicle in the first time period follows includes a noise variance matrix, and the noise variance matrix is used to reflect random situations.

[0016] In some embodiments, predicting multiple long-term trajectories of a vehicle in a third time period according to the short-term trajectory of the vehicle and the historical lane-changing data of the road where the vehicle is located, in combination with multiple sets of fitting parameters, includes:

[0017] According to the historical lane-changing data of the road where the vehicle is located, predict the lateral position of the vehicle at a first time point, where the first time point is the end time of the third time period, and the lateral position indicates the position of the vehicle in the lane at the first time point;

[0018] For each set of fitting parameters, according to the short-term trajectory of the vehicle, the lateral position of the vehicle at the first time point, and the fitting parameters, predict the long-term trajectory of the vehicle in the third time period.

[0019] In some embodiments, predicting the lateral position of the vehicle at a first time point according to the historical lane-changing data of the road where the vehicle is located includes:

[0020] Obtain the lane-changing probability matrix of a road based on the historical lane-changing data of the road where the vehicle is located. The lane-changing probability matrix includes the probability that a vehicle on the road moves from the first lane on the road to the second lane on the road;

[0021] Obtain the probability that the vehicle is in the second lane at the first time point based on the lane-changing probability matrix of the road and the probability that the vehicle is in the first lane at the second time point, where the second time point is before the third time point;

[0022] Multiply the position of each second lane by the probability that the vehicle is in each second lane at the first time point and sum them up to obtain the lateral position of the vehicle at the first time point.

[0023] In some embodiments, the above fitting parameters include lateral fitting parameters and longitudinal fitting parameters. Predicting the long-time domain trajectory of the vehicle within the third time period based on the short-time domain trajectory of the vehicle, the lateral position of the vehicle at the first time point, and the fitting parameters includes:

[0024] Construct a lateral displacement model based on the short-time domain trajectory of the vehicle, the lateral position of the vehicle at the first time point, and the lateral fitting parameters. The lateral displacement model is used to simulate the lateral displacement of the vehicle and the satisfied constraint conditions;

[0025] Obtain the lateral displacements corresponding to multiple time points within the third time period of the vehicle according to the multiple time points within the third time period and the lateral displacement model;

[0026] Construct a longitudinal displacement model based on the short-time domain trajectory of the vehicle and the longitudinal fitting parameters. The longitudinal displacement model is used to simulate the longitudinal displacement of the vehicle and the satisfied constraint conditions;

[0027] Obtain the longitudinal displacements corresponding to multiple time points within the third time period of the vehicle according to the multiple time points within the third time period and the longitudinal displacement model.

[0028] In some embodiments, the above screening of the trajectory of the vehicle from multiple long-time domain trajectories of the vehicle based on the short-time domain trajectory of the vehicle includes:

[0029] For each long-time domain trajectory of the vehicle, extract the trajectory corresponding to the second time period from the long-time domain trajectory;

[0030] Obtain the distance between the trajectory corresponding to the second time period and the short-time domain trajectory of the vehicle;

[0031] Among the trajectories corresponding to multiple second time periods, determine the trajectory with the minimum distance as the trajectory of the vehicle.

[0032] In a second aspect, a vehicle trajectory prediction device is provided. The device includes:

[0033] A short-term domain prediction module, which is used to predict the short-term domain trajectory of a vehicle in a second time period according to the motion data of the vehicle in a first time period through Gaussian process regression. The first time period is any time period before the current moment, and the second time period is any time period after the current moment. The motion data includes the lateral displacement and longitudinal displacement corresponding to multiple time points of the vehicle in the first time period. The short-term domain trajectory includes the lateral displacement and longitudinal displacement corresponding to multiple time points of the vehicle in the second time period;

[0034] A long-term domain prediction module, which is used to predict multiple long-term domain trajectories of the vehicle in a third time period according to the short-term domain trajectory of the vehicle and the historical lane-changing data of the road where the vehicle is located, in combination with multiple sets of fitting parameters. Each long-term domain trajectory corresponds to a set of fitting parameters. The historical lane-changing data includes the historical driving records of multiple vehicles passing on the road. The third time period is any time period after the current moment, and the third time period includes the second time period. The long-term domain trajectory includes the lateral displacement and longitudinal displacement corresponding to multiple time points of the vehicle in the third time period;

[0035] A screening module, which is used to screen out the trajectory of the vehicle from multiple long-term domain trajectories of the vehicle based on the short-term domain trajectory of the vehicle.

[0036] In some embodiments, the above short-term domain prediction module is used for:

[0037] According to the motion data of the vehicle in the first time period, through Gaussian process regression, obtain the prior distribution followed by the lateral displacement of the vehicle in the first time period;

[0038] According to the prior distribution followed by the lateral displacement of the vehicle in the first time period and the longitudinal displacement corresponding to each time point in the second time period, obtain the posterior distribution followed by the lateral displacement corresponding to each time point of the vehicle in the second time period;

[0039] According to the posterior distribution followed by the lateral displacement corresponding to each time point in the second time period, predict the lateral displacement corresponding to multiple time points of the vehicle in the second time period, and obtain the short-term domain trajectory of the vehicle in the second time period.

[0040] In some embodiments, the prior distribution followed by the lateral displacement of the vehicle in the first time period includes a noise variance matrix, and the noise variance matrix is used to reflect random situations.

[0041] In some embodiments, the above long-term domain prediction module includes:

[0042] A lateral position prediction unit, which is used to predict the lateral position of the vehicle at a first time point according to the historical lane-changing data of the road where the vehicle is located. The first time point is the time point when the third time period ends. The lateral position indicates the position of the vehicle in the lane at the first time point;

[0043] A trajectory prediction unit, configured to, for each set of fitting parameters, predict the long-time domain trajectory of a vehicle within a third time period according to the short-time domain trajectory of the vehicle, the lateral position of the vehicle at a first time point, and the fitting parameters.

[0044] In some embodiments, the above-mentioned lateral position prediction unit includes:

[0045] A matrix acquisition sub-unit, configured to acquire a lane-changing probability matrix of a road according to the historical lane-changing data of the road where the vehicle is located, where the lane-changing probability matrix includes the probability that a vehicle on the road moves from a first lane on the road to a second lane on the road;

[0046] A probability acquisition sub-unit, configured to acquire the probability that the vehicle is located in the second lane at the first time point according to the lane-changing probability matrix of the road and the probability that the vehicle is located in the first lane at a second time point, where the second time point is before the third time point;

[0047] A position acquisition sub-unit, configured to multiply the position of each second lane by the probability that the vehicle is located in each second lane at the first time point and sum the results to obtain the lateral position of the vehicle at the first time point.

[0048] In some embodiments, the above-mentioned fitting parameters include a lateral fitting parameter and a longitudinal fitting parameter, and the above-mentioned trajectory prediction unit is configured to:

[0049] Construct a lateral displacement model according to the short-time domain trajectory of the vehicle, the lateral position of the vehicle at the first time point, and the lateral fitting parameter, where the lateral displacement model is used to simulate the lateral displacement of the vehicle and the satisfied constraint conditions;

[0050] Obtain the lateral displacements corresponding to multiple time points within the third time period of the vehicle according to multiple time points within the third time period and the lateral displacement model;

[0051] Construct a longitudinal displacement model according to the short-time domain trajectory of the vehicle and the longitudinal fitting parameter, where the longitudinal displacement model is used to simulate the longitudinal displacement of the vehicle and the satisfied constraint conditions;

[0052] Obtain the longitudinal displacements corresponding to multiple time points within the third time period of the vehicle according to multiple time points within the third time period and the longitudinal displacement model.

[0053] In some embodiments, the above-mentioned screening module is configured to:

[0054] For each long-time domain trajectory of the vehicle, extract the trajectory corresponding to a second time period from the long-time domain trajectory;

[0055] Obtain the distance between the trajectory corresponding to the second time period and the short-time domain trajectory of the vehicle;

[0056] Among the trajectories corresponding to multiple second time periods, determine the trajectory with the minimum distance as the trajectory of the vehicle.

