Vehicle Trajectory Prediction Method and Related Equipment

Abstract

The present invention discloses a vehicle trajectory prediction method and related devices. The method includes: obtaining the actual trajectory sequence of a vehicle and vehicle state quantities, and combining the time stamps corresponding to the vehicle state quantities to establish a historical driving data set for the actual trajectory sequence and the vehicle state quantities; performing segmentation processing on the historical driving data set to obtain a segmented driving data set; calculating a first corner sequence using Ackermann geometric relationships, and obtaining a second corner sequence using an artificial neural network model; establishing a fusion coefficient related to the vehicle speed v x to fuse the first corner sequence and the second corner sequence to obtain a corner prediction sequence; using a vehicle dynamics trajectory model to combine with the corner prediction sequence to obtain a first predicted trajectory, using a trailer kinematics model to combine with a planned prediction trajectory sequence to obtain a second predicted trajectory, and fusing the first predicted trajectory and the second predicted trajectory to obtain a predicted dangerous point trajectory. The embodiments of the present application effectively improve the accuracy of the predicted trajectory of the dangerous point.

Description

Technical Field

[0001] The present invention relates to the field of safe driving, and particularly to a vehicle trajectory prediction method and related devices. Background Art

[0002] During driving, some position points on a vehicle (generally some points on the outer side of the vehicle body) are prone to collide with other vehicles or objects. These position points are generally referred to as vehicle danger points. For example: the points at the four corners of the front and rear of the vehicle.

[0003] By predicting the trajectory of the vehicle danger points, the trajectory of the vehicle danger points can be estimated in advance, and then the risk of vehicle collision can be determined according to this trajectory. It can be seen that predicting the trajectory of the vehicle danger points can effectively improve driving safety.

[0004] Currently, the accuracy of the trajectory of vehicle danger points predicted by the technology for predicting the trajectory of vehicle danger points is not high. Summary of the Invention

[0005] In view of the above problems, the present invention provides a vehicle trajectory prediction method and related devices that overcome the above problems or at least partially solve the above problems. The technical solutions are as follows:

[0006] A vehicle trajectory prediction method, comprising:

[0007] Obtain the actual trajectory sequence of the vehicle and vehicle state quantities, and establish a historical driving data set for the actual trajectory sequence and vehicle state quantities in combination with the time stamps corresponding to the vehicle state quantities;

[0008] Perform segmentation processing on the historical driving data set to obtain a segmented driving data set, and the segmented driving data set is used to calculate and obtain a planned prediction trajectory sequence Λ segi ;

[0009] Calculate a first corner sequence δ using Ackermann geometric relations sw_1 , and input a path curvature sequence τ series_i , vehicle speed v x and load G into an artificial neural network model to obtain a second corner sequence δ sw_2 ;

[0010] Establish a fusion coefficient related to the vehicle speed v x , and fuse the first corner sequence δ sw_1 and the second corner sequence δ sw_2 to obtain a corner prediction sequence

[0011] Use a vehicle dynamics trajectory model in combination with the corner prediction sequence Obtain the first predicted trajectory, and utilize the trailer kinematic model to combine with the planned prediction trajectory sequence Λ segi Obtain the second predicted trajectory, and fuse the first predicted trajectory and the second predicted trajectory to obtain the predicted dangerous point trajectory.

[0012] Optionally, acquire the actual trajectory sequence and vehicle state quantities of the vehicle, and combine with the time stamps corresponding to the vehicle state quantities to establish a historical driving data set for the actual trajectory sequence and vehicle state quantities, including:

[0013] Measure the actual trajectory sequence Λ of the positioning point O of the vehicle a , where Λ a ={ξ a1 , ξ a2 ,… ξ an}, ξ ai =[X ai , Y ai , t ai , X ai and Y ai are respectively the lateral coordinate and longitudinal coordinate of the positioning point O in the global coordinate system, and t ai is the time stamp of the actual trajectory sequence;

[0014] Measure the vehicle state quantities and record the time stamps TimeStamps corresponding to the vehicle state quantities. The vehicle state quantities include the steering wheel angle δ sw , the trailer yaw angular velocity ω r , the center of mass side slip angle β, the longitudinal acceleration a x , the vehicle speed v x and the trailer angle

[0015] Based on the time relationship between the time stamp t ai of the actual trajectory sequence and the time stamps TimeStamps corresponding to the vehicle state quantities, establish the corresponding relationship between the actual trajectory sequence Λ a and the time stamps TimeStamps;

[0016] Based on the actual trajectory sequence Λ a , the vehicle state quantities, the time stamps TimeStamps and the corresponding relationship, establish the historical driving data set Θ, where,

[0017]

[0018] Optionally, perform segmentation processing on the historical driving data set to obtain a segmented driving data set, which is used to calculate and obtain the planned prediction trajectory sequence Λ segi , including:

[0019] Perform segmentation processing on the historical driving dataset Θ to obtain a segmented driving dataset Θ a_segi , where:

[0020]

[0021] is an actual trajectory sequence, is a steering wheel angle sequence, ω ri , β i , a xi , v xi , respectively represent the yaw angular velocity of the trailer, the sideslip angle of the center of mass, the longitudinal acceleration, the vehicle speed, and the trailer angle corresponding to the time stamp TimeStamps_i at time i;

[0022] The obtaining process includes:

[0023] Obtain the actual trajectory sequence Λ in the historical driving dataset Θ a data point ξ corresponding to the time stamp TimeStamps_i at time i of ai , where ξ ai = [X ai , Y ai , t ai , and retrieve N data points backward from the data point ξ ai to obtain an actual trajectory sequence N is the number of path points of the predicted trajectory;

[0024] The obtaining process includes:

[0025] Obtain the steering wheel angle δ in the historical driving dataset Θ sw data point corresponding to the time stamp TimeStamps_i at time i of According to the chronological order, retrieve N data points backward from the data point to obtain a steering wheel angle sequence N is the number of path points of the predicted trajectory.

[0026] Optionally, the first angle sequence δ calculated using the Ackermann geometric relationship sw_1 includes:

[0027] Obtain the predicted trajectory sequence Λ segi , and calculate the path curvature sequence τ based on the predicted trajectory sequence Λ segi , and based on the path curvature sequence τ series_i , series_iThe steering radius sequence is calculated;

[0028] According to the Ackermann geometric relationship

[0029]

[0030] The first cornering angle sequence δ is calculated, where R is the steering radius, I is the steering transmission ratio, and L is the wheelbase of the vehicle. sw_1 , where R is the steering radius, I is the steering transmission ratio, and L is the wheelbase of the vehicle.

