CN116071925B - Track prediction method and device and electronic processing device - Google Patents
Track prediction method and device and electronic processing device Download PDFInfo
- Publication number
- CN116071925B CN116071925B CN202310120681.4A CN202310120681A CN116071925B CN 116071925 B CN116071925 B CN 116071925B CN 202310120681 A CN202310120681 A CN 202310120681A CN 116071925 B CN116071925 B CN 116071925B
- Authority
- CN
- China
- Prior art keywords
- information
- vector information
- environment
- vector
- dynamic
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 95
- 238000012545 processing Methods 0.000 title claims abstract description 34
- 239000013598 vector Substances 0.000 claims abstract description 289
- 238000013528 artificial neural network Methods 0.000 claims abstract description 27
- 239000011159 matrix material Substances 0.000 claims description 68
- 238000005070 sampling Methods 0.000 claims description 27
- 230000003993 interaction Effects 0.000 claims description 26
- 238000000605 extraction Methods 0.000 claims description 17
- 230000007246 mechanism Effects 0.000 claims description 13
- 238000013135 deep learning Methods 0.000 claims description 11
- 239000000284 extract Substances 0.000 claims description 6
- 238000004590 computer program Methods 0.000 claims description 5
- 230000002829 reductive effect Effects 0.000 abstract description 17
- 230000008569 process Effects 0.000 description 33
- 238000004364 calculation method Methods 0.000 description 17
- 238000010586 diagram Methods 0.000 description 15
- 238000009877 rendering Methods 0.000 description 11
- 230000007613 environmental effect Effects 0.000 description 9
- 230000005540 biological transmission Effects 0.000 description 7
- 239000003795 chemical substances by application Substances 0.000 description 7
- 238000013527 convolutional neural network Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 230000003068 static effect Effects 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 238000012549 training Methods 0.000 description 3
- 238000012935 Averaging Methods 0.000 description 2
- 235000004522 Pentaglottis sempervirens Nutrition 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 230000002452 interceptive effect Effects 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 230000003313 weakening effect Effects 0.000 description 2
- 238000003491 array Methods 0.000 description 1
- 230000003542 behavioural effect Effects 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000000670 limiting effect Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0125—Traffic data processing
- G08G1/0133—Traffic data processing for classifying traffic situation
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Abstract
The application provides a track prediction method and device and an electronic processing device. Wherein the method comprises the following steps: acquiring dynamic information and environment information of a target object; encoding the dynamic information and the environment information into dynamic vector information and environment vector information; and extracting the characteristics of the dynamic vector information and the environment vector information through a neural network, and establishing an association relation between the dynamic vector information and the environment vector information so as to predict the track through the association relation. The method and the device convert dynamic information and environment information of the target object into vector information, so that the vector information is input into a neural network to conduct track prediction on the target object. By converting the multi-total information into vector information, the input of the neural network is unified. In addition, all information is converted into vector information for input, so that the traditional image input is replaced, the computing resource cost is reduced, and the prediction speed is improved.
Description
Technical Field
The present invention relates to the field of unmanned aerial vehicle, and in particular, to a trajectory prediction method and apparatus, and an electronic processing apparatus.
Background
In the existing track prediction algorithm, a method based on raster rendering representation is mainly used, the idea of the method is to render the current complex environment information and the historical state of an agent into an image of a bird's eye view as input forms, and a convolutional neural network is used for extracting characteristics and decoding prediction information. In the method, the characteristic of the image layer by using the convolutional neural network is improved to be intensive calculation, so that the calculated amount is large, and the track prediction efficiency is low.
Disclosure of Invention
In view of the foregoing, an object of an embodiment of the present application is to provide a track prediction method, a track prediction device, and an electronic processing device, which can improve track prediction efficiency.
In a first aspect, an embodiment of the present application provides a track prediction method, including: acquiring dynamic information and environment information of a target object; encoding the dynamic information and the environment information into dynamic vector information and environment vector information; and extracting the characteristics of the dynamic vector information and the environment vector information through a neural network, and establishing an association relation between the dynamic vector information and the environment vector information so as to predict the track through the association relation.
In the implementation process, the dynamic information and the environment information of the target object are encoded into vector information, and the vector information is used as the input of the neural network, so that the representation forms of the static information and the dynamic information of the target object are unified, and the difficulty of constructing a neural network model is reduced. In addition, as the input of the neural network is that the vector information replaces the image information, the vector information is processed without rendering or intensive calculation, so that the cost of calculation resources is reduced, and the prediction speed is improved.
In one embodiment, the neural network includes a multi-layer perceptron, the feature extraction is performed on the dynamic vector information and the environment vector information through the neural network, and an association relationship between the dynamic vector information and the environment vector information is established, so as to perform track prediction through the association relationship, and the method includes: extracting intermediate features of the dynamic vector information and the environment vector information through a multi-layer perceptron; and establishing an association relation between the dynamic vector information and the environment vector information according to the intermediate characteristics of the dynamic vector information and the environment vector information so as to predict the track through the association relation.
In the implementation process, the multi-layer perceptron is used for extracting intermediate features of the input dynamic vector information and the environment vector information so as to realize the respective extraction of different vector information, prevent the mutual interference among the vector information and improve the accuracy of feature extraction.
In one embodiment, the neural network further comprises: the self-attention mechanism and the full-connection graph network establish the association relation between the dynamic vector information and the environment vector information according to the intermediate characteristics of the dynamic vector information and the environment vector information so as to predict the track through the association relation, and the method comprises the following steps: acquiring an interaction relation between the intermediate features of the dynamic vector information and the environment vector information through the full connection graph network; and extracting global information from the interaction relationship between the intermediate features through the self-attention mechanism to determine the association relationship between the dynamic vector information and the environment vector information so as to predict the track through the association relationship.
In the implementation process, the self-attention mechanism extracts the global information of the interaction relationship between the intermediate features, so that the interaction information of the target object and the surrounding environment can be accurately captured, and the accuracy of track prediction is improved.
