Method and device for predicting vehicle motion trajectory, storage medium and electronic device

Abstract

The application discloses a vehicle motion trajectory prediction method and device, a storage medium and an electronic device. The method comprises the following steps: acquiring a vehicle portrait feature, a real-time trajectory feature and an electronic map feature of a target vehicle; and using a target prediction model to predict the vehicle portrait feature, the real-time trajectory feature and the electronic map feature, so as to obtain a motion trajectory of the target vehicle. The application solves the technical problem of low prediction accuracy of the vehicle motion trajectory in the related art.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, and more specifically, to a method, apparatus, storage medium, and electronic device for predicting vehicle trajectory. Background Technology

[0002] The rapid increase in the number of autonomous vehicles has spurred the development of intelligent transportation systems, which can effectively reduce human-caused traffic accidents. Planning and control are among the key technologies for autonomous vehicles. These systems need to accurately assess the risks of the current driving scenario and plan the vehicle's trajectory based on the assessment results, thereby ensuring safe and reasonable interaction between the autonomous vehicle and other vehicles. The accuracy of the vehicle's trajectory in the driving scenario has a significant impact on the rationality of planning and control and driving safety.

[0003] In real-world traffic scenarios, vehicle trajectories are influenced not only by prior knowledge such as driving destination, road structure, traffic signs, and traffic rules, but also by uncertain posterior knowledge such as other dynamic traffic participants and the driver's subjective driving intentions. In related technical solutions, rule-based trajectory prediction methods are used to predict vehicle trajectories in simple driving scenarios with a limited prediction range. However, in more complex driving scenarios, deep learning-based prediction methods exhibit superior performance. Nevertheless, deep learning-based trajectory prediction methods, due to their insufficient integration of prior knowledge about the vehicle within the driving scenario, are only applicable to specific scenarios. In real-world traffic situations, their accuracy in predicting vehicle trajectories is low, failing to meet practical needs.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This invention provides a method, apparatus, storage medium, and electronic device for predicting vehicle trajectory, to at least solve the technical problem of low accuracy in predicting vehicle trajectory in related technologies.

[0006] According to one aspect of the present invention, a method for predicting vehicle trajectory is provided, comprising: acquiring vehicle profile features, real-time trajectory features, and electronic map features of a target vehicle, wherein the vehicle profile features include a first feature, a second feature, and a third feature, the first feature being used to characterize attribute information of the target vehicle, the second feature being used to characterize travel information of the target vehicle, and the third feature being used to characterize state information of the target vehicle, and the electronic map features including dynamic map features and static map features; and using a target prediction model to predict the vehicle profile features, real-time trajectory features, and electronic map features to obtain the trajectory of the target vehicle, wherein the target prediction model is trained using multiple sets of data through machine learning, and each set of data includes: vehicle profile sample features, real-time trajectory sample features, and electronic map sample features.

[0007] Optionally, a target prediction model is used to predict the vehicle profile features, real-time trajectory features, and electronic map features to obtain the target vehicle's motion trajectory, which includes: encoding the vehicle profile features, real-time trajectory features, and electronic map features to obtain encoding results; performing feature fusion processing on the encoding results to obtain fusion results; performing decoding processing on the fusion results to obtain decoding results; and determining the target vehicle's motion trajectory based on the decoding results.

[0008] Optionally, the vehicle profile features, real-time trajectory features, and electronic map features are encoded to obtain the encoding results, including: enhancing the vehicle profile features, real-time trajectory features, and electronic map features based on a preset method to obtain the enhanced results; and encoding the enhanced results using a multilayer perceptron structure and a batch normalization structure to obtain the encoded results.

[0009] Optionally, feature fusion processing is performed on the encoding results to obtain the fusion results, including: using a converter structure to repeatedly perform self-attention and cross-attention operations on the time dimension and vehicle dimension of the encoding results to obtain the fusion results.

[0010] Optionally, the fusion result is decoded to obtain the following: adding dimensions to the fusion result to obtain a first intermediate result; performing a transformation operation on the first intermediate result using a multilayer perceptron structure to obtain a transformed result; repeatedly performing self-attention operations on the time dimension and vehicle dimension of the transformed result using a converter structure to obtain a second intermediate result; performing regularization processing on the second intermediate result using a layer normalization structure to obtain a third intermediate result; and decoding the third intermediate result using a multilayer perceptron structure and a batch normalization structure to obtain the decoded result.

[0011] Optionally, determining the motion trajectory of the target vehicle based on the decoding result includes: calculating the similarity of the decoding result to obtain the similarity result; and merging the decoding result using the similarity result to obtain the motion trajectory of the target vehicle.

