Vehicle trajectory prediction method, device, electronic device and readable storage medium

Abstract

The present disclosure provides a vehicle trajectory prediction method, device, electronic device and computer-readable storage medium, which relate to the field of regression prediction technology. Among them, the vehicle trajectory prediction method includes: preprocessing the historical trajectory of the vehicle to obtain the preprocessed trajectory of the vehicle; preprocessing the map information of the vehicle's driving area to obtain the vehicle's drivable lane line; performing trajectory feature processing on the preprocessed trajectory to obtain the driving interaction features between multiple vehicles; performing lane topology extraction on the drivable lane line to obtain the lane line connectivity features; performing autoregressive prediction based on the driving interaction features and the lane line connectivity features to obtain the predicted trajectory of the vehicle. Through the technical solution of the present disclosure, the topological structure of the lane can be fully considered in the process of vehicle trajectory prediction, and the reliability of the generated predicted trajectory can be guaranteed while improving the prediction accuracy, which is conducive to meeting the safety requirements of autonomous driving.

Description

Background Art

[0002] In order to improve the safety of autonomous vehicles, it is necessary to accurately predict the trajectories of surrounding vehicles based on the scene in which they are located.

[0003] In the related art, the time series features of the historical trajectory, the interaction features with the surrounding vehicles, and the driving scene features are extracted through a neural network model. The above features are concatenated or input into a fully connected layer to fuse the above features to obtain a context vector. Finally, the context vector is directly input into a decoder for decoding to obtain the prediction data of the future trajectory. This solution has the following defects:

[0004] Due to the uncertainty of the driver's intention and the scene they are in, there are many possibilities for the vehicle's future trajectory. Considering that drivers tend to drive along lanes, the possible future trajectory of the vehicle is directly related to the lanes it can drive in. Therefore, the model's ability to understand scene information such as the available lanes during the prediction process is crucial. The way of extracting image features based on the neural network model will make the model believe that the closer the area on the map is to the vehicle, the more important it is, resulting in the model being unable to adaptively generate multiple possible trajectories based on the scene the target is in. In complex scenes such as intersections, there is a mode collapse phenomenon in which only one possible trajectory, straight driving, can be predicted, which directly leads to unreliable prediction results.

[0005] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0006] The purpose of the present disclosure is to provide a vehicle trajectory prediction method, device, electronic device and computer-readable storage medium, which at least to some extent overcome the problem of unreliable vehicle trajectory prediction results in related technologies.

[0007] Other features and advantages of the present disclosure will become apparent from the following detailed description, or may be learned in part by the practice of the present disclosure.

[0008] According to one aspect of the present disclosure, a vehicle trajectory prediction method is provided, comprising: preprocessing a historical trajectory of a vehicle to obtain a preprocessed trajectory of the vehicle; preprocessing map information of a vehicle driving area to obtain a drivable lane line of the vehicle; performing trajectory feature processing on the preprocessed trajectory to obtain a plurality of driving interaction features between the vehicles; performing lane topology extraction on the drivable lane lines to obtain lane line connectivity features; performing autoregressive prediction based on the driving interaction features and the lane line connectivity features to obtain a predicted trajectory of the vehicle.

[0009] In one embodiment of the present disclosure, before performing trajectory feature processing on the pre-processed trajectory to obtain the driving interaction features between the multiple vehicles, it also includes: constructing a trajectory feature processing module based on a gated recurrent unit GRU encoder and a graph attention mechanism GAT; performing trajectory feature processing on the pre-processed trajectory to obtain the driving interaction features between the multiple vehicles specifically includes: inputting the pre-processed trajectory into the trajectory feature processing module, and outputting the driving interaction features.

[0010] In one embodiment of the present disclosure, the pre-processed trajectory is input into the trajectory feature processing module, and the driving interaction feature is output. Specifically, the pre-processed trajectory is mapped to the feature space based on a one-dimensional convolutional neural network to obtain a trajectory space feature; the trajectory space feature is encoded based on the GRU encoder to obtain respective context vectors of the multiple vehicles; the current positions of the multiple vehicles are used as nodes, and the respective context vectors of the multiple vehicles are used as node features, and are fully connected to construct a fully connected graph; based on the GAT, the interaction context vectors between the multiple vehicles are extracted from the fully connected graph, and the interaction context vectors are used as the driving interaction feature.

[0011] In one embodiment of the present disclosure, before performing lane topology extraction on the drivable lane line to obtain the lane line connectivity feature, it also includes: constructing a variant lane graph convolutional neural network LaneGCN based on the graph convolutional neural network GCN, and constructing a lane topology extraction module based on the LaneGCN and the first MLP multi-layer perceptron; performing lane topology extraction on the drivable lane line to obtain the lane line connectivity feature specifically includes: inputting the drivable lane line into the lane topology extraction module, and outputting the lane line connectivity feature.

[0012] In one embodiment of the present disclosure, the inputting of the drivable lane line into the lane topology extraction module and the outputting of the lane line connectivity feature specifically include: mapping the drivable lane line and the current moment positions of the multiple vehicles to the feature space based on the first MLP to obtain a spatial feature vector, wherein the spatial feature vector represents the number of lane nodes included in the drivable lane line corresponding to each of the vehicles; constructing a front-to-back adjacency matrix including the lane nodes based on the front-to-back connection relationship between the lane nodes; and extracting the topological connectivity features of the lane nodes from the adjacency matrix based on the variant lane performance of the LaneGCN to obtain a lane line feature vector as the lane line connectivity feature.

[0013] In one embodiment of the present disclosure, before performing autoregressive prediction based on the driving interaction features and the lane line connectivity features to obtain the predicted trajectory of the vehicle, it also includes: constructing an autoregressive prediction module based on a gated recurrent unit GRU decoder, a second MLP multi-layer perceptron, a third MLP multi-layer perceptron, and a lane-based attention mechanism Lane-Attention; performing autoregressive prediction based on the driving interaction features and the lane line connectivity features to obtain the predicted trajectory of the vehicle specifically includes: inputting the driving interaction features and the lane line connectivity features into the autoregressive prediction module, and outputting the predicted trajectory of the vehicle.