[0031] Optionally, establishing a fusion coefficient related to the vehicle speed v x to fuse the first cornering angle sequence δ sw_1 and the second cornering angle sequence δ sw_2 to obtain a cornering angle prediction sequence including:

[0032] Based on the formula

[0033]

[0034] fusing the first cornering angle sequence δ sw_1 and the second cornering angle sequence δ sw_2 to obtain a cornering angle prediction sequence where k is a fusion coefficient related to the vehicle speed v x , k ∈ (0, 1), and k is positively correlated with the absolute value of the vehicle speed v x .

[0035] Optionally, the determination process of the fusion coefficient k includes:

[0036] Based on the functional relationship between the fusion coefficient k and the vehicle speed v x to determine the fusion coefficient k, where

[0037]

[0038] is the vehicle speed when k = 0.5, and a is a calibration quantity that enables the artificial neural network model to converge fastest. 1

[0039] Optionally, using the vehicle dynamics trajectory model in combination with the cornering angle prediction sequence to obtain a first predicted trajectory, and using the trailer kinematic model in combination with the planned prediction trajectory sequence Λ segi to obtain a second predicted trajectory, including:

[0040] Inputting the cornering angle prediction sequence the vehicle speed v x and the load G into the vehicle dynamics trajectory model to calculate the first predicted trajectory P1 ;

[0041] Input the planned prediction trajectory sequence Λ segi into the kinematic model of the trailer to obtain the second predicted trajectory P 2 .

[0042] Optionally, fusing the first predicted trajectory and the second predicted trajectory to obtain the predicted dangerous point trajectory includes:

[0043] Based on the formula

[0044] P = a·P 1 +(1 - a)·P 2

[0045] Calculate the predicted dangerous point trajectory P, where P 1 is the first predicted trajectory, P 2 is the second predicted trajectory, and a is the confidence coefficient of the vehicle dynamics trajectory model, a ∈ [0, 1].

[0046] Optionally, the training process of the artificial neural network model includes:

[0047] Obtain the historical trajectory sequence and historical vehicle state quantities of the vehicle, and combine the time stamps corresponding to the historical vehicle state quantities to establish a training driving data set for the historical trajectory sequence and historical vehicle state quantities;

[0048] Perform segmentation processing on the training driving data set to obtain a segmented training driving data set;

[0049] Train the artificial neural network model based on the segmented training driving data set.

[0050] A vehicle trajectory prediction device includes:

[0051] A data set establishment unit, configured to obtain the actual trajectory sequence and vehicle state quantities of the vehicle, and combine the time stamps corresponding to the vehicle state quantities to establish a historical driving data set for the actual trajectory sequence and vehicle state quantities;

[0052] A data set segmentation unit, configured to perform segmentation processing on the historical driving data set to obtain a segmented driving data set, and the segmented driving data set is used to calculate and obtain the planned prediction trajectory sequence Λ segi ;

[0053] An angle sequence obtaining unit, configured to calculate the first angle sequence δ sw_1 using the Ackermann geometric relationship, and input the path curvature sequence τ series_i , vehicle speed v x and load G into the artificial neural network model to obtain the second angle sequence δsw_2 ;

[0054] A fusion unit, configured to establish a fusion coefficient related to the vehicle speed v x to fuse the first corner sequence δ sw_1 and the second corner sequence δ sw_2 to obtain a corner prediction sequence

[0055] A trajectory acquisition unit, configured to use a vehicle dynamics trajectory model to combine with the corner prediction sequence to obtain a first predicted trajectory, use a trailer kinematic model to combine with the planned prediction trajectory sequence Λ segi to obtain a second predicted trajectory, and fuse the first predicted trajectory and the second predicted trajectory to obtain a predicted dangerous point trajectory.

[0056] An electronic device, comprising:

[0057] One or more processors;

[0058] A storage device storing one or more programs thereon;

[0059] When the one or more programs are executed by the one or more processors, the one or more processors are caused to implement any of the above vehicle trajectory prediction methods.

[0060] A storage medium storing a program thereon, where the program, when executed by a processor, implements any of the above vehicle trajectory prediction methods.

[0061] A processor for running a program, where the program, when running, executes any of the above vehicle trajectory prediction methods.

[0062] By means of the above technical solutions, the vehicle trajectory prediction method and related devices provided by the present invention can respectively use the Ackermann geometric relationship and the artificial neural network model to obtain corner sequences, and fuse the two obtained corner sequences based on a fusion coefficient related to the vehicle speed to obtain a corner prediction sequence. Then, use a vehicle dynamics trajectory model to combine with the corner prediction sequence to obtain a first predicted trajectory, use a trailer kinematic model to combine with the planned prediction trajectory sequence to obtain a second predicted trajectory, and fuse the first predicted trajectory and the second predicted trajectory to obtain a predicted dangerous point trajectory. Through the above two fusion processes, the accuracy of the predicted trajectory of the dangerous point is effectively improved in the embodiments of the present application.

[0063] The above description is only an overview of the technical solution of the present invention. In order to be able to understand the technical means of the present invention more clearly, it can be implemented according to the content of the specification. And in order to make the above and other objects, features, and advantages of the present invention more obvious and understandable, the specific embodiments of the present invention are specifically exemplified below. Description of the Drawings

[0064] By reading the following detailed description of the preferred embodiments, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present invention. Moreover, throughout the drawings, the same reference numerals are used to represent the same components. In the drawings:

[0065] Figure 1 A flowchart of a vehicle trajectory prediction method provided by an embodiment of the present application is shown;

[0066] Figure 2 A schematic diagram of using vehicle speed to determine a trailer kinematic model and obtain a second predicted trajectory provided by an embodiment of the present application is shown;

[0067] Figure 3 A schematic diagram of a function relationship curve between a fusion coefficient and a speed provided by an embodiment of the present application is shown;

[0068] Figure 4 A schematic diagram of the recording and use of a driving data set provided by an embodiment of the present application is shown;

[0069] Figure 5 A flowchart of another vehicle trajectory prediction method provided by an embodiment of the present application is shown;

[0070] Figure 6 A schematic diagram of the structure of a vehicle trajectory prediction device provided by an embodiment of the present application is shown;

[0071] Figure 7 A schematic diagram of the structure of an electronic device provided by an embodiment of the present application is shown. Detailed Embodiments

[0072] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be completely conveyed to those skilled in the art.

[0073] As Figure 1 shown, a vehicle trajectory prediction method provided by an embodiment of the present application may include:

[0074] S100. Obtain the actual trajectory sequence and vehicle state quantities of the vehicle, and establish a historical driving data set for the actual trajectory sequence and vehicle state quantities in combination with the timestamps corresponding to the vehicle state quantities.

[0075] In practical applications, at least one of the actual trajectory sequence and vehicle state quantities can be collected by sensors or obtained through other means (such as from a vehicle manual).