In one embodiment, the track of the target object is predicted to be a plurality of predicted tracks, and after the track prediction is performed by the association relationship, the method includes: determining a probability corresponding to each predicted trajectory based on deep learning; and outputting the predicted track with the highest probability as the final predicted track of the target object.
In the implementation process, when a plurality of predicted tracks are determined, the probability of occurrence of the plurality of predicted tracks is determined through the neural network, so that the predicted track with the largest occurrence probability is used as the final predicted track, and the accuracy of track prediction is improved.
In one embodiment, the environmental information includes lane information including: the lane line and the lane center line, the environment vector information comprises lane vector information, the lane line and the lane center line are constructed into a multi-section broken line form by setting a first sampling point with a first preset distance, the dynamic information and the environment information are respectively encoded into dynamic vector information and environment vector information, and the method comprises the following steps: encoding the position information of each first sampling point in the lane information into a first position vector; and constructing a lane two-dimensional matrix according to the first position vector so as to form lane vector information.
In the implementation process, the lane information is usually a continuous straight line, and when vector encoding is performed, the straight line is converted into a multi-section broken line form, so that the position information of each first sampling point in the lane information is respectively obtained to construct a lane two-dimensional matrix, the lane information is converted into the vector information, the lane information is not required to be extracted in an image processing mode, intensive calculation is reduced, and the calculation resource cost is reduced. In addition, the information loss in the image rendering process is prevented, and the integrity of lane information transmission is improved.
In one embodiment, the environment information includes closed-loop topology information, the environment vector information includes closed-loop topology vector information, the closed-loop topology information is constructed in a multi-segment polyline form of closed-loop connection by setting second sampling points spaced by a second preset distance, and the encoding the dynamic information and the environment information into dynamic vector information and environment vector information respectively includes: encoding the position information of each second sampling point of the closed-loop topology information into a second position vector; constructing a closed-loop topology two-dimensional matrix according to the second position vector to form closed-loop topology vector information; wherein a starting two-dimensional vector in a starting two-dimensional matrix in the closed-loop topology two-dimensional matrix is equal to a last two-dimensional vector in a last two-dimensional matrix in the closed-loop topology two-dimensional matrix.
In the implementation process, the closed-loop topology information is converted into a multi-section broken line form of closed-loop connection when being encoded, so that all characteristic information in the closed-loop topology information can be obtained, and the integrity and accuracy of closed-loop topology information encoding are improved. In addition, the closed-loop topology information is converted into vector information, and the closed-loop topology information is not required to be extracted in an image processing mode, so that intensive calculation is reduced, the cost of calculation resources is reduced, the information loss in the image rendering process is prevented, and the integrity of lane information transmission is improved.
In one embodiment, the dynamic information includes historical track information, the dynamic vector information includes track vector information, and the encoding the dynamic information and the environment information into dynamic vector information and environment vector information, respectively, includes: encoding the position information of each track point in the history track information into a track point position vector; and constructing a track two-dimensional matrix according to the track point position vector to form track vector information.
In the implementation process, because the formed historical track is a set of track points of each time point in the past in the running process of the target object, the track two-dimensional matrix is constructed by encoding each track point as a track point position vector, the historical track information is converted into vector information, extraction of the historical track information in an image processing mode is not needed, intensive calculation is reduced, and calculation resource expenditure is reduced. In addition, the information loss in the image rendering process is prevented, and the integrity of the historical track information transmission is improved.
In a second aspect, an embodiment of the present application further provides a track prediction apparatus, including: the acquisition module is used for: the method comprises the steps of acquiring dynamic information and environment information of a target object; and a coding module: for encoding the dynamic information and the environment information into dynamic vector information and environment vector information; and a prediction module: the method is used for extracting the characteristics of the dynamic vector information and the environment vector information through a neural network, and establishing the association relation between the dynamic vector information and the environment vector information so as to predict the track through the association relation.
In a third aspect, an embodiment of the present application further provides an electronic processing device, including: a processor, a memory storing machine-readable instructions executable by the processor, which when executed by the processor, perform the steps of the method of the first aspect, or any of the possible implementations of the first aspect.
In a fourth aspect, the embodiments of the present application further provide a computer readable storage medium having stored thereon a computer program which, when executed by a processor, performs the steps of the trajectory prediction method of the first aspect, or any one of the possible implementations of the first aspect.
In order to make the above objects, features and advantages of the present application more comprehensible, embodiments accompanied with figures are described in detail below.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are needed in the embodiments will be briefly described below, it being understood that the following drawings only illustrate some embodiments of the present application and therefore should not be considered limiting the scope, and that other related drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic diagram of interaction between a positioning device and a processing device according to an embodiment of the present disclosure;
FIG. 2 is a block diagram of an electronic processing device according to an embodiment of the present disclosure;
FIG. 3 is a flowchart of a track prediction method according to an embodiment of the present disclosure;
FIG. 4 is a schematic diagram of single node feature extraction provided in an embodiment of the present application;
fig. 5 is a schematic diagram of a process for implementing step 203 provided in an embodiment of the present application;
FIG. 6 is a schematic diagram of an adjacency matrix provided in an embodiment of the present application;
FIG. 7 is a schematic diagram of a track prediction result provided in an embodiment of the present application;
Fig. 8 is a schematic functional block diagram of a track prediction apparatus according to an embodiment of the present application.
Detailed Description
The technical solutions in the embodiments of the present application will be described below with reference to the drawings in the embodiments of the present application.
It should be noted that: like reference numerals and letters denote like items in the following figures, and thus once an item is defined in one figure, no further definition or explanation thereof is necessary in the following figures. Meanwhile, in the description of the present application, the terms "first", "second", and the like are used only to distinguish the description, and are not to be construed as indicating or implying relative importance.