[0014] In one embodiment of the present disclosure, the inputting of the driving interaction features and the lane line connectivity features into the autoregressive prediction module and outputting the predicted trajectory of the vehicle specifically includes: determining the first lane node and the second lane node having the minimum distance from the drivable lane line where the vehicle is located according to the current moment position of the vehicle and the key points at the current moment, wherein the key points at the current moment are initially determined based on the lane line feature vector; constructing the relevant lane segment of the vehicle based on the drivable lane line to which the first lane node and the second lane node belong; obtaining the lane segment feature vector of the relevant lane segment based on the second MLP multilayer perceptron; calculating the perception of the relevant lane segment in the lane segment feature vector at the current moment position of the vehicle based on Lane-Attention; updating the current moment position of the vehicle based on the perception to obtain an updated position feature; decoding the updated position feature and the interaction context vector based on the GRU decoder to obtain a hidden layer vector; inputting the hidden layer vector into the third MLP to predict the next predicted trajectory point at the current moment; and generating the predicted trajectory of the vehicle based on multiple predicted next predicted trajectory points.

[0015] In one embodiment of the present disclosure, obtaining the lane segment feature vector of the relevant lane segment based on the second MLP multilayer perceptron specifically includes: determining the distance and angle difference between the lane node on the relevant lane segment and the current position; obtaining the position code of the vehicle based on the distance and the angle difference; and performing two-layer feature fusion on the position code based on the second MLP multilayer perceptron to obtain the lane segment feature vector.

[0016] In one embodiment of the present disclosure, the Lane-Attention-based calculation of the perception of the relevant lane segment in the lane segment feature vector at the current moment position of the vehicle specifically includes: the lane segment feature vector passes through a fully connected layer to form a Key-Value data pair; the current moment position of the vehicle passes through the fully connected layer to obtain a Query; the Key-Value data pair and the Query are input into the Lane-Attention, and the perception is output.

[0017] In one embodiment of the present disclosure, the inputting the hidden layer vector into the third MLP to obtain the trajectory point at the next moment specifically includes: inputting the hidden layer vector into the first sub-MLP and the second sub-MLP in the third MLP respectively; the first sub-MLP outputs the next predicted trajectory point and the corresponding probability; the second sub-MLP outputs the key point at the next moment, and the key point at the next moment and the next predicted trajectory point are used as the updated current moment position and the key point at the current moment when arriving at the next moment.

[0018] In one embodiment of the present disclosure, before performing trajectory feature processing on the preprocessed trajectory to obtain the driving interaction features between the multiple vehicles, it also includes: constructing a model training set and a verification set based on the preprocessed trajectory and the drivable lane lines; training a prediction model based on the model training set and the verification set, and the prediction model includes the trajectory feature processing module, the lane topology extraction module and the autoregressive prediction module.

[0019] In one embodiment of the present disclosure, it also includes: based on the next predicted trajectory point, using a first loss function to perform convergence optimization on the prediction model; based on the corresponding probability, using the first loss function to perform convergence optimization on the prediction model to obtain the optimized prediction model.

[0020] In one embodiment of the present disclosure, the preprocessing of the historical trajectory of the vehicle to obtain the preprocessed trajectory of the vehicle specifically includes: collecting the trajectory coordinates of the vehicle within a preset time period according to a preset collection frequency as the historical trajectory; supplementing the missing coordinates of the vehicle during driving based on linear differences; and generating the preprocessed trajectory based on the trajectory coordinates and the missing coordinates.

[0021] In one embodiment of the present disclosure, the map information of the vehicle driving area is preprocessed to obtain the drivable lane line of the vehicle, specifically including: extracting the lane center line in the map information; sampling the lane center line at equal intervals to obtain multiple lane nodes; determining the lane nodes within a preset range as initial nodes based on the current position of the vehicle; based on the connection relationship between the lanes, extending the initial nodes forward and backward by specified lengths respectively to generate a search interval; connecting the lane nodes searched in the search interval as initial lane lines; based on the historical trajectory and driving direction of the vehicle, screening out unreasonable lane lines from the initial lane lines to obtain the drivable lane lines.

[0022] According to another aspect of the present disclosure, a vehicle trajectory prediction device is provided, including: a first preprocessing module, used to preprocess the historical trajectory of a vehicle to obtain a preprocessed trajectory of the vehicle; a second preprocessing module, used to preprocess the map information of the vehicle's driving area to obtain a drivable lane line of the vehicle; a trajectory feature processing module, used to perform trajectory feature processing on the preprocessed trajectory to obtain a plurality of driving interaction features between the vehicles; a lane topology extraction module, used to perform lane topology extraction on the drivable lane lines to obtain lane line connectivity features; and an autoregressive prediction module, used to perform autoregressive prediction based on the driving interaction features and the lane line connectivity features to obtain a predicted trajectory of the vehicle.

[0023] According to another aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute any one of the above-mentioned vehicle trajectory prediction methods by executing the executable instructions.

[0024] In one embodiment of the present disclosure, the electronic device is an edge computing node.

[0025] According to another aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, any one of the above-mentioned vehicle trajectory prediction methods is implemented.

[0026] The vehicle trajectory prediction solution provided by the embodiments of the present disclosure constructs a vehicle trajectory prediction based on modules such as trajectory feature processing, lane topology extraction, and autoregressive prediction. After preprocessing the original vehicle historical trajectory and map information of the vehicle driving area, a preprocessed trajectory and the vehicle's drivable lane lines are obtained. Based on the preprocessed trajectory, driving interaction features of multiple vehicles during driving are obtained. Based on the drivable lane lines, connectivity features between these drivable lane lines are obtained. The vehicle's driving trajectory is predicted based on the driving interaction features and the connectivity features. The acquisition of the vehicle's drivable lane lines realizes the mining of available semantic information in the map information. In the vehicle trajectory prediction process, the topological structure of the lane can be fully considered, and the reliability of the generated predicted trajectory is guaranteed while improving the prediction accuracy, which is conducive to meeting the safety requirements of autonomous driving.