[0057] In a third aspect, a computer device is provided. The computer device includes a processor and a memory. The memory is used to store at least one segment of computer program, and the at least one segment of computer program is loaded and executed by the processor to implement the operations performed by the vehicle trajectory prediction method provided in the above first aspect or various optional implementation manners of the first aspect.

[0058] In a fourth aspect, a computer-readable storage medium is provided. At least one segment of computer program is stored in the computer-readable storage medium, and the at least one segment of computer program is loaded and executed by a processor to implement the operations performed by the vehicle trajectory prediction method provided in the above first aspect or various optional implementation manners of the first aspect.

[0059] In a fifth aspect, a computer program product or a computer program is provided. The computer program product or the computer program includes computer program code. The computer program code is stored in a computer-readable storage medium. A processor of a computer device reads the computer program code from the computer-readable storage medium, and the processor executes the computer program code, so that the computer device performs the operations performed by the vehicle trajectory prediction method provided in the above first aspect or various optional implementation manners of the first aspect.

[0060] Based on the implementation manners provided in the above aspects of the present application, further combinations can be made to provide more implementation manners. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0062] Figure 1 It is a schematic diagram of the implementation environment of a vehicle trajectory prediction method provided by an embodiment of the present application;

[0063] Figure 2 It is a flowchart of a vehicle trajectory prediction method provided by an embodiment of the present application;

[0064] Figure 3 It is a flowchart of another vehicle trajectory prediction method provided by an embodiment of the present application;

[0065] Figure 4 It is a structural block diagram of a vehicle trajectory prediction device provided by an embodiment of the present application;

[0066] Figure 5It is a schematic structural diagram of a computer device provided by an embodiment of the present application. Detailed implementation manners

[0067] To make the objectives, technical solutions and advantages of the present application clearer, the following will further describe the embodiments of the present application in detail with reference to the accompanying drawings.

[0068] In the present application, terms such as "first" and "second" are used to distinguish identical or similar items with basically the same functions. It should be understood that there is no logical or temporal dependency between "first", "second", and "nth", nor are the quantity and execution order limited.

[0069] In the present application, the term "at least one" means one or more, and the meaning of "a plurality" means two or more.

[0070] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data for analysis, stored data, displayed data, etc.) and signals involved in the present application are all authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions. For example, the data (such as motion data and historical lane change data, etc.) involved in the present application are all obtained under full authorization.

[0071] The vehicle trajectory prediction method provided by the embodiments of the present application can be executed by a computer device. In some embodiments, the computer device is an in-vehicle terminal or a server. First, taking the computer device as a server as an example, the implementation environment of the vehicle trajectory prediction method provided by the embodiments of the present application will be introduced.

[0072] Figure 1 It is a schematic diagram of the implementation environment of a vehicle trajectory prediction method provided by the embodiments of the present application. Refer to Figure 1 , this implementation environment includes an in-vehicle terminal 101 and a server 102. The in-vehicle terminal 101 and the server 102 can be directly or indirectly connected through a wired network or a wireless network, and the present application does not limit this here.

[0073] In some embodiments, an application program supporting vehicle trajectory prediction is installed and run on the in-vehicle terminal 101. The in-vehicle terminal 101 receives the motion data of surrounding vehicles collected by sensors mounted on its own vehicle through the above application program, and sends the motion data of surrounding vehicles to the server 102.

[0074] In some embodiments, the server 102 is an independent physical server, or can also be a server cluster or distributed system composed of multiple physical servers, or can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, Content Delivery Network (CDN), big data, and artificial intelligence platforms. The server 102 is used to provide background services for an application program that supports vehicle trajectory prediction. Schematically, the server 102 receives the motion data of surrounding vehicles sent by the in-vehicle terminal 101, predicts the trajectories of the surrounding vehicles based on the motion data of the surrounding vehicles, and returns the predicted trajectories to the in-vehicle terminal 101.

[0075] In some embodiments, the server 102 undertakes the main computing work, and the in-vehicle terminal 101 undertakes the secondary computing work; or, the server 102 undertakes the secondary computing work, and the in-vehicle terminal 101 undertakes the main computing work; or, the server 102 and the in-vehicle terminal 101 adopt a distributed computing architecture for collaborative computing.

[0076] Secondly, taking the computer device as the in-vehicle terminal as an example, the implementation environment of the vehicle trajectory prediction method provided by the embodiments of the present application is introduced. An application program that supports vehicle trajectory prediction is installed and run on the in-vehicle terminal. The in-vehicle terminal predicts the trajectories of surrounding vehicles based on the motion data of the surrounding vehicles through the above application program.

[0077] Those skilled in the art can know that the number of the above in-vehicle terminals and servers can be more or less. For example, the above in-vehicle terminal can be only one, or the above in-vehicle terminal can be dozens or hundreds, or more. The present application does not limit the number and device types of the in-vehicle terminals, nor the number and device types of the servers.

[0078] In some embodiments, the above-mentioned wireless network or wired network uses standard communication technologies and / or protocols. The network is typically the Internet, but can also be any network, including but not limited to any combination of a local area network (LAN), a metropolitan area network (MAN), a wide area network (WAN), a mobile, wired or wireless network, a private network or a virtual private network. In some embodiments, technologies and / or formats including Hyper Text Mark-up Language (HTML), Extensible Markup Language (XML), etc. are used to represent the data exchanged through the network. In addition, conventional encryption technologies such as Secure Socket Layer (SSL), Transport Layer Security (TLS), Virtual Private Network (VPN), Internet Protocol Security (IPsec), etc. can also be used to encrypt all or some of the links. In other embodiments, custom and / or proprietary data communication technologies can also be used to replace or supplement the above-mentioned data communication technologies.

[0079] Figure 2 is a flowchart of a vehicle trajectory prediction method provided by an embodiment of the present application, as Figure 2 shown, the method includes the following steps:

[0080] 201. The computer device predicts the short-term trajectory of the vehicle in the second time period through Gaussian process regression according to the motion data of the vehicle in the first time period. The first time period is any time period before the current moment, and the second time period is any time period after the current moment. The motion data includes the lateral displacement and longitudinal displacement corresponding to multiple time points of the vehicle in the first time period, and the short-term trajectory includes the lateral displacement and longitudinal displacement corresponding to multiple time points of the vehicle in the second time period.

[0081] Among them, the above vehicle is any vehicle around the vehicle (ego vehicle) served by the computer device. Taking the computer device as an in-vehicle terminal as an example, the above vehicle is the vehicle diagonally in front of the vehicle where the in-vehicle terminal is located. The current moment is any moment during the driving of the ego vehicle. The first time period is the time period before this moment, and the second time period is the time period after this moment. For example, the current moment is 12:00:00, the first time period is 5 minutes before 12:00:00, that is, from 11:55:00 to 12:00:00, and the second time period is 1 second after 12:00:00, that is, 12:00:01. Of course, the durations of the first time period and the second time period can be of any length, and the embodiments of the present application do not limit this. The motion data is also called historical trajectory data or historical time series data, such as the training set TS in the following formula (1). The first time period includes multiple time points, and each time point corresponds to a lateral displacement and a longitudinal displacement of the above vehicle. The lateral displacement and the longitudinal displacement are related to the distance between the position of the above vehicle at the previous time point and the position at the current time point. The lateral displacement is the component of this distance in the direction perpendicular to the road, and the longitudinal displacement is the component of this distance in the road direction. The second time period includes multiple time points. For example, the duration of the second time period is 1 second, and every 0.1 second within this 1 second is a time point. The short-time domain trajectory is the same as this motion data, and the embodiments of the present application will not elaborate on this here.