[0012] According to another aspect of the present invention, a vehicle trajectory prediction device is also provided, comprising: an acquisition module, configured to acquire vehicle profile features, real-time trajectory features, and electronic map features of a target vehicle, wherein the vehicle profile features include a first feature, a second feature, and a third feature, the first feature being used to characterize attribute information of the target vehicle, the second feature being used to characterize travel information of the target vehicle, and the third feature being used to characterize state information of the target vehicle, and the electronic map features including dynamic map features and static map features; and a prediction module, configured to use a target prediction model to predict the vehicle profile features, real-time trajectory features, and electronic map features to obtain the trajectory of the target vehicle, wherein the target prediction model is trained using multiple sets of data through machine learning, and each set of data includes: vehicle profile sample features, real-time trajectory sample features, and electronic map sample features.

[0013] Optionally, the prediction module is also used to: encode vehicle profile features, real-time trajectory features and electronic map features to obtain encoding results; perform feature fusion processing on the encoding results to obtain fusion results; perform decoding processing on the fusion results to obtain decoding results; and determine the motion trajectory of the target vehicle based on the decoding results.

[0014] Optionally, the prediction module is also used to: enhance vehicle profile features, real-time trajectory features and electronic map features based on a preset method to obtain enhanced results; and encode the enhanced results using a multilayer perceptron structure and a batch normalization structure to obtain encoded results.

[0015] Optionally, the prediction module is also used to: repeatedly perform self-attention and cross-attention operations on the time and vehicle dimensions of the encoded result using a converter structure to obtain the fusion result.

[0016] Optionally, the prediction module is further configured to: add dimensions to the fusion result to obtain a first intermediate result; perform a transformation operation on the first intermediate result using a multilayer perceptron structure to obtain a transformed result; repeatedly perform self-attention operations on the time dimension and vehicle dimension of the transformed result using a converter structure to obtain a second intermediate result; perform regularization processing on the second intermediate result using a layer normalization structure to obtain a third intermediate result; and perform decoding processing on the third intermediate result using a multilayer perceptron structure and a batch normalization structure to obtain a decoded result.

[0017] Optionally, the prediction module is also used to: calculate the similarity of the decoding results to obtain similarity results; and use the similarity results to merge the decoding results to obtain the motion trajectory of the target vehicle.

[0018] According to another aspect of the present invention, a non-volatile storage medium is also provided, wherein a computer program is stored in the storage medium, and the computer program is configured to execute the vehicle trajectory prediction method described in any of the above embodiments when running.

[0019] According to another aspect of the present invention, a processor is also provided, the processor being configured to run a program, wherein the program is configured to execute the vehicle trajectory prediction method of any of the above embodiments during runtime.

[0020] According to another aspect of the present invention, an electronic device is also provided, including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the vehicle trajectory prediction method of any of the above embodiments.

[0021] In this embodiment of the invention, by acquiring the vehicle profile features, real-time trajectory features, and electronic map features of the target vehicle, and then using a target prediction model to predict the vehicle profile features, real-time trajectory features, and electronic map features, the movement trajectory of the target vehicle is obtained. This achieves the purpose of combining vehicle profile features, real-time trajectory features, and electronic map features for trajectory prediction, thereby improving the technical effect of predicting the vehicle movement trajectory and solving the technical problem of low prediction accuracy of vehicle movement trajectory in related technologies. Attached Figure Description

[0022] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0023] Figure 1 This is a flowchart of a method for predicting vehicle trajectory according to one embodiment of the present invention;

[0024] Figure 2 This is a schematic diagram of a vehicle profile feature according to one embodiment of this application;

[0025] Figure 3 This is a schematic diagram illustrating a process for obtaining vehicle profile features according to one embodiment of this application;

[0026] Figure 4 This is a schematic diagram of a method for predicting vehicle trajectory according to one embodiment of this application;

[0027] Figure 5 This is a structural block diagram of a vehicle trajectory prediction device according to one embodiment of the present invention. Detailed Implementation

[0028] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0029] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0030] According to one embodiment of the present invention, an embodiment of a method for predicting vehicle trajectory is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0031] This method embodiment can be executed in an electronic device or similar computing device that includes memory and a processor in a vehicle. Taking operation on a vehicle terminal as an example, the vehicle terminal may include one or more processors (processors may include, but are not limited to, central processing units (CPUs), graphics processing units (GPUs), digital signal processing (DSP) chips, microprocessors (MCUs), programmable logic devices (FPGAs), neural network processors (NPUs), tensor processors (TPUs), artificial intelligence (AI) type processors, etc.) and memory for storing data. Optionally, the above-described automotive electronic device may also include transmission devices, input / output devices, and display devices for communication functions. Those skilled in the art will understand that the above structural description is merely illustrative and does not limit the structure of the vehicle terminal. For example, the vehicle terminal may include more or fewer components than those described above, or have a different configuration than those described above.