[0076] Specifically, the trajectory of the vehicle is the trajectory of the horizontal and vertical coordinates of preset positioning points in the vehicle changing with time, which can be detected and obtained by a positioning device (such as a GPS, Beidou positioning device, etc.). The actual trajectory sequence of the vehicle includes multiple trajectory points with time stamps. When the vehicle in this application is a vehicle including a tractor and a trailer, the above-mentioned preset positioning points can be located on the tractor or on the trailer.

[0077] Optionally, the above vehicle state quantities can include various data, such as at least one of multiple data such as steering wheel angle, yaw rate, sideslip angle of the center of mass, longitudinal acceleration, vehicle speed, etc. When the vehicle in this application is a vehicle including a tractor and a trailer, the above driving data set can also include: the trailer angle. When the vehicle in this application is a vehicle including a tractor and a trailer, the above yaw rate can be the yaw rate of the trailer.

[0078] The steering wheel angle is the rotation angle of the steering wheel.

[0079] The yaw rate is the angular velocity of the vehicle mass rotating around the z-axis (vehicle coordinate system), which can be collected by a gyroscope sensor.

[0080] The sideslip angle of the center of mass is the angle between the direction of the vehicle center of mass velocity and the head pointing direction, which can be collected by a gyroscope sensor.

[0081] The longitudinal acceleration is the acceleration of the vehicle along the longitudinal axis direction, which can be collected by an accelerometer sensor.

[0082] The trailer angle refers to the angle between the trailer and the tractor, which can be collected by a laser or vision sensor.

[0083] It can be understood that at least some of the various parameters in the actual trajectory sequence and vehicle state quantities can be collected by sensors. The collection frequencies of different sensors are not the same, which also results in different collection times for different parameters. This application can correspond various parameters in the vehicle state quantities through time stamps. For example: all vehicle state quantities collected within a certain time range are corresponding to the same time stamp, and this time stamp matches this time range. Such as: this time stamp is a moment within this time range or an identifier of this time range. At the same time, this application can also correspond the actual trajectory sequence with this time stamp. In this way, the actual trajectory sequence and vehicle state quantities collected within the same time range can both correspond to this time stamp, thereby establishing a historical driving data set.

[0084] S200. Split the historical driving dataset to obtain a split driving dataset, which is used to calculate the planned prediction trajectory sequence Λ segi .

[0085] The historical driving dataset includes actual trajectory sequences and vehicle state quantities corresponding to multiple timestamps. In this application, the historical driving dataset can be split into multiple split driving datasets. Each split driving dataset can include: the planned prediction trajectory sequence Λ corresponding to the same timestamp segi , the steering wheel angle sequence, the yaw rate of the trailer, the sideslip angle of the center of mass, the longitudinal acceleration, the vehicle speed, and the trailer angle. The above-mentioned planned prediction trajectory sequence Λ segi can specifically include: the trajectory points corresponding to a timestamp and the N trajectory points after this trajectory point. The above-mentioned steering wheel angle sequence can specifically include: the steering wheel angle corresponding to a timestamp and the N steering wheel angles after this steering wheel angle.

[0086] S300. Calculate the first angle sequence δ using the Ackermann geometry relationship sw_1 , and use an artificial neural network model to input the path curvature sequence τ series_i , the vehicle speed v x , and the load G to obtain the second angle sequence δ sw_2 .

[0087] The Ackermann geometry relationship is also known as the Ackermann steering relationship or the Ackermann steering model. The Ackermann geometry relationship is specifically:

[0088]

[0089] where δ sw is the steering wheel angle, R is the turning radius, I is the steering ratio, and L is the vehicle wheelbase.

[0090] It can be seen from this formula that after obtaining the turning radius, the steering ratio, and the vehicle wheelbase, the steering wheel angle can be calculated based on this formula. For a certain fixed vehicle, the steering ratio and the vehicle wheelbase are known quantities. The turning radius, also known as the turning circle radius, refers to the distance from the center of rotation to the contact point of the front outer steering wheel with the ground during the driving process of the vehicle. The steering ratio is the ratio of the steering degree of the steering wheel to the steering degree of the wheel.

[0091] In this application, the path curvature can be calculated based on the trajectory sequence, and then the turning radius can be calculated based on the path curvature. Furthermore, the steering wheel angle can be calculated according to the Ackermann geometry relationship, and the first angle sequence δ can be obtained sw_1 .

[0092] Optionally, the above-mentioned use of the artificial neural network model to input the path curvature sequence τ series_i , the vehicle speed vx and the load G to obtain the second corner sequence δ sw_2 Specifically, it may include:

[0093] Input the path curvature sequence τ series_i , vehicle speed v x and the load G into a pre-trained artificial neural network model to obtain the second corner sequence δ output by the pre-trained artificial neural network model sw_2 .

[0094] Among them, the input of the above artificial neural network model can be: path curvature sequence, vehicle speed and load, and the output of the above artificial neural network model can be: corner sequence. The above artificial neural network model can be represented by model M 2 .

[0095] It can be understood that this application can train the artificial neural network model through the historical trajectory sequence and historical vehicle state quantities.

[0096] S400. Establish a fusion coefficient related to the vehicle speed v x to fuse the first corner sequence δ sw_1 and the second corner sequence δ sw_2 to obtain the corner prediction sequence

[0097] It can be understood that the first corner sequence δ sw_1 is calculated through the Ackermann geometric relationship, and the second corner sequence δ sw_2 is obtained through the artificial neural network model. Therefore, this application will fuse the first corner sequence δ sw_1 and the second corner sequence δ sw_2 to obtain a corner prediction sequence with relatively high accuracy. At the same time, since this application performs the above fusion through a fusion coefficient related to the vehicle speed v x , when the vehicle speed v x is different, the fusion coefficient may also be different, further improving the accuracy of the fused corner prediction sequence.

[0098] Optionally, when the vehicle speed is low, the fusion coefficient can approach 0, and at this time approaches δ sw1 ; when the vehicle speed is high, the fusion coefficient approaches 1, and at this time approaches δ sw2 . In practical applications, the embodiments of this application can determine the fusion coefficient based on the functional relationship between the fusion coefficient and v x .

[0099] S500. Use the vehicle dynamics trajectory model to combine the corner prediction sequence Obtain the first predicted trajectory, and use the trailer kinematic model to combine with the planned predicted trajectory sequence Λ segi Obtain the second predicted trajectory, and fuse the first predicted trajectory and the second predicted trajectory to obtain the predicted dangerous point trajectory.

[0100] It should be noted that steps S100 to S500 in the vehicle trajectory prediction method of this application are not limited to Figure 1 the execution order shown.

[0101] Among them, dynamics mainly studies the relationship between the forces acting on an object and the motion of the object. The vehicle dynamics trajectory model is generally used to analyze the ride comfort and the handling stability of the vehicle, and can obtain the predicted trajectory of the vehicle.

