Automobile track prediction method and system based on high-definition map and GEGNN-LSTM model

Abstract

The invention belongs to the technical field of automobile automatic driving, and particularly relates to an automobile track prediction method and system based on a high-definition map and a GEGNN-LSTM model, and the method comprises the steps: firstly obtaining precise road information, including a road shape, an intersection position and the like, through employing the high-definition map; and then, a GEGNN-LSTM model is adopted to learn the historical track data of the automobile, various factors such as the historical track of the automobile, traffic signal lamp sign information and road information are comprehensively considered, and the future track of the automobile is predicted. The method can be widely applied to the fields of intelligent traffic systems, automatic driving and the like, and provides powerful support for the aspects of vehicle driving safety, road traffic efficiency improvement and the like. Meanwhile, the implementation of the method also helps to promote the further development of the intelligent traffic system, and has important practical value and social benefit.

Description

Technical Field

[0001] The present invention belongs to the technical field of automobile automatic driving, and specifically relates to a method and system for predicting automobile trajectory based on high-definition maps and a GEGNN-LSTM model. Background Art

[0002] In the past, the automotive industry usually used a single LSTM model and its improved model to predict vehicle trajectories. This method has made pioneering progress in the past few years to fill the gap in the smart car industry. It has very good performance for predicting only the historical time series trajectory characteristics of the target vehicle, or for smart cars using high-definition cameras, only using the GNN model and its improved model for spatial prediction of graph data. However, although GNN can effectively process spatial information, it may not be sensitive enough to capture the dynamic changes of vehicle trajectories, such as changes in vehicle speed and acceleration, and may be subject to certain processing time restrictions when processing real-time dynamically changing data. Although LSTM is good at processing time series data, it cannot directly process spatial relationships. Subsequently, in response to the defects of these two models, the industry proposed the GNN-transformer model, which aims to solve the shortcomings of the two models by combining spatial features with time series features. However, in the GNN model, the graph connection layer uses a fully connected method, which leads to a large amount of data in the data set for some unimportant factors in trajectory prediction, which in turn leads to inaccurate trajectory prediction results. Summary of the invention

[0003] The purpose of the present invention is to provide a vehicle trajectory prediction method and system based on high-definition maps and GEGNN-LSTM models to solve the problems raised in the background technology.

[0004] The present invention achieves the above-mentioned purpose through the following technical solutions:

[0005] In a first aspect, the present invention proposes a vehicle trajectory prediction method based on a high-definition map and a GEGNN-LSTM model, the method comprising:

[0006] S1. Construct a GEGNN-LSTM model, wherein the GEGNN-LSTM model includes an Encoder module and a graph social pooling network built based on GEGNN, and a Decoder module built based on GEGNN;

[0007] S2, collecting and encoding the historical tracks and traffic light signs of the target vehicle and surrounding vehicles in the high-definition map bird's-eye view through the Encoder module to form a target data set;

[0008] S3, constructing a vector line graph based on the target data set to represent road information;

[0009] S4, using the graph social pooling network to interactively encode the target vehicle, surrounding vehicles and road information to form interactive information features;

[0010] S5, through the ground Figure 2 The 3D coordinate system is standardized, and an identification label is constructed based on the relationship between the vehicle and the lane centerline, wherein the identification label includes lane information of the target vehicle and surrounding vehicles;

[0011] S6. Based on the Encoder module, interactively concatenate the interactive information features and the lane information to obtain a global information code;

[0012] S7. Encode the global information into the Decoder module for decoding to obtain a predicted trajectory, wherein the predicted trajectory includes a lateral position coordinate and a longitudinal position coordinate of the vehicle in a two-dimensional coordinate system.

[0013] A further improvement is that in step S1, constructing the GEGNN-LSTM model includes: using the GNN model improved by the Gaussian probability model as the feature interaction layer, using graph social pooling to perform interactive feature fusion on the vehicle trajectory information and the traffic light sign information, outputting the fused feature vector and using it as the input of the LSTM model to connect the two models.

[0014] A further improvement is that in step S2, the historical trajectory information of the target vehicle and surrounding vehicles includes the historical position and vehicle status of the target vehicle in the past 5 seconds, and the historical position includes the vehicle's position on the ground. Figure 2 The vehicle state includes the acceleration and speed of the vehicle; the traffic light sign includes the signal state of the traffic light in the past 5 seconds, and the signal state includes the color of the traffic light and the countdown time.