[0032] The memory can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to the information processing method in this embodiment of the invention. The processor executes various functional applications and data processing by running the computer program stored in the memory, thereby implementing the aforementioned information processing method. The memory may include high-speed random access memory (RAM) and non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the mobile terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks (LANs), mobile communication networks, and combinations thereof.

[0033] The transmission device is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the mobile terminal's communication provider. In one example, the transmission device includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0034] The display device can be, for example, a touchscreen liquid crystal display (LCD) and a touch display (also referred to as a "touchscreen" or "touch display"). This LCD allows a user to interact with the user interface of the mobile terminal. In some embodiments, the mobile terminal has a graphical user interface (GUI), which allows the user to interact with the GUI via finger contact and / or gestures on a touch-sensitive surface. Optional human-computer interaction functions include: creating web pages, drawing, word processing, creating electronic documents, playing games, video conferencing, instant messaging, sending and receiving emails, a call interface, playing digital video, playing digital music, and / or web browsing, etc. Executable instructions for performing the above human-computer interaction functions are configured / stored in one or more processor-executable computer program products or readable storage media.

[0035] This embodiment provides a method for predicting the trajectory of a vehicle running on the aforementioned vehicle terminal. Figure 1 This is a flowchart of a method for predicting vehicle trajectory according to one embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps:

[0036] Step S12: Obtain the vehicle profile features, real-time trajectory features, and electronic map features of the target vehicle. The vehicle profile features include a first feature, a second feature, and a third feature. The first feature is used to characterize the attribute information of the target vehicle, the second feature is used to characterize the travel information of the target vehicle, and the third feature is used to characterize the status information of the target vehicle. The electronic map features include dynamic map features and static map features.

[0037] Step S14: The target prediction model is used to predict the vehicle profile features, real-time trajectory features and electronic map features to obtain the movement trajectory of the target vehicle. The target prediction model is trained by machine learning using multiple sets of data. Each set of data includes: vehicle profile sample features, real-time trajectory sample features and electronic map sample features.

[0038] Specifically, the aforementioned target vehicles can be autonomous vehicles. Figure 2 This is a schematic diagram of a vehicle profile feature according to one embodiment of this application, such as... Figure 2As shown, the first, second, and third features in the vehicle profile correspond to the long-term, mid-to-long-term, and short-term features, respectively. Long-term features characterize the target vehicle's attribute information, including its color, size, license plate number, vehicle type, model, and acceleration / deceleration performance. Mid-to-long-term features characterize the target vehicle's travel information, including predicted destinations and historical travel trajectories. Short-term features characterize the target vehicle's status information, including predicted driving conditions and control light information.

[0039] Figure 3 This is a schematic diagram illustrating a process for obtaining vehicle profile features according to one embodiment of this application, such as... Figure 3 As shown, the information in the vehicle profile features can be obtained from the target vehicle's perception module, fusion module, and cloud server to construct vehicle profile tags.

[0040] Specifically, the aforementioned vehicle profile features can be obtained by identifying data collected by sensors in autonomous vehicles using a target deep learning model. For example, the target deep learning model can be trained using an labeled image dataset collected by the target vehicle's perception module. The image data collected by the autonomous vehicle's perception module is labeled with information such as the target vehicle's location, license plate number, vehicle color, vehicle type and model, and vehicle control information. An initial deep learning model is trained using this labeled image dataset to obtain the target deep learning model. This target deep learning model is then deployed on the autonomous vehicle, and based on the real-time image data collected by the vehicle's perception module, the vehicle's location, license plate number, vehicle color (e.g., yellow, blue, green), vehicle type information (e.g., taxis, dump trucks, buses, excavators, passenger cars, commercial vehicles, etc.), vehicle model information, and vehicle control light information (e.g., indicator lights for brake, left turn, right turn, etc.) are obtained in the target scene.

[0041] Based on the acquired vehicle model information, the target vehicle can send a request to the cloud server via the network to obtain and update its own stored vehicle size and acceleration / deceleration performance information. The cloud server contains a vehicle knowledge base that provides functions such as information input, storage, query, and update interfaces. It can pre-store model information, vehicle size, and corresponding acceleration / deceleration performance information for all vehicles on the market, and all information in the vehicle knowledge base is updated according to a preset period.

