Intersection guide line generation method and system based on trajectory spatiotemporal characteristics

Abstract

This invention discloses a method and system for generating intersection guide lines based on trajectory spatiotemporal features. It constructs a conditional variational autoencoder model and uses a trajectory spatiotemporal feature encoding sub-model to encode historical trajectories, obtaining corresponding historical trajectory spatiotemporal feature vectors. The spatiotemporal feature vectors of multiple historical trajectories are fused using a lane feature representation sub-model to generate lane representation vectors. Vectors of a specified dimension are sampled from a standard normal distribution, and these vectors are combined with the lane representation vectors using the trajectory spatiotemporal feature encoding sub-model. The trajectory spatiotemporal feature encoding sub-model iteratively outputs multiple trajectory point coordinates, and the connection layer uses these output trajectory point coordinates as input to predict and generate guide line trajectory point coordinates. By constructing a data-driven trajectory generation model, the generated guide line trajectory conforms to actual road traffic conditions.

Description

Technical Field

[0001] This invention relates to high-precision road scene production technology, and in particular to a method and system for generating intersection guide lines based on trajectory spatiotemporal features. Background Technology

[0002] In crowdsourced high-precision map creation, guide lines are a map element of lanes. Lane-level guide lines are safe, comfortable, and drivable trajectories used to guide autonomous vehicles through a section of lane. In regular spaces with clear road structures, lane centerlines can serve as lane guide lines; however, generating lane guide lines in free spaces at intersections where clear road structures are lacking is a very challenging task.

[0003] Lane guide lines refer to the guide lines trajectories between two connected lanes in the free space of an intersection. They are generated offline using algorithms and models, and need to meet the requirements of safe, comfortable, and feasible driving. As an element of the lane, they are pre-configured in a high-precision map.

[0004] Currently, guideline generation can be mainly divided into two approaches: one is based on mathematical optimization, and the other is based on data-driven methods. The mathematical optimization approach is commonly used in traditional trajectory planning. It models guideline generation as a non-convex mathematical optimization problem, defining an objective function and constraints such as traffic regulations and collision avoidance, and then generating the guideline through curve fitting and numerical optimization algorithms. However, this method relies on the completeness and accuracy of road elements in high-precision maps, such as lane edges and center lines. Furthermore, the trajectories generated by this method are difficult to apply to the free space of intersections. Summary of the Invention

[0005] The purpose of this invention is to overcome the above-mentioned technical deficiencies and propose a method and system for generating intersection guide lines based on trajectory spatiotemporal characteristics, thereby solving the problem that existing technologies cannot generate guide lines that conform to actual road traffic conditions in the free space of intersections.

[0006] To achieve the above-mentioned technical objectives, the first aspect of the technical solution of the present invention provides a method for generating intersection guide lines based on trajectory spatiotemporal features, which includes the following steps:

[0007] A conditional variational autoencoder model is constructed, which includes a trajectory spatiotemporal feature encoding sub-model and a lane feature representation sub-model.

[0008] Historical trajectories are encoded using a trajectory spatiotemporal feature encoding sub-model to obtain corresponding historical trajectory spatiotemporal feature vectors.

[0009] The spatiotemporal feature vectors of multiple historical trajectories are fused through the lane feature representation sub-model to generate a lane representation vector;

[0010] A vector of a specified dimension is sampled from a standard normal distribution, and the vector of the specified dimension is combined with the lane representation vector through a trajectory spatiotemporal feature encoding sub-model.

[0011] The trajectory spatiotemporal feature encoding sub-model cyclically outputs the coordinates of multiple trajectory points. The connection layer uses the output trajectory point coordinates as input to predict and generate the coordinates of the guide line trajectory points.

[0012] A second aspect of the present invention provides an intersection guide line generation system based on trajectory spatiotemporal features, which includes the following functional modules:

[0013] The model building module is used to build a conditional variational autoencoder model, which includes a trajectory spatiotemporal feature encoding sub-model and a lane feature representation sub-model.

[0014] The spatiotemporal coding module is used to encode historical trajectories using a trajectory spatiotemporal feature coding sub-model, thereby obtaining the corresponding historical trajectory spatiotemporal feature vectors.