[0102] Kinematics studies the motion laws of objects from a geometric perspective, including the changes in the position, speed, etc. of an object in space over time. Therefore, the trailer kinematic model is a mathematical model that can reflect the relationship between parameters such as the position, speed, and acceleration of the trailer and time.

[0103] Optionally, the vehicle dynamics trajectory model may include: a tractor model and a trailer model.

[0104] The trailer kinematic model in the embodiment of this application can be a two-degree-of-freedom vehicle model. Considering the constant-speed assumption of this model, as Figure 2 shown, the embodiment of this application can build multiple trailer kinematic models 201 according to different vehicle speeds. The embodiment of this application can use a switch 202 to select different trailer kinematic models 201 according to the vehicle speed, and then use the selected trailer kinematic model 201 to process the predicted trajectory sequence Λ segi for calculation to obtain the second predicted trajectory. It can be understood that the more trailer kinematic models 201 are built, the more accurate the second predicted trajectory obtained based on the trailer kinematic model will be. In practical applications, this application can equivalently convert the current trailer angle into the steering angle input of the front wheels, and online estimate parameters such as the side slip stiffness of the trailer tires and the position of the trailer center of mass. This application can update the parameters of the vehicle dynamics trajectory model in real time to improve the accuracy of the model.

[0105] Among them, using the trailer kinematic model to calculate the predicted trajectory sequence Λ segi to obtain the second predicted trajectory of the vehicle is an existing mature technology, and those skilled in the art are aware of its specific solution, so it will not be elaborated here.

[0106] The vehicle trajectory prediction method provided by the present invention can obtain the steering angle sequence by using the Ackermann geometric relationship and the artificial neural network model respectively, and fuse the two obtained steering angle sequences based on a fusion coefficient related to the vehicle speed to obtain a steering angle prediction sequence. Then, the vehicle dynamics trajectory model is used to combine with the steering angle prediction sequence to obtain a first predicted trajectory, and the trailer kinematic model is used to combine with the planned prediction trajectory sequence to obtain a second predicted trajectory, and the first predicted trajectory and the second predicted trajectory are fused to obtain a predicted dangerous point trajectory. Through the above two fusion processes, the accuracy of the predicted trajectory of the dangerous point is effectively improved in the embodiments of the present application.

[0107] Compared with the existing methods, the vehicle trajectory prediction method provided by the present invention applies the vehicle dynamics trajectory model in the calculation process. At the same time, the method can continuously improve the model based on the accumulation of historical driving data, and improves the accuracy of the trailer dangerous point trajectory prediction model.

[0108] Optionally, in other embodiments, Figure 1 Step S100 in the method shown can specifically include:

[0109] Measure the actual trajectory sequence Λ of the positioning point O of the vehicle a , where Λ a ={ξ a1 , ξ a2 ,… ξ an}, ξ ai =[X ai , Y ai , t ai , X ai and Y ai are respectively the lateral coordinate and the longitudinal coordinate of the positioning point O in the global coordinate system, and t ai is the time stamp of the actual trajectory sequence;

[0110] Measure the vehicle state quantity and record the time stamp TimeStamps corresponding to the vehicle state quantity. The vehicle state quantity includes the steering wheel angle δ sw , the trailer yaw rate ω r , the center of mass side slip angle β, the longitudinal acceleration a x , the vehicle speed v x and the trailer angle

[0111] Based on the time relationship between the time stamp t ai of the actual trajectory sequence and the time stamp TimeStamps corresponding to the vehicle state quantity, establish the corresponding relationship between the actual trajectory sequence Λ a and the time stamp TimeStamps;

[0112] Based on the actual trajectory sequence Λ a, vehicle state variables, time stamps TimeStamps, and corresponding relationships to establish a historical driving dataset Θ, where

[0113]

[0114] The above positioning point O can be the center point of the rear axle of the vehicle, or the center point of the front axle of the vehicle, the geometric center point of the vehicle, or the center point of the vehicle head, etc. The above rear axle and front axle can be the rear axle and front axle of the tractor, or the rear axle and front axle of the trailer.

[0115] It can be understood that each trajectory point in the trajectory sequence will correspond to a trajectory moment to indicate at what moment the positioning point of the vehicle reaches this trajectory point. In addition to the trajectory of the vehicle, the historical driving dataset can also include vehicle state variables. At least some of the data in these vehicle state variables can be collected by sensors. Therefore, at least some of the data in the above vehicle state variables in the historical driving dataset also correspond to collection moments. The collection moments of each sensor are different, and at the same time, the collection moment of the above vehicle state variables may also be different from the trajectory moment. Therefore, in order to correspond the various data in the historical driving dataset collected or obtained within a period of time, this application uses time stamps TimeStamps. Optionally, the historical driving dataset can be stored in a database, and the time stamp TimeStamps can be the time stamp in the database. This time stamp TimeStamps is a unique binary number automatically generated in the database and is usually used as a mechanism for adding a version stamp to table rows. Optionally, over time, this application can collect multiple historical driving datasets. The collection moments of the data in one historical driving dataset can be not completely the same, and each historical driving dataset corresponds to a time stamp TimeStamps, so that each historical driving dataset can correspond to multiple different time stamps TimeStamps respectively.

[0116] Optionally, in other embodiments Figure 1 The step S200 in the shown method can specifically include:

[0117] Perform a segmentation process on the historical driving dataset Θ to obtain a segmented driving dataset Θ a_segi , where:

[0118]

[0119] is an actual trajectory sequence is a sequence of steering wheel angles, ω ri , β i , a xi , v xi , respectively represent the yaw angular velocity, sideslip angle of the center of mass, longitudinal acceleration, vehicle speed, and trailer angle corresponding to the time stamp TimeStamps_i at time i;

[0120] The obtaining process of

[0121] obtain the actual trajectory sequence Λ in the historical driving data set Θ a the data point ξ corresponding to the time stamp TimeStamps_i at time i of ai , where ξ ai =[X ai , Y ai , t ai , and retrieve N data points backward from the data point ξ ai to obtain an actual trajectory sequence N is the number of path points of the predicted trajectory;

[0122] The obtaining process of

[0123] obtain the steering wheel angle δ in the historical driving data set Θ sw the data point corresponding to the time stamp TimeStamps_i at time i of According to the time sequence, from the data point retrieve N data points backward to obtain a steering wheel angle sequence N is the number of path points of the predicted trajectory.

[0124] Among them, TimeStamps_i represents the time stamp at time i. Among them, N can be 99 or other natural numbers.

[0125] It can be understood that through the above segmentation process, the segmented driving data set Θ obtained in this application a_segi not only includes various data corresponding to the time stamp at time i, but also includes N trajectory points and N steering wheel angles after time i. Therefore, this application can calculate the planned prediction trajectory sequence Λ according to the segmented driving data set segi .