The intelligent body track prediction under the dynamic complex environment is an important problem in the fields of intelligent unmanned vehicles and automatic driving. In the track prediction task, the algorithm needs to predict a plurality of possible tracks and corresponding probabilities of the intelligent agent in a certain time period in the future according to the historical track information of the intelligent agent in the current environment and the surrounding road environment. For complex environments, the agent needs to consider not only historical track information, but also interaction information and road topology information which may be generated by surrounding agents.
The inventor of the application finds that the conventional track prediction method by using an image rendering method has the following defects through long-term study: first, there is a loss of information in the rendering process of mapping all information to the bird's eye view, such as timing information, it is difficult to accurately represent the information in a unified image. Second, feature extraction at the image level using convolutional neural networks is computationally intensive, and is prohibitively expensive on edge devices with limited computational resources. Third, the ability of convolutional neural networks to extract features is limited by receptive fields and does not give good consideration to global information to learn advanced interaction processes between agents. Therefore, the current trajectory prediction method has low prediction efficiency.
In addition, since the behavioral intent of the agent is not observable, its trajectory is highly free and constantly changing, requiring the model to have the ability to characterize such high degree of freedom outputs. Second, the input of the model is usually multi-modal heterogeneous information, such as static and dynamic mixed information of historical tracks, road maps, obstacle categories and the like, which makes it difficult for the model to extract effective information in multi-modal input features. Therefore, it is difficult to achieve accurate and efficient trajectory prediction.
In view of this, the inventor of the present application proposes a track prediction method, which converts dynamic information and environmental information of a target object into vector information, inputs the vector information into a multi-layer perceptron and a self-attention mechanism, performs intermediate feature extraction to establish an association relationship between each vector information and a global state, completes track prediction, converts input of a neural network into a vector, unifies a representation form of dynamic and static input, and simultaneously replaces input of an image level with new input representation, thereby greatly reducing computing resource expenditure and improving model computing speed.
For the convenience of understanding the present embodiment, a detailed description will be first given of an operating environment in which a trajectory prediction method disclosed in the embodiments of the present application is executed.
Fig. 1 is a schematic diagram illustrating interaction between a positioning device and a processing device according to an embodiment of the present application. The processing device 100 is communicatively coupled to one or more positioning devices 200 for data communication or interaction via a network. The processing device 100 may be a web server, a database server, etc., or may be a personal computer (personal computer, PC), a tablet computer, a smart phone, a personal digital assistant (personal digital assistant, PDA), etc. The processing device 100 may also be a module or the like integrated in the control system of the target object. The setting mode of the processing device can be adjusted according to actual conditions, and the setting mode is not particularly limited.
The positioning device 200 may be a GPS navigator, a beidou navigation system, etc. The positioning device 200 is disposed on the target object, and is used for acquiring the position information of the target object in real time.
Optionally, the positioning device 200 may also be disposed on other objects around the target object to obtain location information of the other objects around the target object.
It will be appreciated that the positioning device 200, upon acquiring the location information, transmits the location information to the processing device 100. The processing device 100 predicts the trajectory of the target object by processing the position information.
In order to facilitate understanding of the present embodiment, a processing device that performs the trajectory prediction method disclosed in the embodiment of the present application will be described in detail below.
As shown in fig. 2, a block schematic diagram of the electronic processing device is shown. The electronic processing device 100 may include a memory 111 and a processor 113. It will be appreciated by those of ordinary skill in the art that the configuration shown in fig. 2 is merely illustrative and is not intended to limit the configuration of electronic processing device 100. For example, electronic processing device 100 may also include more or fewer components than shown in FIG. 2, or have a different configuration than shown in FIG. 2.
The memory 111 and the processor 113 are directly or indirectly electrically connected to each other to realize data transmission or interaction. For example, the components may be electrically connected to each other via one or more communication buses or signal lines. The processor 113 is used to execute executable modules stored in the memory.
The Memory 111 may be, but is not limited to, a random access Memory (Random Access Memory, RAM), a Read Only Memory (ROM), a programmable Read Only Memory (Programmable Read-Only Memory, PROM), an erasable Read Only Memory (Erasable Programmable Read-Only Memory, EPROM), an electrically erasable Read Only Memory (Electric Erasable Programmable Read-Only Memory, EEPROM), etc. The memory 111 is configured to store a program, and the processor 113 executes the program after receiving an execution instruction, and a method executed by the electronic processing device 100 defined by the process disclosed in any embodiment of the present application may be applied to the processor 113 or implemented by the processor 113.
The processor 113 may be an integrated circuit chip having signal processing capabilities. The processor 113 may be a general-purpose processor, including a central processing unit (Central Processing Unit, CPU for short), a network processor (Network Processor, NP for short), etc.; but also digital signal processors (digital signal processor, DSP for short), application specific integrated circuits (Application Specific Integrated Circuit, ASIC for short), field Programmable Gate Arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The disclosed methods, steps, and logic blocks in the embodiments of the present application may be implemented or performed. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
The electronic processing device 100 in this embodiment may be used to perform each step in each method provided in the embodiments of the present application. The implementation of the trajectory prediction method is described in detail below by means of several embodiments.
Referring to fig. 3, a flowchart of a track prediction method according to an embodiment of the present application is shown. The specific flow shown in fig. 3 will be described in detail.
Step 201, acquiring dynamic information and environment information of a target object.
The target object here may be an intelligent car, a robot, an automatic dispenser, an intelligent handling device, or the like.
The environment information comprises lane information, wherein the lane information comprises lane lines of a lane where a target object is located, lane center line information, closed loop topology information and the like. The closed-loop topology information is closed-loop information on the driving track of the target object, for example, intersection information. The dynamic information includes historical track information, which may include historical track information of the target object itself, historical track information of other objects around the target object, and the like.
It can be understood that the positioning device is arranged on the target object to position the target object, so that the position information of the target object at any moment can be obtained, and the historical track information of the target object can be determined according to the position information of the target object at any moment.