[0015] The feature fusion module is used to fuse the spatiotemporal feature vectors of multiple historical trajectories through the lane feature representation sub-model to generate a lane representation vector;

[0016] The vector union module is used to sample vectors of a specified dimension from a standard normal distribution and unite the vectors of the specified dimension with the lane representation vectors through the trajectory spatiotemporal feature encoding sub-model.

[0017] The guide line generation module is used to cyclically output the coordinates of multiple trajectory points through the trajectory spatiotemporal feature encoding sub-model. The connection layer uses the output trajectory point coordinates as input to predict and generate the coordinates of the guide line trajectory points.

[0018] A third aspect of the present invention provides a server, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method for generating intersection guide lines based on trajectory spatiotemporal features.

[0019] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for generating intersection guide lines based on trajectory spatiotemporal features.

[0020] Compared with existing technologies, the intersection guide line generation method and system based on trajectory spatiotemporal features of the present invention constructs a conditional variational autoencoder model and uses a trajectory spatiotemporal feature encoding sub-model to encode historical trajectories, obtaining corresponding historical trajectory spatiotemporal feature vectors respectively; the spatiotemporal feature vectors of multiple historical trajectories are fused through a lane feature representation sub-model to generate a lane representation vector; vectors of a specified dimension are sampled from a standard normal distribution, and the specified dimension vectors are combined with the lane representation vectors through the trajectory spatiotemporal feature encoding sub-model; multiple trajectory point coordinates are cyclically output through the trajectory spatiotemporal feature encoding sub-model, and the connection layer uses the output trajectory point coordinates as input to predict and generate guide line trajectory point coordinates; by constructing a data-driven trajectory generation model, the generated guide line trajectory conforms to the actual road traffic conditions. Attached Figure Description

[0021] Figure 1 This is a flowchart of the intersection guide line generation method based on trajectory spatiotemporal features according to an embodiment of the present invention;

[0022] Figure 2 This is a training graph of the conditional variational autoencoder model in the intersection guide line generation method based on trajectory spatiotemporal features described in the embodiments of the present invention;

[0023] Figure 3 This is a derivation diagram of the conditional variational autoencoder model in the intersection guide line generation method based on trajectory spatiotemporal features described in the embodiments of the present invention;

[0024] Figure 4 This is a schematic diagram of an intersection guide line generated by the intersection guide line generation method based on trajectory spatiotemporal features as described in an embodiment of the present invention;

[0025] Figure 5 This is a block diagram of the intersection guide line generation system based on trajectory spatiotemporal features as described in an embodiment of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0027] like Figure 1 As shown, an embodiment of the present invention provides a method for generating intersection guide lines based on trajectory spatiotemporal features, which includes the following steps:

[0028] S1. Construct a conditional variational autoencoder model, which includes a trajectory spatiotemporal feature encoding sub-model and a lane feature representation sub-model.

[0029] S2. The historical trajectory is encoded using the trajectory spatiotemporal feature encoding sub-model to obtain the corresponding historical trajectory spatiotemporal feature vectors.

[0030] S3. The spatiotemporal feature vectors of multiple historical trajectories are fused through the lane feature representation sub-model to generate a lane representation vector;

[0031] S4. Sample vectors of a specified dimension from the standard normal distribution, and combine the vectors of the specified dimension with the lane representation vectors through the trajectory spatiotemporal feature encoding sub-model;

[0032] S5. The trajectory spatiotemporal feature encoding sub-model outputs multiple trajectory point coordinates in a loop. The connection layer uses the output trajectory point coordinates as input to predict and generate the guide line trajectory point coordinates.

[0033] The guide line described in this invention is a trajectory composed of several trajectory points, which can be represented as T = {p1, p2, ..., p...} n}, where p i Let p represent the i-th trajectory point. To simplify the implementation of the method, this invention only considers the longitude and latitude characteristics of the trajectory point, and does not consider characteristics such as speed and heading angle. i Represented as (x i ,y i ), x i ,y i They represent p respectively i The longitude and latitude values. The generation of the guide line can be established by the mathematical model of the following formula (1):

[0034] (x i+1 ,y i+1 )=g(z,V,(x i ,y i )) Formula (1)

[0035] Where z is a noise vector of a specified dimension randomly generated from a standard normal distribution, used to enhance the robustness of the model. V is a vectorized representation of lane structure and driving behavior, also known as the lane representation vector; g is a prediction function for predicting the position of the trajectory point. The physical meaning of formula (1) is that the function g will predict the position of the current trajectory point (x i ,y i Using the lane representation vector V and the noise vector z as input, predict the position (x) of the next point. i+1 ,y i+1 ).