[0042] For example, the aforementioned target deep learning model can also be trained using the computational result dataset of the target vehicle fusion module. This dataset includes the target vehicle's historical travel trajectory, the time and location information of surrounding vehicles' appearance and disappearance, and their location information. The initial deep learning model is trained using this dataset to obtain the target deep learning model. Based on the target deep learning model, predictions are made using the real-time computational results of the fusion module to obtain the predicted destination and driving state of the target vehicle. Specifically, by using the target deep learning model to predict the time and location information of the appearance and disappearance of surrounding vehicles, the target vehicle's destination in the future can be determined. By using the target deep learning model to predict the target vehicle's historical travel trajectory information, the target vehicle's driving intentions in the future can be determined, such as lane keeping, changing lanes to the left or right, decelerating while keeping lanes, decelerating while changing lanes to the left or right, and making a U-turn. Based on these driving intentions, the predicted driving state of the target vehicle can then be determined as either a stable or aggressive state.

[0043] The aforementioned electronic map features include dynamic map features and static map features. For example, dynamic map features may include traffic light information and road condition information, while static map features may include road sign information, such as lane center lines, lane boundaries, road boundaries, intersections, speed limit signs, no-stopping signs, etc.

[0044] Figure 4 This is a schematic diagram of a method for predicting vehicle trajectory according to one embodiment of this application, as shown below. Figure 4 As shown, the target prediction model can be a graph convolutional network model. This model is trained using multiple sets of data through machine learning. Each set of data includes: vehicle profile sample features, real-time trajectory sample features, and electronic map sample features. The process of obtaining the vehicle profile sample features, real-time trajectory sample features, and electronic map sample features used to train the target prediction model can refer to the implementation process of obtaining vehicle profile features, real-time trajectory features, and electronic map features in the above embodiment, and will not be elaborated further. The target prediction model obtained by machine learning training using multiple sets of data can be derived as an Open Neural Network Exchange (ONNX) structure model. This ONNX structure model is deployed on autonomous vehicles for predicting vehicle trajectories.

[0045] Based on steps S12 to S14 above, by acquiring the vehicle profile features, real-time trajectory features, and electronic map features of the target vehicle, and then using a target prediction model to predict the vehicle profile features, real-time trajectory features, and electronic map features, the motion trajectory of the target vehicle is obtained. This achieves the goal of combining vehicle profile features, real-time trajectory features, and electronic map features for trajectory prediction, thereby improving the technical effect of predicting vehicle motion trajectory and solving the technical problem of low prediction accuracy of vehicle motion trajectory in related technologies.

[0046] Optionally, in step S14, a target prediction model is used to predict the vehicle profile features, real-time trajectory features, and electronic map features to obtain the target vehicle's motion trajectory, including:

[0047] Step S141: Encode the vehicle profile features, real-time trajectory features, and electronic map features to obtain the encoding results;

[0048] Step S142: Perform feature fusion processing on the encoding result to obtain the fusion result;

[0049] Step S143: Decode the fusion result to obtain the decoded result;

[0050] Step S144: Determine the motion trajectory of the target vehicle based on the decoding result.

[0051] Based on steps S141 to S144 above, the vehicle profile features, real-time trajectory features and electronic map features are encoded to obtain an encoding result. Then, the encoding result is fused to obtain a fusion result. Subsequently, the fusion result is decoded to obtain a decoding result. Finally, the motion trajectory of the target vehicle is determined based on the decoding result. This method can combine vehicle profile features, real-time trajectory features and electronic map features to make more accurate trajectory prediction and further improve the reliability of the predicted trajectory.

[0052] Optionally, in step S141, the vehicle profile features, real-time trajectory features, and electronic map features are encoded to obtain the following encoding results:

[0053] Step S1411: Enhance the vehicle profile features, real-time trajectory features, and electronic map features based on a preset method to obtain the enhancement result;

[0054] Step S1412: Encode the enhancement result using a multilayer perceptron structure and a batch normalization structure to obtain the encoded result.

[0055] Specifically, downsampling is performed on static map features in the electronic map to reduce redundant map element information, resulting in enhanced static map features. When enhancing vehicle profile features, target vehicles in the target driving scene are randomly dropped, resulting in enhanced vehicle profile features. When enhancing real-time trajectory features, the vehicle trajectories in the target driving scene data are randomly rotated based on a preset angle, resulting in enhanced real-time trajectory features.