In addition, after the position information of the target object is determined, the environment information of the target object can be determined according to the position information. For example, the position information may be input to an AI model, which stores all environmental characteristics of the target object travel section in advance. After the AI model determines a corresponding location based on the location information of the target device, environmental information surrounding the location may be determined based on the location.
In some embodiments, an information acquisition device may be further disposed on the target object to acquire environmental information around the target object in real time. The information acquisition device includes, but is not limited to, an image acquisition device, an infrared sensing device, a laser device, and the like.
Step 202, the dynamic information and the environment information are encoded into dynamic vector information and environment vector information.
The environment vector information comprises lane vector information, closed-loop topology vector information and the like. The motion vector information may include track vector information of the target object itself, track vector information of other objects around the target object, and the like.
Because the environmental characteristics of the lane line, the lane center line and the like are continuous line information, when the environmental information of the lane line, the lane center line and the like is encoded, the lane line and the lane center line can be constructed into a multi-section broken line form by arranging first sampling points at intervals of a first preset distance so as to respectively take the position coordinates of each first sampling point to encode as a first position vector, a plurality of first position vectors in each broken line are constructed into a two-dimensional matrix vector, and the two-dimensional matrix vectors of the broken lines are spliced on the 0 th dimension (namely spliced in the form of points) so as to obtain the lane two-dimensional matrix. The multi-section broken line coded by the coding mode is a non-directional line segment.
Similarly, when the closed-loop topology information is encoded, the closed-loop topology information can be also constructed into a multi-section broken line form of closed-loop connection by setting second sampling points spaced by a second preset distance, so that the position coordinates of each second sampling point are respectively encoded into second position vectors, a plurality of second position vectors in each broken line are constructed into two-dimensional matrix vectors, the two-dimensional matrix vectors of the broken lines are spliced in the 0 th dimension to obtain a broken line two-dimensional matrix, and the corresponding semantic information in the closed-loop topology information is spliced on the basis of the broken line two-dimensional matrix to determine the closed-loop topology two-dimensional matrix.
The historical track information of the target object is determined by the positions of a plurality of track points in the running process of the target object, when the historical track information is encoded, the position coordinates of each track point in the running process of the target object are encoded into track point position vectors, and track two-dimensional matrix vectors are constructed through the track point position vectors. For example, the location of a trace point may be represented by (x, y), and the trace two-dimensional matrix vector may be: [ (x 1, y 1), (x 2, y 2) … (xn, yn) ].
In some embodiments, the context information further includes point information (e.g., traffic lights, signboards, etc.), which is information that cannot be represented by vectors, and which can be encoded as vector information with a starting point coincident with an ending point. For example, the position of the point information may be represented by (x, y), and the two-dimensional matrix vector of the point information is: [ (x 1, y 1), (x 1, y 1) … (x 1, y 1), (0, 1) ]. Wherein, (0, 1) is semantic information corresponding to the attribute in the point information.
It will be appreciated that the above-mentioned encoding methods of the environmental information and the dynamic information are merely exemplary, and the encoding methods of the environmental information and the dynamic information may be adjusted according to practical situations, which are not particularly limited in this application.
The lane two-dimensional matrix, the closed-loop topology two-dimensional matrix, the trajectory two-dimensional matrix vector, and the dot information two-dimensional matrix vector are presented in the same form.
And 203, extracting features of the dynamic vector information and the environment vector information through a neural network, and establishing an association relation between the dynamic vector information and the environment vector information so as to predict the track through the association relation.
The neural network can comprise multiple deep learning models such as a multi-layer perceptron, a self-attention mechanism, a full-connection graph network and the like. The multi-layer perceptron is used for extracting and characteristic extracting input dynamic vector information and environment vector information, and the self-attention mechanism and the full-connection graph network are used for establishing each local characteristic and the association relation between the local characteristic and the global characteristic.
In the implementation process, the dynamic information and the environment information of the target object are encoded into vector information, and the vector information is used as the input of the neural network, so that the representation forms of the static information and the dynamic information of the target object are unified, and the difficulty of constructing a neural network model is reduced. In addition, as the input of the neural network is that the vector information replaces the image information, the vector information is processed without rendering or intensive calculation, so that the cost of calculation resources is reduced, and the prediction speed is improved.
In one possible implementation, step 203 includes: extracting intermediate features of the dynamic vector information and the environment vector information through a multi-layer perceptron; and establishing an association relation between the dynamic vector information and the environment vector information according to the intermediate characteristics of the dynamic vector information and the environment vector information so as to predict the track through the association relation.
It will be appreciated that, when processing the above-mentioned motion vector information and environment vector information, feature extraction may be performed first according to the motion vector information and environment vector information, respectively, so as to extract motion vector information features and environment vector information features, respectively. When there are a plurality of pieces of information in the motion vector information or the environment vector information, the plurality of pieces of information can be extracted separately. The multi-layer perceptron can be provided with a plurality of channels, and each channel respectively extracts the characteristics of corresponding information. For example, if the lane vector information includes a plurality of pieces of lane line information, each piece of lane line information is input to a corresponding channel when the lane vector information is extracted, and each piece of lane line information is extracted.
The intermediate feature extraction of the motion vector information and the environment vector information can be simultaneously extracted through different channels, or can be sequentially extracted. Illustratively, as shown in FIG. 4, when a single node extracts features, the multi-layer perceptron performs hierarchical feature extraction on multiple pieces of information for the single node.
After the multi-layer perceptron performs feature extraction on the dynamic vector information and the environment vector information, the obtained intermediate features can be used for interaction among different information. That is, all or part of the intermediate features of the environment vector information and the motion vector information are connected to each other to determine the association between the various information. For example, after the track feature of the target object is extracted, the track feature may be connected with the left lane line, the right lane line, the track of the left object, and the track of the right object of the track feature through a deep learning network of the neural network, so as to determine the association relationship between the track feature and other features.