[0015] A further improvement is that step S3 comprises:

[0016] S301, set a feature vector X = [x, y, v, a, s, t], where: x, y are the horizontal and vertical coordinates of the vehicle, v is the vehicle speed, a is the vehicle acceleration, s is the traffic light sign, and t is the traffic light countdown time;

[0017] S302, incorporate the feature vector X into the feature vector graph to construct a vector line graph. A vector graph contains multiple feature vectors X: Grah = [X (1) , X (2) , X (3) , …, X (n)], n<200.

[0018] A further improvement is that step S4 comprises:

[0019] S401, determine at time t1, the characteristic vector of the target vehicle is The feature vectors of the surrounding vehicles are

[0020] S402: construct a node i on the graph, where each node i represents a vehicle or a traffic light, and the edges between the nodes represent the distance, relative speed, and Gaussian features between the vehicles. Specifically, to form, among which represents the physical distance between nodes i and j at time t1, represents the relative speed between nodes i and j, Represents the Gaussian feature between nodes i and j;

[0021] S403, calculating Gaussian features, that is, calculating the distance and relative speed between the nodes by calculating the mean and standard deviation of the Gaussian distribution between two nodes;

[0022] S404: Calculate the weight of the edge between two nodes using the Gaussian expression of the edge weight, wherein the weight is adjusted according to the traffic light sign to reflect the different effects of red light, green light and yellow light on vehicle driving.

[0023] A further improvement is that step S5 comprises:

[0024] S501, placing the origin of the two-dimensional coordinate system at the position of the last point of the target vehicle's historical trajectory time point, and taking the direction of the target vehicle at that time point as the positive direction of the X-axis of the two-dimensional coordinate system;

[0025] S502: The target vehicle lane labels are divided into four types, which respectively represent the first lane closest to the left side of the lane centerline, the first lane closest to the right side of the lane centerline, the second lane on the left side of the lane centerline, and the second lane on the right side of the lane centerline, thereby constructing a lane id of: Indicates the lane ID information of the target vehicle and surrounding vehicles.

[0026] A further improvement is that step S6 comprises:

[0027] S601, interactively splicing the encoded historical trajectory of the target vehicle, the historical trajectory of surrounding vehicles, traffic light sign information, and the interactive information between the target vehicle and surrounding vehicles, specifically including updating the basic formula of the feature vector: Therefore, when constructing the graph network later, after incorporating the lane information features, the neighbor feature aggregation formula is used to form preliminary splicing information:

[0028]

[0029] in: is the feature of node i at layer l, W (l) is the weight matrix of the lth layer, σ is the activation function, is the set of neighbor nodes of node i, d ij is the actual distance between nodes i and j, μ ij and σ ij are the mean and standard deviation of the Gaussian distribution of edge (i, j), α s is the weight factor adjusted according to the traffic light sign;

[0030] S602, the preliminary splicing information is then spliced ​​with the road label information to obtain global splicing information; all nodes are aggregated to obtain global splicing information, and the formula is: H final =Aggregate(H l ), where: Aggregate is the aggregation function, H l is the output feature of the last layer, H final It is the final global splicing information.

[0031] A further improvement is that the step S7 comprises: final Arrange the node features in chronological order and finally add them to the input of the LSTM model. Collect data according to the 5-second historical trajectory, divide the first 3 seconds of data into a training set, and the last 2 seconds into a test set. Use the LSTM network to decode and output the target vehicle trajectory within the next 2 seconds.

[0032] In a second aspect, the present invention proposes a vehicle trajectory prediction system based on a high-definition map and a GEGNN-LSTM model, which is applied to execute any of the above-mentioned vehicle trajectory prediction methods, and the system comprises:

[0033] A model building module is configured to build a GEGNN-LSTM model, wherein the GEGNN-LSTM model includes an Encoder module and a graph social pooling network built based on GEGNN, and a Decoder module built based on;

[0034] The data acquisition module is configured to collect and encode the historical tracks and traffic light signs of the target vehicle and surrounding vehicles in the high-definition map bird's-eye view through the Encoder module to form a target data set; and construct a vector line graph based on the target data set to represent road information;

[0035] A feature interaction module is configured to interactively encode target vehicle, surrounding vehicles and road information using the graph social pooling network to form interactive information features;

[0036] Tag building blocks are configured to pass ground Figure 2 The 3D coordinate system is standardized, and an identification label is constructed based on the relationship between the vehicle and the lane centerline, wherein the identification label includes lane information of the target vehicle and surrounding vehicles;

[0037] An information splicing module is configured to interactively splice the interactive information features and the lane information based on the Encoder module to obtain a global information encoding;

[0038] The trajectory prediction module is configured to encode the global information into the Decoder module for decoding to obtain a predicted trajectory, wherein the predicted trajectory includes a lateral position coordinate and a longitudinal position coordinate of the vehicle in a two-dimensional coordinate system.