[0056] Furthermore, the augmentation results are encoded using the Multilayer Perceptron (MLP) structure and BatchNorm structure in the target prediction model to obtain encoded results. For example, encoding the augmentation results corresponding to vehicle profile features using the MLP structure and BatchNorm structure yields encoded result A; encoding the augmentation results corresponding to static map features yields encoded result B; and encoding the augmentation results corresponding to real-time trajectory features yields encoded result C.

[0057] Based on the above steps S1411 to S1412, the vehicle profile features, real-time trajectory features and electronic map features are enhanced by a preset method to obtain the enhancement result. Then, the enhancement result is encoded using a multilayer perceptron structure and a batch normalization structure to obtain the encoding result. This can efficiently process the vehicle profile features, real-time trajectory features and electronic map features, and effectively improve the prediction performance of the target prediction model.

[0058] Optionally, in step S142, feature fusion processing is performed on the encoding result to obtain the fusion result, which includes: using a converter structure to repeatedly perform self-attention operation and cross-attention operation on the time dimension and vehicle dimension of the encoding result to obtain the fusion result.

[0059] Specifically, the Transformer structure of the target prediction model is used to repeatedly perform self-attention and cross-attention operations on the time and vehicle dimensions of the encoded results to obtain the fusion result. For example, based on the Transformer structure, the following operations are performed: Self-attention is applied to the vehicle dimension of encoded result A to obtain transition result D; self-attention is applied to the time dimension of transition result D to obtain transition result E; cross-attention is applied to the time dimensions of encoded result B and transition result E to obtain transition result F; cross-attention is applied to the time dimensions of encoded result C and transition result F to obtain transition result G; self-attention is applied to the vehicle dimension of transition result G to obtain transition result H; self-attention is applied to the time dimension of transition result H to obtain transition result I; cross-attention is applied to the time dimensions of encoded result B and transition result I to obtain transition result J; cross-attention is applied to the time dimensions of encoded result C and transition result J to obtain transition result K; self-attention is applied to the vehicle dimension of transition result K to obtain transition result L; and self-attention is applied to the time dimension of transition result L to obtain fusion result M.

[0060] Based on the above steps, a converter structure is used to repeatedly perform self-attention and cross-attention operations on the time and vehicle dimensions of the encoded results to obtain fusion results. This allows for feature fusion of vehicle profile features, real-time trajectory features, and electronic map features to obtain more accurate predicted trajectories and further improve the prediction performance of the target prediction model.

[0061] Optionally, in step S143, the fusion result is decoded to obtain the following decoded result:

[0062] Step S1431: Add dimensions to the fusion result to obtain the first intermediate result;

[0063] Step S1432: Perform a transformation operation on the first intermediate result using a multilayer perceptron structure to obtain the transformation result;

[0064] Step S1433: The self-attention operation is repeatedly performed on the time dimension and vehicle dimension of the transformation result using the converter structure to obtain the second intermediate result;

[0065] Step S1434: Regularize the second intermediate result using the layer normalization structure to obtain the third intermediate result;

[0066] Step S1435: The third intermediate result is decoded using a multilayer perceptron structure and a batch normalization structure to obtain the decoding result.

[0067] Specifically, the fusion result M is augmented with a dimension to obtain the first intermediate result N. Based on the MLP structure, the first intermediate result N is transformed to obtain the transformed result O. The Transformer structure is used to perform the following operations: a self-attention operation is performed on the time dimension of the transformed result O to obtain the transition result P; a self-attention operation is performed on the vehicle dimension of the transition result P to obtain the transition result Q; a self-attention operation is performed on the time dimension of the transition result Q to obtain the transition result R; and a self-attention operation is performed on the vehicle dimension of the transition result R to obtain the second intermediate result S. The second intermediate result S is regularized based on the LayerNorm structure of the target prediction model to obtain the third intermediate result T. The third intermediate result T is then decoded using the MLP structure and the BatchNorm structure to obtain the decoded result U.

[0068] Based on steps S1431 to S1435 above, a first intermediate result is obtained by adding dimensions to the fusion result. Then, a transformation operation is performed on the first intermediate result using a multilayer perceptron structure to obtain a transformed result. A self-attention operation is then repeatedly performed on the time dimension and vehicle dimension of the transformed result using a converter structure to obtain a second intermediate result. Subsequently, a layer normalization structure is used to perform regularization processing on the second intermediate result to obtain a third intermediate result. Finally, a multilayer perceptron structure and a batch normalization structure are used to decode the third intermediate result to obtain a decoded result, thereby further ensuring the accuracy of the predicted trajectory.