Illustratively, as shown in FIG. 5, a plurality of vector information V is shown in FIG. 5 1 l ......V i l ..........V l
u is input into a multi-layer perceptron, and then intermediate feature extraction is carried out by the multi-layer perceptron to obtain eV 1 l ......eV i l ..........eV u l . And respectively connecting all intermediate features or part of intermediate features of the environment vector information and the dynamic vector information to respectively determine the association relation between various information, and finally connecting the intermediate features with the association relation based on the association relation to obtain eV (electronic voltage) i i+l 。
It can be understood that the neural network herein may have a deep learning capability, after determining a local feature of a certain information, determining the local feature according to a deep learning result to primarily determine other features with higher relevance, and respectively establishing association relations between the local feature and other features with higher relevance, so as to determine global relations of all the information, so as to perform track prediction. Of course, the neural network may not have the ability of deep learning, and if the neural network does not have the ability of deep learning, after determining the local characteristics of a certain information, the local characteristics are respectively associated with all other characteristics, so as to determine the global relation of all the information, so as to perform track prediction.
In the implementation process, the multi-layer perceptron is used for extracting intermediate features of the input dynamic vector information and the environment vector information so as to realize the respective extraction of different vector information, prevent the mutual interference among the vector information and improve the accuracy of feature extraction.
In one possible implementation, step 203 includes: acquiring an interaction relationship between intermediate features of the dynamic vector information and the environment vector information through a full connection graph network; and extracting global information from the interaction relationship between the intermediate features through a self-attention mechanism to determine the association relationship between the dynamic vector information and the environment vector information so as to predict the track through the association relationship.
The interaction relationship here includes interaction of the environment information with the target object, interaction between the target objects, interaction of the target object history track information with the environment information, and the like. The interaction relationship may be constructed by maxpool, concat or the like.
The interactive relation between the intermediate features of the dynamic vector information and the environment vector information is obtained through the fully connected graph network, and the interactive relation comprises information interaction of all features through the fully connected adjacency matrix coding. The adjacency matrix may be as shown in fig. 6, wherein a connection between adjacent features in the adjacency matrix indicates that there is a connection between the two adjacent features, and a non-connection between adjacent features in the adjacency matrix indicates that there is no connection between the two adjacent features. As shown in fig. 6, the adjacency matrix is diagonally symmetrical to indicate whether there is interaction between the respective target object trajectories or lane lines.
After determining the adjacency matrix, extracting global information from the adjacency matrix, the dynamic vector information and the environment vector information through a self-attention mechanism.
The self-attention mechanism described above can be represented by the following structure:
wherein P is Q 、P K 、P V To orthographically project vectors in each inverse direction for the full connection matrix P, GAN (Graph Attention Networks, chinese name: graph self-attention network) is the graph self-attention network.
In the implementation process, the self-attention mechanism extracts the global information of the interaction relationship between the intermediate features, so that the interaction information of the target object and the surrounding environment can be accurately captured, and the accuracy of track prediction is improved.
In one possible implementation, after step 203, the method further includes: determining a probability corresponding to each predicted trajectory based on deep learning; and outputting the predicted track with the highest probability as the final predicted track of the target object.
It will be appreciated that after the track prediction is performed on the target object in the above manner, there may be more than one predicted track (as shown in fig. 7, 5 predicted tracks are shown in fig. 7). The probability that each predicted track of the target object is likely to occur is further determined through the deep learning network. And taking the predicted track with the highest probability as the final predicted track of the target object.
In some embodiments, after determining the probability that each predicted track of the target object may occur, the deep learning network may directly output all the predicted tracks and the corresponding probabilities, so that the target object pair may select the corresponding final predicted track based on the predicted track and the corresponding probabilities.
It can be understood that the predicted track in the embodiment of the present application is determined by vector information, and when outputting the predicted track and the corresponding probability, the predicted track and the probability need to be decoded to be output in the form of a passing track.
In the implementation process, when a plurality of predicted tracks are determined, the probability of occurrence of the plurality of predicted tracks is determined through the neural network, so that the predicted track with the largest occurrence probability is used as the final predicted track, and the accuracy of track prediction is improved.
In one possible implementation, step 202 includes: encoding the position information of each first sampling point in the lane information into a first position vector; and constructing a lane two-dimensional matrix according to the first position vector to form lane vector information.
Wherein, the multi-section broken line in the lane information is a non-directional line segment. When the lane information is processed, the direction information in the lane information can be subjected to the processes of averaging, filtering, weakening and the like so as to obtain the lane information without the direction information. Or when the lane information is encoded, the direction information of the lane information is not encoded.
It will be appreciated that the lane information is not direction sensitive and that the start point of the lane information may select any first sampling point when training the lane information. The first position vectors of the lane two-dimensional matrix constructed by different initial first sampling points are the same, but the lane two-dimensional matrix expression forms can be different.
In the implementation process, the lane information is usually a continuous straight line, and when vector encoding is performed, the straight line is converted into a multi-section broken line form, so that the position information of each first sampling point in the lane information is respectively obtained to construct a lane two-dimensional matrix, the lane information is converted into the vector information, the lane information is not required to be extracted in an image processing mode, intensive calculation is reduced, and the calculation resource cost is reduced. In addition, the information loss in the image rendering process is prevented, and the integrity of lane information transmission is improved.
In one possible implementation, step 202 further includes: encoding the position information of each second sampling point of the closed-loop topology information into a second position vector; and constructing a closed-loop topology two-dimensional matrix according to the second position vector to form closed-loop topology vector information.
The starting two-dimensional vector in the starting two-dimensional matrix in the closed-loop topology two-dimensional matrix is equal to the last two-dimensional vector in the last two-dimensional matrix in the closed-loop topology two-dimensional matrix, namely the starting point and the ending point in the closed-loop topology information are consistent, so that the closed-loop two-dimensional vector is formed.