[0039] In a third aspect, the present invention proposes an electronic device, comprising: a processor; and a memory for storing instructions executable by the processor; wherein the processor is configured to execute the instructions to implement the prediction method as described above.

[0040] In a fourth aspect, the present invention proposes a computer-readable storage medium. When instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to perform the prediction method as described above.

[0041] The beneficial effects of the present invention are:

[0042] (1) The present invention makes full use of the precise road information provided by high-definition maps, making the prediction results more consistent with the actual road conditions and improving the accuracy and reliability of the prediction. The application of the GEGNN-LSTM model enables the solution to comprehensively consider multiple factors such as the vehicle's historical trajectory, traffic light sign information, and road information, and realizes the intelligent prediction of the vehicle's future trajectory, providing strong support for the construction of intelligent transportation systems.

[0043] (2) The present invention also has good scalability and adaptability, and can be flexibly adjusted and optimized for different road types and traffic conditions to meet the application requirements in different scenarios. The implementation of this solution not only helps to improve road traffic efficiency and safety, but also provides important data support for urban planning, traffic management and other fields, and promotes the further development of intelligent transportation systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 is a flow chart of the prediction method of the present invention;

[0045] Figure 2 is a structural diagram of the prediction system in the present invention;

[0046] Figure 3 It is a comparison case diagram between the prediction system and the comparison algorithm system in the present invention. DETAILED DESCRIPTION

[0047] The following description provides specific application scenarios and requirements of this specification, with the purpose of enabling those skilled in the art to make and use the contents of this specification. Various local modifications to the disclosed embodiments will be apparent to those skilled in the art, and the general principles defined herein may be applied to other embodiments and applications without departing from the spirit and scope of this specification. Therefore, this specification is not limited to the embodiments shown, but to the widest scope consistent with the claims.

[0048] The terms used herein are only used for the purpose of describing specific example embodiments and are not restrictive. For example, unless the context clearly indicates otherwise, as used herein, the singular forms "a", "an" and "the" may also include plural forms. When used in this specification, the terms "include", "comprise" and / or "contain" mean that the associated integers, steps, operations, elements and / or components exist, but do not exclude the existence of one or more other features, integers, steps, operations, elements, components and / or groups or that other features, integers, steps, operations, elements, components and / or groups may be added in the system / method.

[0049] In view of the following description, these and other features of the present specification, as well as the operation and function of the related elements of the structure, and the economy of the combination and manufacture of the parts can be significantly improved. Reference is made to the accompanying drawings, all of which form a part of this specification. However, it should be clearly understood that the drawings are for illustration and description purposes only and are not intended to limit the scope of this specification. It should also be understood that the drawings are not drawn to scale.

[0050] The flowcharts used in this specification illustrate the operations implemented by the system according to some embodiments in this specification. It should be clearly understood that the operations of the flowcharts may not be implemented in sequence. On the contrary, the operations may be implemented in reverse order or simultaneously. In addition, one or more other operations may be added to the flowchart. One or more operations may be removed from the flowchart.

[0051] Example 1

[0052] See also Figure 1 , a vehicle trajectory prediction method based on a high-definition map and a GEGNN-LSTM model, the method comprising:

[0053] S1. Build the GEGNN-LSTM model. The GEGNN-LSTM model includes the Encoder module and graph social pooling network built based on GEGNN, and the Decoder module built based on GEGNN.

[0054] Preferably, the GNN model improved by the Gaussian probability model is used as the feature interaction layer, and the vehicle trajectory information and the traffic light sign information are interactively fused by using graph social pooling. The fused feature vector is output and used as the input of the LSTM model to connect the two models.