[0126] Optionally, in other embodiments, Figure 1 in the method shown in sw_1 step S300 uses the Ackermann geometric relationship to calculate the first angle sequence δ

[0127] obtain the predicted trajectory sequence Λ segi , based on the predicted trajectory sequence Λ segi calculate the path curvature sequence τ series_i , based on the path curvature sequence τ series_iThe steering radius sequence is calculated;

[0128] According to the Ackermann geometric relationship

[0129]

[0130] the first steering angle sequence δ is calculated sw_1 , where R is the steering radius, I is the steering transmission ratio, and L is the wheelbase of the vehicle.

[0131] In the steady-state steering situation with very low vehicle speed, the movement of the vehicle simply follows the Ackermann geometric relationship, that is, the curvature of the vehicle trajectory is proportional to the steering wheel angle. By predicting the trajectory sequence Λ segi , the path curvature sequence τ can be calculated series_i , and then the first steering angle sequence δ can be obtained from the Ackermann geometric relationship sw_1 . The Ackermann geometric relationship can be represented by the model M 1 . The model M 1 has the advantages of small computational complexity and fast computational speed; however, when the vehicle speed is high, the accuracy of the model M 1 will be greatly affected.

[0132] Optionally, in other embodiments, Figure 1 step S400 in the method shown may include:

[0133] Based on the formula

[0134]

[0135] the first steering angle sequence δ sw_1 and the second steering angle sequence δ sw_2 are fused to obtain the steering angle prediction sequence where k is a fusion coefficient related to the vehicle speed v x , k ∈ (0, 1), and k is positively correlated with the absolute value of the vehicle speed v x .

[0136] This application can achieve the fusion of the prediction results of the two models M 1 , M 2 through the fusion coefficient k.

[0137] The fusion coefficient k is related to the vehicle speed v x . When the vehicle speed is low, the fusion coefficient k approaches 0, and at this time is close to δ sw1 ; while when the vehicle speed is high, the fusion coefficient k approaches 1, and at this time is close to δ sw2 .

[0138] Optionally, the determination process of the fusion coefficient k includes:

[0139] Based on the fusion coefficient k and the vehicle speed v x of the functional relationship

[0140]

[0141] Determine the fusion coefficient k, where is the vehicle speed when k = 0.5, a 1 is the calibration quantity for the fastest convergence of the artificial neural network model.

[0142] Such as Figure 3 in is 5 m / s. Can affect the shape of the function relationship curve. In practical applications, those skilled in the art can set the value of a 1 is the calibration quantity for calibrating to obtain a suitable fusion function relationship. In an alternative embodiment, the functional relationship curve between the fusion coefficient k and the speed v x can be as Figure 3 shown.

[0143] Figure 3 The abscissa in is the speed, with the unit of m / s; the ordinate is the fusion coefficient. Figure 3 The three curves in are the fusion function relationship curves when a 1 is 1, 1.8, and 2.6 respectively. The process of using a 1 for calibration can be: First, set an initial value of a 1 , and train the model M 2 according to the historical driving data set, and observe the convergence of the model M 2 ; then select another initial value of a 1 , continue the training and observe the convergence of the model M 2 , and repeat this process until a value of a 1 for the fastest convergence is found. The value of the fastest convergence a 1 is the value of a 1 .

[0144] Optionally, in other embodiments, Figure 1 in the method shown in step S500, the vehicle dynamics trajectory model is combined with the corner prediction sequence to obtain the first predicted trajectory, and the trailer kinematic model is combined with the planned prediction trajectory sequence Λ segi to obtain the second predicted trajectory, which may include:

[0145] Input the corner prediction sequence the vehicle speed v x and the load G into the vehicle dynamics trajectory model, and calculate to obtain the first predicted trajectory P1 ;

[0146] Input the planned prediction trajectory sequence Λ segi into the trailer kinematic model to obtain the second predicted trajectory P 2 .

[0147] Optionally, in other embodiments Figure 1 in the method shown, step S500 of fusing the first predicted trajectory and the second predicted trajectory to obtain the predicted dangerous point trajectory may include:

[0148] Based on the formula

[0149] P = a·P 1 +(1 - a)·P 2

[0150] calculate to obtain the predicted dangerous point trajectory P, where P 1 is the first predicted trajectory, P 2 is the second predicted trajectory, and a is the confidence coefficient of the vehicle dynamics trajectory model, a ∈ [0, 1].

[0151] Embodiments of the present application can use the driving data set with historical timestamps to calculate P 1 and P 2 respectively with the proximity between P, and then according to this proximity, use a simple linear relationship and normalization to obtain the determined confidence coefficient a ∈ [0, 1]. During the process of adjusting the vehicle dynamics trajectory model, a starts from 0, and as the accuracy of the vehicle dynamics trajectory model improves, a gradually approaches 1. When a is greater than 0.5, it indicates that the accuracy of the predicted trajectory of the dangerous point obtained using the vehicle dynamics trajectory model exceeds the accuracy of the predicted trajectory of the dangerous point obtained using the trailer kinematic model.

[0152] Optionally, the training process of the artificial neural network model in the present application may include:

[0153] Obtain the historical trajectory sequence and historical vehicle state quantities of the vehicle, and combine the timestamps corresponding to the historical vehicle state quantities to establish a training driving data set for the historical trajectory sequence and historical vehicle state quantities;

[0154] Perform segmentation processing on the training driving data set to obtain a segmented training driving data set;

[0155] Train the artificial neural network model based on the segmented training driving data set.

[0156] The present application can train the artificial neural network model during the vehicle driving process.

[0157] Optionally, the acquisition method of the historical trajectory sequence and historical vehicle state quantities during the training process of the artificial neural network model can be the same asFigure 1 In the method shown, the way to obtain the actual trajectory sequence and vehicle state quantity in step S100 is the same, so it will not be elaborated here.

[0158] Optionally, the specific way of segmentation processing during the training process of the artificial neural network model can be the same as Figure 1 the specific way of segmentation processing in step S200 in the method shown, so it will not be elaborated here.

[0159] In practical applications, as Figure 4 shown, this application can first collect a driving data set for a period of time and use it as the initial driving data set. Starting from the starting point of the initial driving data set, in the same acquisition manner as in step S100, the actual trajectory sequence and vehicle state quantity of the vehicle are obtained from the initial driving data set, and a training driving data set is established. Then, a segmented training driving data set is obtained from the training driving data set to train the artificial neural network model. At the same time, during the normal driving process of the vehicle, this application can continue to record driving data and add the recorded data to the initial driving data set, and then obtain a training driving data set from the initial driving data set with added data and train the artificial neural network model again. Each obtained training driving data set can be the data after the previously obtained training driving data set. This application can continuously improve the training driving data set and train the model M online 2 . When the confidence coefficient a prediction accuracy of the vehicle dynamics trajectory model is large enough (for example, it has exceeded the accuracy calculated by kinematics), it can be considered that the steering wheel angle sequence is reliable, and the data set is perfect enough, and it is no longer necessary to collect driving data.