[0027] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0029] Figure 1 A flow chart showing a vehicle trajectory prediction method according to an embodiment of the present disclosure is shown;

[0030] Figure 2 A flow chart showing another vehicle trajectory prediction method in an embodiment of the present disclosure;

[0031] Figure 3 A flow chart showing another vehicle trajectory prediction method in an embodiment of the present disclosure is shown;

[0032] Figure 4 A flow chart showing another vehicle trajectory prediction method in an embodiment of the present disclosure;

[0033] Figure 5 A schematic diagram showing a vehicle trajectory prediction model in an embodiment of the present disclosure;

[0034] Figure 6 A flow chart showing another vehicle trajectory prediction method in an embodiment of the present disclosure;

[0035] Figure 7 A flow chart showing another vehicle trajectory prediction method in an embodiment of the present disclosure;

[0036] Figure 8A flow chart showing another vehicle trajectory prediction method in an embodiment of the present disclosure;

[0037] Fig. 9 A flow chart showing another vehicle trajectory prediction method in an embodiment of the present disclosure;

[0038] Fig.10 A schematic diagram showing another vehicle trajectory prediction model in an embodiment of the present disclosure;

[0039] Fig.11 A schematic diagram showing the deployment of edge computing nodes in an embodiment of the present disclosure is shown;

[0040] Fig.12 A schematic diagram showing a vehicle trajectory prediction device according to an embodiment of the present disclosure;

[0041] Fig.13 A schematic diagram of an electronic device in an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0042] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as being limited to the examples set forth herein; on the contrary, these embodiments are provided so that the present disclosure will be more comprehensive and complete, and the concepts of the example embodiments are fully conveyed to those skilled in the art. The described features, structures, or properties may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure may be practiced while omitting one or more of the certain details, or other components, devices, steps, etc. may be adopted. In other cases, known technical solutions are not shown or described in detail to avoid obscuring various aspects of the present disclosure.

[0043] In addition, the accompanying drawings are only schematic diagrams of the present disclosure, and the same reference numerals in the drawings represent the same or similar components, so their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities, which do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0044] The exemplary embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.

[0045] GRU (Gate Recurrent Unit) is a type of recurrent neural network (RNN). GRU has two gates, a reset gate and an update gate. The reset gate determines how to combine new input information with previous memory, and the update gate defines the amount of previous memory saved to the current time step.

[0046] GAT (Graph Attention Network): It aggregates neighbor nodes through the self-attention mechanism, realizes adaptive matching of weights of different neighbors, and thus improves the accuracy of the model.

[0047] GCN (Graph Convolutional Network): GCN cleverly designed a method to extract features from graph data, so that we can use these features to perform node classification, graph classification, and link prediction on graph data, and also get graph embedding.

[0048] MLP (Multi-Layer Perception): also called Artificial Neural Network (ANN). In addition to the input and output layers, there can be one or more hidden layers in the middle. The layers of the multilayer perceptron are fully connected (that is, any neuron in the previous layer is connected to all neurons in the next layer). The hidden layer is fully connected to the input layer. Assuming that the input layer is represented by vector X, the output of the hidden layer is f(W1X+b1), W1 is the weight (also called the connection coefficient), b1 is the bias, and the function f can be a commonly used sigmoid function or tanh function. The hidden layer to the output layer can be regarded as a multi-category logistic regression, that is, Softmax regression, so the output of the output layer is softmax(W2X1+b2), and X1 represents the output of the hidden layer f(W1X+b1).

[0049] Attention: Imagine the elements in Source as a series of<Key,Value> The data pairs are composed of a certain element Query in the Target. By calculating the similarity or correlation between the Query and each Key, the weight coefficient of the Value corresponding to each Key is obtained, and then the Value is weighted and summed to obtain the final Attention value. In essence, the Attention mechanism is to perform a weighted sum of the Value values ​​of the elements in the Source, and the Query and Key are used to calculate the weight coefficient of the corresponding Value.

[0050] like Figure 1 As shown, the server executes the vehicle trajectory prediction method, including the following steps:

[0051] Step S102, preprocessing the historical trajectory of the vehicle to obtain a preprocessed trajectory of the vehicle.

[0052] Step S104, pre-processing the map information of the vehicle driving area to obtain the drivable lane lines of the vehicle.

[0053] In the present disclosure, the vehicle trajectory prediction model includes a trajectory feature processing module, a lane topology extraction module and an autoregressive prediction module.

[0054] Step S106, performing trajectory feature processing on the pre-processed trajectory to obtain driving interaction features between multiple vehicles.

[0055] Step S108: extract lane topology from drivable lane lines to obtain lane line connectivity features.

[0056] Step S110, performing autoregressive prediction based on the driving interaction features and the lane line connectivity features to obtain a predicted trajectory of the vehicle.

[0057] In this embodiment, a vehicle trajectory prediction is constructed based on modules such as trajectory feature processing, lane topology extraction, and autoregressive prediction. After preprocessing the original vehicle historical trajectory and map information of the vehicle driving area, a preprocessed trajectory and a drivable lane line of the vehicle are obtained. Based on the preprocessed trajectory, driving interaction features of multiple vehicles during driving are obtained. Based on the drivable lane lines, connectivity features between these drivable lane lines are obtained. The vehicle's driving trajectory is predicted based on the driving interaction features and the connectivity features. The acquisition of the drivable lane lines of the vehicle realizes the mining of available semantic information in the map information. In the process of vehicle trajectory prediction, the topological structure of the lane can be fully considered, and the understanding of the scene in which the vehicle is located is enhanced in the process of trajectory prediction. While improving the prediction accuracy, the reliability of the generated predicted trajectory is guaranteed, which is conducive to meeting the safety requirements of autonomous driving.

[0058] In one embodiment of the present disclosure, before performing trajectory feature processing on the preprocessed trajectory to obtain driving interaction features between multiple vehicles, it also includes: constructing a trajectory feature processing module based on a gated recurrent unit GRU encoder and a graph attention mechanism GAT; performing trajectory feature processing on the preprocessed trajectory to obtain driving interaction features between multiple vehicles, specifically including: inputting the preprocessed trajectory into the trajectory feature processing module, and outputting the driving interaction features.

[0059] Specifically, the trajectory feature processing module includes a gated recurrent unit (GRU) encoder and a graph attention mechanism (GAT).

[0060] like Figure 2 As shown, in one embodiment of the present disclosure, step S106, performing trajectory feature processing on the pre-processed trajectory to obtain a specific implementation of the driving interaction features between multiple vehicles, includes:

[0061] Step S202, mapping the preprocessed trajectory to the feature space based on a one-dimensional convolutional neural network to obtain trajectory space features.

[0062] Step S204: Encode the trajectory spatial features based on the GRU encoder to obtain respective context vectors of multiple vehicles.

[0063] Specifically, the preprocessed trajectory is input into a one-dimensional CNN convolutional neural network to map it to the feature space, and then sequentially input into a two-layer GRU encoder to obtain the context vector H = [h1,h2,...,h N ], as the motion characteristics of each vehicle.

[0064] Where N is the number of vehicles and h is the hidden context vector.

[0065] Step S206, taking the current positions of the multiple vehicles as nodes and the respective context vectors of the multiple vehicles as node features, fully connect them to construct a fully connected graph.

[0066] Step S208: extracting interaction context vectors between multiple vehicles from the fully connected graph based on GAT, and using the interaction context vectors as driving interaction features.

