Intersection guide line generation method and system based on online learning, and storage medium

Abstract

The application discloses a kind of intersection guide line generation method, system and storage medium based on online learning, its method includes the following steps: obtaining the historical traffic trajectory and historical guide line trajectory in target road intersection;The trajectory point of the historical traffic trajectory is converted with the trajectory point of the historical guide line trajectory Coordinate value;Based on conditional variational autoencoder, the historical traffic trajectory and the historical guide line trajectory after coordinate conversion are network trained, and the target guide line trajectory is obtained;Based on Carla simulation platform, the trajectory point of the target guide line trajectory is adjusted, and the final guide line trajectory is obtained;It can automatically adapt to new intersection scene and plan trajectory, and a reasonable controllable guide line can be planned for two connected lanes in the case of no clear road structure in road intersection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of high-precision map making, and in particular relates to a method and system for generating intersection guide lines based on online learning, and a storage medium. BACKGROUND

[0002] In crowd-sourced high-precision map making, a guide line is a map element of a lane and is a safe, comfortable and drivable trajectory that can be used to guide an intelligent driving vehicle through the lane; the guide line is a static guide line in motion planning and is usually generated offline and preset in a high-precision map. In a lane with clear and accurate ordinary and structured features, the lane center line is usually taken as the guide line of the lane, but it is a very challenging task to plan a guide line for two connected lanes in a road intersection without clear road structures.

[0003] Current methods for generating intersection guide lines can be mainly divided into two categories. One category is a method based on a mathematical model, which models a non-convex mathematical optimization problem, formulates an optimization objective function and constraint conditions such as traffic rules and collision avoidance, and generates a safe and comfortable guide line through curve fitting and numerical optimization algorithms. This method heavily depends on the completeness and accuracy of road elements in the map, such as lane boundary lines and lane center lines, and is not suitable for free space with unclear road structures, and the generated trajectory is difficult to meet the comfort of humans and the followability of vehicle control. The other category is a learning method based on data driving, represented by a deep neural network, which constructs a large-scale trajectory data set, trains a trajectory planning model, and generates a guide line trajectory that is more consistent with human driving behavior. However, the current offline training model based on the neural network planning model has a slow model update cycle and is difficult to adapt to new intersection scenarios in high-precision map making. SUMMARY

[0004] The present application provides a method and system for generating intersection guide lines based on online learning, which can automatically adapt to new intersection scenarios for trajectory planning and plan a reasonable and controllable guide line for two connected lanes in a road intersection without clear road structures.

[0005] In a first aspect, a method for generating intersection guide lines based on online learning is provided, comprising the following steps:

[0006] acquiring historical passing trajectories and historical guide line trajectories in a target road intersection;

[0007] converting coordinate values of trajectory points of the historical passing trajectories and trajectory points of the historical guide line trajectories;

[0008] training the historical passing trajectory and the historical guide line trajectory after the coordinate conversion based on a conditional variational autoencoder to obtain a target guide line trajectory;

[0009] adjusting the trajectory points of the target guide line trajectory based on a Carla simulation platform to obtain a final guide line trajectory.

[0010] According to the first aspect, in a first possible implementation manner of the first aspect, the step of "converting the trajectory points of the historical passing trajectory and the trajectory points of the historical guide line trajectory in coordinate values" specifically includes the following steps:

[0011] converting the global coordinate values of the trajectory points of the historical passing trajectory and the global coordinate values of the trajectory points of the historical guide line trajectory into local coordinate values, respectively; or,

[0012] converting the global coordinate values of the trajectory points of the historical passing trajectory and the global coordinate values of the trajectory points of the historical guide line trajectory into local coordinate values, respectively, and rotating a local coordinate system in which the local coordinate values of the trajectory points are located based on a preset rotation angle; or,

[0013] converting the global coordinate values of the trajectory points of the historical passing trajectory and the global coordinate values of the trajectory points of the historical guide line trajectory into behavior coordinate values.