[0069] Optionally, in step S144, determining the trajectory of the target vehicle based on the decoding result includes:

[0070] Step S1441: Calculate the similarity of the decoding results to obtain the similarity results;

[0071] Step S1442: Merge the decoding results using the similarity results to obtain the motion trajectory of the target vehicle.

[0072] Based on the above steps S1441 to S1442, similarity results are obtained by calculating the similarity of the decoding results. Then, the similarity results are used to merge the decoding results to obtain the motion trajectory of the target vehicle. This allows for the merging of similar predicted motion trajectories, thereby obtaining a more accurate motion trajectory of the target vehicle.

[0073] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0074] This embodiment also provides a vehicle trajectory prediction device, which is used to implement the above embodiments and preferred embodiments, and will not be repeated as already described. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0075] Figure 5 This is a structural block diagram of a vehicle trajectory prediction device according to one embodiment of the present invention, such as... Figure 5 As shown, the vehicle trajectory prediction device 500 includes:

[0076] The acquisition module 501 is used to acquire the vehicle profile features, real-time trajectory features and electronic map features of the target vehicle. The vehicle profile features include a first feature, a second feature and a third feature. The first feature is used to characterize the attribute information of the target vehicle, the second feature is used to characterize the travel information of the target vehicle, and the third feature is used to characterize the status information of the target vehicle. The electronic map features include dynamic map features and static map features.

[0077] The prediction module 502 is used to predict the vehicle profile features, real-time trajectory features and electronic map features using a target prediction model to obtain the movement trajectory of the target vehicle. The target prediction model is trained by machine learning using multiple sets of data. Each set of data includes: vehicle profile sample features, real-time trajectory sample features and electronic map sample features.

[0078] Optionally, the prediction module 502 is further configured to: encode the vehicle profile features, real-time trajectory features, and electronic map features to obtain an encoding result; perform feature fusion processing on the encoding result to obtain a fusion result; perform decoding processing on the fusion result to obtain a decoding result; and determine the motion trajectory of the target vehicle based on the decoding result.

[0079] Optionally, the prediction module 502 is further configured to: enhance vehicle profile features, real-time trajectory features and electronic map features based on a preset method to obtain an enhancement result; and encode the enhancement result using a multilayer perceptron structure and a batch normalization structure to obtain an encoding result.

[0080] Optionally, the prediction module 502 is further configured to: repeatedly perform self-attention and cross-attention operations on the time dimension and vehicle dimension of the encoded result using a converter structure to obtain the fusion result.

[0081] Optionally, the prediction module 502 is further configured to: add dimensions to the fusion result to obtain a first intermediate result; perform a transformation operation on the first intermediate result using a multilayer perceptron structure to obtain a transformed result; repeatedly perform self-attention operations on the time dimension and vehicle dimension of the transformed result using a converter structure to obtain a second intermediate result; perform regularization processing on the second intermediate result using a layer normalization structure to obtain a third intermediate result; and perform decoding processing on the third intermediate result using a multilayer perceptron structure and a batch normalization structure to obtain a decoded result.

[0082] Optionally, the prediction module 502 is further configured to: calculate the similarity of the decoding results to obtain similarity results; and use the similarity results to merge the decoding results to obtain the motion trajectory of the target vehicle.

[0083] It should be noted that the above modules can be implemented by software or hardware. For the latter, they can be implemented in the following ways, but are not limited to: all the above modules are located in the same processor; or, the above modules are located in different processors in any combination.

[0084] Embodiments of the present invention also provide a non-volatile storage medium storing a computer program, wherein the computer program is configured to execute the steps in any of the above method embodiments when running.

[0085] Optionally, in this embodiment, the storage medium may be configured to store a computer program for performing the following steps:

[0086] S1, acquire the vehicle profile features, real-time trajectory features and electronic map features of the target vehicle. The vehicle profile features include a first feature, a second feature and a third feature. The first feature is used to characterize the attribute information of the target vehicle, the second feature is used to characterize the travel information of the target vehicle, and the third feature is used to characterize the status information of the target vehicle. The electronic map features include dynamic map features and static map features.

[0087] S2 uses a target prediction model to predict the vehicle profile features, real-time trajectory features, and electronic map features to obtain the target vehicle's movement trajectory. The target prediction model is trained using multiple sets of data through machine learning. Each set of data includes: vehicle profile sample features, real-time trajectory sample features, and electronic map sample features.

[0088] Optionally, in this embodiment, the storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0089] Embodiments of the present invention also provide a processor configured to run a computer program to perform the steps in any of the above method embodiments.