[0036] This invention considers the strong feature extraction and function fitting capabilities of DNN models and implements a conditional variational autoencoder (CVAE)-based model. Therefore, a conditional variational autoencoder model is constructed; for example... Figure 2 As shown, the Conditional Variational Autoencoder (CVAE) model includes a trajectory spatiotemporal feature encoding sub-model (ST-block module), a lane feature representation sub-model (Fusion-block), and a trajectory decoding model. The trajectory spatiotemporal feature encoding sub-model consists of fully connected layers (FC layers), convolutional layers (CONV layers), and long short-term memory artificial neural network layers (LSTM layers), used to extract high-level spatiotemporal vectors with semantic features. The lane feature representation sub-model (Fusion block) is used to extract lane representation vectors that represent lane structure and driving behavior. Multiple historical traffic trajectories exist within a lane, each providing information to the lane representation vector. The lane representation vector is obtained by fusing the spatiotemporal feature vectors of multiple trajectories. The trajectory decoding model decodes the input feature vector into a sequence of trajectory points; this model is an autoregressive model that relies on the previously predicted trajectory point positions and lane representation vectors.

[0037] Before generating guide lines using the Conditional Variational Autoencoder (CVAE) model, it needs to be trained first, such as... Figure 2 As shown, the specific training process is as follows:

[0038] Training and test sets are constructed based on all historical trajectories within the two connecting lanes of the intersection, and the conditional variational autoencoder model is trained accordingly. The format of a single data sample is ({C1,...,C...). i ,...,C m},X), where C i X represents all historical traffic trajectories within the two connected lanes of the intersection, and X represents the guide line trajectory. The entire dataset is divided into a training set and a test set in a 4:1 ratio.

[0039] Before inputting data samples into the Conditional Variational Autoencoder (CVAE) model, the original trajectory points need to be transformed into a data format suitable for input to the neural network model. The longitude and latitude in the original trajectory are Earth coordinate values, which cannot be used as model input. Therefore, the global Earth coordinate values ​​need to be converted into local coordinate values ​​or other equivalent representations. This invention designs the following three coordinate transformation methods:

[0040] 1) Local coordinate transformation in the same direction as the global coordinate system, as shown in formula (2), where

[0041]

[0042] Where x and y represent longitude and latitude values ​​respectively, the superscripts l and g represent local and global coordinates respectively, and the subscript i represents the i-th sampling point.

[0043] 2) The direction in which the vehicle enters the intersection is the positive y-axis, and the right-hand side of the y-axis is a Cartesian coordinate system with the positive y-axis direction, as shown in formulas (3) and (4). This method requires rotating the coordinates.

[0044] x new =x old cosθ-y old sinθ formula (3)

[0045] y new =x old sinθ+y old cosθ formula (4)

[0046] Global coordinates are converted to local coordinates using the method described in 1), where x new ,y new Let x represent the longitude and latitude of the new local coordinates, respectively. old ,y old These represent the longitude and latitude of the old local coordinates, respectively.

[0047] 3) Trajectory behavior sequence, that is, the behavior coordinate of the current sampling point is its coordinate value minus the coordinate value of the previous sampling point, as shown in formula (5), where the meaning of the letter symbols is the same as in 1), and the behavior coordinate of the first sampling point is itself minus.

[0048]

[0049] In this embodiment of the invention, formulas (1) and (2) are used to translate the global Earth coordinate values ​​to a local coordinate system in the same direction with the first point of the trajectory as the origin, for use as input data for the model.

[0050] The guide line trajectory X is encoded using the ST-block module to obtain its spatiotemporal feature vector V. X .

[0051] The ST-block module is used to analyze other historical traffic trajectories within the lane. i Encoding yields the corresponding spatiotemporal feature vectors V. i Specifically, the spatial features of historical travel trajectories are extracted through the CONV layer of the ST-block module; then, the spatial features of historical travel trajectories are used as input through the LSTM layer of the ST-block module to extract the spatiotemporal features of historical travel trajectories, resulting in the spatiotemporal feature vector V of historical travel trajectories. i .

[0052] The spatiotemporal feature vectors V of multiple trajectories i Lane representation vector V is generated through Fusion-block fusion. s .

[0053] V X and V s Together, a multilayer FC network is used to predict the mean μ and the logarithm of the squared variance logσ of a prior multivariate normal distribution. 2 .

[0054] Calculate the prior distribution of the sample, i.e., assume the sample follows a normal distribution N(μ, σ). 2 The mean and variance of the normal distribution are calculated. During the training of CVAE, the output of the encoding stage is the mean and variance of the normal distribution, which are calculated using the FC layer. Specifically, a vector of a specified dimension is sampled from the standard normal distribution ∈ ~ N(0,1), with the dimension consistent with the dimension of the prior multivariate normal distribution. The vector z = ∈ × σ + μ is calculated as the input for trajectory decoding, and the specified dimension vector is combined with the lane representation vector through the FC layer of the ST-block module.

[0055] The LSTM network of the ST-block module outputs multiple values ​​in a loop, and the FC layer uses the output of the LSTM as input to predict the generated trajectory points.

[0056] During the training of the Conditional Variational Autoencoder (CVAE) model, loss functions are defined using the mini-batch BP algorithm and the Adam algorithm, and the loss functions are used to correct the CVAE model.

[0057] The formula for calculating the loss function is as follows:

[0058]

[0059] After the Conditional Variational Autoencoder (CVAE) model is trained, as follows: Figure 3 As shown, the process of generating guide lines using the Conditional Variational Autoencoder (CVAE) model is as follows:

[0060] The historical traffic trajectories within the two connecting lanes of the intersection are used as model input, and the global coordinate values ​​of the trajectory points are transformed into local coordinate values. Specifically, the global Earth coordinate values ​​are translated into a local coordinate system in the same direction with the first point of the trajectory as the origin.

[0061] Historical travel trajectory C is analyzed using the ST-block module. i Encode the corresponding spatiotemporal feature vectors V. i .

[0062] The spatiotemporal feature vectors V of multiple historical travel trajectories i Lane representation vector V is generated through Fusion-block fusion. s ;

[0063] A vector z of a specified dimension is sampled from the standard normal distribution z ~ N(0,1). The vector z is then passed through an FC layer and interposed with the lane representation vector V. s joint.

[0064] The LSTM network iteratively outputs multiple values. The FC layer takes the LSTM output as input to predict the generated guide line trajectory points. These guide line trajectory points are then combined to form the intersection guide line. Figure 4 As shown.

[0065] This invention constructs a conditional variational autoencoder model and uses a trajectory spatiotemporal feature encoding sub-model to encode historical trajectories, obtaining corresponding historical trajectory spatiotemporal feature vectors. The spatiotemporal feature vectors of multiple historical trajectories are fused using a lane feature representation sub-model to generate a lane representation vector. Vectors of a specified dimension are sampled from a standard normal distribution, and these vectors are combined with the lane representation vectors using the trajectory spatiotemporal feature encoding sub-model. Multiple trajectory point coordinates are cyclically output by the trajectory spatiotemporal feature encoding sub-model, and the connection layer uses these output trajectory point coordinates as input to predict and generate guide line trajectory point coordinates. By constructing a data-driven trajectory generation model, the generated guide line trajectory conforms to actual road traffic conditions.

[0066] like Figure 5 As shown in the figure, this invention also discloses an intersection guide line generation system based on trajectory spatiotemporal features, which includes the following functional modules:

[0067] Model building module 10 is used to build a conditional variational autoencoder model, which includes a trajectory spatiotemporal feature encoding sub-model and a lane feature representation sub-model.

[0068] The spatiotemporal coding module 20 is used to encode historical trajectories using a trajectory spatiotemporal feature coding sub-model to obtain corresponding historical trajectory spatiotemporal feature vectors.

[0069] The feature fusion module 30 is used to fuse the spatiotemporal feature vectors of multiple historical trajectories through the lane feature representation sub-model to generate a lane representation vector;

[0070] The vector union module 40 is used to sample vectors of a specified dimension from a standard normal distribution and to unite the vectors of the specified dimension with the lane representation vectors through the trajectory spatiotemporal feature encoding sub-model.