[0014] According to the first aspect, in a second possible implementation manner of the first aspect, the step of "training the historical passing trajectory and the historical guide line trajectory after the coordinate conversion based on a conditional variational autoencoder to obtain a target guide line trajectory" specifically includes the following steps:

[0015] encoding feature vectors of the historical passing trajectory and the historical guide line trajectory in the conditional variational autoencoder;

[0016] predicting a prior normal distribution of the historical passing trajectory and the historical guide line trajectory after the feature vector encoding;

[0017] decoding the historical passing trajectory and the prior normal distribution result after the feature vector encoding to obtain a target guide line trajectory.

[0018] According to the second possible implementation manner of the first aspect, in a third possible implementation manner of the first aspect, the step of "encoding feature vectors of the historical passing trajectory and the historical guide line trajectory" specifically includes the following steps:

[0019] extracting space-time feature vectors of the historical passing trajectory and the historical guide line trajectory based on a trajectory space-time feature encoding sub-model, to correspondingly obtain passing space-time feature vectors and guide line space-time feature vectors;

[0020] The lane representation vector is obtained based on the lane feature representation sub-model and all the traffic space-time feature vectors of the historical traffic trajectory.

[0021] According to a fourth possible implementation of the first aspect, in the fourth possible implementation of the first aspect, the step of "encoding the feature vector of the historical traffic trajectory and the historical guide line trajectory to predict a prior normal distribution" specifically includes the following steps:

[0022] The mean and variance of the prior normal distribution are predicted based on the full connection layer neural network combined with the guide line space-time feature vector and the lane representation vector, and a sampling vector satisfying a preset dimension condition in the standard normal distribution is selected;

[0023] The trajectory decoding input vector is obtained according to the mean, the variance and the sampling vector.

[0024] According to a fifth possible implementation of the first aspect, in the fourth possible implementation of the first aspect, the step of "trajectory decoding the historical traffic trajectory with the encoded feature vector and the prior normal distribution result to obtain a target guide line trajectory" specifically includes the following steps:

[0025] The trajectory decoding input vector and the lane representation vector are combined based on the full connection layer neural network;

[0026] The combined trajectory decoding input vector and the lane representation vector are output recurrently trained based on the long short-term memory neural network;

[0027] The output value of the long short-term memory neural network is trained and the target guide line trajectory is predicted based on the full connection layer neural network.

[0028] According to the first aspect, in a sixth possible implementation of the first aspect, the step of "adjusting the trajectory points of the target guide line trajectory based on the Carla simulation platform to obtain a final guide line trajectory" specifically includes the following steps:

[0029] The trajectory points of the target guide line trajectory are subjected to preset condition constraints based on the MPC model to obtain a constraint guide line trajectory;

[0030] The similarity between the trajectory points of the target guide line trajectory and the trajectory points of the constraint guide line trajectory is judged based on the twin network, and the trajectory points of the target guide line trajectory are adjusted according to the similarity to obtain a final guide line trajectory.

[0031] In a seventh possible implementation manner of the first aspect, in the sixth possible implementation manner of the first aspect, the step of adjusting the trajectory points of the target guide line trajectory according to the similarity and obtaining a final guide line trajectory specifically comprises the following steps:

[0032] When it is judged that one of the trajectory points of the target guide line trajectory is similar to one of the trajectory points of the corresponding constraint guide line trajectory, the one of the trajectory points of the target guide line trajectory is the trajectory point of the final guide line trajectory.

[0033] When it is judged that one of the trajectory points of the target guide line trajectory is not similar to one of the trajectory points of the corresponding constraint guide line trajectory, the one of the trajectory points of the constraint guide line trajectory is the trajectory point of the final guide line trajectory.

[0034] In a second aspect, a guide line generation system for a road intersection based on online learning is provided, comprising:

[0035] a historical trajectory acquisition module configured to acquire historical passing trajectories and historical guide line trajectories in a target road intersection;

[0036] a coordinate conversion module in communication connection with the historical trajectory acquisition module and configured to convert coordinate values of trajectory points of the historical passing trajectories and trajectory points of the historical guide line trajectories;

[0037] a target guide line trajectory acquisition module in communication connection with the coordinate conversion module and configured to perform network training on the historical passing trajectories and the historical guide line trajectories after the coordinate conversion based on a conditional variational autoencoder (CVAE) to obtain a target guide line trajectory;

[0038] a final guide line trajectory acquisition module in communication connection with the target guide line trajectory acquisition module and configured to adjust trajectory points of the target guide line trajectory based on a Carla simulation platform to obtain a final guide line trajectory.