[0082] In some embodiments, the above motion data is represented by the following formula (1):

[0083] TS = [T, X, Y] = [(t 1 , t 2 , …, t n ), (x 1 , x 2 , …, x n ), (y 1 , y 2 , …, y n )] (1)

[0084] Among them, TS is used to represent the motion data of the vehicle, T is used to represent the time point sequence, X is used to represent the longitudinal displacement sequence, Y is used to represent the lateral displacement sequence, t n is used to represent the moment corresponding to the nth time point, x n is used to represent the longitudinal displacement corresponding to the nth time point, y n is used to represent the lateral displacement corresponding to the nth time point, and n is used to represent the number of multiple time points included in the first time period.

[0085] The Gaussian process refers to a set of random variables, and any finite number of random variables in this set follow a joint Gaussian distribution. The Gaussian process is represented by the following formula (2):

[0086]

[0087] Among them, x and x' are used to represent any random variables, m(x) is used to represent the mean function, and k(x, x') is used to represent the covariance function.

[0088] In the embodiments of the present application, it is assumed that the trajectory of the vehicle follows a Gaussian distribution. The computer device predicts the short-term trajectory of the vehicle in the second time period through Gaussian process regression based on the lateral displacement and longitudinal displacement corresponding to each time point in the movement data of the vehicle in the first time period.

[0089] 202. The computer device predicts multiple long-term trajectories of the vehicle in the third time period according to the short-term trajectory of the vehicle and the historical lane-changing data of the road where the vehicle is located, in combination with multiple sets of fitting parameters. Each long-term trajectory corresponds to a set of fitting parameters. The historical lane-changing data includes the historical driving records of multiple vehicles passing on the road. The third time period is any time period after the current moment. The third time period includes the second time period. The long-term trajectory includes the lateral displacement and longitudinal displacement corresponding to multiple time points of the vehicle in the third time period.

[0090] Among them, the historical lane-changing data includes the historical lane-changing data corresponding to multiple time points in the historical time period. The historical lane-changing data corresponding to each time point includes the driving records of multiple vehicles on the road at this time point. The driving record includes the lateral speed and lateral position of the vehicle at this time point. The lateral speed indicates the speed of the vehicle, and the lateral position indicates the position of the lane where the vehicle is located, which can be represented by the center line of the lane. The historical lane-changing data of the road reflects the lane-changing behavior of the vehicle and indicates the number of lanes crossed by the vehicle when changing lanes in the historical time period. Taking the road including 3 lanes and the vehicle changing from lane i to lane j as an example, the lane-changing behavior of the vehicle is as shown in the following formula (3):

[0091]

[0092] The above-mentioned fitting parameters are parameters set manually according to historical experience and are used to calculate the lateral displacement and longitudinal displacement corresponding to this time point according to the time point in the third time period. The third time period includes multiple time points in the second time period. The long-term trajectory is the same as the above-mentioned short-term trajectory, and the embodiments of the present application will not elaborate here.

[0093] In the embodiments of the present application, for each set of fitting parameters in the multiple sets of fitting parameters, the computer device predicts the lateral displacement and longitudinal displacement corresponding to each time point of the vehicle in the third time period according to the short-term trajectory of the vehicle, the historical lane-changing data of the road where the vehicle is located, and this set of fitting parameters, and obtains the long-term trajectory of the vehicle in the third time period.

[0094] 203. The computer device filters out the vehicle's trajectory from multiple long-time domain trajectories of the vehicle based on the short-time domain trajectory of the vehicle.

[0095] In the embodiment of the present application, the computer device filters out the trajectory with the minimum distance from the short-time domain trajectory from multiple long-time domain trajectories of the vehicle, and determines this trajectory as the vehicle's trajectory.

[0096] The vehicle trajectory prediction method provided by the embodiment of the present application predicts the short-time domain trajectory of the vehicle in the second time period according to the motion data of the vehicle in the first time period. And according to the historical lane-changing data of the road where the vehicle is located, it predicts multiple long-time domain trajectories of the vehicle in the third time period. Then, according to the short-time domain trajectory, it filters out the vehicle's trajectory from the multiple long-time domain trajectories, realizing the prediction of the vehicle's trajectory by combining long and short time domains through a mathematical method. Compared with predicting the trajectory through a neural network model, it reduces the computational amount consumed by the prediction and reduces the cost of trajectory prediction.

[0097] Figure 3 is a flowchart of a vehicle trajectory prediction method provided by the embodiment of the present application. As Figure 3 shown, the method includes the following steps.

[0098] 301. The computer device obtains the prior distribution that the lateral displacement of the vehicle follows in the first time period through Gaussian process regression according to the motion data of the vehicle in the first time period. The first time period is any time period before the current moment, the second time period is any time period after the current moment, and the motion data includes the lateral displacement and longitudinal displacement corresponding to multiple time points of the vehicle in the first time period.

[0099] Among them, the lateral displacement of the vehicle in the first time period is also called the observed value, and the prior distribution that the lateral displacement of the vehicle in the first time period follows is expressed as the following formula (4):

[0100] y~N(0,K(Z,Z))(4)

[0101] Among them, formula (4) indicates that the lateral displacement of the vehicle follows a Gaussian distribution with a mean of 0 and a variance of K(Z,Z). K(Z,Z) is an N×N order symmetric positive definite covariance matrix, which is expressed as the following formula (5):

[0102]

[0103] Among them, z is used to represent the longitudinal displacement and lateral displacement corresponding to any time point. For example, z 1 is used to represent the longitudinal displacement and lateral displacement corresponding to the first time point, that is, z 1 =(x 1 ,y 1 ), kij = k(z i , z j ) is used to represent the correlation between z i and z j . n is used to represent the number of time points in the first time period. The covariance function k(z i , z j ) is calculated by the following formula (6):

[0104]

[0105] where δ(z i , z j ) is used to represent the Kronecker delta function, and σ f , v n and l are hyperparameters in the covariance function. The hyperparameters are represented as the hyperparameter set θ = [σ f , l, σ n . The hyperparameter set θ is obtained by the maximum likelihood method shown in the following calculation process:

[0106] According to the Bayesian principle, the following formula (7) is constructed:

[0107]

[0108] When p(θ|y, z) is the largest, the estimate of θ is the maximum a posteriori estimate. Assume that the prior distribution is a uniform distribution, that is, θ can take any value, then:

[0109]

[0110] where, is used to represent the ideal hyperparameter set, and p(y|z, θ) is used to represent the marginal likelihood. The likelihood function is the following formula (9)

[0111]

[0112] Derive the hyperparameter set θ in the above formula (9), and use the conjugate gradient method to minimize the partial derivative obtained by the derivation to obtain the ideal hyperparameter set

[0113] In the embodiments of the present application, the computer device obtains the hyperparameters in the covariance function according to the motion data of the vehicle in the first time period. The computer device obtains the covariance between the displacements corresponding to each time point in the first time period through the above formula (6) according to the hyperparameters and the motion data of the vehicle in the first time period, and obtains the variance in the prior distribution that the lateral displacement of the vehicle obeys as shown in formula (5). According to the variance, the prior distribution that the lateral displacement of the vehicle obeys in the first time period is obtained.

[0114] In some embodiments, the prior distribution followed by the lateral displacement of the vehicle within the first time period includes a noise variance matrix, and the noise variance matrix is used to reflect random situations. The prior distribution is shown in the following formula (10):

[0115]

[0116] Wherein, is used to represent the noise variance matrix, which is preset, and I N is used to represent the N×N identity matrix. By setting the noise variance matrix in the prior distribution, the prior distribution can be made more in line with the actual situation, thereby improving the accuracy of the short-time domain trajectory obtained according to the prior distribution.

[0117] 302. The computer device obtains the posterior distribution followed by the lateral displacement of the vehicle at each time point within the second time period according to the prior distribution followed by the lateral displacement of the vehicle within the first time period and the longitudinal displacement corresponding to each time point within the second time period.