[0160] In this way, this application realizes the update and expansion of training data, and also realizes the continuous optimization of the artificial neural network model, further improving the accuracy of the predicted trajectory of dangerous points.

[0161] This application applies a vehicle dynamics trajectory model including a tractor model and a trailer model, and improves the prediction accuracy compared with the current method of calculating by kinematic model.

[0162] This application can learn online based on the recorded historical data to improve the calculation accuracy of the trailer dangerous point trajectory, and define a confidence coefficient to quantify this process.

[0163] Using the positioning point trajectory and steering wheel angle measured by the sensor as the original data for training actually takes into account the tracking control error and reduces the influence of the error between the positioning point planned trajectory and the positioning point actual trajectory on the trailer dangerous point trajectory prediction accuracy.

[0164] As the basis of the subsequent collision detection module, this method is beneficial to improving the driving safety of autonomous vehicles.

[0165] For ease of understanding, the solution of this application will be described below in conjunction with Figure 5 the following:

[0166] As Figure 5 shown, a vehicle trajectory prediction method provided by this application may include:

[0167] Obtain the actual trajectory sequence of the vehicle and the vehicle state quantity, and establish a historical driving data set for the actual trajectory sequence and the vehicle state quantity in combination with the time stamp corresponding to the vehicle state quantity; the vehicle state quantity of the vehicle includes: vehicle speed and load;

[0168] Perform segmentation processing on the historical driving data set to obtain a segmented driving data set, and the segmented driving data set is used to calculate and obtain a planned prediction trajectory sequence;

[0169] Calculate the path curvature of the planned prediction trajectory sequence to obtain a path curvature sequence;

[0170] Use the Ackermann geometric relationship to calculate the path curvature sequence to obtain a first steering angle sequence;

[0171] Use an artificial neural network model, input the path curvature sequence, vehicle speed, and load to obtain a second steering angle sequence;

[0172] Establish a fusion coefficient related to the vehicle speed, and fuse the first steering angle sequence and the second steering angle sequence to obtain a steering angle prediction sequence;

[0173] Use the vehicle dynamics trajectory model to combine the steering angle prediction sequence to obtain a first prediction trajectory P 1 ;

[0174] Use the trailer kinematic model to combine the planned prediction trajectory sequence to obtain a second prediction trajectory P 2 ;

[0175] Based on the formula

[0176] P = a·P 1 +(1 - a)·P 2

[0177] Calculate to obtain a predicted dangerous point trajectory P, where P 1 is the first prediction trajectory, P 2 is the second prediction trajectory, and a is the confidence coefficient of the vehicle dynamics trajectory model, a ∈ [0, 1].

[0178] Among them, as Figure 5 shown, the confidence coefficient a can be calculated based on the historical driving data set, the first prediction trajectory P 1 and the second prediction trajectory P 2 obtained.

[0179] Corresponding to the method embodiment shown, the present application also provides a vehicle trajectory prediction device. Figure 1 As shown, a vehicle trajectory prediction device provided by an embodiment of the present application may include:

[0180] A data set establishment unit 100, configured to obtain the actual trajectory sequence and vehicle state quantity of the vehicle, and establish a historical driving data set for the actual trajectory sequence and vehicle state quantity in combination with the time stamp corresponding to the vehicle state quantity; Figure 6

[0181] A data set segmentation unit 200, configured to perform segmentation processing on the historical driving data set to obtain a segmented driving data set, and the segmented driving data set is used to calculate and obtain a planned prediction trajectory sequence Λ

[0182] ; segi

[0183] An angle sequence obtaining unit 300, configured to calculate a first angle sequence δ by using Ackermann geometric relationship, sw_1 input a path curvature sequence τ series_i vehicle speed v x and load G into an artificial neural network model to obtain a second angle sequence δ sw_2 ;

[0184] A fusion unit 400, configured to establish a fusion coefficient related to the vehicle speed v x and fuse the first angle sequence δ sw_1 and the second angle sequence δ sw_2 to obtain an angle prediction sequence

[0185] A trajectory obtaining unit 500, configured to obtain a first predicted trajectory by using a vehicle dynamics trajectory model in combination with the angle prediction sequence obtain a second predicted trajectory by using a trailer kinematic model in combination with the planned prediction trajectory sequence Λ segi and fuse the first predicted trajectory and the second predicted trajectory to obtain a predicted dangerous point trajectory.

[0186] Optionally, the data set establishment unit 100 includes: a first measurement subunit, a second measurement subunit, a corresponding subunit, and an establishment subunit,

[0187] The first measurement subunit is configured to measure and obtain the actual trajectory sequence Λ of the positioning point O of the vehicle, a where Λ a ={ξ a1 , ξ a2 ,…ξ an}, ξ ai =[X ai , Y​​ai , t ai , X ai and Y ai are respectively the horizontal coordinate and the vertical coordinate of the positioning point O in the global coordinate system, t ai is the timestamp of the actual trajectory sequence;

[0188] The second measurement subunit is configured to measure vehicle state quantities and record the timestamps TimeStamps corresponding to the vehicle state quantities, where the vehicle state quantities include the steering wheel angle δ sw , the yaw rate ω of the trailer r , the sideslip angle β of the center of mass, the longitudinal acceleration a x , the vehicle speed v x and the trailer angle

[0189] The corresponding subunit is configured to establish the corresponding relationship between the actual trajectory sequence Λ ai and the timestamp TimeStamps corresponding to the vehicle state quantities based on the time relationship between the timestamp t a of the actual trajectory sequence and the timestamp TimeStamps corresponding to the vehicle state quantities;

[0190] The establishment subunit is configured to establish a historical driving data set Θ based on the actual trajectory sequence Λ a , the vehicle state quantities, the timestamp TimeStamps, and the corresponding relationship, where

[0191]

[0192] Optionally, the data set splitting unit 200 includes: a first obtaining subunit, a second obtaining subunit, and a splitting subunit,

[0193] The first obtaining subunit is configured to obtain the data point ξ a corresponding to the timestamp TimeStamps_i at the moment i of the actual trajectory sequence Λ ai in the historical driving data set Θ, where ξ ai = [X ai , Y ai , t ai , and retrieve N data points backward from the data point ξ ai in chronological order to obtain a segment of the actual trajectory sequence N is the number of path points of the predicted trajectory;

[0194] The second obtaining subunit is configured to obtain the data point sw corresponding to the timestamp TimeStamps_i at the moment i of the steering wheel angle δ Retrieve N data points backward from the said data points in chronological order to obtain a sequence of steering wheel angles where N is the number of path points of the predicted trajectory

[0195] A segmentation subunit, configured to segment the historical driving data set Θ to obtain a segmented driving data set Wherein:

[0196]