[0090] Optionally, in this embodiment, the processor can be configured to perform the following steps via a computer program:

[0091] S1, acquire the vehicle profile features, real-time trajectory features and electronic map features of the target vehicle. The vehicle profile features include a first feature, a second feature and a third feature. The first feature is used to characterize the attribute information of the target vehicle, the second feature is used to characterize the travel information of the target vehicle, and the third feature is used to characterize the status information of the target vehicle. The electronic map features include dynamic map features and static map features.

[0092] S2 uses a target prediction model to predict the vehicle profile features, real-time trajectory features, and electronic map features to obtain the target vehicle's movement trajectory. The target prediction model is trained using multiple sets of data through machine learning. Each set of data includes: vehicle profile sample features, real-time trajectory sample features, and electronic map sample features.

[0093] Optionally, in this embodiment, the processor can be configured to perform the following steps via a computer program:

[0094] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0095] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0096] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0097] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0098] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0099] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0100] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for predicting vehicle trajectory, characterized in that, include: The system acquires vehicle profile features, real-time trajectory features, and electronic map features of a target vehicle. The vehicle profile features include a first feature, a second feature, and a third feature. The first feature characterizes the target vehicle's attribute information, including its color, size, license plate number, vehicle type, model, and acceleration / deceleration performance. The second feature characterizes the target vehicle's travel information, including a predicted destination and historical travel trajectory. The third feature characterizes the target vehicle's status information, including a predicted driving status and control light information. The electronic map features include dynamic map features and static map features. A target prediction model is used to predict the vehicle profile features, the real-time trajectory features, and the electronic map features to obtain the movement trajectory of the target vehicle. The target prediction model is trained using multiple sets of data through machine learning. Each set of data includes: vehicle profile sample features, real-time trajectory sample features, and electronic map sample features. The process of predicting the vehicle profile features, real-time trajectory features, and electronic map features using the target prediction model to obtain the motion trajectory of the target vehicle includes: enhancing the vehicle profile features, real-time trajectory features, and electronic map features to obtain enhancement results, wherein the enhancement results include enhancement results corresponding to static map features, enhancement results corresponding to vehicle profile features, and enhancement results corresponding to real-time trajectory features; enhancing the vehicle profile features, real-time trajectory features, and electronic map features to obtain enhancement results includes: performing a downsampling operation on the static map features in the electronic map features to obtain enhancement results corresponding to the static map features; and enhancing the vehicle... The vehicle profile features are enhanced, and target vehicles in the target driving scene are randomly discarded to obtain the enhanced result corresponding to the vehicle profile features. The vehicle trajectory in the target driving scene data is randomly rotated based on a preset angle to obtain the enhanced result corresponding to the real-time trajectory features. The enhanced result is encoded to obtain an encoded result, which includes encoded result A corresponding to the vehicle profile features, encoded result B corresponding to the static map features, and encoded result C corresponding to the real-time trajectory features. The encoded result is then fused to obtain a fusion result. The fusion result is then decoded to obtain a decoded result. The motion trajectory of the target vehicle is determined based on the decoded result. The encoding results are subjected to feature fusion processing to obtain the fusion result, which includes: performing a self-attention operation on the vehicle dimension of the encoding result A using a converter structure to obtain transition result D; performing a self-attention operation on the time dimension of transition result D to obtain transition result E; performing a cross-attention operation on the time dimensions of the encoding result B and transition result E to obtain transition result F; performing a cross-attention operation on the time dimensions of the encoding result C and transition result F to obtain transition result G; performing a self-attention operation on the vehicle dimension of transition result G to obtain transition result H; performing a self-attention operation on the time dimension of transition result H to obtain transition result I; performing a cross-attention operation on the time dimensions of the encoding result B and transition result I to obtain transition result J; performing a cross-attention operation on the time dimensions of the encoding result C and transition result J to obtain transition result K; performing a self-attention operation on the vehicle dimension of transition result K to obtain transition result L; and performing a self-attention operation on the time dimension of transition result L to obtain the fusion result. Decoding the fusion result to obtain the decoding result includes: adding dimensions to the fusion result to obtain a first intermediate result; performing a transformation operation on the first intermediate result using a multilayer perceptron structure to obtain a transformed result; repeatedly performing self-attention operations on the time dimension and vehicle dimension of the transformed result using a converter structure to obtain a second intermediate result; performing regularization processing on the second intermediate result using a layer normalization structure to obtain a third intermediate result; and decoding the third intermediate result using a multilayer perceptron structure and a batch normalization structure to obtain the decoding result.