[0067] Specifically, the current position of each vehicle is regarded as a point on the graph, and the respective context vector is the node feature. A fully connected graph is constructed, and GAT is applied to extract the interaction features between vehicles to obtain the updated context vector after interaction. Subsequent processing will target the interaction context vector of each vehicle Do it in parallel.

[0068] In this embodiment, the trajectory spatial features corresponding to the preprocessed trajectory are encoded by the GRU encoder to obtain the motion features of multiple vehicles. By taking each vehicle as a neighbor node and aggregating them using GAT, adaptive matching of weights of different neighbors is achieved, and then the interactive motion features between multiple vehicles are obtained. The key information of motion-related features during vehicle driving is extracted, which is beneficial to improving the accuracy of trajectory prediction.

[0069] In one embodiment of the present disclosure, before performing lane topology extraction on the drivable lane lines to obtain lane line connectivity features, it also includes: constructing a variant lane graph convolutional neural network LaneGCN based on the graph convolutional neural network GCN, and constructing a lane topology extraction module based on LaneGCN and the first MLP multi-layer perceptron; performing lane topology extraction on the drivable lane lines to obtain lane line connectivity features, specifically including: inputting the drivable lane lines into the lane topology extraction module, and outputting the lane line connectivity features.

[0070] Specifically, the lane topology extraction module includes a variant lane graph convolutional neural network LaneGCN and a first MLP multi-layer perceptron.

[0071] like Figure 3 As shown, in one embodiment of the present disclosure, step S108, performing lane topology extraction on the drivable lane lines to obtain lane line connectivity features, specifically includes:

[0072] Step S302: Based on the first MLP, the drivable lane lines and the current positions of multiple vehicles are mapped to the feature space to obtain a spatial feature vector, where the spatial feature vector represents the number of lane nodes included in the drivable lane lines corresponding to each vehicle.

[0073] Specifically, the drivable lane line composed of lane nodes arranged from back to front is mapped to the feature space through two layers of MLP to obtain the spatial feature vector M i,j =[L1,L2,...,L W ].

[0074] Among them, i represents the i-th vehicle, j represents the j-th drivable lane line, and W represents the number of lane nodes contained in the lane line.

[0075] Step S304: construct a front-and-back adjacency matrix including the lane nodes based on the front-and-back connection relationship between the lane nodes.

[0076] Step S306: Based on the variant lane performance of LaneGCN, the topological connectivity features of the lane nodes are extracted from the adjacency matrix to obtain the lane line feature vector as the lane line connectivity feature.

[0077] Specifically, since the lane nodes on a single lane line only have a front-to-back connection relationship, two adjacency matrices are used to represent the front-to-back connection relationship respectively. In order to increase the receptive field, the front-to-back adjacency matrices of five scales (1, 2, 4, 6, 8) are generated with reference to the dilated convolution, and GCN is applied based on the adjacency matrix to extract the topological connectivity between the lane nodes on the drivable lane line to obtain the updated lane line feature vector

[0078] In this embodiment, by using modules such as the variant lane graph convolutional neural network LaneGCN and the first MLP, semantic information such as lane topology structure can be effectively extracted, and then by applying the above information to the trajectory prediction process, the generalization ability of prediction in complex traffic scenarios is improved.

[0079] In one embodiment of the present disclosure, before performing autoregressive prediction based on driving interaction features and lane connectivity features to obtain a predicted trajectory of the vehicle, it also includes: constructing an autoregressive prediction module based on a gated recurrent unit GRU decoder, a second MLP multilayer perceptron, a third MLP multilayer perceptron, and a lane-based attention mechanism Lane-Attention; performing autoregressive prediction based on driving interaction features and lane connectivity features to obtain a predicted trajectory of the vehicle, specifically including: inputting the driving interaction features and lane connectivity features into the autoregressive prediction module, and outputting the predicted trajectory of the vehicle.

[0080] like Figure 4 As shown, in one embodiment of the present disclosure, step S110, performing autoregressive prediction based on driving interaction features and lane line connectivity features to obtain a specific implementation of the predicted trajectory of the vehicle includes:

[0081] Step S402, according to the current position of the vehicle and the key point at the current moment, determine the first lane node and the second lane node that have the minimum distance to the drivable lane line where the vehicle is located.

[0082] Specifically, according to the current position and the current key point Find the first lane node a and the second lane node b that are closest to the current drivable lane line respectively.

[0083] Among them, at the initial stage, the key point at the current moment is determined based on the lane line feature vector, and when at least one round of prediction is completed, the key point at the current moment is generated in the previous round of prediction.

[0084] Step S404: construct relevant lane segments for vehicle driving based on the drivable lane lines to which the first lane node and the second lane node belong.

[0085] The relevant lane sections t is the time.

[0086] Step S406: obtaining a lane segment feature vector of the relevant lane segment based on the second MLP multi-layer perceptron.

[0087] Step S408: Calculate the perception A of the relevant lane segment in the lane segment feature vector at the current position of the vehicle based on Lane-Attention. i .

[0088] Step S410: Update the current position of the vehicle based on the perception to obtain updated position features

[0089]

[0090] Implement Lane-Attention to obtain updated scene understanding location features

[0091] like Figure 5 As shown, in step S412, the updated position feature and the interaction context vector are decoded based on the GRU decoder to obtain a hidden layer vector.

[0092] Specifically, the GRU decoder is composed of the interaction context vector at time t during recursive prediction The position feature after Lane-Attention at the previous moment Get the final hidden layer vector at the current moment

[0093] Step S414, input the hidden layer vector into the third MLP to predict the next predicted trajectory point at the current moment.

[0094] like Figure 5 As shown, in one embodiment of the present disclosure, further, it also includes: step S414, inputting the hidden layer vector into the third MLP to obtain the trajectory point at the next moment, specifically including: inputting the hidden layer vector into the first sub-MLP, i.e., MLP1, and the second sub-MLP, i.e., MLP2, in the third MLP; the first sub-MLP outputs the next predicted trajectory point And the corresponding probability; the second sub-MLP outputs the key point of the next moment The key point of the next moment and the next predicted trajectory point are used as the updated current moment position and the key point of the current moment when arriving at the next moment.

[0095] In this embodiment, the hidden layer vector is passed through two independent three-layer MLPs to obtain the predicted value at the current moment. As the output trajectory point and its probability, the key point of the next moment The new first lane node and the second lane node are re-determined by taking the current position of the vehicle and the key point of the current moment as the autoregressive prediction at the next moment.