[0118] Wherein, the second time period includes multiple time points, and the lateral displacement corresponding to each time point follows a posterior distribution respectively. The longitudinal displacement corresponding to each time point within the second time period is a known value, and the longitudinal displacement corresponding to each time point can be preset for this time point, or can be calculated by the computer device according to the speed of the vehicle and this time point. The embodiments of the present application do not limit this. The lateral displacement corresponding to each time point within the second time period is also called the predicted value.

[0119] In the embodiments of the present application, for each time point within the second time period, the computer device constructs the joint distribution of the lateral displacement within the first time period and the lateral displacement within the second time period as shown in the following formula (11) according to the prior distribution followed by the lateral displacement of the vehicle within the first time period, the longitudinal displacement and the lateral displacement corresponding to this time point within the second time period:

[0120]

[0121] Wherein, z i =(x i , y i ), y i is used to represent the lateral displacement corresponding to the i-th time point within the second time period, x i is used to represent the longitudinal displacement corresponding to the i-th time point within the second time period, and this lateral displacement is an unknown value. K(z i , Z)=K(Z, z i ) T , and K(z i , Z) is represented by the following formula (12):

[0122] K(z i ,Z) = [k(z i ,z 1 ) k(z i ,z 2 )… k(z i ,z n )] (12)

[0123] Among them, n is used to represent the number of time points in the first time period. K(z i ,Z) is the covariance matrix between the test matrix point z i and the training set Z. k(z i ,z i ) is the covariance of the test matrix point z i itself.

[0124] After that, according to the above joint distribution, the computer device obtains the posterior distribution that the lateral displacement of the vehicle follows at this time point through the following formulas (13) and (14), and this posterior distribution is shown in the following formula (15).

[0125]

[0126] Among them, μ(y i ) is used to represent the mean of y i .

[0127] Among them, cov(y i ) is used to represent the mean of y i .

[0128] y i |(X,Y,x i ) ~ N(μ(y i ),cov(y i ))(15)

[0129] Among them, y i |(X,Y,x i ) is used to represent the distribution of y i under the given conditions of X, Y, and x i .

[0130] 303. The computer device predicts the lateral displacements corresponding to multiple time points of the vehicle in the second time period according to the posterior distribution that the lateral displacement corresponding to each time point in the second time period follows, and obtains the short-time domain trajectory of the vehicle in the second time period. The short-time domain trajectory includes the lateral displacements and longitudinal displacements corresponding to multiple time points of the vehicle in the second time period.

[0131] In the embodiments of the present application, for each time point within the second time period, the computer device predicts the lateral displacement corresponding to this time point of the vehicle within the second time period by using methods such as maximum likelihood estimation according to the posterior distribution followed by the lateral displacement corresponding to this time point. By predicting the lateral displacements corresponding to multiple time points within the second time period, the short-time domain trajectory of the vehicle within the second time period is obtained. This short-time domain trajectory can be represented by Tr LS =[Tr LS (1), Tr LS (2), …, Tr LS (n)] = [(x 1 , y 1 ), (x 2 , y 2 ), …, (x n , y n )], where n is the number of time points within the second time period. Taking the duration of the second time period as 1 second and each 0.1 second within the second time period being a time point as an example, the computer device obtains the short-time domain trajectory Tr LS =[(x 1 , y 1 ), (x 2 , y 2 ), …, (x 10 , y 10 )] by rolling iteration and looping through the above process 10 times.

[0132] The above steps 301 to 303 are a possible implementation manner for the computer device to predict the short-time domain trajectory of the vehicle within the second time period based on the motion data of the vehicle within the first time period through Gaussian process regression. This possible implementation manner constructs a Gaussian process regression model as shown in the following formula (16):

[0133] y i =f(x i ) + ε (16)

[0134] where ε is Gaussian noise with a preset mean of 0 and a variance of . This formula (16) constructs the expression relationship between the input x i and the output y i . By using the squared exponential covariance shown in formula (6) as the kernel function, mapping the input to a high-dimensional space, and then adding a noise factor, the above Gaussian process regression model is obtained.

[0135] This process combines historical data and time series information to classify and predict the behaviors of surrounding vehicles, and can fully utilize the distribution state of prior data to achieve accurate prediction of data changes in the future short time, and has extremely high accuracy for short-time prediction.

[0136] In the above process, the computer device establishes a trajectory prediction model within a short time based on Gaussian process regression. During the application of Gaussian process regression, it is assumed that the lateral displacement of the vehicle within the first time period follows the prior probability of a Gaussian process, and the optimal hyperparameters in the Gaussian process regression model are obtained through the maximum likelihood method. The computer device inputs the longitudinal displacement corresponding to each time point within the second time period into the Gaussian process regression model to apply the above Gaussian regression model and perform the operations shown in steps 302 to 303 above, that is, combining Bayesian theory to obtain the corresponding posterior probability, and predicting the lateral displacement based on this posterior probability.

[0137] 304. The computer device obtains a lane-changing probability matrix of the road according to the historical lane-changing data of the road where the vehicle is located. The lane-changing probability matrix includes the probability that a vehicle on the road moves from the first lane on the road to the second lane on the road. The historical lane-changing data includes the historical driving records of multiple vehicles passing on the road.

[0138] Among them, the lane-changing probability matrix (state transition matrix) corresponding to any time point within the historical time period is related to the lateral speeds of multiple vehicles on the road at this time point. Each lateral speed corresponds to a lane-changing probability matrix. The following formula (17) is an example of a lane-changing probability matrix:

[0139]

[0140] Among them, T(v) is used to represent the lane-changing probability matrix corresponding to the lateral speed v at time t, and a ij is used to represent the probability that the vehicle switches from lane i to lane j, and this probability is expressed as the following formula (18):

[0141] a ij = P(Lane t+1 = j | Lane t = i) (18)

[0142] Among them, Lane t+1 is used to represent the lane where the vehicle is located at time t + 1, and Lane t is used to represent the lane where the vehicle is located at time t.

[0143] In the embodiments of the present application, the computer device obtains the historical driving records of vehicles with the same lateral speed as the current lateral speed of the vehicle from the historical lane-changing data corresponding to the first historical time point in the historical time period, and obtains the historical driving records of these vehicles from the historical lane-changing data corresponding to the second historical time point in the historical time period. The second historical time point is the next time point after the first historical time point. The computer device obtains the lane-changing behavior of these vehicles according to the lateral positions of these vehicles in the historical driving records corresponding to the first historical time point and the lateral positions of these vehicles in the historical driving records corresponding to the second historical time point. The computer device statistically analyzes the lane-changing behavior of these vehicles to obtain a lane-changing probability matrix corresponding to the current lateral speed of the vehicle.

[0144] In some embodiments, taking the historical time period including the t time point and the t + 1 time point as an example, the t + 1 time point is the next time point after the t time point. The computer device obtains the lane-changing probability matrix corresponding to the t + 1 time point according to the lane-changing probability matrix corresponding to the t time point through the following formula (19).

[0145]

[0146] Wherein, T t+1 (v) is used to represent the lane-changing probability matrix corresponding to the lateral speed v at the t + 1 time point, and T t (v) is used to represent the lane-changing probability matrix corresponding to the lateral speed v at the t time point. is used to represent the Gaussian distribution function corresponding to the lane-changing behavior. Taking the lane-changing behavior including not changing lanes, changing lanes once, and changing lanes twice as an example, this Gaussian distribution function can be obtained through the following formula (20):

[0147]

[0148] Wherein, m is used to represent the lane-changing behavior, 0 represents not changing lanes, 1 represents changing lanes once, and 2 represents changing lanes twice. μ m is used to represent the expectation of the Gaussian distribution function of the lateral speed corresponding to the corresponding lane-changing behavior. is used to represent the variance of the Gaussian distribution function of the lateral speed corresponding to the corresponding lane-changing behavior.

[0149] After obtaining the lane-changing probability matrix corresponding to the t + 1 time point through the above formula (20), the computer device normalizes the probabilities in the lane-changing probability matrix through the following formula (21) to obtain the final lane-changing probability matrix corresponding to the t + 1 time point.