2. The method for predicting vehicle trajectory according to claim 1, characterized in that, The vehicle profile features, the real-time trajectory features, and the electronic map features are encoded to obtain the encoding results, including: The vehicle profile features, real-time trajectory features, and electronic map features are enhanced using a preset method to obtain the enhancement result. The enhancement result is encoded using a multilayer perceptron structure and a batch normalization structure to obtain the encoded result.

3. The method for predicting vehicle trajectory according to claim 1, characterized in that, Determining the motion trajectory of the target vehicle based on the decoding result includes: The similarity of the decoding results is calculated to obtain the similarity result; The similarity results are used to merge the decoding results to obtain the motion trajectory of the target vehicle.

4. A device for predicting vehicle trajectory, characterized in that, include: The acquisition module is used to acquire vehicle profile features, real-time trajectory features, and electronic map features of the target vehicle. The vehicle profile features include a first feature, a second feature, and a third feature. The first feature characterizes the attribute information of the target vehicle, including its color, size, license plate number, vehicle type, model, and acceleration / deceleration performance. The second feature characterizes the travel information of the target vehicle, including a predicted destination and historical travel trajectory. The third feature characterizes the status information of the target vehicle, including a predicted driving status and control light information. The electronic map features include dynamic map features and static map features. The prediction module is used to predict the vehicle profile features, the real-time trajectory features, and the electronic map features using a target prediction model to obtain the motion trajectory of the target vehicle. The target prediction model is trained using multiple sets of data through machine learning, and each set of data includes: vehicle profile sample features, real-time trajectory sample features, and electronic map sample features. The prediction module is further configured to enhance the vehicle profile features, the real-time trajectory features, and the electronic map features to obtain enhancement results, wherein the enhancement results include enhancement results corresponding to static map features, enhancement results corresponding to vehicle profile features, and enhancement results corresponding to real-time trajectory features; enhancing the vehicle profile features, the real-time trajectory features, and the electronic map features to obtain enhancement results includes: performing downsampling on the static map features in the electronic map features to obtain enhancement results corresponding to the static map features; enhancing the vehicle profile features and randomly discarding target vehicles in the target driving scenario to obtain enhancement results corresponding to the vehicle profile features; randomly rotating the vehicle trajectory of the target driving scenario data based on a preset angle to obtain enhancement results corresponding to the real-time trajectory features; encoding the enhancement results to obtain encoding results, wherein the encoding results include encoding result A corresponding to the vehicle profile features, encoding result B corresponding to the static map features, and encoding result C corresponding to the real-time trajectory features; performing feature fusion on the encoding results to obtain fusion results; decoding the fusion results to obtain decoding results; and determining the motion trajectory of the target vehicle based on the decoding results. The prediction module is further configured to perform a self-attention operation on the vehicle dimension of the encoded result A using a converter structure to obtain transition result D; perform a self-attention operation on the time dimension of transition result D to obtain transition result E; perform a cross-attention operation on the time dimensions of encoded result B and transition result E to obtain transition result F; perform a cross-attention operation on the time dimensions of encoded result C and transition result F to obtain transition result G; perform a self-attention operation on the vehicle dimension of transition result G to obtain transition result H; perform a self-attention operation on the time dimension of transition result H to obtain transition result I; perform a cross-attention operation on the time dimensions of encoded result B and transition result I to obtain transition result J; perform a cross-attention operation on the time dimensions of encoded result C and transition result J to obtain transition result K; perform a self-attention operation on the vehicle dimension of transition result K to obtain transition result L; and perform a self-attention operation on the time dimension of transition result L to obtain the fusion result. The prediction module is further configured to add dimensions to the fusion result to obtain a first intermediate result; perform a transformation operation on the first intermediate result using a multilayer perceptron structure to obtain a transformed result; repeatedly perform self-attention operations on the time dimension and vehicle dimension of the transformed result using a converter structure to obtain a second intermediate result; perform regularization processing on the second intermediate result using a layer normalization structure to obtain a third intermediate result; and perform decoding processing on the third intermediate result using a multilayer perceptron structure and a batch normalization structure to obtain the decoded result.

5. A non-volatile storage medium, characterized in that, The storage medium stores a computer program, wherein the computer program is configured to execute the vehicle trajectory prediction method according to any one of claims 1 to 3 when it is run.

6. A processor, characterized in that, The processor is used to run a program, wherein the program is configured to execute the vehicle trajectory prediction method according to any one of claims 1 to 3 when running.

7. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to perform the vehicle trajectory prediction method according to any one of claims 1 to 3.

Images