[0096] Continue to calculate the perception A of the relevant lane segment in the lane segment feature vector at the current position of the vehicle based on Lane-Attention i ; Update the current position of the vehicle based on the perception and obtain the updated position feature

[0097] Step S416, generating a predicted trajectory of the vehicle based on multiple predicted next predicted trajectory points.

[0098] Specifically, by repeating S402 to S414, multiple predicted trajectories of each vehicle in the future Ns can be obtained.

[0099] like Figure 6 As shown, in one embodiment of the present disclosure, in step S406, obtaining a lane segment feature vector of the relevant lane segment based on the second MLP multilayer perceptron specifically includes:

[0100] Step S602, determining the distance and angle difference between the lane node on the relevant lane segment and the current position.

[0101] Step S604, obtaining the position code of the vehicle based on the distance and the angle difference.

[0102] Step S606: Perform two-layer feature fusion on the position encoding based on the second MLP multi-layer perceptron to obtain a lane segment feature vector.

[0103] In this embodiment, considering that the vehicle tends to travel along the lane line, the distance D between the lane node on the lane segment and the current position of the vehicle is j and the angle difference Δθ j Get position code: R j =MLP(D j ,Δθ j ).

[0104] Feature fusion through two layers of MLP Get lane segment feature vector

[0105] like Figure 7 As shown, in one embodiment of the present disclosure, in step S408, a specific implementation method of calculating the perception of the relevant lane segment in the lane segment feature vector at the current moment position of the vehicle based on Lane-Attention includes:

[0106] Step S702: The lane segment feature vector is transformed into a Key-Value data pair through a fully connected layer.

[0107] Step S704: The current position of the vehicle passes through the fully connected layer to obtain a Query.

[0108] Step S705: Input the Key-Value data pair and the Query into Lane-Attention and output the perception A i .

[0109] In this embodiment, the lane segment feature vector The fully connected layer forms the Key and Value in the attention mechanism, and the vehicle's current position features The fully connected layer is used as a query, and the attention mechanism is applied to calculate the perception A of the relevant lane segment. i :

[0110]

[0111] Among them, a is the first lane node, b is the second lane node, i represents the lane point, l represents the lane segment, Q i is the matrix of Query, is the matrix of Key, V i,l is the matrix of Value, d k is the normalization parameter.

[0112] In one embodiment of the present disclosure, before performing trajectory feature processing on the preprocessed trajectory to obtain driving interaction features between multiple vehicles, it also includes: constructing a model training set and a verification set based on the preprocessed trajectory and the drivable lane lines; training a prediction model based on the model training set and the verification set, and the prediction model includes a trajectory feature processing module, a lane topology extraction module and an autoregressive prediction module.

[0113] In one embodiment of the present disclosure, it also includes: based on the next predicted trajectory point, using the first loss function to perform convergence optimization on the prediction model; based on the corresponding probability, using the first loss function to perform convergence optimization on the prediction model to obtain an optimized prediction model.

[0114] Specifically, assuming that a vehicle has K lanes, the prediction result for the vehicle is Where T is the future time range, and the three dimensions correspond to (x coordinate, y coordinate) s i and probability p i, only the probability of the final predicted point is taken as the probability of the trajectory, the first loss function is applied to the predicted coordinates, that is, the next predicted trajectory point, and the second loss function is applied to the probability, corresponding to the regression subtask and the classification subtask. The regression subtask can use loss functions such as mean square error, L2 norm, Huber, etc., and the classification subtask can use loss functions such as cross entropy and KL divergence. The total loss function of the model is:

[0115] L total =L reg +λL cls (3)

[0116] Among them, L reg is the first loss function, L cls is the second loss function, λ is a hyperparameter, which depends on the importance of the two subtasks. The total loss function is used to train the model proposed in the present invention end-to-end. When the training is completed, the current network weights are saved for subsequent testing and reasoning.

[0117] Specifically, the regression subtask only applies Huber Loss to the drivable lane line that is closest to the true trajectory:

[0118]

[0119] Among them, M is the number of trajectories generated by the corresponding lane line, is a constant value, and s i is the predicted trajectory coordinate, s gt is the real trajectory coordinate.

[0120] The classification subtask is also only applied to the drivable lane line closest to the true trajectory, using the KL divergence loss function:

[0121]

[0122]

[0123]

[0124] Where D(s i ,s gt )=||s i,T -s gt,T ||2 represents the Euclidean distance between the point at time T on the i-th predicted trajectory and the GT. The KL divergence between the two in the classification subtask loss function indicates that we hope that the distribution of the predicted score is close to the distribution of the true trajectory, that is, the closer the overall distance to the true trajectory, the higher the score.

[0125] λ is taken as 0.1 based on experience. During training, Adam is used as the optimizer, batchsize is 25, dropout=0.1 is introduced on the last layer of GRU, ReLU is used as the activation function, the hidden feature dimensions are all 128, the initial learning rate is 0.01, and the total loss function is used to train the model proposed in the present invention end-to-end. During the training process, if the loss function value on the validation set does not decrease for three consecutive cycles, the learning rate is multiplied by 0.5. If the loss function value on the validation set does not decrease for five consecutive cycles, the model is judged to have converged, the training is completed, and the current network weights are saved for subsequent testing and reasoning.

[0126] Save the trained and converged model, input the preprocessed test set data into the trained model, and predict the future trajectory coordinates of the vehicle.

[0127] Preferably, considering the limited computing power on the vehicle side, relying on the high reliability and low latency characteristics of the 5G network, edge computing nodes can be deployed on both sides of the road. The historical observation data of all relevant vehicles in the scene, such as trajectory coordinates and the road topology of the current scene, are collected on the node, and the trajectory prediction task is deployed at the same time. The edge computing node collects the historical trajectory coordinates of all vehicles within the coverage area and completes the prediction based on the high-precision bird's-eye view map of the current scene. Subsequently, downstream tasks such as path planning and collision risk assessment of each vehicle can be completed based on the prediction results. Finally, the results of each vehicle's planned path are sent to the vehicle terminal to complete the vehicle trajectory prediction and ensure driving safety.

[0128] like Figure 8 As shown, in one embodiment of the present disclosure, the historical trajectory of the vehicle is preprocessed to obtain the preprocessed trajectory of the vehicle, specifically including:

[0129] Step S802 , collecting the trajectory coordinates of the vehicle within a preset time period according to a preset collection frequency as a historical trajectory.

[0130] Specifically, historical trajectories and map information of vehicle driving areas can also be stored in the Argoverse dataset, which is a dataset used to support 3DTracking and Motion Forecasting research for autonomous vehicles.

[0131] Step S804, supplementing the missing coordinates during the vehicle's travel based on the linear difference.

[0132] Step S806: Generate a preprocessing trajectory based on the trajectory coordinates and the missing coordinates.

[0133] In this embodiment, the Argoverse dataset provides a bird's-eye view map of all vehicles and corresponding scenes in multiple scenes. Assuming that the edge computing node has completed the collection of vehicle GPS trajectory points and bird's-eye view map information, the Argoverse dataset can be reasonably used as the data collected by the edge computing node for simulation. The trajectory input is the GPS positioning trajectory coordinates of all vehicles with a historical 2s sampling frequency of 10Hz. The missing coordinate data is supplemented by linear interpolation to complete the preprocessing of the trajectory data and obtain the preprocessed trajectory.

[0134] like Fig. 9 As shown, in one embodiment of the present disclosure, the map information of the vehicle driving area is preprocessed to obtain the drivable lane line of the vehicle, which specifically includes:

[0135] Step S902: extracting the lane centerline in the map information.

[0136] Step S904: sampling the lane centerline at equal intervals to obtain a plurality of lane nodes.

[0137] Specifically, the map information is presented in the form of vectorized lane lines, and all lane center lines are sampled at equal intervals of 1m as lane nodes L i ={x i ,y i ,id,pre,suc,left,right}.

[0138] Among them, x and y represent the longitude and latitude coordinates, id is the serial number of the node, pre, suc, left, and right represent the node IDs that have four connection relationships with the lane node: predecessor, successor, left neighbor, and right neighbor. Lane nodes with different connection relationships constitute lane lines.

[0139] Step S906: determining a lane node within a preset range as an initial node based on the current position of the vehicle.

[0140] For example, according to the current position of each vehicle in the map, an initial node is selected within a range of 2m.

[0141] Step S908, based on the connection relationship between lanes, the initial node is extended forward and backward by a specified length to generate a search interval.

[0142] Among them, according to the connection relationship between lane nodes, the initial lane node is extended forward and backward by a certain length. The search length of this search interval is a constant acceleration motion model:

[0143]

[0144] Among them, the buffer distance D0 is 20m forward and 10m backward after experimental verification.

[0145] Step S910: Connect the lane nodes found in the search interval to form initial lane lines.

[0146] Step S912: based on the historical trajectory and driving direction of the vehicle, unreasonable lane lines are screened out from the initial lane lines to obtain drivable lane lines.

[0147] Specifically, the searched lane nodes are finally connected into lane lines, and unreasonable lane lines are screened out based on whether the lane lines contain the vehicle's historical trajectory and whether the direction is consistent with the vehicle's driving direction, to obtain the final drivable lane lines and complete the preprocessing of map information.

[0148] In this embodiment, the features of the drivable lane lines obtained by the above-mentioned preprocessing method can ensure the reliability of lane topology feature extraction.

[0149] like Fig.10 As shown, the prediction model includes a trajectory feature processing module, a lane topology extraction module and an autoregressive prediction module.

[0150] In the trajectory feature processing module, the preprocessed trajectory obtained by preprocessing the historical trajectory passes through the GRU encoder to generate the context vector H = [h1,h2,...,h N ], taking the current positions of multiple vehicles as nodes, taking the respective context vectors of multiple vehicles as node features, fully connecting, constructing a fully connected graph, and extracting the interaction context vectors between multiple vehicles from the fully connected graph based on GAT

[0151] In the lane topology extraction module, the map information is preprocessed to obtain the drivable lane lines. Based on MLP, the drivable lane lines and the current positions of multiple vehicles are mapped to the feature space to obtain the spatial feature vector M i,j =[L1,L2,...,L W ], based on the variant lane performance of LaneGCN, the topological connectivity features of lane nodes are extracted from the adjacency matrix to obtain the lane line feature vector

[0152] In the autoregressive prediction module, the GRU decoder is composed of the context vector at time t during recursive prediction. The position feature after Lane-Attention at the previous moment Get the final hidden layer vector at the current moment The hidden layer vector is passed through two independent MLP1 and MLP2 to obtain the predicted value P at the current moment. t refineAs the output trajectory point and its probability, the key point of the next moment

[0153] like Fig.11 As shown, an edge computing node 1140 is deployed near the road where the vehicle 1120 is traveling. The edge computing node communicates with the cloud server through the core network. The historical observation data of all relevant vehicles in the scene, such as trajectory coordinates and the road topology of the current scene, are collected on the node, and the trajectory prediction task is deployed at the same time. The edge computing node collects the historical trajectory coordinates of all vehicles within the coverage area and completes the prediction based on the high-precision bird's-eye view map of the current scene. Subsequently, downstream tasks such as path planning and collision risk assessment of each vehicle can be completed based on the prediction results. Finally, the results such as the planned path of each vehicle are sent to the vehicle terminal to complete the vehicle trajectory prediction and ensure driving safety.

[0154] It should be noted that the above figures are only schematic illustrations of the processes included in the method according to an exemplary embodiment of the present invention, and are not intended to be limiting. It is easy to understand that the processes shown in the above figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be performed synchronously or asynchronously, for example, in multiple modules.

[0155] It will be appreciated by those skilled in the art that various aspects of the present invention may be implemented as a system, method or program product. Therefore, various aspects of the present invention may be specifically implemented in the following forms, namely: a complete hardware implementation, a complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software, which may be collectively referred to herein as a "circuit", "module" or "system".

[0156] Refer to the following Fig.12 The vehicle trajectory prediction device 1200 according to this embodiment of the present invention is described. Fig.12 The vehicle trajectory prediction device 1200 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0157] The vehicle trajectory prediction device 1200 is in the form of a hardware module. The components of the vehicle trajectory prediction device 1200 may include but are not limited to: a first preprocessing module 1202, which is used to preprocess the historical trajectory of the vehicle to obtain the preprocessed trajectory of the vehicle; a second preprocessing module 1204, which is used to preprocess the map information of the vehicle driving area to obtain the drivable lane line of the vehicle; a trajectory feature processing module 1206, which is used to perform trajectory feature processing on the preprocessed trajectory to obtain the driving interaction features between multiple vehicles; a lane topology extraction module 1208, which is used to perform lane topology extraction on the drivable lane line to obtain the lane line connectivity features; an autoregressive prediction module 1210, which is used to perform autoregressive prediction based on the driving interaction features and the lane line connectivity features to obtain the predicted trajectory of the vehicle.

[0158] Refer to the following Fig.13 An electronic device 1300 according to this embodiment of the present invention is described. Fig.13 The electronic device 1300 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0159] As a preferred implementation, the electronic device is an edge computing node.

[0160] like Fig.13 As shown, the electronic device 1300 is in the form of a general computing device. The components of the electronic device 1300 may include but are not limited to: at least one processing unit 1310, at least one storage unit 1320, and a bus 1330 connecting different system components (including the storage unit 1320 and the processing unit 1310).

[0161] The storage unit stores program codes, which can be executed by the processing unit 1310, so that the processing unit 1310 performs the steps of various exemplary embodiments of the present invention described in the above “Exemplary Method” section of this specification. For example, the processing unit 1310 can perform the following steps: Figure 1 Steps S102 to S110 shown in , and other steps defined in the vehicle trajectory prediction method of the present disclosure.

[0162] The storage unit 1320 may include a readable medium in the form of a volatile storage unit, such as a random access storage unit (RAM) 13201 and / or a cache storage unit 13202 , and may further include a read-only storage unit (ROM) 13203 .

[0163] The storage unit 1320 may also include a program / utility 13204 having a set (at least one) of program modules 13205, such program modules 13205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0164] Bus 1330 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0165] The electronic device 1300 may also communicate with one or more external devices 1370 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device, and / or may communicate with any device that enables the electronic device 1300 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 1350. Furthermore, the electronic device 1300 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 1360. As shown, the network adapter 1360 communicates with other modules of the electronic device 1300 via a bus 1330. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0166] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the implementation of the present disclosure.

[0167] In an exemplary embodiment of the present disclosure, a computer-readable storage medium is also provided, on which a program product capable of implementing the above method of the present specification is stored. In some possible implementations, various aspects of the present invention may also be implemented in the form of a program product, which includes a program code, and when the program product is run on a terminal device, the program code is used to enable the terminal device to execute the steps according to various exemplary embodiments of the present invention described in the above "Exemplary Method" section of the present specification.

[0168] The program product for implementing the above method according to an embodiment of the present invention may adopt a portable compact disk read-only memory (CD-ROM) and include program code, and may be run on a terminal device, such as a personal computer. However, the program product of the present invention is not limited thereto, and in this document, a readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, apparatus, or device.

[0169] Computer readable signal media may include data signals propagated in baseband or as part of a carrier wave, in which readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Readable signal media may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0170] The program code embodied on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the foregoing.

[0171] Program code for performing the operations of the present invention may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., through the Internet using an Internet service provider).

[0172] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into multiple modules or units to be embodied.

[0173] In addition, although the steps of the method in the present disclosure are described in a specific order in the drawings, this does not require or imply that the steps must be performed in this specific order, or that all the steps shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps, etc.

[0174] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a mobile terminal, or a network device, etc.) to execute the method according to the implementation of the present disclosure.

[0175] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any modification, use or adaptation of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the appended claims.

Claims

1. A vehicle trajectory prediction method, characterized in that: include: Preprocessing the historical trajectory of the vehicle to obtain a preprocessed trajectory of the vehicle; Preprocessing the map information of the vehicle driving area to obtain the drivable lane line of the vehicle; Inputting the preprocessed trajectory into a trajectory feature processing module, outputting the driving interaction features between the multiple vehicles, the trajectory feature processing module is constructed based on a gated recurrent unit GRU encoder and a graph attention mechanism GAT, including: encoding the trajectory space features corresponding to the preprocessed trajectory based on the GRU encoder to obtain the motion features of the multiple vehicles, taking the multiple vehicles as neighbor nodes, and aggregating them using the GAT to obtain the interactive motion features between the multiple vehicles; Performing lane topology extraction on the drivable lane lines to obtain lane line connectivity features; The driving interaction features and the lane line connectivity features are input into an autoregressive prediction module, and a predicted trajectory of the vehicle is output. The autoregressive prediction module is constructed based on a gated recurrent unit GRU decoder, a second MLP multilayer perceptron, a third MLP multilayer perceptron, and a lane-based attention mechanism Lane-Attention, including: based on the GRU decoder, a hidden layer vector at the current moment is obtained by the interaction context vector and the updated position feature during recursive prediction, the hidden layer vector is input into the third MLP to obtain a next predicted trajectory point, and the predicted trajectory of the vehicle is generated based on multiple predicted next predicted trajectory points, and the updated position feature is obtained by the lane segment feature vector after the Lane-Attention, and the lane segment feature vector is obtained based on the second MLP multilayer perceptron.

2. The vehicle trajectory prediction method according to claim 1, characterized in that: The step of inputting the preprocessed trajectory into the trajectory feature processing module and outputting the driving interaction feature specifically includes: Mapping the preprocessed trajectory to a feature space based on a one-dimensional convolutional neural network to obtain trajectory space features; Encoding the trajectory spatial features based on the GRU encoder to obtain respective context vectors of the plurality of vehicles; Taking the current positions of the multiple vehicles as nodes and the respective context vectors of the multiple vehicles as node features, performing full connection, and constructing a fully connected graph; Based on the GAT, a plurality of interaction context vectors between the vehicles are extracted from the fully connected graph, and the interaction context vectors are used as the driving interaction features.

3. The vehicle trajectory prediction method according to claim 2, characterized in that: Before extracting lane topology from the drivable lane lines to obtain lane line connectivity features, the method further includes: A variant lane graph convolutional neural network LaneGCN is constructed based on the graph convolutional neural network GCN, and a lane topology extraction module is constructed based on the LaneGCN and the first MLP multi-layer perceptron; The extracting lane topology of the drivable lane line to obtain the lane line connectivity feature specifically includes: The drivable lane line is input into the lane topology extraction module, and the lane line connectivity feature is output.

4. The vehicle trajectory prediction method according to claim 3, characterized in that: The step of inputting the drivable lane line into the lane topology extraction module and outputting the lane line connectivity feature specifically includes: Mapping the drivable lane line and the current positions of the plurality of vehicles to a feature space based on the first MLP to obtain a spatial feature vector, wherein the spatial feature vector represents the number of lane nodes included in the drivable lane line corresponding to each of the vehicles; Based on the front-to-back connection relationship between the lane nodes, construct a front-to-back adjacency matrix including the lane nodes; Based on the variant lane performance of the LaneGCN, the topological connectivity features of the lane nodes are extracted from the adjacency matrix to obtain a lane line feature vector as the lane line connectivity feature.

5. The vehicle trajectory prediction method according to claim 4, characterized in that: The step of inputting the driving interaction feature and the lane line connectivity feature into the autoregressive prediction module and outputting the predicted trajectory of the vehicle specifically includes: Determine, according to the current position of the vehicle and the key point at the current moment, a first lane node and a second lane node having a minimum distance from the drivable lane line where the vehicle is located, wherein initially, the key point at the current moment is determined based on the lane line feature vector; Constructing a relevant lane segment for the vehicle to travel based on the drivable lane line to which the first lane node and the second lane node belong; Obtaining a lane segment feature vector of the relevant lane segment based on the second MLP multilayer perceptron; Calculating, based on Lane-Attention, a perception of the relevant lane segment in the lane segment feature vector at the current moment position of the vehicle; Based on the perception, the current position of the vehicle is updated to obtain an updated position feature; Decoding the updated position feature and the interaction context vector based on the GRU decoder to obtain a hidden layer vector; Input the hidden layer vector into the third MLP to predict the next predicted trajectory point at the current moment; A predicted trajectory of the vehicle is generated based on a plurality of predicted next predicted trajectory points.

6. The vehicle trajectory prediction method according to claim 5, characterized in that: The obtaining of the lane segment feature vector of the relevant lane segment based on the second MLP multilayer perceptron specifically includes: Determine the distance and angle difference between the lane node on the relevant lane segment and the current position; Obtaining a position code of the vehicle based on the distance and the angle difference; Based on the second MLP multi-layer perceptron, two-layer feature fusion is performed on the position encoding to obtain the lane segment feature vector.

7. The vehicle trajectory prediction method according to claim 5, characterized in that: The Lane-Attention-based calculation of the perception of the relevant lane segment in the lane segment feature vector at the current moment position of the vehicle specifically includes: The lane segment feature vector is formed into a Key-Value data pair through a fully connected layer; The current position of the vehicle passes through the fully connected layer to obtain a Query; The Key-Value data pair and the Query are input into the Lane-Attention, and the perception is output.

8. The vehicle trajectory prediction method according to claim 5, characterized in that: The step of inputting the hidden layer vector into the third MLP to obtain the trajectory point at the next moment specifically includes: Inputting the hidden layer vector into the first sub-MLP and the second sub-MLP in the third MLP respectively; The first sub-MLP outputs the next predicted trajectory point and the corresponding probability; The second sub-MLP outputs the key point at the next moment, and the key point at the next moment and the next predicted trajectory point are used as the updated current moment position and the key point at the current moment when arriving at the next moment.

9. The vehicle trajectory prediction method according to claim 8, characterized in that: Before performing trajectory feature processing on the preprocessed trajectory to obtain the driving interaction features between the plurality of vehicles, the method further includes: Constructing a model training set and a validation set based on the preprocessed trajectory and the drivable lane line; A prediction model is trained based on the model training set and the validation set, wherein the prediction model includes the trajectory feature processing module, the lane topology extraction module and the autoregressive prediction module.

10. The vehicle trajectory prediction method according to claim 9, characterized in that: Also includes: Based on the next predicted trajectory point, using a first loss function to perform convergence optimization on the prediction model; Based on the corresponding probability, the prediction model is converged and optimized using a first loss function to obtain the optimized prediction model.

11. The vehicle trajectory prediction method according to any one of claims 1 to 10, characterized in that: The preprocessing of the historical trajectory of the vehicle to obtain the preprocessed trajectory of the vehicle specifically includes: Collecting the trajectory coordinates of the vehicle within a preset time period according to a preset collection frequency as the historical trajectory; Supplementing missing coordinates during the vehicle's travel based on linear differences; The pre-processed trajectory is generated based on the trajectory coordinates and the missing coordinates.

12. The vehicle trajectory prediction method according to claim 11, characterized in that: The preprocessing of the map information of the vehicle driving area to obtain the drivable lane line of the vehicle specifically includes: Extracting lane center lines from the map information; Sampling the lane centerline at equal intervals to obtain a plurality of lane nodes; Determining the lane node within a preset range as an initial node based on the current position of the vehicle; Based on the connection relationship between the lanes, the initial node is extended forward and backward by a specified length to generate a search interval; Connecting the lane nodes searched in the search interval as initial lane lines; Based on the historical trajectory and driving direction of the vehicle, unreasonable lane lines are screened out from the initial lane lines to obtain the drivable lane lines.

13. A vehicle trajectory prediction device, characterized in that: include: A first preprocessing module, used for preprocessing the historical trajectory of the vehicle to obtain the preprocessed trajectory of the vehicle; A second preprocessing module is used to preprocess the map information of the vehicle driving area to obtain the drivable lane line of the vehicle; A trajectory feature processing module is used to input the preprocessed trajectory into the trajectory feature processing module, and output the driving interaction features between the multiple vehicles. The trajectory feature processing module is constructed based on a gated recurrent unit GRU encoder and a graph attention mechanism GAT, including: encoding the trajectory space features corresponding to the preprocessed trajectory based on the GRU encoder to obtain the motion features of the multiple vehicles, taking the multiple vehicles as neighbor nodes, and aggregating them using the GAT to obtain the interactive motion features between the multiple vehicles; A lane topology extraction module is used to extract lane topology from the drivable lane lines to obtain lane line connectivity features; An autoregressive prediction module is used to input the driving interaction feature and the lane line connectivity feature into the autoregressive prediction module, and output the predicted trajectory of the vehicle. The autoregressive prediction module is constructed based on a gated recurrent unit GRU decoder, a second MLP multilayer perceptron, a third MLP multilayer perceptron, and a lane-based attention mechanism Lane-Attention, including: based on the GRU decoder, during recursive prediction, a hidden layer vector at the current moment is obtained by using the interaction context vector and the updated position feature, the hidden layer vector is input into the third MLP to obtain a next predicted trajectory point, and the predicted trajectory of the vehicle is generated based on multiple predicted next predicted trajectory points. The updated position feature is obtained by a lane segment feature vector after the Lane-Attention, and the lane segment feature vector is obtained based on the second MLP multilayer perceptron.

14. An electronic device, characterized in that: include: processor; as well as A memory, configured to store executable instructions of the processor; Wherein, the processor is configured to execute the vehicle trajectory prediction method described in any one of claims 1 to 12 by executing the executable instructions.

15. The electronic device according to claim 14, characterized in that: The electronic device is an edge computing node.

16. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the vehicle trajectory prediction method according to any one of claims 1 to 12 is implemented.

Images