[0039] In a third aspect, a computer readable storage medium having a computer program stored thereon is provided, and the computer program is executed by a processor to implement the guide line generation method for a road intersection based on online learning as described in any one of the above aspects.

[0040] Compared with the prior art, the advantages of the present application are as follows: the trajectory generation model based on CVAE is data-driven, can automatically adapt to new intersection scenarios for trajectory planning, and can plan a reasonable and controllable guide line for two connected lanes in the case of no clear road structure at a road intersection. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 is a flowchart of an embodiment of a guide line generation method for a road intersection based on online learning.

[0042] Figure 2 is a flowchart of a process for generating a target guide line trajectory based on a CVAE model according to the present application;

[0043] Figure 3 is a structural diagram of a guide line generation system for a road intersection based on online learning according to the present application. BRIEF DESCRIPTION OF DRAWINGS

[0045] 100, a guide line generation system for a road intersection based on online learning; 110, a historical trajectory acquisition module; 120, a coordinate conversion module; 130, a target guide line trajectory acquisition module; 140, a final guide line trajectory acquisition module. DETAILED DESCRIPTION

[0046] Reference will now be made in detail to the present application, examples of which are illustrated in the accompanying drawings. While the present application will be described in conjunction with the specific embodiments, it will be understood that the present application is not limited to the embodiments set forth herein but instead covers all alternatives, modifications, and equivalents falling within the spirit and scope of the present application as defined by the appended claims. It should be noted that the method steps described herein can all be implemented by any functional block or functional arrangement, and any functional block or functional arrangement can be implemented as a physical entity or a logical entity, or a combination of both.

[0047] In order for those skilled in the art to better understand the present application, the present application will be further described in detail below in conjunction with the drawings and specific embodiments.

[0048] Note: The examples to be introduced next are only specific examples and are not intended to limit the embodiments of the present application to the specific steps, values, conditions, data, sequences, etc. Those skilled in the art can use the concept of the present application to construct more embodiments not mentioned in the present specification by reading the present specification.

[0049] Referring to Figure 1 The embodiments of the present application provide a guide line generation method for a road intersection based on online learning, as shown in the drawings, comprising the following steps:

[0050] S100, acquiring historical passing trajectories in a target road intersection and historical guide line trajectories X;

[0051] S200, performing coordinate value conversion on the trajectory points of the historical passing trajectories and the trajectory points of the historical guide line trajectories;

[0052] S300, performing network training on the historical passing trajectories and the historical guide line trajectories after coordinate conversion based on a conditional variational autoencoder to acquire target guide line trajectories;

[0053] S400, adjusting the track points of the target guide line track based on the Carla simulation platform to obtain a final guide line track.

[0054] Specifically, in the embodiment, since the existing data-driven track planning mainly adopts a neural network model, offline training needs to construct a scene data set in advance, and the production cost is high. In view of the above problems, the application designs a track planning model based on a conditional variational autoencoder (CVAE), and the basic mathematical form is described as: The meaning of the formula is that the guide line track generated by the function is obtained under the condition that , , , The point sequence composed of latitude and longitude, is a random vector of fixed dimension subject to Gaussian distribution, is a semantic vector of lane driving behavior extracted by the model, is a generation model. Therefore, the application is a data-driven track generation model based on CVAE, which can automatically adapt to new intersection scenarios for track planning, and can plan a reasonable and controllable guide line for two connected lanes in the case of no clear road structure at the road intersection.