[0071] The guide line generation module 50 is used to cyclically output multiple trajectory point coordinates through the trajectory spatiotemporal feature encoding sub-model. The connection layer uses the output trajectory point coordinates as input to predict and generate guide line trajectory point coordinates.

[0072] The execution method of the intersection guide line generation system based on trajectory spatiotemporal features in this embodiment is basically the same as the intersection guide line generation method based on trajectory spatiotemporal features described above, so it will not be described in detail.

[0073] In this embodiment, the server is a device that provides computing services, typically referring to a computer with high computing power that is provided to multiple consumers via a network. The server in this embodiment includes a memory, a processor, and a system bus. The memory includes executable programs stored thereon. Those skilled in the art will understand that the terminal device structure of this embodiment does not constitute a limitation on the terminal device and may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0074] The memory can be used to store software programs and modules. The processor executes various terminal functions and data processing by running the software programs and modules stored in the memory. The memory can mainly include a program storage area and a data storage area. The program storage area can store the operating system, application programs required for at least one function (such as sound playback, image playback, etc.), etc.; the data storage area can store data created according to the use of the terminal (such as audio data, phonebook, etc.). In addition, the memory can include high-speed random access memory, and can also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0075] An executable program for generating intersection guide lines based on trajectory spatiotemporal features is contained in a memory. This executable program can be divided into one or more modules / units, which are stored in the memory and executed by a processor to complete the information acquisition and implementation process. Each module / unit can be a series of computer program instruction segments capable of performing a specific function, describing the execution process of the computer program on the server. For example, the computer program can be divided into a model building module 10, a spatiotemporal encoding module 20, a feature fusion module 30, a vector union module 40, and a guide line generation module 50.

[0076] The processor is the control center of the server, connecting various parts of the terminal device through various interfaces and lines. It performs various terminal functions and processes data by running or executing software programs and / or modules stored in memory, and by calling data stored in memory, thereby providing overall monitoring of the terminal. Optionally, the processor may include one or more processing units; preferably, the processor may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, applications, etc., and the modem processor mainly handles wireless communication. It is understood that the modem processor may also not be integrated into the processor.

[0077] The system bus connects various functional components within a computer, transmitting data, address, and control information. Types of system buses include PCI, ISA, and VESA. Processor instructions are transmitted to memory via the bus, and memory sends data back to the processor. The system bus handles the data and instruction exchange between the processor and memory. Of course, the system bus can also connect to other devices, such as network interfaces and display devices.

[0078] The server should include at least a CPU, chipset, memory, and disk system; other components will not be described in detail here.

[0079] In this embodiment of the invention, the executable program executed by the processor included in the terminal is specifically: a method for generating intersection guide lines based on trajectory spatiotemporal features, which includes the following steps:

[0080] A conditional variational autoencoder model is constructed, which includes a trajectory spatiotemporal feature encoding sub-model and a lane feature representation sub-model.

[0081] Historical trajectories are encoded using a trajectory spatiotemporal feature encoding sub-model to obtain corresponding historical trajectory spatiotemporal feature vectors.

[0082] The spatiotemporal feature vectors of multiple historical trajectories are fused through the lane feature representation sub-model to generate a lane representation vector;

[0083] A vector of a specified dimension is sampled from a standard normal distribution, and the vector of the specified dimension is combined with the lane representation vector through a trajectory spatiotemporal feature encoding sub-model.

[0084] The trajectory spatiotemporal feature encoding sub-model cyclically outputs the coordinates of multiple trajectory points. The connection layer uses the output trajectory point coordinates as input to predict and generate the coordinates of the guide line trajectory points.

[0085] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0086] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0087] Those skilled in the art will recognize that the modules, units, and / or method steps of the various embodiments described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0088] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features thereof can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating intersection guide lines based on trajectory spatiotemporal features, characterized in that, Includes the following steps: A conditional variational autoencoder model is constructed, which includes a trajectory spatiotemporal feature encoding sub-model and a lane feature representation sub-model. Historical trajectories are encoded using a trajectory spatiotemporal feature encoding sub-model to obtain corresponding historical trajectory spatiotemporal feature vectors. The spatiotemporal feature vectors of multiple historical trajectories are fused through the lane feature representation sub-model to generate a lane representation vector; A vector of a specified dimension is sampled from a standard normal distribution, and the vector of the specified dimension is combined with the lane representation vector through a trajectory spatiotemporal feature encoding sub-model. The trajectory spatiotemporal feature encoding sub-model cyclically outputs the coordinates of multiple trajectory points. The connection layer uses the output trajectory point coordinates as input to predict and generate the coordinates of the guide line trajectory points.

2. The intersection guide line generation method based on trajectory spatiotemporal features according to claim 1, characterized in that, The conditional variational autoencoder model is trained based on all historical trajectories within the two connecting lanes of the intersection.

3. The intersection guide line generation method based on trajectory spatiotemporal features according to claim 2, characterized in that, Before training a conditional variational autoencoder model based on all historical trajectories within two connected lanes at an intersection, it is necessary to first convert the trajectory points of the historical trajectories from global coordinate values ​​to local coordinate values.

4. The intersection guide line generation method based on trajectory spatiotemporal features according to claim 3, characterized in that, The process of converting the trajectory points of the historical trajectory from global coordinate values ​​to local coordinate values ​​specifically involves shifting the global Earth coordinate values ​​of the trajectory points in the historical trajectory to a local coordinate system in the same direction with the first point of the historical trajectory as the origin.

5. The intersection guide line generation method based on trajectory spatiotemporal features according to claim 2, characterized in that, During the training of the conditional variational autoencoder model, loss functions are defined using the mini-batch BP algorithm and the Adam algorithm, and the conditional variational autoencoder model is corrected using the loss functions.

6. The intersection guide line generation method based on trajectory spatiotemporal features according to claim 1, characterized in that, The trajectory spatiotemporal feature encoding sub-model consists of fully connected layers, convolutional layers, and long short-term memory artificial neural network layers; and In the process of combining a vector of a specified dimension with a lane representation vector through a trajectory spatiotemporal feature encoding sub-model, the vector of the specified dimension is combined with the lane representation vector through a fully connected layer of the trajectory spatiotemporal feature encoding sub-model. In the process of cyclically outputting multiple values ​​through the trajectory spatiotemporal feature encoding sub-model, the specific method is to cyclically output multiple values ​​through the long short-term memory artificial neural network of the trajectory spatiotemporal feature encoding sub-model.

7. The intersection guide line generation method based on trajectory spatiotemporal features according to claim 6, characterized in that, The historical trajectory is encoded using a trajectory spatiotemporal feature encoding sub-model to obtain corresponding historical trajectory spatiotemporal feature vectors; specifically including: Spatial features of historical trajectories are extracted through the convolutional layer of the trajectory spatiotemporal feature encoding sub-model; By using the long short-term memory artificial neural network layer of the trajectory spatiotemporal feature encoding sub-model as input, the spatiotemporal features of the historical trajectory are extracted to obtain the spatiotemporal feature vector of the historical trajectory.

8. A system for generating intersection guide lines based on trajectory spatiotemporal characteristics, characterized in that, Includes the following functional modules: The model building module is used to build a conditional variational autoencoder model, which includes a trajectory spatiotemporal feature encoding sub-model and a lane feature representation sub-model. The spatiotemporal coding module is used to encode historical trajectories using a trajectory spatiotemporal feature coding sub-model, thereby obtaining the corresponding historical trajectory spatiotemporal feature vectors. The feature fusion module is used to fuse the spatiotemporal feature vectors of multiple historical trajectories through the lane feature representation sub-model to generate a lane representation vector; The vector union module is used to sample vectors of a specified dimension from a standard normal distribution and unite the vectors of the specified dimension with the lane representation vectors through the trajectory spatiotemporal feature encoding sub-model. The guide line generation module is used to cyclically output the coordinates of multiple trajectory points through the trajectory spatiotemporal feature encoding sub-model. The connection layer uses the output trajectory point coordinates as input to predict and generate the coordinates of the guide line trajectory points.

9. A server comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the intersection guide line generation method based on trajectory spatiotemporal features as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the intersection guide line generation method based on trajectory spatiotemporal features as described in any one of claims 1 to 7.

Images