[0197] is an actual trajectory sequence is a sequence of steering wheel angles, ω ri , β i , a xi , v xi , respectively represent the yaw angular velocity of the trailer, the sideslip angle of the center of mass, the longitudinal acceleration, the vehicle speed, and the included angle of the trailer corresponding to the time stamp TimeStamps_i at time i

[0198] Optionally, the angle sequence obtaining unit 300 calculates a first angle sequence δ using the Ackermann geometry relationship sw_1 , specifically set as:

[0199] Obtain a predicted trajectory sequence Λ segi , based on the predicted trajectory sequence Λ segi Calculate a path curvature sequence τ series_i , based on the path curvature sequence τ series_i Calculate a steering radius sequence;

[0200] According to the Ackermann geometry relationship

[0201]

[0202] Calculate a first angle sequence δ sw_1 , where R is the steering radius, I is the steering ratio, and L is the vehicle wheelbase

[0203] Optionally, the fusion unit 400 is specifically configured to:

[0204] Based on the formula

[0205]

[0206] Fuse the first angle sequence δ sw_1 and the second angle sequence δ sw_2 to obtain an angle prediction sequence where k is related to the vehicle speed v xThe relevant fusion coefficient, k ∈ (0, 1), and k is positively correlated with the absolute value of the vehicle speed v x is positively correlated.

[0207] Optionally, the fusion unit 400 determines that the specific configuration of the fusion coefficient k is:

[0208] Based on the functional relationship between the fusion coefficient k and the vehicle speed v x the functional relationship

[0209]

[0210] determine the fusion coefficient k, where is the vehicle speed when k = 0.5, a 1 is the calibration quantity that enables the artificial neural network model to converge fastest.

[0211] Optionally, the trajectory acquisition unit 500 uses the vehicle dynamics trajectory model in combination with the steering angle prediction sequence to obtain a first predicted trajectory, and uses the trailer kinematic model in combination with the planned prediction trajectory sequence Λ segi to obtain a second predicted trajectory, and the specific configuration is:

[0212] Input the steering angle prediction sequence the vehicle speed v x and the load G into the vehicle dynamics trajectory model, and calculate to obtain a first predicted trajectory P 1 ;

[0213] Input the planned prediction trajectory sequence Λ segi into the trailer kinematic model to obtain a second predicted trajectory P 2 .

[0214] Optionally, the trajectory acquisition unit 500 fuses the first predicted trajectory and the second predicted trajectory to obtain a predicted dangerous point trajectory, and the specific configuration is:

[0215] Based on the formula

[0216] P = a·P 1 +(1 - a)·P 2

[0217] Calculate to obtain a predicted dangerous point trajectory P, where P 1 is the first predicted trajectory, P 2 is the second predicted trajectory, and a is the confidence coefficient of the vehicle dynamics trajectory model, a ∈ [0, 1].

[0218] Optionally, Figure 6The device shown also includes: a training unit, configured to obtain a historical trajectory sequence and historical vehicle state quantities of a vehicle, establish a training driving data set for the historical trajectory sequence and historical vehicle state quantities in combination with the time stamps corresponding to the historical vehicle state quantities; perform segmentation processing on the training driving data set to obtain a segmented training driving data set; and train the artificial neural network model based on the segmented training driving data set.

[0219] The vehicle trajectory prediction device provided by the present invention can respectively use the Ackermann geometric relationship and the artificial neural network model to obtain an angle sequence, fuse the two obtained angle sequences based on a fusion coefficient related to the vehicle speed to obtain an angle prediction sequence, then use the vehicle dynamics trajectory model in combination with the angle prediction sequence to obtain a first prediction trajectory, use the trailer kinematic model in combination with the planned prediction trajectory sequence to obtain a second prediction trajectory, and fuse the first prediction trajectory and the second prediction trajectory to obtain a predicted dangerous point trajectory. Through the above two fusion processes in the embodiments of the present application, the accuracy of the predicted trajectory of the dangerous point is effectively improved.

[0220] As Figure 7 shown, an embodiment of the present invention further provides an electronic device, which may include:

[0221] One or more processors 701;

[0222] A storage device 702, on which one or more programs are stored;

[0223] When the one or more programs are executed by the one or more processors 701, the one or more processors 701 are caused to implement any vehicle trajectory prediction method provided by the embodiments of the present application.

[0224] An embodiment of the present invention further provides a storage medium, on which a program is stored, and when the program is executed by a processor, any vehicle trajectory prediction method provided by the embodiments of the present application is implemented.

[0225] An embodiment of the present invention further provides a processor, which is used to run a program, and when the program runs, any vehicle trajectory prediction method provided by the embodiments of the present application is executed.

[0226] It should be noted that in this text, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprising", "including" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the existence of additional identical elements in the process, method, article or device comprising the said element.

[0227] Each embodiment in this specification is described in a related manner. For the same or similar parts among the embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments. In particular, for system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and reference can be made to the relevant parts of the method embodiments for the relevant content.

[0228] The above are only the preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention are all included in the protection scope of the present invention.

Claims

1. A vehicle trajectory prediction method, characterized in that, comprising: Obtaining the actual trajectory sequence of the vehicle and vehicle state quantities, and combining the time stamps corresponding to the vehicle state quantities to establish a historical driving data set for the actual trajectory sequence and the vehicle state quantities; The vehicle state quantities include at least one of steering wheel angle, yaw rate, center of mass side slip angle, longitudinal acceleration, and vehicle speed; the historical driving data set includes the actual trajectory sequences and vehicle state quantities corresponding to multiple time stamps; The historical driving data set is segmented to obtain a segmented driving data set, and the segmented driving data set is used to calculate a planned prediction trajectory sequence Λ segi ; The first cornering angle sequence δ is calculated using Ackermann geometric relationships sw_1 , and using an artificial neural network model, the path curvature sequence τ series_i , vehicle speed v x and load G are input to obtain the second cornering angle sequence δ sw_2 ; Based on the functional relationship between the fusion coefficient k and the vehicle speed v x establish the fusion coefficient k related to the vehicle speed v x and fuse the first corner sequence δ sw_1 and the second corner sequence δ sw_2 to obtain a corner prediction sequence wherein, the functional relationship between the fusion coefficient k and the vehicle speed v x is as follows: where k ∈ (0, 1), is the vehicle speed when k = 0.5, and a 1 is a calibration quantity that enables the artificial neural network model to converge fastest; Using the vehicle dynamics trajectory model in combination with the predicted steering angle sequence to obtain a first predicted trajectory, using the trailer kinematic model in combination with the planned predicted trajectory sequence Λ segi to obtain a second predicted trajectory, and fusing the first predicted trajectory and the second predicted trajectory to obtain a predicted dangerous point trajectory.

2. The method according to claim 1, characterized in that, The obtaining the actual trajectory sequence of the vehicle and vehicle state quantities, and combining the time stamps corresponding to the vehicle state quantities to establish a historical driving data set for the actual trajectory sequence and the vehicle state quantities includes: The actual trajectory sequence Λ of the positioning point O of the vehicle is measured a , where Λ a = {ξ a1 , ξ a2 , … ξ an}, ξ ai = [X ai , Y ai , t ai , X ai and Y ai are the lateral coordinate and longitudinal coordinate of the positioning point O in the global coordinate system respectively, and t ai is the timestamp of the actual trajectory sequence; Measure the vehicle state quantity and record the time stamp TimeStamps corresponding to the vehicle state quantity, where the vehicle state quantity includes the steering wheel angle δ sw , the yaw angular velocity ω of the trailer r , the sideslip angle β of the center of mass, the longitudinal acceleration a x , the vehicle speed v x and the trailer angle Based on the time stamp t of the actual trajectory sequence ai and the time relationship of the time stamp TimeStamps corresponding to the vehicle state quantity, establish the actual trajectory sequence ∧ a corresponding relationship with the time stamp TimeStamps; Based on the actual trajectory sequence Λ a , the vehicle state quantity, the time stamp TimeStamps, and the corresponding relationship, a historical driving data set Θ is established, where 3. The method according to claim 2, characterized in that, The historical driving data set is segmented to obtain a segmented driving data set, and the segmented driving data set is used to calculate a planned prediction trajectory sequence Λ segi , including: Perform a segmentation process on the historical driving dataset Θ to obtain a segmented driving dataset Θ a_segi , where: is an actual trajectory sequence, is a steering wheel angle sequence, ω ri , β i , a xi , v xi , respectively represent the yaw rate, center of mass side slip angle, longitudinal acceleration, vehicle speed, and trailer angle corresponding to the time stamp TimeStamps_i at time i; The said obtaining process includes: Obtain the actual trajectory sequence Λ in the historical driving dataset Θ a The data point ξ corresponding to the timestamp TimeStamps_i at moment i ai , where ξ ai = [X ai , Y ai , t ai . In chronological order, retrieve N data points backward from the data point ξ ai to obtain an actual trajectory sequence N is the number of path points of the predicted trajectory; The said obtaining process includes: Obtain the steering wheel angle δ in the historical driving dataset Θ sw The data point corresponding to the timestamp TimeStamps_i at moment i In chronological order, from the said data point Retrieve N data points backward to obtain a sequence of steering wheel angles N is the number of path points of the predicted trajectory.

4. The method according to claim 1, characterized in that, The first rotation angle sequence δ calculated using Ackermann geometric relationships sw_1 , includes: Obtain the predicted trajectory sequence Λ se , based on the predicted trajectory sequence Λ segi Calculate the path curvature sequence τ series_i , based on the path curvature sequence τ series_i Calculate the turning radius sequence; According to the Ackermann geometry relationship Calculate the first cornering angle sequence δ sw_1 , where R is the turning radius, I is the steering transmission ratio, and L is the vehicle wheelbase.

5. The method according to claim 1, characterized in that, Establishing a fusion coefficient related to the vehicle speed v x to fuse the first cornering sequence δ sw_1 and the second cornering sequence δ sw_2 to obtain a corner prediction sequence including: Based on the formula For the first corner angle sequence δ sw_1 and the second corner angle sequence δ sw_2 are fused to obtain a corner angle prediction sequence where k is a fusion coefficient related to the vehicle speed v x , k ∈ (0, 1), and k is positively correlated with the absolute value of the vehicle speed v x .

6. The method according to claim 1, characterized in that, The use of the vehicle dynamics trajectory model in combination with the predicted steering angle sequence to obtain a first predicted trajectory, and the use of the trailer kinematic model in combination with the planned predicted trajectory sequence Λ segi to obtain a second predicted trajectory, including: Input the corner prediction sequence the vehicle speed v x and the load G into the vehicle dynamics trajectory model, and calculate the first predicted trajectory P 1 ; Input the planned prediction trajectory sequence Λ segi into the kinematic model of the trailer to obtain the second predicted trajectory P 2 .

7. The method according to claim 1, characterized in that, The fusing the first predicted trajectory and the second predicted trajectory to obtain a predicted dangerous point trajectory includes: Based on the formula P = α·P 1 +(1 - a)·P 2 The predicted dangerous point trajectory P is calculated, where P 1 is the first predicted trajectory, and P 2 is the second predicted trajectory, and a is the confidence coefficient of the vehicle dynamics trajectory model, and a ∈ [0, 1].

8. The method according to claim 1, characterized in that, The training process of the artificial neural network model includes: Obtaining the historical trajectory sequence of the vehicle and historical vehicle state quantities, and combining the time stamps corresponding to the historical vehicle state quantities to establish a training driving data set for the historical trajectory sequence and the historical vehicle state quantities; Performing a segmentation process on the training driving data set to obtain a segmented training driving data set; Training the artificial neural network model based on the segmented training driving data set.

9. A vehicle trajectory prediction device, characterized in that, comprising: A data set establishment unit, configured to obtain the actual trajectory sequence of the vehicle and vehicle state quantities, and combine the time stamps corresponding to the vehicle state quantities to establish a historical driving data set for the actual trajectory sequence and the vehicle state quantities; The vehicle state quantities include at least one of steering wheel angle, yaw rate, center of mass side slip angle, longitudinal acceleration, and vehicle speed; the historical driving data set includes the actual trajectory sequences and vehicle state quantities corresponding to multiple time stamps; A dataset splitting unit for splitting the historical driving dataset to obtain a split driving dataset, where the split driving dataset is used to calculate a planned prediction trajectory sequence Λ segi ; A corner angle sequence obtaining unit is configured to calculate a first corner angle sequence δ by using Ackermann geometric relationships sw_1 , and by using an artificial neural network model, input a path curvature sequence τ series_i , a vehicle speed v x and a load G to obtain a second corner angle sequence δ sw_2 ; Fusion unit, used to calculate the vehicle speed v based on the fusion coefficient k x The functional relationship is established with the vehicle speed v x The relevant fusion coefficient k, for the first corner sequence δ sw_1 and the second corner sequence δ sw_2 Fusion to obtain the corner prediction sequence Among them, the fusion coefficient k and the vehicle speed v x The functional relationship is: Among them, k∈(0,1), is the vehicle speed when k = 0.5, a 1 To make the artificial neural network model converge as quickly as possible. A trajectory acquisition unit, configured to obtain a first predicted trajectory by using a vehicle dynamics trajectory model in combination with the predicted steering angle sequence and obtain a second predicted trajectory by using a trailer kinematics model in combination with the planned and predicted trajectory sequence Λ segi and fuse the first predicted trajectory and the second predicted trajectory to obtain a predicted dangerous point trajectory.

Images