[0055] Preferably, in another embodiment of the application, the step of “S200, converting the track points of the historical passing track and the track points of the historical guide line track into coordinate values” specifically includes the following steps:

[0056] S210, respectively converting the global coordinate values of the track points of the historical passing track and the global coordinate values of the track points of the historical guide line track into local coordinate values; or,

[0057] S220, respectively converting the global coordinate values of the track points of the historical passing track and the global coordinate values of the track points of the historical guide line track into local coordinate values, and rotating the local coordinate system in which the local coordinate values of the track points are located based on a preset rotation angle; or,

[0058] S230, respectively converting the global coordinate values of the track points of the historical passing track and the global coordinate values of the track points of the historical guide line track into behavior coordinate values.

[0059] Specifically, in this embodiment, the longitude and latitude of the trajectory points of the historical travel trajectory are Earth coordinate values, which cannot be used as input to the CVAE model. It is necessary to convert the global Earth coordinate values ​​into local coordinate values ​​or other equivalent (behavioral coordinates) representations. This patent designs three methods for representing trajectory points:

[0060] S210, the local coordinate transformation in the same direction as the global coordinate system is as follows:

[0061]

[0062] in These represent longitude and latitude values ​​respectively, with superscripts. Representing local and global coordinates respectively, subscripts Indicates the first There are several trajectory points, with the subscript 0 indicating the first sampling point of the trajectory.

[0063] S220, the direction for vehicles entering the intersection is... Positive axis direction The right side of the axis is The Cartesian coordinate system in the positive direction of the axis, and the local coordinate system based on the local coordinate values ​​of the rotation trajectory points at the preset rotation angle, are transformed as follows:

[0064]

[0065] These represent the longitude and latitude of the new local coordinates, respectively. These represent the longitude and latitude of the old local coordinates (local coordinate values ​​converted using the S210 conversion formula). This is the preset rotation angle.

[0066] S230, the behavior coordinates of the current sampling point are its coordinates minus the coordinates of the previous sampling point, as shown in the following formula, where the meanings of the letters are the same as those in the above formula, and the behavior coordinates of the first trajectory point are subtracted from itself;

[0067]

[0068] See Figure 2 As shown, preferably, in another embodiment of this application, the step "S300, performing network training on the historical travel trajectory and the historical guide line trajectory after coordinate transformation based on a conditional variational autoencoder to obtain the target guide line trajectory" specifically includes the following steps:

[0069] S310, In the conditional variational autoencoder, feature vector encoding is performed on the historical travel trajectory and the historical guide line trajectory;

[0070] The trajectory point is encoded into a high-dimensional feature vector, i.e., the trajectory point is a two-dimensional independent feature of longitude and latitude, which needs to be encoded into a higher-dimensional feature vector to improve the performance of the CVAE model.

[0071] S320, the history passing trajectory and the history guide line trajectory after the feature vector encoding are predicted to be normal distribution;

[0072] S330, the history passing trajectory and the prior normal distribution result after the feature vector encoding are trajectory decoded to obtain the target guide line trajectory;

[0073] The trajectory decoding model is to decode the input feature vector into a point sequence of the trajectory.

[0074] Preferably, in another embodiment of the application, the step of "S310, feature vector encoding of the history passing trajectory and the history guide line trajectory" specifically comprises the following steps:

[0075] S311, extracting the spatial-temporal feature vector of the history passing trajectory and the history guide line trajectory based on the trajectory spatial-temporal feature encoding sub-model, corresponding to obtaining the passing spatial-temporal feature vector and the guide line spatial-temporal feature vector;

[0076] S312, fusing all the passing spatial-temporal feature vectors of the history passing trajectory based on the lane feature representation sub-model to obtain a lane representation vector.

[0077] Specifically, in this embodiment, the trajectory spatial-temporal feature encoding sub-model (Spatial-Temporal Block, STblock) is to extract the spatial-temporal feature vector of the trajectory, and the point sequence of the trajectory has spatial features and time sequence features; the STblock is composed of a fully connected neural network (Fully Connected, FC) FC layer, a convolution layer (Convolution Layer, CONV), and a long short-term memory neural network (Long Short-Term Memory, LSTM).

[0078] The history guide line trajectory is encoded based on the trajectory spatial-temporal feature encoding sub-model ST-block to obtain the guide line spatial-temporal feature vector of the history guide line trajectory .The history passing trajectory in the lane is encoded by the trajectory spatial-temporal feature encoding sub-model ST-block to obtain the corresponding passing spatial-temporal feature vector . .

[0079] The lane feature representation sub-model Fusion block, i.e., extracting a lane representation vector capable of representing a lane structure and driving behavior. There are multiple historical passing trajectories in the lane, and each trajectory provides a certain amount of information to the lane representation vector. The spatiotemporal feature vectors of multiple trajectories are fused to obtain the lane representation vector.

[0080] Therefore, all the passing spatiotemporal feature vectors of the historical passing trajectories The lane representation vector is generated by fusion through the Fusion-block .

[0081] Preferably, in another embodiment of the present application, the step of "S320, predicting a prior normal distribution of the historical passing trajectory and the historical guide line trajectory after encoding the feature vector" specifically comprises the following steps:

[0082] S321, based on a fully connected layer neural network, jointly predicting the mean and variance of the prior normal distribution, and selecting a sampling vector in the standard normal distribution that meets a preset dimension condition, based on the guide line spatiotemporal feature vector and the lane representation vector;

[0083] S322, obtaining a trajectory decoding input vector according to the mean, the variance, and the sampling vector.

[0084] Specifically, in the present embodiment, the prior distribution of the sample is calculated, i.e., assuming that the sample is subject to a normal distribution , the mean and variance of the normal distribution are calculated; in the training process of the CVAE, the output of the encoding stage is the mean and variance of the normal distribution, which is calculated through the FC layer. Therefore, based on the fully connected layer neural network FC, the guide line spatiotemporal feature vector and the lane representation vector , the mean and the variance square of the prior multivariate normal distribution are predicted, a vector of a preset dimension is sampled from the standard normal distribution , the dimension is consistent with the dimension of the prior multivariate normal distribution, and a trajectory decoding input vector is calculated as the input of trajectory decoding.

[0085] Preferably, in another embodiment of the present application, the step of "S330, trajectory decoding the historical passing trajectory after encoding the feature vector and the prior normal distribution result to obtain a target guide line trajectory" specifically comprises the following steps:

[0086] S331, based on a fully connected layer neural network, jointly predicting the mean and variance of the prior normal distribution, and selecting a sampling vector in the standard normal distribution that meets a preset dimension condition, based on the guide line spatiotemporal feature vector and the lane representation vector;

[0087] S332, based on the long short-term memory neural network, the combined trajectory decoding input vector and the lane representation vector are cyclically output trained;

[0088] S333, based on the full connection layer neural network, the output value of the long short-term memory neural network is trained and the target guide line trajectory is predicted.

[0089] Specifically, in the embodiment, based on the full connection layer neural network FC, the trajectory decoding input vector and the lane representation vector are combined; based on the long short-term memory neural network LSTM, a plurality of values are cyclically output trained; based on the full connection layer neural network FC, the output of the LSTM is taken as input to predict and generate the target guide line trajectory T The first trajectory point of the model generates the target guide line trajectory), and the point sequence consists of:

[0090]

[0091] Meanwhile, in the CVAE model training process, the loss function needs to be defined as follows:

[0092]

[0093] Wherein represents the recognition network, which learns the posterior distribution, is the prior network, which learns the prior distribution, is the noise vector, is the historical guide line trajectory, is the first trajectory point of the historical guide line trajectory, is the target guide line trajectory T, is the first trajectory point of the model generating the target guide line trajectory, , represents the first historical passing trajectory in the lane; in the training process, the target guide line trajectory point generated by the model and the input historical guide line trajectory are used to calculate the loss function, and the gradient is calculated to update the CVAE model.

[0094] Preferably, in another embodiment of the present application, the step of "S400, adjusting the trajectory points of the target guide line trajectory based on the Carla simulation platform to obtain the final guide line trajectory" specifically comprises the following steps:

[0095] S410, based on the MPC model, the trajectory points of the target guide line trajectory are pre-conditionally constrained to obtain a constrained guide line trajectory;

[0096] S420, Based on the Siamese network, determine the similarity between the trajectory points of the target guide line trajectory and the trajectory points of the constraint guide line trajectory, and adjust the trajectory points of the target guide line trajectory according to the similarity to obtain the final guide line trajectory.

[0097] Specifically, in this embodiment, the simulation submodule mainly adopts the Carla simulation platform, which is an open-source simulator for autonomous driving that supports the development, training, and verification of autonomous driving systems. Based on the MPC (Model Predictive Control) model, preset lateral force coefficients and curvature constraints are set to generate a controllable trajectory composed of a sequence of points. .

[0098] LSTM-based twin networks are used for T and Guide line trajectory scoring, judgment and The similarity.

[0099] Preferably, in another embodiment of this application, the step "S420, adjusting the trajectory points of the target guide line trajectory based on similarity to obtain the final guide line trajectory" specifically includes the following steps:

[0100] S421, when it is determined that one of the trajectory points of the target guide line trajectory is similar to one of the trajectory points of the corresponding constraint guide line trajectory, then one of the trajectory points of the target guide line trajectory is the trajectory point of the final guide line trajectory.

[0101] S422, when it is determined that one of the trajectory points of the target guide line trajectory is not similar to one of the trajectory points of the corresponding constraint guide line trajectory, then one of the trajectory points of the constraint guide line trajectory is the trajectory point of the final guide line trajectory.

[0102] Specifically, in this embodiment, because It controls the feasible trajectory, therefore, if and If they are not similar, then replace ,Add to To the new dataset; if and If similarity is found, it indicates that the guide line trajectory generated by the trajectory planning model conforms to the control parameters and is feasible; therefore, T is added to the new dataset.

[0103] See also Figure 3 As shown, this embodiment of the invention also provides an intersection guide line generation system 100 based on online learning, including: a historical trajectory acquisition module 110, a coordinate transformation module 120, a target guide line trajectory acquisition module 130, and a final guide line trajectory acquisition module 140;

[0104] a historical trajectory acquisition module 110, configured to acquire historical passing trajectories and historical guide line trajectories in a target road intersection;

[0105] a coordinate conversion module 120, in communication connection with the historical trajectory acquisition module 110, configured to perform coordinate value conversion on trajectory points of the historical passing trajectories and trajectory points of the historical guide line trajectories;

[0106] a target guide line trajectory acquisition module 130, in communication connection with the coordinate conversion module 120, configured to perform network training on the historical passing trajectories and the historical guide line trajectories after coordinate conversion based on a conditional variational autoencoder, and acquire a target guide line trajectory;

[0107] a final guide line trajectory acquisition module 140, in communication connection with the target guide line trajectory acquisition module 130, configured to adjust trajectory points of the target guide line trajectory based on a Carla simulation platform, and acquire a final guide line trajectory.

[0108] Therefore, the application is a data-driven trajectory generation model based on CVAE, which can automatically adapt to new intersection scenarios for trajectory planning, and can plan a reasonable and controllable guide line for two connected lanes in the case of no clear road structure at the road intersection.

[0109] Specifically, the present embodiment corresponds to the above-mentioned method embodiment one by one, and the functions of each module have been described in detail in the corresponding method embodiment, so they will not be repeated here.

[0110] Based on the same inventive concept, the present embodiment also provides a computer readable storage medium having a computer program stored thereon, and the computer program is executed by a processor to implement all method steps or part of the method steps of the above method.

[0111] The present application can implement all or part of the above method, and can also be completed by a computer program to instruct related hardware. The computer program can be stored in a computer readable storage medium. When the computer program is executed by a processor, the steps of each method embodiment can be implemented. The computer program includes computer program code, which can be in the form of source code, object code, executable files or some intermediate forms. The computer readable medium can include any entity or device that can carry computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0112] Based on the same inventive concept, the embodiments of the present application further provide an electronic device, comprising a memory and a processor, the memory storing a computer program running on the processor, and the processor implements all the method steps or part of the method steps in the above method when executing the computer program.

[0113] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor and the like, and the processor is a control center of the computer device, which connects all parts of the computer device through various interfaces and lines.

[0114] The memory can be used to store computer programs and / or modules, and the processor realizes various functions of the computer device by running or executing the computer programs and / or modules stored in the memory, and calling data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application program required by a function (such as a sound playing function, an image playing function, etc.); and the data storage area can store data created according to the use of the mobile phone (such as audio data, video data, etc.). In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one magnetic disk storage device, a flash memory device, or other volatile solid-state memory devices.

[0115] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, a server or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program code.

[0116] The present application is described in reference to the appended drawings figures and / or block diagrams of methods, apparatus (systems), servers, and computer program products according to embodiments of the application. It will be understood that each block of the flowchart and / or block diagrams, and combinations of blocks in the flowchart and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more flowcharts and / or blocks Figure 1 one or more blocks or combinations of blocks in the flowchart and / or block diagram.

[0117] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the flowchart and / or block diagram block or blocks. Figure 1 one or more flowcharts and / or blocks Figure 1 one or more blocks or combinations of blocks in the flowchart and / or block diagram.

[0118] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the flowchart and / or block diagram block or blocks. Figure 1 one or more flowcharts and / or blocks Figure 1 one or more blocks or combinations of blocks in the flowchart and / or block diagram.

[0119] Obviously, numerous modifications and variations of the present application are possible in light of the above teachings. It is therefore to be understood that within the scope of the appended claims and their equivalents, the application can be practiced otherwise than as specifically described.

Claims

1. A method for generating intersection guide lines based on online learning, characterized in that, Includes the following steps: Obtain historical traffic trajectories and historical guide line trajectories within the target road intersection; The coordinate values ​​of the trajectory points of the historical travel trajectory and the trajectory points of the historical guide line trajectory are converted. Based on the conditional variational autoencoder, network training is performed on the historical travel trajectory and the historical guide line trajectory after coordinate transformation to obtain the target guide line trajectory. The trajectory points of the target guide line trajectory are adjusted based on the Carla simulation platform to obtain the final guide line trajectory; The step of "training a network based on a conditional variational autoencoder to obtain the target guide line trajectory by analyzing the historical travel trajectory and the historical guide line trajectory after coordinate transformation" specifically includes the following steps: In the conditional variational autoencoder, the historical travel trajectory and the historical guide line trajectory are encoded using feature vectors; The historical travel trajectory and the historical guide line trajectory, after being encoded with feature vectors, are predicted to follow a prior normal distribution. The historical travel trajectory encoded by feature vectors is decoded with the prior normal distribution results to obtain the target guide line trajectory; The step of "encoding the historical travel trajectory and the historical guide line trajectory with feature vectors" specifically includes the following steps: Based on the trajectory spatiotemporal feature coding sub-model, the spatiotemporal feature vectors of the historical travel trajectory and the historical guide line trajectory are extracted, and the corresponding travel spatiotemporal feature vector and guide line spatiotemporal feature vector are obtained. Based on the lane feature representation sub-model, all the spatiotemporal feature vectors of the historical traffic trajectory are fused to obtain the lane representation vector.

2. The intersection guide line generation method based on online learning as described in claim 1, characterized in that, The step of "converting the coordinate values ​​of the trajectory points of the historical travel trajectory and the trajectory points of the historical guide line trajectory" specifically includes the following steps: Convert the global coordinate values ​​of the trajectory points of the historical travel trajectory and the global coordinate values ​​of the trajectory points of the historical guide line trajectory into local coordinate values ​​respectively; or... The global coordinate values ​​of the trajectory points of the historical travel trajectory and the global coordinate values ​​of the trajectory points of the historical guide line trajectory are converted into local coordinate values, respectively, and the local coordinate system containing the local coordinate values ​​of the trajectory points is rotated based on a preset rotation angle; or, The global coordinate values ​​of the trajectory points of the historical travel trajectory and the global coordinate values ​​of the trajectory points of the historical guide line trajectory are respectively converted into behavioral coordinate values.

3. The intersection guide line generation method based on online learning as described in claim 1, characterized in that, The step of "predicting a prior normal distribution between the historical travel trajectory after feature vector encoding and the historical guide line trajectory" specifically includes the following steps: Based on a fully connected neural network that combines the spatiotemporal feature vector of the guide line with the lane representation vector, the mean and variance of the prior normal distribution are predicted, and a sampling vector that meets the preset dimension condition is selected from the standard normal distribution. The trajectory decoding input vector is obtained based on the mean, the variance, and the sampling vector.

4. The intersection guide line generation method based on online learning as described in claim 3, characterized in that, The step of "decoding the historical travel trajectory after feature vector encoding and the prior normal distribution result to obtain the target guide line trajectory" specifically includes the following steps: The trajectory decoding input vector and the lane representation vector are combined based on a fully connected layer neural network; The trajectory decoding input vector and the lane representation vector are cyclically output and trained based on a long short-term memory neural network; The target guide line trajectory is predicted by training the output value of the long short-term memory neural network based on the fully connected layer neural network.

5. The intersection guide line generation method based on online learning as described in claim 1, characterized in that, The step of "adjusting the trajectory points of the target guide line trajectory based on the Carla simulation platform to obtain the final guide line trajectory" specifically includes the following steps: Based on the MPC model, the trajectory points of the target guide line trajectory are subject to preset condition constraints to obtain the constrained guide line trajectory; The similarity between the trajectory points of the target guide line trajectory and the trajectory points of the constraint guide line trajectory is determined based on the Siamese network, and the trajectory points of the target guide line trajectory are adjusted according to the similarity to obtain the final guide line trajectory.

6. The intersection guide line generation method based on online learning as described in claim 5, characterized in that, The step of "adjusting the trajectory points of the target guide line trajectory based on similarity to obtain the final guide line trajectory" specifically includes the following steps: When it is determined that one of the trajectory points of the target guide line trajectory is similar to one of the trajectory points of the corresponding constraint guide line trajectory, then one of the trajectory points of the target guide line trajectory is the trajectory point of the final guide line trajectory. When it is determined that one of the trajectory points of the target guide line trajectory is not similar to one of the trajectory points of the corresponding constraint guide line trajectory, then one of the trajectory points of the constraint guide line trajectory becomes the trajectory point of the final guide line trajectory.

7. A road guide line generation system based on online learning, characterized in that, include: The historical trajectory acquisition module is used to acquire historical traffic trajectories and historical guide line trajectories within the target road intersection; The coordinate transformation module is communicatively connected to the historical trajectory acquisition module and is used to transform the coordinate values ​​of the trajectory points of the historical travel trajectory and the trajectory points of the historical guide line trajectory. The target guide line trajectory acquisition module is communicatively connected to the coordinate transformation module and is used to perform network training on the coordinate-transformed historical travel trajectory and the historical guide line trajectory based on the conditional variational autoencoder to acquire the target guide line trajectory. The final guide line trajectory acquisition module is communicatively connected to the target guide line trajectory acquisition module and is used to adjust the trajectory points of the target guide line trajectory based on the Carla simulation platform to obtain the final guide line trajectory. The step of "training a network based on a conditional variational autoencoder to obtain the target guide line trajectory by analyzing the historical travel trajectory and the historical guide line trajectory after coordinate transformation" specifically includes the following steps: In the conditional variational autoencoder, the historical travel trajectory and the historical guide line trajectory are encoded using feature vectors; The historical travel trajectory and the historical guide line trajectory, after being encoded with feature vectors, are predicted to follow a prior normal distribution. The historical travel trajectory encoded by feature vectors is decoded with the prior normal distribution results to obtain the target guide line trajectory; The step of "encoding the historical travel trajectory and the historical guide line trajectory with feature vectors" specifically includes the following steps: Based on the trajectory spatiotemporal feature coding sub-model, the spatiotemporal feature vectors of the historical travel trajectory and the historical guide line trajectory are extracted, and the corresponding travel spatiotemporal feature vector and guide line spatiotemporal feature vector are obtained. Based on the lane feature representation sub-model, all the spatiotemporal feature vectors of the historical traffic trajectory are fused to obtain the lane representation vector.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the intersection guide line generation method based on online learning as described in any one of claims 1 to 6.

Images