In some embodiments, semantic information is also encoded in the closed-loop two-dimensional vector, which may be represented using a binary code. For example, the position of the second sampling point may be represented by (x, y), and the lane line corresponds to the attribute as a double yellow line, and the semantic information corresponding to the attribute is: [ 'double', 'yellow' ], can be encoded as [0,1]. The coded semantic information can be spliced with the vector to obtain a closed-loop topological two-dimensional matrix of [ (x 1, y 1), (x 2, y 2) … (xn, yn), (0, 1) ].
Wherein, the multi-section broken line in the closed-loop topology information is a non-directional line segment. When the closed-loop topology information is processed, the direction information in the closed-loop topology information can be subjected to the processes of averaging, filtering, weakening and the like so as to obtain the closed-loop topology information without the direction information. Or when the closed-loop topology information is encoded, the direction information of the closed-loop topology information is not encoded.
It will be appreciated that the closed-loop topology information is not direction sensitive, and that any second sampling point may be selected from the start of the closed-loop topology information when training the closed-loop topology information. The second position vector information of the closed-loop topology two-dimensional matrix constructed by the different starting second sampling points is the same, but the closed-loop topology two-dimensional matrix expression forms can be different.
In the implementation process, the closed-loop topology information is converted into a multi-section broken line form of closed-loop connection when being encoded, so that all characteristic information in the closed-loop topology information can be obtained, and the integrity and accuracy of closed-loop topology information encoding are improved. In addition, the closed-loop topology information is converted into vector information, and the closed-loop topology information is not required to be extracted in an image processing mode, so that intensive calculation is reduced, the cost of calculation resources is reduced, the information loss in the image rendering process is prevented, and the integrity of lane information transmission is improved.
In one possible implementation, step 202 further includes: encoding the position information of each track point in the historical track information into a track point position vector; and constructing a track two-dimensional matrix according to the track point position vector to form track vector information.
The historical track information is directed information. In encoding the historical track information, it is necessary to encode the track point position vector with the direction information. When training the historical track information, the starting point of the historical track information is a fixed track point.
In the implementation process, because the formed historical track is a set of track points of each time point in the past in the running process of the target object, the track two-dimensional matrix is constructed by encoding each track point as a track point position vector, the historical track information is converted into vector information, extraction of the historical track information in an image processing mode is not needed, intensive calculation is reduced, and calculation resource expenditure is reduced. In addition, the information loss in the image rendering process is prevented, and the integrity of the historical track information transmission is improved.
Based on the same application conception, the embodiment of the present application further provides a trajectory prediction device corresponding to the trajectory prediction method, and since the principle of solving the problem by the device in the embodiment of the present application is similar to that of the embodiment of the trajectory prediction method, the implementation of the device in the embodiment of the present application may refer to the description in the embodiment of the method, and the repetition is omitted.
Fig. 8 is a schematic functional block diagram of a track prediction apparatus according to an embodiment of the present application. The respective modules in the trajectory prediction apparatus in the present embodiment are configured to perform the respective steps in the above-described method embodiments. The track prediction device comprises an acquisition module 301, a coding module 302 and a prediction module 303;
wherein,
the acquisition module 301 is configured to acquire dynamic information and environment information of a target object.
The encoding module 302 is configured to encode the dynamic information and the environment information into dynamic vector information and environment vector information.
The prediction module 303 is configured to perform feature extraction on the motion vector information and the environment vector information through a neural network, and establish an association relationship between the motion vector information and the environment vector information, so as to perform track prediction through the association relationship.
In a possible implementation, the prediction module 303 is further configured to: extracting intermediate features of the dynamic vector information and the environment vector information through a multi-layer perceptron; and establishing an association relation between the dynamic vector information and the environment vector information according to the intermediate characteristics of the dynamic vector information and the environment vector information so as to predict the track through the association relation.
In a possible implementation manner, the prediction module 303 is specifically configured to: acquiring an interaction relation between the intermediate features of the dynamic vector information and the environment vector information through the full connection graph network; and extracting global information from the interaction relationship between the intermediate features through the self-attention mechanism to determine the association relationship between the dynamic vector information and the environment vector information so as to predict the track through the association relationship.
In a possible implementation manner, the track prediction device further comprises a determining module, configured to determine a probability corresponding to each predicted track based on deep learning; and outputting the predicted track with the highest probability as the final predicted track of the target object.
In a possible implementation, the encoding module 302 is further configured to: encoding the position information of each first sampling point in the lane information into a first position vector; and constructing a lane two-dimensional matrix according to the first position vector so as to form lane vector information.
In a possible implementation, the encoding module 302 is further configured to: encoding the position information of each second sampling point of the closed-loop topology information into a second position vector; constructing a closed-loop topology two-dimensional matrix according to the second position vector to form closed-loop topology vector information; wherein a starting two-dimensional vector in a starting two-dimensional matrix in the closed-loop topology two-dimensional matrix is equal to a last two-dimensional vector in a last two-dimensional matrix in the closed-loop topology two-dimensional matrix.
In a possible implementation, the encoding module 302 is further configured to: encoding the position information of each track point in the history track information into a track point position vector; and constructing a track two-dimensional matrix according to the track point position vector to form track vector information.
Furthermore, the present application provides a computer readable storage medium, on which a computer program is stored, which when executed by a processor performs the steps of the trajectory prediction method described in the above method embodiments.
The computer program product of the track prediction method provided in the embodiments of the present application includes a computer readable storage medium storing program codes, where the instructions included in the program codes may be used to execute the steps of the track prediction method described in the method embodiments, and specifically, reference may be made to the method embodiments described above, and details thereof are not repeated herein.
In the several embodiments provided in this application, it should be understood that the disclosed apparatus and method may be implemented in other manners as well. The apparatus embodiments described above are merely illustrative, for example, flow diagrams and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
In addition, the functional modules in the embodiments of the present application may be integrated together to form a single part, or each module may exist alone, or two or more modules may be integrated to form a single part.