[0150]

[0151] Wherein, is used to represent the probability before normalization. is used to represent the probability after normalization, and n is used to represent the number of lanes.

[0152] It should be noted that the above road corresponds to multiple lane-changing behaviors, and each lane-changing behavior corresponds to a lane-changing probability matrix. In the process of obtaining the lane-changing probability matrix corresponding to the time point t + 1 based on the lane-changing probability matrix corresponding to the time point t, for each lane-changing behavior, the lane-changing probability matrix of the lane-changing behavior corresponding to the time point t is added to the Gaussian distribution function of the lane-changing behavior to obtain the lane-changing probability matrix of the lane-changing behavior corresponding to the time point t + 1. The initial lane-changing probability matrix can be obtained by statistically analyzing the lane-changing behaviors of multiple vehicles passing on the road within a preset time period. For example, for each vehicle passing on the road within the preset time period, record the lane where the vehicle is located and the lane-changing behavior within the preset time period to obtain a large amount of vehicle data. According to this vehicle data, count the number of times of transferring from lane i to lane j, denoted as N i，j , calculate the total number of times the vehicles in lane i transfer to other lanes Then divide the number of times the vehicle transfers from lane i to lane j by the total number of times the vehicles in lane i transfer to other lanes to obtain the lane-changing probability T in the lane-changing probability matrix i，j = N i，j / N i .

[0153] In some embodiments, the computer device constructs a lane-changing behavior multi-classification model as shown in the following formula (22) according to the above formulas (17) and (18):

[0154]

[0155] where v i is used to represent the lateral speed of the vehicle at the i-th time point within the second time period, and MC θ (v i ) is used to represent the lane-changing behavior multi-classification model with the parameter θ corresponding to v i . p(Lane i = n|v i ; θ) is used to represent the sub-model corresponding to lane n, which is used to output the probability that the vehicle changes lanes to lane n. r is a value related to the parameter θ, and the parameter θ is obtained through training. Through the above formula (22), the probabilities of the vehicle changing lanes from the current lane to other lanes can be obtained. is the unnormalized weight of each lane. These weights are usually associated with the state of the vehicle (such as lateral speed) and the eigenvalue related to the lane. The exponential function amplifies the influence of the high-weight categories, weakens the influence of the low-weight categories, and enhances the classification ability of the model.

[0156] In some embodiments, the computer device can use the Gaussian distribution function based on lateral velocity shown in the following formula (23) or the Gaussian cumulative function based on lateral velocity shown in (24) to replace the above-mentioned

[0157]

[0158] where μ and σ are the mean and standard deviation of the lateral velocity, respectively.

[0159] The above process simplifies the recognition of the driving intention of surrounding vehicles into a problem of selecting the target lane, introduces the Softmax regression strategy for classifying vehicle lane-changing behaviors, sets the state transition matrix based on the lateral velocity, and constructs a multi-classification model for lane-changing behaviors.

[0160] In some embodiments, the above historical lane-changing data extracts the data of non-lane-changing, one-time lane-changing, and two-time lane-changing of vehicles from the US-101 section in the NGSIM database. According to this data, the Gaussian distribution function corresponding to different lane-changing behaviors is statistically obtained, and then the lane-changing probability matrix is updated and calculated according to the Gaussian distribution function. The embodiments of the present application do not limit this.

[0161] 305. The computer device obtains the probability that the vehicle is in the second lane at the first time point according to the lane-changing probability matrix of the road and the probability that the vehicle is in the first lane at the second time point. The second time point is before the first time point, the first time point is the end time point of the third time period, the third time period is any time period after the current moment, and the third time period includes the second time period.

[0162] where both the first lane and the second lane are any lane on the road, and the first lane and the second lane are the same lane or different lanes. The embodiments of the present application do not limit this.

[0163] In the embodiments of the present application, the computer device multiplies the probability that the vehicle switches from the first lane to the second lane in the lane-changing probability matrix of the road by the probability that the vehicle is in the first lane at the second time point to obtain the probability that the vehicle is in the second lane at the first time point.

[0164] In some embodiments, when the computer device obtains the lane-changing probability matrix at the t+1 time point according to the lane-changing probability matrix at the t time point, the computer device uses the lane-changing probability matrix at the t+1 time point to obtain the probability that the vehicle is in the second lane at the first time point, and uses the lane-changing probability matrix at the t time point to obtain the probability that the vehicle is in the second lane at the second time point. The computer device uses the following formula (25) to obtain the probability that the vehicle is in the second lane at the first time point by using the lane-changing probability matrix at the t+1 time point.

[0165]

[0166] Among them, It is used to represent the probability that the vehicle is in lane j at the first time point obtained by obtaining the lane-changing probability matrix at time point t + 1 based on the lane-changing probability matrix at time point t. a ij (v) It is used to represent the probability that the vehicle switches from lane i to lane j in the lane-changing probability matrix at time point t + 1 corresponding to the lateral velocity v. It is used to represent the probability that the vehicle is in lane i at the second time point obtained by obtaining the lane-changing probability matrix at time point t. The computer device performs multiple operations using the above formula (25) to obtain the probability sequence of the vehicle changing to lane j via the lane-changing probability matrix.

[0167] In some embodiments, when there are multiple lane-changing probability matrices corresponding to multiple lane-changing behaviors for the lateral velocity v, the computer device obtains the maximum probability that the vehicle switches from lane i to lane j from the multiple lane-changing probability matrices, and performs the operation shown in the above formula (25) according to this probability.

[0168] 306. The computer device multiplies the position of each second lane by the probability that the vehicle is in each second lane at the first time point and sums them up to obtain the lateral position of the vehicle at the first time point, and the lateral position indicates the position of the lane where the vehicle is located at the first time point.

[0169] Among them, the lateral position is represented as the position of the center line of the lane on the road, etc., and the embodiments of the present application do not limit this.

[0170] In some embodiments, when the computer device obtains the lane-changing probability matrix at time point t + 1 according to the lane-changing probability matrix at time point t, the computer device uses the following formula (26) to obtain the lateral position of the vehicle at the first time point.

[0171]

[0172] Among them, y t+1 It is used to represent the lateral position of the vehicle at the first time point, t + 1 is used to represent that this lateral position is obtained according to the lane-changing probability matrix at time point t + 1, and n is used to represent the number of lanes on the road. It is used to represent the probability that the vehicle is in lane j at the first time point. It is used to represent the lateral position of lane j.

[0173] The above steps 304 to 306 are a possible implementation for the computer device to predict the lateral position of the vehicle at the first time point based on the historical lane-changing data of the road where the vehicle is located. In this possible implementation, the process of predicting the lateral position of the vehicle is a Markov process. In this process, at any time point, the state transition relationship between different lanes of the vehicle is a Markov chain, and the state sequence of this Markov chain is shown in the following formula (27):

[0174]

[0175] where S t is used to represent the state sequence at time t, is used to represent the lateral speed of the vehicle on the nth lane at time t, is used to represent the lateral position of the vehicle on the nth lane at time t, is used to represent the probability of the vehicle on lane i at time t.

[0176] The dynamic changes in the lateral position of the vehicle (including lateral speed and lane-changing probability, etc.) are modeled through the above formula (27) to predict the possible target lateral position of the vehicle at the next moment. The process of predicting the target lateral position of the vehicle is a dynamic process, which not only depends on the current lateral position and speed, but is also affected by the historical state. The state sequence of the Markov chain can capture the transfer relationship between different lanes of the vehicle, as well as the position and speed information of the vehicle on each lane, providing a complete dynamic description of the vehicle's behavior. By calculating and updating the lane-changing probability matrix, the above state sequence can predict the target lane where the vehicle is most likely to be at the next moment and the specific position (next state) on that lane. Next state calculation: current state × lane-changing probability matrix.

[0177] The above process introduces the Softmax strategy, constructs a multi-classification model for lane-changing behavior, extracts actual lane-changing trajectory data to construct a lane-changing probability matrix, and updates and calculates the lane-changing probability matrix in each recognition period, improving the accuracy of adapting to the transition relationship between different lane-changing behaviors and being able to accurately predict the lateral position of the vehicle.

[0178] It should be noted that the above steps 304 to 306 can be executed before steps 301 to 303 or executed simultaneously with the above steps 301 to 303, and the embodiments of the present application do not make any limitations in this regard.

[0179] 307. For each set of fitting parameters, the computer device predicts the long-time domain trajectory of the vehicle within the third time period according to the short-time domain trajectory of the vehicle, the lateral position of the vehicle at the first time point, and the fitting parameters. Each set of fitting parameters includes a lateral fitting parameter and a longitudinal fitting parameter, and the long-time domain trajectory includes the lateral displacement and longitudinal displacement corresponding to multiple time points within the third time period of the vehicle.

[0180] In the embodiments of the present application, let the first time point (initial time) of the second time period be t 0 , and the last time point of the second time period be t 1 , and the last time point of the third time period be t e . When the duration of the second time period is 1 second, t 1 ∈ [0, 1], t e ∈ [1, t max , and t max is used to represent the duration of the third time period, also known as the maximum lane-changing time.

[0181] For each set of fitting parameters, in the process of obtaining the lateral displacement according to the lateral fitting parameters, the computer device constructs the lateral displacement models shown in the following formulas (28) and (29) based on the short-time domain trajectory of the vehicle, the lateral position of the vehicle at the first time point, and the lateral fitting parameters. The lateral displacement models are used to simulate the lateral displacement of the vehicle and the satisfied constraint conditions:

[0182] y(t) = a 0 + a 1 t + a 2 t 2 + a 3 t 3 + a 4 t 4 + a 5 t 5 (28)

[0183]

[0184] Among them, y(t) is used to represent the lateral displacement corresponding to the vehicle at the time point t, and a 0 to a 5 are used to represent the lateral fitting parameters, y 0 is used to represent the lateral displacement corresponding to the first time point of the second time period in the short-time domain trajectory, is used to represent the speed at the time point t 0 , is used to represent the acceleration of the vehicle at the time point t 0 , y 1 is used to represent the lateral displacement corresponding to the last time point of the second time period in the short-time domain trajectory, y e is used to represent the lateral position obtained through the above step 306, is used to represent the speed at the time point t e .

[0185] Based on multiple time points within the third time period and the lateral displacement model, the computer device obtains the lateral displacements corresponding to multiple time points of the vehicle within the third time period. Correspondingly, for each time point within the third time period, the computer device obtains the lateral displacement corresponding to that time point through the above-mentioned lateral displacement model. During the obtaining process, if the data obtained during the obtaining process meets the constraint conditions shown in formula (29), the obtained lateral displacement is determined as the lateral displacement corresponding to that time point under the lateral fitting parameter. If the data obtained during the obtaining process does not meet the constraint conditions shown in formula (29), the lateral fitting parameter is updated, and based on the updated lateral fitting parameter, the lateral displacement is obtained.

[0186] During the process of obtaining the lateral displacement according to the lateral fitting parameter, the computer device constructs the longitudinal displacement models shown in the following formulas (30) and (31) based on the short-time domain trajectory of the vehicle and the longitudinal fitting parameter. The longitudinal displacement model is used to simulate the longitudinal displacement of the vehicle and the satisfied constraint conditions:

[0187] x(t) = b 0 + b 1 t + b 2 t 2 + b 3 t 3 + b 4 t 4 + b 5 t 5 (30)

[0188]

[0189] Wherein, x(t) is used to represent the longitudinal displacement of the vehicle corresponding to the time point t, b 0 to t 5 are used to represent the lateral fitting parameters, x 0 is used to represent the longitudinal displacement corresponding to the first time point of the second time period in the short-time domain trajectory, is used to represent the speed at the time point t 0 a is used to represent the acceleration of the vehicle at the time point t 0 is used to represent the speed of the vehicle at the time point t 0 a is used to represent the acceleration of the vehicle at the time point t 1 is used to represent the lateral displacement corresponding to the last time point of the second time period in the short-time domain trajectory, is used to represent the speed of the vehicle at the time point t e a is used to represent the acceleration of the vehicle at the time point t is used to represent the speed of the vehicle at the time point t e a is used to represent the acceleration of the vehicle at the time point t.

[0190] Based on multiple time points within the third time period and the longitudinal displacement model, the computer device obtains the longitudinal displacements corresponding to multiple time points of the vehicle within the third time period. Correspondingly, for each time point within the third time period, the computer device obtains the longitudinal displacement corresponding to this time point through the above longitudinal displacement model. During the obtaining process, if the data obtained during the obtaining process meets the constraint conditions shown in formula (31), the obtained longitudinal displacement is determined as the longitudinal displacement corresponding to this time point under this longitudinal fitting parameter. If the data obtained during the obtaining process does not meet the constraint conditions shown in formula (31), the longitudinal fitting parameter is updated, and based on the updated longitudinal fitting parameter, the longitudinal displacement is obtained.

[0191] The long-time-domain trajectory obtained through the above process can adopt Tr L =[Tr L (1), Tr L (2), …, Tr L (n)] = [(x 1 , y 1 ), (x 2 , y 2 ), …, (x n , y n )] to represent. This trajectory is also called a candidate trajectory, and n is the number of time points within the third time period.

[0192] It should be noted that both the above lateral displacement model and longitudinal displacement model are fitted according to the fifth-degree polynomial curve, and relatively accurate lateral displacement and longitudinal displacement can be obtained. Of course, the computer device can also adopt other polynomial curves for fitting, and the embodiments of the present application do not limit this.

[0193] The above step 307 is a possible implementation manner for the computer device to predict the long-time-domain trajectory of the vehicle within the third time period based on the short-time-domain trajectory of the vehicle, the lateral position of the vehicle at the first time point, and the fitting parameter. The above steps 304 to 307 are a possible implementation manner for the computer device to predict multiple long-time-domain trajectories of the vehicle within the third time period based on the short-time-domain trajectory of the vehicle and the historical lane-changing data of the road where the vehicle is located, in combination with multiple sets of fitting parameters. Each long-time-domain trajectory corresponds to a set of fitting parameters. The above method combines the lateral position and the short-time-domain trajectory, and combines multiple sets of fitting parameters to obtain multiple long-time-domain trajectories, expanding the prediction range of the trajectory and being able to improve the accuracy of trajectory prediction.

[0194] 308. For each long-time-domain trajectory of the vehicle, the computer device extracts the trajectory corresponding to the second time period from the long-time-domain trajectory and obtains the distance between the trajectory corresponding to the second time period and the short-time-domain trajectory of the vehicle.

[0195] In an embodiment of the present application, for each long-time domain trajectory of a vehicle, the computer device extracts the lateral displacement and longitudinal displacement corresponding to each time point within the second time period from the long-time domain trajectory according to a plurality of time points included in the second time period, and obtains the trajectory corresponding to the second time period in the long-time domain trajectory. The computer device extracts each long-time domain trajectory to obtain trajectories corresponding to a plurality of second time periods. For the trajectory corresponding to each second time period, the computer device calculates the distance between the trajectory corresponding to the second time period and the short-time domain trajectory. The distance may be an Euclidean distance or the like, and the embodiment of the present application does not limit this.

[0196] 309. Among the trajectories corresponding to a plurality of second time periods, the computer device determines the trajectory with the minimum distance as the trajectory of the vehicle.

[0197] In an embodiment of the present application, the computer device determines the trajectory with the minimum distance from the trajectories corresponding to a plurality of second time periods through the following formula (32), and determines this trajectory as the trajectory of the vehicle:

[0198]

[0199] where Tr is used to represent the trajectory of the vehicle, m is used to represent the number of groups of fitting parameters, n is used to represent the number of time points within the second time period, Tr L (j)(t) is used to represent the lateral displacement and longitudinal displacement corresponding to the t-th time point under the j-th group of fitting parameters, Tr LS (t) is used to represent the lateral displacement and longitudinal displacement corresponding to the t-th time point in the short-time domain trajectory.

[0200] The above steps 308 to 309 are a possible implementation manner for the computer device to screen out the trajectory of the vehicle from multiple long-time domain trajectories of the vehicle based on the short-time domain trajectory of the vehicle. This possible implementation manner realizes screening out the trajectory closest to the short-time domain trajectory from multiple long-time domain trajectories by comparing the trajectory extracted from the long-time domain trajectory with the short-time domain trajectory, that is, the trajectory with the lowest average deviation degree relative to the short-time domain trajectory, and taking this trajectory as the trajectory of the vehicle. The above process has a relatively high prediction trajectory accuracy in a short time, effectively represents the driving trajectory of the actual vehicle in a long time, can obtain accurate prediction trajectories at different moments, can accurately and effectively predict the movement trajectories of surrounding vehicles, and improves the accuracy of the predicted trajectory.

[0201] The vehicle trajectory prediction method provided by the embodiment of the present application predicts the short-time domain trajectory of a vehicle in a second time period according to the motion data of the vehicle in a first time period. Further, according to the historical lane-changing data of the road where the vehicle is located, multiple long-time domain trajectories of the vehicle in a third time period are predicted. Then, according to the short-time domain trajectory, the trajectory of the vehicle is screened out from the multiple long-time domain trajectories, realizing the combination of short and long time domain predictions of the vehicle trajectory by a mathematical method. Compared with predicting the trajectory through a neural network model, the computational amount consumed by the prediction is reduced, and the cost of performing trajectory prediction is reduced.

[0202] Figure 4 It is a structural block diagram of a vehicle trajectory prediction device provided by an embodiment of the present application. This device is used to execute the steps when the above vehicle trajectory prediction method is executed. Refer to Figure 4 This vehicle trajectory prediction device includes:

[0203] The short-time domain prediction module 401 is used to predict the short-time domain trajectory of the vehicle in a second time period through Gaussian process regression according to the motion data of the vehicle in a first time period. The first time period is any time period before the current moment, the second time period is any time period after the current moment, the motion data includes the lateral displacement and longitudinal displacement corresponding to multiple time points of the vehicle in the first time period, and the short-time domain trajectory includes the lateral displacement and longitudinal displacement corresponding to multiple time points of the vehicle in the second time period;

[0204] The long-time domain prediction module 402 is used to predict multiple long-time domain trajectories of the vehicle in a third time period according to the short-time domain trajectory of the vehicle and the historical lane-changing data of the road where the vehicle is located, in combination with multiple sets of fitting parameters. Each long-time domain trajectory corresponds to a set of fitting parameters. The historical lane-changing data includes the historical driving records of multiple vehicles passing on the road. The third time period is any time period after the current moment, and the third time period includes the second time period. The long-time domain trajectory includes the lateral displacement and longitudinal displacement corresponding to multiple time points of the vehicle in the third time period;

[0205] The screening module 403 is used to screen out the trajectory of the vehicle from the multiple long-time domain trajectories of the vehicle based on the short-time domain trajectory of the vehicle.

[0206] In some embodiments, the above short-time domain prediction module 401 is used to:

[0207] Obtain the prior distribution followed by the lateral displacement of the vehicle in the first time period through Gaussian process regression according to the motion data of the vehicle in the first time period;

[0208] Obtain the posterior distribution followed by the lateral displacement of the vehicle at each time point in the second time period according to the prior distribution followed by the lateral displacement of the vehicle in the first time period and the longitudinal displacement corresponding to each time point in the second time period;

[0209] Predict the lateral displacements corresponding to multiple time points of the vehicle within the second time period according to the posterior distribution followed by the lateral displacements corresponding to each time point within the second time period, so as to obtain the short-time domain trajectory of the vehicle within the second time period.

[0210] In some embodiments, the prior distribution followed by the lateral displacement of the vehicle within the first time period includes a noise variance matrix, and the noise variance matrix is used to reflect random situations.

[0211] In some embodiments, the long-time domain prediction module 402 includes:

[0212] A lateral position prediction unit, configured to predict the lateral position of the vehicle at a first time point according to the historical lane-changing data of the road where the vehicle is located. The first time point is the end time point of the third time period, and the lateral position indicates the position of the vehicle in the lane at the first time point;

[0213] A trajectory prediction unit, configured to, for each set of fitting parameters, predict the long-time domain trajectory of the vehicle within the third time period according to the short-time domain trajectory of the vehicle, the lateral position of the vehicle at the first time point, and the fitting parameters.

[0214] In some embodiments, the lateral position prediction unit includes:

[0215] A matrix acquisition sub-unit, configured to acquire a lane-changing probability matrix of the road according to the historical lane-changing data of the road where the vehicle is located. The lane-changing probability matrix includes the probability that a vehicle on the road moves from a first lane on the road to a second lane on the road;

[0216] A probability acquisition sub-unit, configured to acquire the probability that the vehicle is located in the second lane at the first time point according to the lane-changing probability matrix of the road and the probability that the vehicle is located in the first lane at a second time point, where the second time point is before the third time point;

[0217] A position acquisition sub-unit, configured to multiply the position of each second lane by the probability that the vehicle is located in each second lane at the first time point and sum them to obtain the lateral position of the vehicle at the first time point.

[0218] In some embodiments, the fitting parameters include a lateral fitting parameter and a longitudinal fitting parameter, and the trajectory prediction unit is configured to:

[0219] Construct a lateral displacement model according to the short-time domain trajectory of the vehicle, the lateral position of the vehicle at the first time point, and the lateral fitting parameter. The lateral displacement model is used to simulate the lateral displacement of the vehicle and the satisfied constraint conditions;

[0220] Obtain the lateral displacements corresponding to multiple time points of the vehicle within the third time period according to multiple time points within the third time period and the lateral displacement model;

[0221] Construct a longitudinal displacement model based on the short-time domain trajectory and longitudinal fitting parameters of the vehicle. The longitudinal displacement model is used to simulate the longitudinal displacement of the vehicle and the satisfied constraint conditions;

[0222] Obtain the longitudinal displacements corresponding to multiple time points of the vehicle within the third time period according to the multiple time points within the third time period and the longitudinal displacement model.

[0223] In some embodiments, the above-mentioned screening module 403 is used for:

[0224] For each long-time domain trajectory of the vehicle, extract the trajectory corresponding to the second time period from the long-time domain trajectory;

[0225] Obtain the distance between the trajectory corresponding to the second time period and the short-time domain trajectory of the vehicle;

[0226] Among the trajectories corresponding to multiple second time periods, determine the trajectory with the minimum distance as the trajectory of the vehicle.

[0227] It should be noted that when the device provided in the above embodiments predicts the vehicle trajectory, only the division of the above-mentioned functional modules is used for illustration. In practical applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the device provided in the above embodiments and the method embodiments belong to the same concept. For the specific implementation process, please refer to the method embodiments, which will not be repeated here.

[0228] Figure 5 It is a schematic structural diagram of a computer device provided by an embodiment of the present application. The computer device 500 may vary greatly due to configuration or performance differences, and may include one or more CPUs (Central Processing Units, processors) 501 and one or more memories 502. Among them, at least one computer program is stored in the memory 502, and the at least one computer program is loaded and executed by the processor 501 to implement the vehicle trajectory prediction method provided by each of the above method embodiments. Of course, the computer device may also have components such as wired or wireless network interfaces, keyboards, and input / output interfaces for input / output. The computer device may also include other components for implementing device functions, which will not be elaborated here.

[0229] The embodiments of the present application further provide a computer-readable storage medium, in which at least one segment of computer program is stored. The at least one segment of computer program is loaded and executed by a processor of a computer device to implement the operations performed by the computer device in the vehicle trajectory prediction method in the above embodiments. For example, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, floppy disk, and optical data storage device, etc.

[0230] In some embodiments, the computer program involved in the embodiments of the present application can be deployed to be executed on a single computer device, or on multiple computer devices located at one place. Or, it can be executed on multiple computer devices distributed at multiple places and interconnected through a communication network. The multiple computer devices distributed at multiple places and interconnected through a communication network can form a blockchain system.

[0231] The embodiments of the present application further provide a computer program product or a computer program. The computer program product or the computer program includes computer program code, and the computer program code is stored in a computer-readable storage medium. The processor of the computer device reads the computer program code from the computer-readable storage medium, and the processor executes the computer program code, so that the computer device executes the vehicle trajectory prediction method provided in the above various optional implementation manners.

[0232] Those of ordinary skill in the art can understand that all or part of the steps to implement the above embodiments can be completed by hardware, or can be completed by a program instructing relevant hardware. The program can be stored in a computer-readable storage medium. The storage medium mentioned above can be a read-only memory, a magnetic disk, or an optical disc, etc.

[0233] The above are only the optional embodiments of the present application, and are not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A vehicle trajectory prediction method, characterized in that: The method comprises: Predicting the short-time domain trajectory of the vehicle in a second time period by Gaussian process regression according to the motion data of the vehicle in a first time period, the first time period being any time period before a current moment, the second time period being any time period after the current moment, the motion data comprising the lateral displacement and the longitudinal displacement of the vehicle corresponding to a plurality of time points in the first time period, and the short-time domain trajectory comprising the lateral displacement and the longitudinal displacement of the vehicle corresponding to a plurality of time points in the second time period; According to the short-time-domain trajectory of the vehicle and the historical lane-changing data of the road where the vehicle is located, in combination with multiple groups of fitting parameters, multiple long-time-domain trajectories of the vehicle in a third time period are predicted, each of the long-time-domain trajectories corresponds to a group of the fitting parameters, the historical lane-changing data includes historical driving records of multiple vehicles passing on the road, the third time period is any time period after the current moment, the third time period includes the second time period, and the long-time-domain trajectory includes the lateral displacement and the longitudinal displacement of the vehicle corresponding to multiple time points in the third time period; Based on the short-time-domain trajectory of the vehicle, a trajectory of the vehicle is screened out from a plurality of long-time-domain trajectories of the vehicle.

2. The method according to claim 1, characterized in that The predicting the short-term trajectory of the vehicle in the second time period by Gaussian process regression according to the motion data of the vehicle in the first time period includes: According to the motion data of the vehicle in the first time period, a prior distribution obeyed by the lateral displacement of the vehicle in the first time period is obtained by Gaussian process regression; According to the prior distribution obeyed by the lateral displacement of the vehicle in the first time period and the longitudinal displacement corresponding to each time point in the second time period, obtaining the posterior distribution obeyed by the lateral displacement of the vehicle at each time point in the second time period; The lateral displacements of the vehicle corresponding to multiple time points in the second time period are predicted according to the posterior distribution obeyed by the lateral displacement corresponding to each time point in the second time period, so as to obtain the short-term time-domain trajectory of the vehicle in the second time period.

3. The method according to claim 2, characterized in that The prior distribution obeyed by the lateral displacement of the vehicle in the first time period includes a noise variance matrix, and the noise variance matrix is ​​used to reflect random situations.

4. The method according to claim 1, characterized in that The predicting of a plurality of long-term trajectories of the vehicle in a third time period according to the short-term trajectory of the vehicle and the historical lane-changing data of the road where the vehicle is located, in combination with a plurality of sets of fitting parameters, comprises: predicting, based on historical lane change data of the road where the vehicle is located, a lateral position of the vehicle at a first time point, the first time point being a time point at which the third time period ends, the lateral position indicating a position of the vehicle in the lane where the vehicle is located at the first time point; For each set of fitting parameters, a long-term trajectory of the vehicle in the third time period is predicted according to the short-term trajectory of the vehicle, the lateral position of the vehicle at the first time point, and the fitting parameters.

5. The method according to claim 4, characterized in that The predicting the lateral position of the vehicle at a first time point according to the historical lane change data of the road where the vehicle is located comprises: Acquire a lane change probability matrix of the road according to historical lane change data of the road where the vehicle is located, wherein the lane change probability matrix includes a probability of a vehicle on the road moving from a first lane on the road to a second lane on the road; Obtaining a probability that the vehicle is located in the second lane at the first time point according to the lane change probability matrix of the road and a probability that the vehicle is located in the first lane at a second time point, the second time point being before the third time point; The position of each second lane and the probability that the vehicle is located in each second lane at the first time point are multiplied and summed to obtain the lateral position of the vehicle at the first time point.

6. The method according to claim 4, characterized in that The fitting parameters include a lateral fitting parameter and a longitudinal fitting parameter, and predicting the long-time-domain trajectory of the vehicle in the third time period according to the short-time-domain trajectory of the vehicle, the lateral position of the vehicle at the first time point, and the fitting parameters includes: Constructing a lateral displacement model according to the short-time trajectory of the vehicle, the lateral position of the vehicle at the first time point and the lateral fitting parameters, wherein the lateral displacement model is used to simulate the lateral displacement of the vehicle and the constraints satisfied; According to the multiple time points in the third time period and the lateral displacement model, obtaining the lateral displacement of the vehicle corresponding to the multiple time points in the third time period; Constructing a longitudinal displacement model according to the short-time-domain trajectory of the vehicle and the longitudinal fitting parameters, wherein the longitudinal displacement model is used to simulate the longitudinal displacement of the vehicle and the constraints satisfied; According to the multiple time points in the third time period and the longitudinal displacement model, the longitudinal displacements of the vehicle corresponding to the multiple time points in the third time period are obtained.

7. The method according to claim 1, characterized in that The step of selecting the trajectory of the vehicle from a plurality of long-time-domain trajectories of the vehicle based on the short-time-domain trajectory of the vehicle comprises: For each long-term trajectory of the vehicle, extracting a trajectory corresponding to the second time period from the long-term trajectory; Obtaining a distance between the trajectory corresponding to the second time period and the short-time domain trajectory of the vehicle; Among the multiple trajectories corresponding to the second time period, the trajectory with the smallest distance is determined as the trajectory of the vehicle.

8. A vehicle trajectory prediction device, characterized in that: The device comprises: a short-time domain prediction module, configured to predict the short-time domain trajectory of the vehicle in a second time period by Gaussian process regression based on the motion data of the vehicle in a first time period, wherein the first time period is any time period before the current moment, and the second time period is any time period after the current moment, wherein the motion data includes the lateral displacement and longitudinal displacement corresponding to multiple time points of the vehicle in the first time period, and the short-time domain trajectory includes the lateral displacement and longitudinal displacement corresponding to multiple time points of the vehicle in the second time period; a long-time domain prediction module, configured to predict multiple long-time domain trajectories of the vehicle in a third time period according to the short-time domain trajectory of the vehicle and the historical lane-changing data of the road where the vehicle is located, in combination with multiple groups of fitting parameters, wherein each of the long-time domain trajectories corresponds to a group of the fitting parameters, the historical lane-changing data includes historical driving records of multiple vehicles passing on the road, the third time period is any time period after the current moment, the third time period includes the second time period, and the long-time domain trajectory includes the lateral displacement and the longitudinal displacement of the vehicle corresponding to multiple time points in the third time period; A screening module is used to screen out a trajectory of the vehicle from a plurality of long-time-domain trajectories of the vehicle based on the short-time-domain trajectory of the vehicle.

9. A computer device, characterized in that: The computer device comprises a processor and a memory, wherein the memory is used to store at least one computer program, and the at least one computer program is loaded by the processor to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store at least one computer program, and the at least one computer program is used to execute the method according to any one of claims 1 to 7.

11. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images