[0055] S2. The historical tracks and traffic light signs of the target vehicle and surrounding vehicles in the high-definition map bird's-eye view are collected and encoded through the Encoder module to form a target data set. The specific implementation plan is as follows:

[0056] S201, extract the historical trajectory information of the target vehicle and encode it. The historical trajectory information of the target vehicle mainly includes the historical position and vehicle status of the target vehicle in the past 5 seconds. The historical position includes the vehicle’s location on the ground. Figure 2 The vehicle state includes the horizontal coordinate and the longitudinal coordinate in the 3D coordinate system, and the vehicle state includes the acceleration and speed of the vehicle.

[0057] S202, extracting and encoding the historical trajectory information of the surrounding vehicles. The historical trajectory information of the target vehicle mainly includes the historical position and vehicle status of the target vehicle in the past 5 seconds. The historical position includes the vehicle’s location on the ground. Figure 2 The vehicle state includes the horizontal coordinate and the longitudinal coordinate in the 3D coordinate system, and the vehicle state includes the acceleration and speed of the vehicle.

[0058] S203, extracting traffic light sign information, mainly including the signal status of the traffic light in the past 5 seconds, the signal status including the color of the traffic light and the countdown time.

[0059] S204, encode and preprocess the extracted data, use normalization, use Python to store the data processing results in the data array, use the data.dropna function to process missing values, use Python's internal function to delete outliers, and delete the values ​​of speed and acceleration that are 2% on both sides of the value, that is, use the following code operation:

[0060] data=data[data['v'].between(data['v'].quantile(0.02),data['v'].quantile(0.98))]data=data[data['a'].between(data['a'].quantile(0.02),data['a'].quantile(0.98))]

[0061] At the same time, for categorical variables such as traffic light signs, one-hot encoding is used to convert them into binary, that is,

[0062] 'traffic_light':['red', 'yellow', 'green'], and its one-hot encoding forms the following table:

[0063] traffic_light red yellow green red 1.0 0.0 0.0 yellow 0.0 1.0 0.0 green 0.0 0.0 1.0

[0064] Use scaler = MinMaxScaler (feature_range = (0, 1)) function to normalize the data, use scaler.fit to fit the scaler only on the training set, and then use scaler.transform to get the same minimum and maximum values ​​on the test set to ensure that the test set is processed in the same way as the training set. Finally, update the processed data in the data dataset.

[0065] S3. Construct a vector line graph based on the target data set to represent road information. The specific scheme is as follows:

[0066] S301, set a feature vector X = [x, y, v, a, s, t], where: x, y are the horizontal and vertical coordinates of the vehicle, v is the vehicle speed, a is the vehicle acceleration, s is the traffic light sign, and t is the traffic light countdown time;

[0067] S302, incorporate the feature vector X into the feature vector graph to construct a vector line graph. A vector graph contains multiple feature vectors X: Grah = [X (1) , X (2) , X (3) , …, X (n) ], n<200. Contains at most 200 feature vectors. If the number of elements is insufficient, it will be supplemented with 0.

[0068] S4. Use the graph social pooling network to interactively encode the target vehicle, surrounding vehicles and road information to form interactive information features. The specific scheme is as follows:

[0069] S401, determine at time t1, the characteristic vector of the target vehicle is The feature vectors of the surrounding vehicles are

[0070] S402: construct a node i on the graph, where each node i represents a vehicle or a traffic light, and the edges between the nodes represent the distance, relative speed, and Gaussian features between the vehicles. Specifically, to form, among which represents the physical distance between nodes i and j at time t1, represents the relative speed between nodes i and j, Represents the Gaussian feature between nodes i and j;

[0071] S403, calculating Gaussian features, that is, calculating the distance and relative speed between the nodes by calculating the mean and standard deviation of the Gaussian distribution between two nodes;

[0072] S404. The weight of the edge between two nodes is calculated using the Gaussian expression of the edge weight. The weight is adjusted according to the traffic light sign to reflect the different effects of red light, green light and yellow light on vehicle driving. More specifically, red light: higher weight, because it usually means stopping or slowing down. Green light: lower weight, because it usually means normal driving. Yellow light: medium weight, because it indicates an imminent change of state.

[0073] S5, through the ground Figure 2 The 3D coordinate system is standardized and an identification label is constructed based on the relationship between the vehicle and the lane centerline. The identification label includes the lane information of the target vehicle and surrounding vehicles. The specific scheme is as follows:

[0074] S501, placing the origin of the two-dimensional coordinate system at the position of the last point of the target vehicle's historical trajectory time point, and taking the direction of the target vehicle at that time point as the positive direction of the X-axis of the two-dimensional coordinate system;

[0075] S502: The target vehicle lane labels are divided into four types, which respectively represent the first lane closest to the left side of the lane centerline, the first lane closest to the right side of the lane centerline, the second lane on the left side of the lane centerline, and the second lane on the right side of the lane centerline, thereby constructing a lane id of: Indicates the lane ID information of the target vehicle and surrounding vehicles.

[0076] S6. Based on the Encoder module, interactive information features and lane information are interactively spliced ​​to obtain global information encoding; the specific scheme is as follows:

[0077] S601, interactively splicing the encoded historical trajectory of the target vehicle, the historical trajectory of surrounding vehicles, traffic light sign information, and the interactive information between the target vehicle and surrounding vehicles, specifically including updating the basic formula of the feature vector: Therefore, when constructing the graph network later, after incorporating the lane information features, the neighbor feature aggregation formula is used to form preliminary splicing information:

[0078]

[0079] in: is the feature of node i at layer l, W (l) is the weight matrix of the lth layer, σ is the activation function, is the set of neighbor nodes of node i, d ij is the actual distance between nodes i and j, μ ij and σ ij are the mean and standard deviation of the Gaussian distribution of edge (i, j), α s is the weight factor adjusted according to the traffic light sign;

[0080] S602, the preliminary splicing information is then spliced ​​with the road label information to obtain global splicing information; all nodes are aggregated to obtain global splicing information, and the formula is: H final =Aggregate(H l ), where: Aggregate is the aggregation function, H l is the output feature of the last layer, H final It is the final global splicing information.

[0081] S7, encode the global information into the Decoder module for decoding, and obtain the predicted trajectory, which includes the lateral position coordinates and longitudinal position coordinates of the vehicle in the two-dimensional coordinate system. The specific scheme is as follows: final Arrange node features in chronological order and add them to the input of the LSTM model. Collect data according to the 5-second historical trajectory, divide the first 3 seconds of data into a training set, and the last 2 seconds into a test set. Use the LSTM network to decode and output the target vehicle trajectory within the next 2 seconds. Use the Matplotlib library in python to draw the historical trajectory and the predicted trajectory. At the same time, compare the predicted future trajectory with the actual future trajectory, calculate the value of the loss function, and update the corresponding network hyperparameters. After the training is completed, the corresponding weight parameters should be retained.

[0082] See also Figure 2 This embodiment proposes a vehicle trajectory prediction system based on a high-definition map and a GEGNN-LSTM model, which is applied to execute the above-mentioned vehicle trajectory prediction method. The system includes:

[0083] The model building module is configured to build a GEGNN-LSTM model, which includes an Encoder module and a graph social pooling network built based on GEGNN, and a Decoder module built based on ;

[0084] The data acquisition module is configured to collect and encode the historical tracks and traffic light signs of the target vehicle and surrounding vehicles in the high-definition map bird's-eye view through the Encoder module to form a target data set; and construct a vector line graph based on the target data set to represent the road information;

[0085] A feature interaction module is configured to interactively encode target vehicle, surrounding vehicles and road information using a graph social pooling network to form interactive information features;

[0086] Tag building blocks are configured to pass ground Figure 2 The 3D coordinate system is standardized and an identification label is constructed based on the relationship between the vehicle and the lane centerline. The identification label includes the lane information of the target vehicle and surrounding vehicles.

[0087] An information concatenation module is configured to concatenate interactive information features and lane information based on the Encoder module to obtain a global information encoding;

[0088] The trajectory prediction module is configured to encode the global information into the Decoder module for decoding to obtain a predicted trajectory, which includes a lateral position coordinate and a longitudinal position coordinate of the vehicle in a two-dimensional coordinate system.

[0089] Case section:

[0090] According to the vehicle trajectory prediction method based on high-definition map and GEGNN-LSTM model proposed in the above embodiment of the present invention, case 1 is predicted, the vehicle trajectory prediction based on LSTM model is used as case 2, and the vehicle trajectory prediction based on GNN model is used as case 3; by inputting S-shaped road data into different case algorithms, 80% of the road data is used as a training set, and 20% of the road data is used as a test set. Through algorithm training and simulation, the trajectory case diagram of the terminal S-shaped road vehicle under different prediction schemes is derived, as shown in FIG. Figure 3 The MAE values ​​under different cases are exported for accuracy comparison, as shown in the following table:

[0091] MAE value:

[0092] Prediction Methods LSTM GNN GEGNN+LSTM MAE 0.178 0.121 0.076

[0093] Combining the MAE values ​​under the above different prediction methods, it can be concluded that the prediction algorithm in this case can obtain more accurate prediction accuracy.

[0094] This embodiment proposes an electronic device, including: a processor; a memory for storing instructions executable by the processor; wherein the processor is configured to execute the instructions to implement the above-mentioned prediction method. And a computer-readable storage medium is also proposed, when the instructions in the computer-readable storage medium are executed by the processor of the electronic device, the electronic device can execute the above-mentioned prediction method.

[0095] The above description is only a preferred embodiment of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of disclosure involved in the present disclosure is not limited to the technical solutions formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above disclosed concept. For example, the above features are replaced with the technical features with similar functions disclosed in the present disclosure (but not limited to) by each other to form a technical solution.

[0096] In addition, although each operation is described in a specific order, this should not be understood as requiring these operations to be performed in the specific order shown or in a sequential order. Under certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although some specific implementation details are included in the above discussion, these should not be interpreted as limiting the scope of the present disclosure. Some features described in the context of a separate embodiment can also be implemented in a single embodiment in combination. On the contrary, the various features described in the context of a single embodiment can also be implemented in multiple embodiments individually or in any suitable sub-combination mode.

[0097] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A vehicle trajectory prediction method based on high-definition maps and GEGNN-LSTM model, characterized in that: The method comprises: S1. Construct a GEGNN-LSTM model, wherein the GEGNN-LSTM model includes an Encoder module and a graph social pooling network built based on GEGNN, and a Decoder module built based on GEGNN; S2, collecting and encoding the historical tracks and traffic light signs of the target vehicle and surrounding vehicles in the high-definition map bird's-eye view through the Encoder module to form a target data set; S3, constructing a vector line graph based on the target data set to represent road information; S4, using the graph social pooling network to interactively encode the target vehicle, surrounding vehicles and road information to form interactive information features; S5. By standardizing the two-dimensional coordinate system of the map, an identification label is constructed according to the relationship between the vehicle and the center line of the lane, wherein the identification label includes lane information of the target vehicle and surrounding vehicles; S6. Based on the Encoder module, interactively concatenate the interactive information features and the lane information to obtain a global information code; S7. Encode the global information into the Decoder module for decoding to obtain a predicted trajectory, wherein the predicted trajectory includes a lateral position coordinate and a longitudinal position coordinate of the vehicle in a two-dimensional coordinate system.

2. The vehicle trajectory prediction method based on high-definition map and GEGNN-LSTM model according to claim 1 is characterized in that: In step S1, constructing the GEGNN-LSTM model includes: using the GNN model improved by the Gaussian probability model as the feature interaction layer, using graph social pooling to perform interactive feature fusion on vehicle trajectory information and traffic light sign information, outputting the fused feature vector and using it as the input of the LSTM model to connect the two models.

3. The vehicle trajectory prediction method based on high-definition map and GEGNN-LSTM model according to claim 1 is characterized in that: In step S2, the historical trajectory information of the target vehicle and surrounding vehicles includes the historical position and vehicle status of the target vehicle in the past 5 seconds, the historical position includes the horizontal coordinate and vertical coordinate of the vehicle in the two-dimensional coordinate system of the map, and the vehicle status includes the acceleration and speed of the vehicle; the traffic light sign includes the signal status of the traffic light in the past 5 seconds, and the signal status includes the color of the traffic light and the countdown time.

4. The vehicle trajectory prediction method based on high-definition map and GEGNN-LSTM model according to claim 3 is characterized by: The step S3 comprises: S301, set a feature vector X = [x, y, v, a, s, t], where: x, y are the horizontal and vertical coordinates of the vehicle, v is the vehicle speed, a is the vehicle acceleration, s is the traffic light sign, and t is the traffic light countdown time; S302, incorporate the feature vector X into the feature vector graph to construct a vector line graph. A vector graph contains multiple feature vectors X: Grah = [X (1) , X (2) , X (3) , …, X (n) ], n<200.

5. The vehicle trajectory prediction method based on high-definition map and GEGNN-LSTM model according to claim 4 is characterized in that: The step S4 comprises: S401, determine at time t1, the characteristic vector of the target vehicle is The feature vectors of the surrounding vehicles are S402: construct a node i on the graph, where each node i represents a vehicle or a traffic light, and the edges between the nodes represent the distance, relative speed, and Gaussian features between the vehicles. Specifically, to form, among which represents the physical distance between nodes i and j at time t1, represents the relative speed between nodes i and j, Represents the Gaussian feature between nodes i and j; S403, calculating Gaussian features, that is, calculating the distance and relative speed between the nodes by calculating the mean and standard deviation of the Gaussian distribution between two nodes; S404: Calculate the weight of the edge between two nodes using the Gaussian expression of the edge weight, wherein the weight is adjusted according to the traffic light sign to reflect the different effects of red light, green light and yellow light on vehicle driving.

6. The vehicle trajectory prediction method based on high-definition map and GEGNN-LSTM model according to claim 5 is characterized by: The step S5 comprises: S501, placing the origin of the two-dimensional coordinate system at the position of the last point of the target vehicle's historical trajectory time point, and taking the direction of the target vehicle at that time point as the positive direction of the X-axis of the two-dimensional coordinate system; S502: The target vehicle lane labels are divided into four types, which respectively represent the first lane closest to the left side of the lane centerline, the first lane closest to the right side of the lane centerline, the second lane on the left side of the lane centerline, and the second lane on the right side of the lane centerline, thereby constructing a lane id of: Indicates the lane ID information of the target vehicle and surrounding vehicles.

7. The vehicle trajectory prediction method based on high-definition map and GEGNN-LSTM model according to claim 6 is characterized by: The step S6 comprises: S601, interactively splicing the encoded historical trajectory of the target vehicle, the historical trajectory of surrounding vehicles, traffic light sign information, and the interactive information between the target vehicle and surrounding vehicles, specifically including updating the basic formula of the feature vector: Therefore, when constructing the graph network later, after incorporating the lane information features, the neighbor feature aggregation formula is used to form preliminary splicing information: in: is the feature of node i at layer l, W (l) is the weight matrix of the lth layer, σ is the activation function, N(i) is the set of neighbor nodes of node i, d ij is the actual distance between nodes i and j, μ ij and σ ij are the mean and standard deviation of the Gaussian distribution of edge (i, j), α s is the weight factor adjusted according to the traffic light sign; S602, the preliminary splicing information is then spliced ​​with the road label information to obtain global splicing information; all nodes are aggregated to obtain global splicing information, and the formula is: H final =Aggregate(H l ), where: Aggregate is the aggregation function, H l is the output feature of the last layer, H final It is the final global splicing information.

8. The vehicle trajectory prediction method based on high-definition map and GEGNN-LSTM model according to claim 7 is characterized in that: The step S7 comprises: final Arrange the node features in chronological order and finally add them to the input of the LSTM model. Collect data according to the 5-second historical trajectory, divide the first 3 seconds of data into a training set, and the last 2 seconds into a test set. Use the LSTM network to decode and output the target vehicle trajectory within the next 2 seconds.

9. A vehicle trajectory prediction system based on high-definition maps and GEGNN-LSTM model, characterized in that: Applied to executing the vehicle trajectory prediction method according to any one of claims 1 to 8, the system comprises: A model building module is configured to build a GEGNN-LSTM model, wherein the GEGNN-LSTM model includes an Encoder module and a graph social pooling network built based on GEGNN, and a Decoder module built based on; The data acquisition module is configured to collect and encode the historical tracks and traffic light signs of the target vehicle and surrounding vehicles in the high-definition map bird's-eye view through the Encoder module to form a target data set; and construct a vector line graph based on the target data set to represent road information; A feature interaction module is configured to interactively encode target vehicle, surrounding vehicles and road information using the graph social pooling network to form interactive information features; A label construction module is configured to construct an identification label according to the relationship between the vehicle and the lane centerline by normalizing the two-dimensional coordinate system of the map, wherein the identification label includes lane information of the target vehicle and surrounding vehicles; An information splicing module is configured to interactively splice the interactive information features and the lane information based on the Encoder module to obtain a global information encoding; The trajectory prediction module is configured to encode the global information into the Decoder module for decoding to obtain a predicted trajectory, wherein the predicted trajectory includes a lateral position coordinate and a longitudinal position coordinate of the vehicle in a two-dimensional coordinate system.

Images