The functions, if implemented in the form of software functional modules and sold or used as a stand-alone product, may be stored in a computer-readable storage medium. Based on such understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes. It is noted that relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising … …" does not exclude the presence of other like elements in a process, method, article or apparatus that comprises the element.
The foregoing description is only of the preferred embodiments of the present application and is not intended to limit the same, but rather, various modifications and variations may be made by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application. It should be noted that: like reference numerals and letters denote like items in the following figures, and thus once an item is defined in one figure, no further definition or explanation thereof is necessary in the following figures.
The foregoing is merely specific embodiments of the present application, but the scope of the present application is not limited thereto, and any person skilled in the art can easily think about changes or substitutions within the technical scope of the present application, and the changes and substitutions are intended to be covered by the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.
Claims (6)
1. A track prediction method, comprising:
acquiring dynamic information and environment information of a target object;
encoding the dynamic information and the environment information into dynamic vector information and environment vector information;
Extracting features of the dynamic vector information and the environment vector information through a neural network, and establishing an association relation between the dynamic vector information and the environment vector information so as to predict a track through the association relation;
the neural network comprises a multi-layer perceptron, the characteristic extraction is carried out on the dynamic vector information and the environment vector information through the neural network, and the association relation between the dynamic vector information and the environment vector information is established so as to carry out track prediction through the association relation, and the method comprises the following steps:
extracting intermediate features of the dynamic vector information and the environment vector information through a multi-layer perceptron;
establishing an association relation between the dynamic vector information and the environment vector information according to the intermediate characteristics of the dynamic vector information and the environment vector information so as to predict a track through the association relation;
the establishing the association relationship between the dynamic vector information and the environment vector information according to the intermediate characteristics of the dynamic vector information and the environment vector information comprises the following steps: connecting all or part of intermediate features of the environment vector information and the dynamic vector information respectively to determine association relations between the dynamic vector information and the environment vector information respectively;
When the dynamic vector information or the environment vector information is multiple, each channel extracts the characteristics of corresponding information respectively so as to extract the multiple information respectively;
the environment information includes lane information and closed-loop topology information, the lane information including: the lane line and the lane center line, the environment vector information comprises lane vector information and closed-loop topology vector information, the lane line and the lane center line are constructed into a multi-section broken line form by setting first sampling points with a first preset distance, the closed-loop topology information is constructed into a multi-section broken line form of closed-loop connection by setting second sampling points with a second preset distance, and the dynamic information and the environment information are respectively encoded into dynamic vector information and environment vector information, and the method comprises the following steps:
encoding the position information of each first sampling point in the lane information into a first position vector;
constructing a lane two-dimensional matrix according to the first position vector to form lane vector information;
encoding the position information of each second sampling point of the closed-loop topology information into a second position vector;
constructing a closed-loop topology two-dimensional matrix according to the second position vector to form closed-loop topology vector information;
Wherein a starting two-dimensional vector in a starting two-dimensional matrix in the closed-loop topology two-dimensional matrix is equal to a last two-dimensional vector in a last two-dimensional matrix in the closed-loop topology two-dimensional matrix;
the dynamic information includes historical track information, the dynamic vector information includes track vector information, the dynamic information and the environment information are respectively encoded into dynamic vector information and environment vector information, and the method further includes:
encoding the position information of each track point in the history track information into a track point position vector;
and constructing a track two-dimensional matrix according to the track point position vector to form track vector information.
2. The method of claim 1, wherein the neural network further comprises: the self-attention mechanism and the full-connection graph network establish the association relation between the dynamic vector information and the environment vector information according to the intermediate characteristics of the dynamic vector information and the environment vector information so as to predict the track through the association relation, and the method comprises the following steps:
acquiring an interaction relation between the intermediate features of the dynamic vector information and the environment vector information through the full connection graph network;
And extracting global information from the interaction relationship between the intermediate features through the self-attention mechanism to determine the association relationship between the dynamic vector information and the environment vector information so as to predict the track through the association relationship.
3. The method according to claim 1 or 2, wherein the track of the target object is predicted as a plurality of predicted tracks, and the method comprises, after the track prediction by the association relation:
determining a probability corresponding to each predicted trajectory based on deep learning;
and outputting the predicted track with the highest probability as the final predicted track of the target object.
4. A trajectory prediction device, comprising:
the acquisition module is used for: the method comprises the steps of acquiring dynamic information and environment information of a target object;
and a coding module: for encoding the dynamic information and the environment information into dynamic vector information and environment vector information;
and a prediction module: the method comprises the steps of extracting characteristics of the dynamic vector information and the environment vector information through a neural network, and establishing an association relation between the dynamic vector information and the environment vector information so as to predict a track through the association relation;
The prediction module is further configured to: extracting intermediate features of the dynamic vector information and the environment vector information through a multi-layer perceptron; establishing an association relation between the dynamic vector information and the environment vector information according to the intermediate characteristics of the dynamic vector information and the environment vector information so as to predict a track through the association relation;
the prediction module is further configured to: connecting all or part of intermediate features of the environment vector information and the dynamic vector information respectively to determine association relations between the dynamic vector information and the environment vector information respectively;
when the dynamic vector information or the environment vector information is multiple, each channel extracts the characteristics of corresponding information respectively so as to extract the multiple information respectively; the environment information includes lane information and closed-loop topology information, the lane information including: the system comprises a lane line and a lane central line, wherein the environment vector information comprises lane vector information and closed-loop topology vector information, the lane line and the lane central line are constructed into a multi-section broken line form by arranging first sampling points with a first preset distance, and the closed-loop topology information is constructed into a multi-section broken line form of closed-loop connection by arranging second sampling points with a second preset distance;
The coding module is further configured to: encoding the position information of each first sampling point in the lane information into a first position vector; constructing a lane two-dimensional matrix according to the first position vector to form lane vector information;
the coding module is further configured to: encoding the position information of each second sampling point of the closed-loop topology information into a second position vector; constructing a closed-loop topology two-dimensional matrix according to the second position vector to form closed-loop topology vector information; wherein a starting two-dimensional vector in a starting two-dimensional matrix in the closed-loop topology two-dimensional matrix is equal to a last two-dimensional vector in a last two-dimensional matrix in the closed-loop topology two-dimensional matrix;
the coding module is further configured to: encoding the position information of each track point in the historical track information into a track point position vector; and constructing a track two-dimensional matrix according to the track point position vector to form track vector information.
5. An electronic processing device, comprising: a processor, a memory storing machine-readable instructions executable by the processor, which when executed by the processor perform the steps of the method of any one of claims 1 to 3.
6. A computer-readable storage medium, characterized in that it has stored thereon a computer program which, when executed by a processor, performs the steps of the method according to any of claims 1 to 3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310120681.4A CN116071925B (en) | 2023-02-13 | 2023-02-13 | Track prediction method and device and electronic processing device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310120681.4A CN116071925B (en) | 2023-02-13 | 2023-02-13 | Track prediction method and device and electronic processing device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116071925A CN116071925A (en) | 2023-05-05 |
| CN116071925B true CN116071925B (en) | 2024-04-12 |
Family
ID=86179947
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310120681.4A Active CN116071925B (en) | 2023-02-13 | 2023-02-13 | Track prediction method and device and electronic processing device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116071925B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113128381A (en) * | 2021-04-06 | 2021-07-16 | 浙江吉利控股集团有限公司 | Obstacle trajectory prediction method, system and computer storage medium |
| CN114707630A (en) * | 2022-02-16 | 2022-07-05 | 大连理工大学 | Multi-modal trajectory prediction method by paying attention to scenes and states |
| CN114802296A (en) * | 2022-03-17 | 2022-07-29 | 浙江工业大学 | Vehicle track prediction method based on dynamic interaction graph convolution |
| CN115221970A (en) * | 2022-07-28 | 2022-10-21 | 重庆长安汽车股份有限公司 | Target trajectory prediction method, system, device, and medium |
| WO2022222095A1 (en) * | 2021-04-22 | 2022-10-27 | 深圳元戎启行科技有限公司 | Trajectory prediction method and apparatus, and computer device and storage medium |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20210174668A1 (en) * | 2019-12-10 | 2021-06-10 | Samsung Electronics Co., Ltd. | Systems and methods for trajectory prediction |
-
2023
- 2023-02-13 CN CN202310120681.4A patent/CN116071925B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113128381A (en) * | 2021-04-06 | 2021-07-16 | 浙江吉利控股集团有限公司 | Obstacle trajectory prediction method, system and computer storage medium |
| WO2022222095A1 (en) * | 2021-04-22 | 2022-10-27 | 深圳元戎启行科技有限公司 | Trajectory prediction method and apparatus, and computer device and storage medium |
| CN114707630A (en) * | 2022-02-16 | 2022-07-05 | 大连理工大学 | Multi-modal trajectory prediction method by paying attention to scenes and states |
| CN114802296A (en) * | 2022-03-17 | 2022-07-29 | 浙江工业大学 | Vehicle track prediction method based on dynamic interaction graph convolution |
| CN115221970A (en) * | 2022-07-28 | 2022-10-21 | 重庆长安汽车股份有限公司 | Target trajectory prediction method, system, device, and medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116071925A (en) | 2023-05-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Mozaffari et al. | Deep learning-based vehicle behavior prediction for autonomous driving applications: A review | |
| Hou et al. | Interactive trajectory prediction of surrounding road users for autonomous driving using structural-LSTM network | |
| KR102539942B1 (en) | Method and apparatus for training trajectory planning model, electronic device, storage medium and program | |
| CN110597086A (en) | Simulation scene generation method and unmanned system test method | |
| CN111316286A (en) | Trajectory prediction method and device, storage medium, driving system and vehicle | |
| CN113159403B (en) | Intersection pedestrian track prediction method and device | |
| CN113947764B (en) | Image processing method, device, equipment and storage medium | |
| Gao et al. | A data-driven lane-changing behavior detection system based on sequence learning | |
| Katariya et al. | Deeptrack: Lightweight deep learning for vehicle trajectory prediction in highways | |
| Xie et al. | Congestion-aware multi-agent trajectory prediction for collision avoidance | |
| CN115909749B (en) | Vehicle running road risk early warning method, device, equipment and storage medium | |
| CN115690153A (en) | Intelligent agent track prediction method and system | |
| Kolekar et al. | Behavior prediction of traffic actors for intelligent vehicle using artificial intelligence techniques: A review | |
| CN114792148A (en) | Method and device for predicting motion trail | |
| Kawasaki et al. | Multimodal trajectory predictions for autonomous driving without a detailed prior map | |
| Zhang et al. | Step attention: Sequential pedestrian trajectory prediction | |
| Gross et al. | Route and stopping intent prediction at intersections from car fleet data | |
| Xie et al. | A cognition‐inspired trajectory prediction method for vehicles in interactive scenarios | |
| Selvaraj et al. | Edge learning of vehicular trajectories at regulated intersections | |
| CN116071925B (en) | Track prediction method and device and electronic processing device | |
| CN113119996B (en) | Trajectory prediction method and apparatus, electronic device and storage medium | |
| CN115719547A (en) | Traffic participant trajectory prediction method and system based on multiple interactive behaviors | |
| Gao et al. | Deep learning‐based hybrid model for the behaviour prediction of surrounding vehicles over long‐time periods | |
| Popescu et al. | A lane assessment method using visual information based on a dynamic bayesian network | |
| Jazayeri | Predicting Vehicle Trajectories at Intersections Using Advanced Machine Learning Techniques |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant |