CN119415828B - Track prediction method based on denoising and related equipment - Google Patents

Track prediction method based on denoising and related equipment

Info

Publication number
CN119415828B
CN119415828B CN202411444348.XA CN202411444348A CN119415828B CN 119415828 B CN119415828 B CN 119415828B CN 202411444348 A CN202411444348 A CN 202411444348A CN 119415828 B CN119415828 B CN 119415828B
Authority
CN
China
Prior art keywords
trajectory
historical
target vehicle
future
code
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
Application number
CN202411444348.XA
Other languages
Chinese (zh)
Other versions
CN119415828A (en
Inventor
王森章
王梓辰
王任之
王建新
张健
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Central South University
Original Assignee
Central South University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Central South University filed Critical Central South University
Priority to CN202411444348.XA priority Critical patent/CN119415828B/en
Publication of CN119415828A publication Critical patent/CN119415828A/en
Application granted granted Critical
Publication of CN119415828B publication Critical patent/CN119415828B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/10—Pre-processing; Data cleansing
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G06N3/0455—Auto-encoder networks; Encoder-decoder networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/0985—Hyperparameter optimisation; Meta-learning; Learning-to-learn

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

本申请涉及轨迹预测技术领域，提供了一种基于去噪的轨迹预测方法及相关设备，该方法包括：对目标车辆的历史轨迹进行编码，得到目标车辆的历史轨迹隐藏编码，并对每个邻居车辆的历史轨迹进行编码，得到每个邻居车辆的历史轨迹隐藏编码；根据所有邻居车辆的历史轨迹隐藏编码，计算目标车辆的最终社会编码；基于目标车辆的历史轨迹隐藏编码和最终社会编码，对目标车辆进行轨迹预测，得到未来轨迹分布；对未来轨迹分布进行采样，得到噪声未来轨迹，并利用所有历史轨迹对噪声未来轨迹进行去噪，得到目标车辆的最终未来轨迹。本申请的方法能够提高轨迹预测的准确性。

The present application relates to the field of trajectory prediction technology and provides a denoising-based trajectory prediction method and related equipment. The method comprises: encoding the historical trajectory of a target vehicle to obtain a hidden code for the target vehicle's historical trajectory, encoding the historical trajectory of each neighboring vehicle to obtain a hidden code for the historical trajectory of each neighboring vehicle; calculating the final social code of the target vehicle based on the hidden codes of the historical trajectories of all neighboring vehicles; predicting the trajectory of the target vehicle based on the hidden codes of the historical trajectory and the final social code to obtain a future trajectory distribution; sampling the future trajectory distribution to obtain a noisy future trajectory, and denoising the noisy future trajectory using all historical trajectories to obtain the final future trajectory of the target vehicle. The method of the present application can improve the accuracy of trajectory prediction.

Description

Track prediction method based on denoising and related equipment
Technical Field
The application relates to the technical field of track prediction, in particular to a track prediction method based on denoising and related equipment.
Background
The purpose of vehicle trajectory prediction is to predict the future trajectory of the target vehicle itself and its surrounding neighbors based on its historical trajectories. Accurate prediction of future trajectories of vehicles is critical for many autopilot applications, including optimizing driving path planning, making accurate driving in dynamic environments, and improving driving safety. Traditionally, statistical models predict future trajectories from historical trajectories of individual agents. However, these models do not take into account interactions between the target agent and surrounding neighbors, thereby degrading predictive performance. To address this problem, various deep learning-based models have been proposed to simulate the spatial interactions between vehicles. However, existing deep learning-based methods cannot model the uncertainty of the trajectory, which is common in real-world driving scenarios. Generally, uncertainty in automatic driving can be roughly divided into uncertainty in driving behavior and uncertainty in driving scene, how to design a track prediction model perceived by the uncertainty is still not fully explored, and the problem of low accuracy of track prediction due to uncertainty is still a research topic to be solved at present.
Disclosure of Invention
The application provides a track prediction method based on denoising and related equipment, which can solve the problem of low track prediction accuracy.
In a first aspect, an embodiment of the present application provides a denoising-based trajectory prediction method, where the trajectory prediction method includes:
Acquiring a historical track of a target vehicle, and acquiring a historical track of each neighbor vehicle of the target vehicle;
encoding the historical track of the target vehicle to obtain a historical track hiding code of the target vehicle, and encoding the historical track of each neighbor vehicle to obtain a historical track hiding code of each neighbor vehicle;
Calculating a final social code of the target vehicle according to the historical track hidden codes of all the neighbor vehicles, wherein the final social code is used for describing interaction information of the target vehicle and all the neighbor vehicles;
Track prediction is carried out on the target vehicle based on the historical track hiding code and the final social code of the target vehicle to obtain future track distribution, wherein the future track distribution is used for describing the future track generated by each maneuvering mode executed by the target vehicle at the future moment and the probability of executing each maneuvering mode executed by the target vehicle at the future moment;
and sampling the future track distribution to obtain a noise future track, and denoising the noise future track by utilizing all the historical tracks to obtain a final future track of the target vehicle.
Optionally, encoding the historical track of the target vehicle to obtain a historical track hidden code of the target vehicle, including:
encoding the historical track of the target vehicle to obtain a historical track code of the target vehicle;
And performing secondary encoding on the historical track codes to obtain the historical track hidden codes.
Optionally, calculating a final social code of the target vehicle according to the historical track hidden codes of all neighboring vehicles, including:
respectively aiming at each neighbor vehicle, generating social codes corresponding to the neighbor vehicles based on historical track hidden codes of the neighbor vehicles, and calculating amplitude embedding and phase embedding of the neighbor vehicles by using the social codes;
calculating a spatial characterization of each neighboring vehicle based on all amplitude embeddings and all phase embeddings;
constructing an initial social code of the target vehicle according to all the spatial characterizations;
And calculating the initial social code by using a attention mechanism to obtain the final social code of the target vehicle.
Optionally, calculating the amplitude embedding and the phase embedding of the neighboring vehicle using social codes includes:
By the formula:
zj＝Plain-FC(hj,Wz)
θj＝Plain-FC(hj,Wθ)
Calculating an amplitude embedding z j and a phase embedding theta j of the j-th neighbor vehicle;
Where h j represents social coding of the j-th neighbor vehicle, plain-FC () represents na iotave full connection, W z represents computing an amplitude-embedded learnable parameter matrix, W θ represents computing a phase-embedded learnable parameter matrix, j=1, 2.
Optionally, calculating a spatial representation of each neighboring vehicle based on all amplitude embeddings and all phase embeddings includes:
By the formula:
calculating a spatial representation o j of the j-th neighbor vehicle;
wherein, the Representing a complex value of the complex, Representing the amplitude embedding and phase embedding of all neighboring vehicles,Representing a combination of amplitude embedding and phase embedding for the 1 st neighbor vehicle,Representing a combination of amplitude embedding and phase embedding for the 2 nd neighbor vehicle,Representing a combination of amplitude embedding and phase embedding for the nth neighbor vehicle, M pos represents a neighbor vehicle position mask matrix, W t,AndAll represent a learnable weight matrix, z k represents the amplitude embedding of the kth neighbor vehicle, theta k represents the phase embedding of the kth neighbor vehicle, and z k⊙cosθk represents a complex valueZ k⊙sinθk represents a complex valueIs a virtual value of k e (1, 2, once again, n), k+.j.
Optionally, track prediction is performed on the target vehicle based on the historical track hiding code and the final social code of the target vehicle to obtain future track distribution, including:
adding the historical track hiding code and the final social code of the target vehicle and carrying out standardization processing to obtain an interaction information vector;
Carrying out self-adaptive fusion on the interaction information vector and the mapping matrix of all maneuvering modes to obtain a fusion vector;
multiplying the fusion vector and the interaction information vector to obtain a final vector, and calculating the final vector to obtain future track distribution Wherein, the Representing the mean value of the target vehicle's position at a future time,Representing the variance of the target vehicle's position at a future time,Representing the correlation coefficient.
Optionally, adaptively fusing the interaction information and the mapping matrix of all maneuvering modes to obtain a fusion vector, including:
By the formula:
Computing fusion vectors
Wherein c t-t′ represents an element corresponding to the t-t' th historical moment in the interaction information, A combined vector representing the mapping matrix of all maneuver modalities, Representing the elements corresponding to the first T h historical moments in the mapping matrix containing all maneuver modalities,Representing the element corresponding to the first 2 historical moments in the mapping matrix containing all maneuver modalities,Representing the element corresponding to the first 1 historic moment in the mapping matrix containing all maneuver modalities, The vector of interaction information is represented as such,The interactive information corresponding to the T-T h historical time is represented, c t-2 represents the interactive information corresponding to the T-2 historical time, c t-1 represents the interactive information corresponding to the T-1 historical time, and T represents the current time.
Optionally, sampling the future track distribution to obtain a noise future track includes:
And taking the maneuver mode corresponding to the probability of the largest value in the future track distribution as a final maneuver mode, and taking the future track generated by the target vehicle executing the final maneuver mode as a noise future track.
Optionally, denoising the future track of noise by using all the historical tracks to obtain a final future track of the target vehicle, including:
By the formula:
Calculating future track of noise after denoising in the r step
Wherein alpha r,All of which represent parameters of the diffusion process,Represents the future track of the noise after the r+1 step denoising,Represents estimated noise, z represents noise, z-N (z; 0,I), I represents identity matrix, f ∈ () represents noise estimation model,Representing spatiotemporal embedding, X tar represents the historical track of the target vehicle,Representing the historic trajectories of all neighbor vehicles, f context () representing the information encoder, r=1, 2,..r, R representing the number of steps of the denoising process, and when r=1, denoising the 1 st step of the future trajectory of the noiseAs the final future trajectory of the target vehicle.
In a second aspect, an embodiment of the present application provides a denoising-based trajectory prediction apparatus, including:
the acquisition module acquires the historical track of the target vehicle and acquires the historical track of each neighbor vehicle of the target vehicle;
the system comprises an encoding module, a history track hiding module and a dynamic information processing module, wherein the encoding module encodes the history track of the target vehicle to obtain a history track hiding code of the target vehicle, and encodes the history track of each neighbor vehicle to obtain a history track hiding code of each neighbor vehicle;
the calculation module is used for calculating the final social code of the target vehicle according to the historical track hidden codes of all the neighbor vehicles, wherein the final social code is used for describing the interaction information of the target vehicle and all the neighbor vehicles;
The track prediction module is used for predicting the track of the target vehicle based on the historical track hiding code and the final social code of the target vehicle to obtain future track distribution, wherein the future track distribution is used for describing the future track generated by each maneuvering mode executed by the target vehicle at the future moment and the probability of executing each maneuvering mode executed by the target vehicle at the future moment;
The sampling module is used for sampling the future track distribution to obtain a noise future track, and denoising the noise future track by utilizing all the historical tracks to obtain a final future track of the target vehicle.
In a third aspect, an embodiment of the present application provides a terminal device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, where the processor implements the denoising-based trajectory prediction method described above when executing the computer program.
In a fourth aspect, embodiments of the present application provide a computer readable storage medium storing a computer program which, when executed by a processor, implements the denoising-based trajectory prediction method described above.
The scheme of the application has the following beneficial effects:
In the embodiment of the application, the historical track of the target vehicle is obtained, the historical track of each neighbor vehicle of the target vehicle is obtained, then the historical track of the target vehicle is encoded to obtain the historical track hiding code of the target vehicle, the historical track of each neighbor vehicle is encoded to obtain the historical track hiding code of each neighbor vehicle, then the final social code of the target vehicle is calculated according to the historical track hiding codes of all neighbor vehicles, then the track prediction is carried out on the target vehicle based on the historical track hiding codes and the final social code of the target vehicle to obtain future track distribution, finally the future track distribution is sampled to obtain the future track of noise, and the future track of noise is denoised by utilizing all the historical tracks to obtain the final future track of the target vehicle. The method comprises the steps of calculating the final social code of the target vehicle according to the historical track hidden code of the neighbor vehicle, and fully analyzing the interaction information between the target vehicle and the neighbor vehicle, so that the final social code can accurately describe the interaction information between the target vehicle and all the neighbor vehicles, the accuracy of future track distribution is high according to the accurate final social code calculation, then the future track distribution is sampled and denoised, the uncertainty of the track can be reduced, and the accuracy of track prediction is further improved.
In addition, the process of obtaining future track distribution and then obtaining the final future track through sampling and denoising is realized, track prediction from thick to thin stages is realized, interaction between vehicles can be well captured, multi-mode tracks of the vehicles can be modeled, and most importantly, uncertainty of the future track can be gradually reduced through sampling and gradual denoising.
Other advantageous effects of the present application will be described in detail in the detailed description section which follows.
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 or the description of the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and that other drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flowchart of a denoising-based trajectory prediction method according to an embodiment of the present application;
fig. 2 is a schematic structural diagram of a denoising-based trajectory prediction apparatus according to an embodiment of the present application;
fig. 3 is a schematic structural diagram of a terminal device according to an embodiment of the present application.
Detailed Description
In the following description, for purposes of explanation and not limitation, specific details are set forth such as the particular system architecture, techniques, etc., in order to provide a thorough understanding of the embodiments of the present application. It will be apparent, however, to one skilled in the art that the present application may be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present application with unnecessary detail.
It should be understood that the terms "comprises" and/or "comprising," when used in this specification and the appended claims, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.
It should also be understood that the term "and/or" as used in the present specification and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes such combinations.
As used in the present description and the appended claims, the term "if" may be interpreted as "when..once" or "in response to a determination" or "in response to detection" depending on the context. Similarly, the phrase "if a determination" or "if a [ described condition or event ] is detected" may be interpreted in the context of meaning "upon determination" or "in response to determination" or "upon detection of a [ described condition or event ]" or "in response to detection of a [ described condition or event ]".
Furthermore, the terms "first," "second," "third," and the like in the description of the present specification and in the appended claims, are used for distinguishing between descriptions and not necessarily for indicating or implying a relative importance.
Reference in the specification to "one embodiment" or "some embodiments" or the like means that a particular feature, structure, or characteristic described in connection with the embodiment is included in one or more embodiments of the application. Thus, appearances of the phrases "in one embodiment," "in some embodiments," "in other embodiments," and the like in the specification are not necessarily all referring to the same embodiment, but mean "one or more but not all embodiments" unless expressly specified otherwise. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless expressly specified otherwise.
Aiming at the problem of low accuracy of the existing track prediction, the embodiment of the application provides a track prediction method based on denoising, which comprises the steps of obtaining a historical track of a target vehicle, obtaining a historical track of each neighbor vehicle of the target vehicle, then encoding the historical track of the target vehicle to obtain a historical track hidden code of the target vehicle, encoding the historical track of each neighbor vehicle to obtain a historical track hidden code of each neighbor vehicle, calculating a final social code of the target vehicle according to the historical track hidden codes of all neighbor vehicles, then predicting the track of the target vehicle based on the historical track hidden codes and the final social code of the target vehicle to obtain future track distribution, finally sampling the future track distribution to obtain a noise future track, and denoising the noise future track by utilizing all the historical tracks to obtain the final future track of the target vehicle. The method comprises the steps of calculating the final social code of the target vehicle according to the historical track hidden code of the neighbor vehicle, and fully analyzing the interaction information between the target vehicle and the neighbor vehicle, so that the final social code can accurately describe the interaction information between the target vehicle and all the neighbor vehicles, the accuracy of future track distribution is high according to the accurate final social code calculation, then the future track distribution is sampled and denoised, the uncertainty of the track can be reduced, and the accuracy of track prediction is further improved.
In addition, the process of obtaining future track distribution and then obtaining the final future track through sampling and denoising is realized, track prediction from thick to thin stages is realized, interaction between vehicles can be well captured, multi-mode tracks of the vehicles can be modeled, and most importantly, uncertainty of the future track can be gradually reduced through sampling and gradual denoising.
The following describes an exemplary denoising-based trajectory prediction method provided by the present application.
As shown in fig. 1, the track prediction method based on denoising provided by the application comprises the following steps:
Step 11, acquiring a historical track of the target vehicle, and acquiring a historical track of each neighbor vehicle of the target vehicle.
The target vehicle is a vehicle needing track prediction, the neighboring vehicle is a vehicle adjacent to the target vehicle, the history track is a motion track of the vehicle in a history time period before the current moment, if the current moment is 9 points, the history track can be a motion track of a half-vehicle from 7 points to 8 points, and the neighboring vehicle is a neighboring vehicle of the target vehicle in the history time period.
In some embodiments of the present application, the historical track may be obtained using a positioning system of the target vehicle and the neighboring vehicle.
And step 12, encoding the historical track of the target vehicle to obtain a historical track hiding code of the target vehicle, and encoding the historical track of each neighbor vehicle to obtain a historical track hiding code of each neighbor vehicle.
The history track hiding code is used for describing dynamic information of the history track, such as moving direction, speed and the like of the vehicle when the history track runs.
In some embodiments of the present application, the step of encoding the historical track of the target vehicle to obtain the hidden encoding of the historical track of the target vehicle specifically includes:
First, the historical track of the target vehicle is encoded, and the historical track code of the target vehicle is obtained.
For example, the historical track of the target vehicle may be encoded by using a multi-layer perceptron to obtain a historical track code of the target vehicle.
And secondly, performing secondary coding on the historical track codes to obtain the historical track hidden codes.
For example, the historical track code may be secondarily encoded using a long and short term memory network to obtain a historical track hidden code.
It should be noted that, the process of encoding the history track of each neighboring vehicle to obtain the history track hidden code of each neighboring vehicle is the same as the process of obtaining the history track hidden code of the target vehicle, that is, encoding the history track of the neighboring vehicle to obtain the history track code of the neighboring vehicle, and then secondarily encoding the history track code to obtain the history track hidden code.
And step 13, calculating the final social code of the target vehicle according to the historical track hidden codes of all the neighbor vehicles.
The final social code is used for describing interaction information of the target vehicle and all the neighbor vehicles, such as the distance between the neighbor vehicles and the target vehicle, the relative speed between the neighbor vehicles and the target vehicle, and the like.
In some embodiments of the present application, the step of calculating the final social code of the target vehicle according to the historical track hidden codes of all neighboring vehicles specifically includes:
the method comprises the steps of firstly, respectively aiming at each neighbor vehicle, generating social codes corresponding to the neighbor vehicles based on historical track hiding codes of the neighbor vehicles, and calculating amplitude embedding and phase embedding of the neighbor vehicles by using the social codes.
In some embodiments of the present application, the step of generating the social code corresponding to the neighboring vehicles based on the historical track hiding code of the neighboring vehicles specifically includes expanding a mask matrix mask representing a positional relationship between the target vehicle and each neighboring vehicle to a shape consistent with a historical time period corresponding to the historical track. A new tensor representation social code is then created, which is initialized to an all zero tensor of the same shape as the mask matrix. And then, using a mask_scanner_function of pytorch to fill the historical track hiding codes of the neighbor vehicles into the social codes according to a mask matrix. Specifically, mask_mask_allows a specific value of one tensor to be bit-inserted into another tensor according to one boolean mask, i.e. for the position where the mask is True, the corresponding value in the historical track hidden code of the neighboring vehicle is filled into the social code, and for the position where the mask is False, the social code remains the same.
The steps of calculating the amplitude embedding and the phase embedding of the neighbor vehicle by using the social codes specifically comprise:
By the formula:
zj＝Plain-FC(hj,Wz)
θj＝Plain-FC(hj,Wθ)
The amplitude embedding z j and phase embedding θ j of the j-th neighbor vehicle are calculated.
Where h j represents social coding of the j-th neighbor vehicle, plain-FC () represents naive full connection, W z represents amplitude embedded learnable weight matrix, W θ represents phase embedded learnable weight matrix, j=1, 2.
Second, a spatial characterization of each neighboring vehicle is computed based on all amplitude embeddings and all phase embeddings.
Specifically, the formula is as follows:
The spatial characterization o j of the j-th neighbor vehicle is calculated.
Wherein, the Representing a complex value of the complex, Representing the amplitude embedding and phase embedding of all neighboring vehicles,Representing a combination of amplitude embedding and phase embedding for the 1 st neighbor vehicle,Representing a combination of amplitude embedding and phase embedding for the 2 nd neighbor vehicle,Representing a combination of amplitude embedding and phase embedding for the nth neighbor vehicle, M pos represents a neighbor vehicle position mask matrix, W t,AndAll represent a learnable weight matrix, z k represents the amplitude embedding of the kth neighbor vehicle, theta k represents the phase embedding of the kth neighbor vehicle, and z k⊙cosθk represents a complex valueZ k⊙sinθk represents a complex valueIs a virtual value of k e (1, 2, once again, n), k+.j.
The above-mentioned surrouding-FC is calculated by extracting phase embedding and amplitude embedding for the neighboring vehicle through convolution operation. The phase embedding and amplitude embedding are then developed by the euler formula to capture the relative interactions and spatiotemporal relationships between vehicles. These features are then re-weighted using adaptive pooling so that the individual features can be automatically adjusted according to relative importance. Finally, further processing is performed by a multi-layer perceptron (MLP) to generate a spatial representation with enhanced phase information and interaction characteristics.
And thirdly, constructing an initial social code of the target vehicle according to all the spatial characterizations.
It should be noted that, for the spatial representation o j, it is a vector formed by a plurality of elements, where the plurality of elements are in one-to-one correspondence with a plurality of historical moments, and are used to describe spatial information of neighboring vehicles at each historical moment. For a historical moment, integrating elements corresponding to the historical moment in all the spatial characterizations to obtain social interaction characterizations, and integrating the social interaction characterizations at all the historical moments to obtain an initial social code Wherein, the Representing social interaction characterization at the T-T h historical time, h t-2 representing social interaction characterization at the T-2 historical time, and h t-1 representing social interaction characterization at the T-1 historical time.
And fourthly, calculating the initial social code by using an attention mechanism to obtain the final social code of the target vehicle.
Specifically, the initial social code is used as the input of the attention mechanism, a query matrix (Q), a key matrix (K) and a value matrix (V) are calculated based on the self-attention mechanism, and then the dot product of the query matrix and the key matrix is calculated to obtain the attention score, so that the dependency relationship between the historical moments, namely the relationship between the characteristics of different historical moments, is captured. Then, normalizing the attention score by using softmax to ensure that the attention distribution at each historical moment is a probability distribution and the sum is 1, weighting and summing the value matrix according to the calculated attention score to obtain a new historical moment characteristic, and finally combining the results of different attention heads to obtain the final social codeThe expression is:
Wherein Concat denotes a splicing operation, head 1 denotes an output of the 1 st attention head, head 2 denotes an output of the 2 nd attention head, and head k denotes an output of the k th attention head.
It is worth mentioning that the final social code of the target vehicle is calculated according to the historical track hidden code of the neighbor vehicle, so that the interaction information between the target vehicle and the neighbor vehicles is fully analyzed, and the final social code can accurately describe the interaction information between the target vehicle and all the neighbor vehicles.
And 14, carrying out track prediction on the target vehicle based on the historical track hiding code and the final social code of the target vehicle to obtain future track distribution.
The future track profile described above is used to describe the future track generated by the target vehicle at the future time instant for each maneuver mode, and the probability that the target vehicle will execute each maneuver mode at the future time instant. The maneuver modes are used to describe the motion conditions of the target vehicle, such as left lane change, right lane change, lane keeping, acceleration, speed keeping, and the like.
In some embodiments of the present application, the step of predicting the track of the target vehicle based on the historical track hiding code and the final social code of the target vehicle to obtain the future track distribution specifically includes:
And the first step is to add the historical track hiding code and the final social code of the target vehicle and perform normalization processing to obtain an interaction information vector.
Illustratively, the normalization process may be performed using existing normalization formulas.
And secondly, carrying out self-adaptive fusion on the interaction information vector and the mapping matrix of all maneuvering modes to obtain a fusion vector.
Specifically, the formula is as follows:
Computing fusion vectors
Wherein c t-t′ represents an element corresponding to the t-t' th historical moment in the interaction information, A combined vector representing the mapping matrix of all maneuver modalities, Representing the elements corresponding to the first T h historical moments in the mapping matrix containing all maneuver modalities,Representing the element corresponding to the first 2 historical moments in the mapping matrix containing all maneuver modalities,Representing the element corresponding to the first 1 historic moment in the mapping matrix containing all maneuver modalities, The vector of interaction information is represented as such,The interactive information corresponding to the T-T h historical time is represented, c t-2 represents the interactive information corresponding to the T-2 historical time, c t-1 represents the interactive information corresponding to the T-1 historical time, and T represents the current time.
It should be noted that, the mapping matrix of the maneuver mode is preset according to the maneuver condition of the maneuver mode, where the mapping matrix has a plurality of elements, and the plurality of elements are in one-to-one correspondence with a plurality of historical moments.
Thirdly, multiplying the fusion vector and the interaction information vector to obtain a final vector, and calculating the final vector to obtain future track distributionWherein, the Representing the mean value of the target vehicle's position at a future time,Representing the variance of the target vehicle's position at a future time,Representing the correlation coefficient.
For example, the final vector may be calculated using a long and short term memory network to obtain the future track distribution. The future trajectory distribution is a gaussian distribution.
It is worth mentioning that the accuracy of the future track distribution obtained by calculating according to the accurate final social coding is high, and the probability of each maneuvering mode of the target vehicle to be executed at the future moment can be accurately described.
And step 15, sampling future track distribution to obtain a noise future track, and denoising the noise future track by utilizing all the historical tracks to obtain a final future track of the target vehicle.
In some embodiments of the present application, the step of sampling the future track distribution to obtain a noise future track, and denoising the noise future track by using all the historical tracks to obtain a final future track of the target vehicle specifically includes:
And in the first step, sampling future track distribution to obtain a noise future track.
Specifically, the maneuver mode corresponding to the probability of the largest value in the future track distribution is taken as the final maneuver mode, and the future track generated by the target vehicle executing the final maneuver mode is taken as the noise future track.
And secondly, denoising the future track of the noise by using all the historical tracks to obtain the final future track of the target vehicle.
Specifically, the formula is as follows:
Calculating future track of noise after denoising in the r step
Wherein alpha r,All of which represent parameters of the diffusion process,Represents the future track of the noise after the r+1 step denoising,Represents estimated noise, z represents noise, z-N (z; 0,I), I represents identity matrix, f ∈ () represents noise estimation model,Representing spatiotemporal embedding, X tar represents the historical track of the target vehicle,Representing the historic trajectories of all neighbor vehicles, f context () representing the information encoder, r=1, 2,..r, R representing the number of steps of the denoising process, and when r=1, denoising the 1 st step of the future trajectory of the noiseAs the final future trajectory of the target vehicle.
In the above formula, when r=r,Future trajectories for the incoming noise. The above described denoising process is a reverse process from the last step to the 1 st step (a reverse diffusion process similar to the diffusion model). The information encoder may be a transducer-based encoder.
For example, the method of the present application may be implemented on an NVIDIA 3090 GPU (Instrida 3090 graphics processing unit) server using a Pytorch framework. The parameters were set by using a 13 x5 grid defined centered on the target vehicle with each column corresponding to one lane, each row being 15 feet apart. When the hidden codes of the historical track are calculated, a multi-layer perceptron and a long-short-term memory network are adopted, the hidden characteristic of the multi-layer perceptron is set to be 32, and the activation function is ReL u. To train the framework corresponding to the method of the present application, a two-stage training strategy is considered, wherein the first stage trains the denoising module (i.e., the formula involved in step 15 of the present application), and the second stage focuses on training the spatiotemporal interaction module (i.e., the formulas, models, etc. involved in steps 11 to 14 of the present application). Each historical track is segmented over a sensing range (i.e., 8 s) containing past (3 s) and future (5 s) locations of 5 Hz. And dividing the data set formed by all the history tracks after being divided into segments into a training set, a verification set and a test set, wherein the dividing ratio is 7:2:1, and the data set is used for executing the method of the application and performing training, verification and test.
It is worth mentioning that the final social code of the target vehicle is calculated according to the historical track hidden code of the neighbor vehicle, so that the interaction information between the target vehicle and the neighbor vehicles is fully analyzed, the final social code can accurately describe the interaction information between the target vehicle and all the neighbor vehicles, the accuracy of future track distribution is high according to the accurate final social code calculation, then the future track distribution is sampled and denoised, the uncertainty of the track can be reduced, and the accuracy of track prediction is further improved.
In addition, the process of obtaining future track distribution and then obtaining the final future track through sampling and denoising is realized, track prediction from thick to thin stages is realized, interaction between vehicles can be well captured, multi-mode tracks of the vehicles can be modeled, and most importantly, uncertainty of the future track can be gradually reduced through sampling and gradual denoising.
The following describes an exemplary denoising-based trajectory prediction apparatus provided by the present application.
As shown in fig. 2, an embodiment of the present application provides a denoising-based trajectory prediction apparatus 200, which includes:
an acquisition module 201 that acquires a history track of the target vehicle and acquires a history track of each neighboring vehicle of the target vehicle;
The encoding module 203 encodes the historical track of the target vehicle to obtain a historical track hiding code of the target vehicle, and encodes the historical track of each neighboring vehicle to obtain a historical track hiding code of each neighboring vehicle;
The calculation module 204 is used for calculating the final social code of the target vehicle according to the historical track hidden codes of all the neighbor vehicles, wherein the final social code is used for describing the interaction information of the target vehicle and all the neighbor vehicles;
the track prediction module 205 predicts the track of the target vehicle based on the historical track hiding code and the final social code of the target vehicle to obtain future track distribution, wherein the future track distribution is used for describing the future track generated by each maneuvering mode executed by the target vehicle at the future moment and the probability of executing each maneuvering mode executed by the target vehicle at the future moment;
the sampling module 206 samples the future track distribution to obtain a noise future track, and denoises the noise future track by using all the historical tracks to obtain a final future track of the target vehicle.
It should be noted that, because the content of information interaction and execution process between the above devices/units is based on the same concept as the method embodiment of the present application, specific functions and technical effects thereof may be referred to in the method embodiment section, and will not be described herein.
It will be apparent to those skilled in the art that, for convenience and brevity of description, only the above-described division of the functional units and modules is illustrated, and in practical application, the above-described functional distribution may be performed by different functional units and modules according to needs, i.e. the internal structure of the apparatus is divided into different functional units or modules to perform all or part of the above-described functions. The functional units and modules in the embodiment may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit, where the integrated units may be implemented in a form of hardware or a form of a software functional unit. In addition, the specific names of the functional units and modules are only for distinguishing from each other, and are not used for limiting the protection scope of the present application. The specific working process of the units and modules in the above system may refer to the corresponding process in the foregoing method embodiment, which is not described herein again.
As shown in fig. 3, an embodiment of the present application provides a terminal device D10 of the embodiment comprising at least one processor D100 (only one processor is shown in fig. 3), a memory D101 and a computer program D102 stored in the memory D101 and executable on the at least one processor D100, the processor D100 implementing the steps of any of the respective method embodiments described above when executing the computer program D102.
Specifically, when the processor D100 executes the computer program D102, the historical track of the target vehicle is obtained, the historical track of each neighboring vehicle of the target vehicle is obtained, then the historical track of the target vehicle is encoded to obtain the historical track hiding code of the target vehicle, the historical track of each neighboring vehicle is encoded to obtain the historical track hiding code of each neighboring vehicle, then the final social code of the target vehicle is calculated according to the historical track hiding codes of all neighboring vehicles, then the track prediction is performed on the target vehicle based on the historical track hiding codes and the final social codes of the target vehicle to obtain the future track distribution, finally the future track distribution is sampled to obtain the future track of noise, and the future track of noise is denoised by utilizing all the historical tracks to obtain the final future track of the target vehicle. The method comprises the steps of calculating the final social code of the target vehicle according to the historical track hidden code of the neighbor vehicle, and fully analyzing the interaction information between the target vehicle and the neighbor vehicle, so that the final social code can accurately describe the interaction information between the target vehicle and all the neighbor vehicles, the accuracy of future track distribution is high according to the accurate final social code calculation, then the future track distribution is sampled and denoised, the uncertainty of the track can be reduced, and the accuracy of track prediction is further improved.
The Processor D100 may be a central processing Unit (CPU, centralProcessing Unit), and the Processor D100 may also be other general purpose processors, digital signal processors (DSP, digital Signal Processor), application SPECIFIC INTEGRATED Circuits (ASIC), off-the-shelf programmable gate arrays (FPGA, field-Programmable GateArray) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or the like. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
The memory D101 may in some embodiments be an internal storage unit of the terminal device D10, for example a hard disk or a memory of the terminal device D10. The memory D101 may also be an external storage device of the terminal device D10 in other embodiments, for example, a plug-in hard disk, a smart memory card (SMC, smartMedia Card), a Secure Digital (SD) card, a flash memory card (FLASH CARD) or the like, which are provided on the terminal device D10. Further, the memory D101 may also include both an internal storage unit and an external storage device of the terminal device D10. The memory D101 is used for storing an operating system, an application program, a boot loader (BootLoader), data, other programs, etc., such as program codes of the computer program. The memory D101 may also be used to temporarily store data that has been output or is to be output.
Embodiments of the present application also provide a computer readable storage medium storing a computer program which, when executed by a processor, implements steps for implementing the various method embodiments described above.
Embodiments of the present application provide a computer program product enabling a terminal device to carry out the steps of the method embodiments described above when the computer program product is run on the terminal device.
The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the present application may implement all or part of the flow of the method of the above embodiments, and may be implemented by a computer program to instruct related hardware, where the computer program may be stored in a computer readable storage medium, and when the computer program is executed by a processor, the computer program may implement the steps of each of the method embodiments described above. Wherein the computer program comprises computer program code which may be in source code form, object code form, executable file or some intermediate form etc. The computer readable medium can include at least any entity or device capable of carrying computer program code to a denoising-based trajectory prediction method device/terminal device, a recording medium, a computer memory, a Read-only memory (ROM), a random access memory (RAM, random Access Memory), an electrical carrier signal, a telecommunications signal, and a software distribution medium. Such as a U-disk, removable hard disk, magnetic or optical disk, etc. In some jurisdictions, computer readable media may not be electrical carrier signals and telecommunications signals in accordance with legislation and patent practice.
In the foregoing embodiments, the descriptions of the embodiments are emphasized, and in part, not described or illustrated in any particular embodiment, reference is made to the related descriptions of other embodiments.
Those of ordinary skill in the art will appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.
While the foregoing is directed to the preferred embodiments of the present application, it will be appreciated by those skilled in the art that various modifications and adaptations can be made without departing from the principles of the present application, and such modifications and adaptations are intended to be comprehended within the scope of the present application.

Claims (8)

1.一种基于去噪的轨迹预测方法，其特征在于，包括：1. A trajectory prediction method based on denoising, characterized by comprising: 获取目标车辆的历史轨迹，并获取所述目标车辆的每个邻居车辆的历史轨迹；Obtain a historical trajectory of a target vehicle and a historical trajectory of each neighboring vehicle of the target vehicle; 对所述目标车辆的历史轨迹进行编码，得到所述目标车辆的历史轨迹隐藏编码，并对每个所述邻居车辆的历史轨迹进行编码，得到每个所述邻居车辆的历史轨迹隐藏编码；所述历史轨迹隐藏编码用于描述历史轨迹的动态信息；Encoding the historical trajectory of the target vehicle to obtain a hidden historical trajectory code of the target vehicle, and encoding the historical trajectory of each of the neighboring vehicles to obtain a hidden historical trajectory code of each of the neighboring vehicles; the hidden historical trajectory code is used to describe dynamic information of the historical trajectory; 根据所有邻居车辆的历史轨迹隐藏编码，计算所述目标车辆的最终社会编码；所述最终社会编码用于描述所述目标车辆与所有邻居车辆的交互信息；Calculate the final social code of the target vehicle based on the hidden codes of the historical trajectories of all neighboring vehicles; the final social code is used to describe the interaction information between the target vehicle and all neighboring vehicles; 基于所述目标车辆的历史轨迹隐藏编码和最终社会编码，对所述目标车辆进行轨迹预测，得到未来轨迹分布；所述未来轨迹分布用于描述所述目标车辆在未来时刻执行每种机动模态生成的未来轨迹，以及所述目标车辆在未来时刻执行每种机动模态的概率；Based on the target vehicle's historical trajectory hidden code and final social code, the target vehicle's trajectory is predicted to obtain a future trajectory distribution; the future trajectory distribution is used to describe the future trajectory generated by the target vehicle executing each maneuver mode at a future moment, as well as the probability of the target vehicle executing each maneuver mode at a future moment; 对所述未来轨迹分布进行采样，得到噪声未来轨迹，并利用所有历史轨迹对所述噪声未来轨迹进行去噪，得到所述目标车辆的最终未来轨迹；Sampling the future trajectory distribution to obtain a noisy future trajectory, and denoising the noisy future trajectory using all historical trajectories to obtain a final future trajectory of the target vehicle; 其中，所述对所述未来轨迹分布进行采样，得到噪声未来轨迹，包括：The sampling of the future trajectory distribution to obtain the noisy future trajectory includes: 将所述未来轨迹分布中值最大的概率对应的机动模态作为最终机动模态，并将目标车辆执行所述最终机动模态生成的未来轨迹作为所述噪声未来轨迹；The maneuver mode corresponding to the maximum probability of the median value of the future trajectory distribution is used as the final maneuver mode, and the future trajectory generated by the target vehicle executing the final maneuver mode is used as the noise future trajectory; 所述利用所有历史轨迹对所述噪声未来轨迹进行去噪，得到所述目标车辆的最终未来轨迹，包括：Denoising the noisy future trajectory using all historical trajectories to obtain the final future trajectory of the target vehicle includes: 通过公式：By formula: 计算第r步去噪后的噪声未来轨迹 Calculate the future trajectory of the noise after denoising in step r 其中，αr、均表示扩散过程的参数，表示第r+1步去噪后的噪声未来轨迹，表示估计噪声，z表示噪声，z～N(z；0,I)，I表示单位矩阵，f∈()表示噪声估计模型，表示时空嵌入，Xtar表示目标车辆的历史轨迹，表示所有邻居车辆的历史轨迹，fcontext()表示信息编码器，r＝1,2,...,R，R表示去噪过程的步数，当r＝1时，将第1步去噪后的噪声未来轨迹作为所述目标车辆的最终未来轨迹。Among them, α r , are the parameters of the diffusion process, represents the future trajectory of the noise after the r+1th step denoising, Represents estimated noise, z represents noise, z～N(z;0,I), I represents the unit matrix, f ∈ () represents the noise estimation model, represents spatiotemporal embedding, X tar represents the historical trajectory of the target vehicle, represents the historical trajectory of all neighboring vehicles, fcontext () represents the information encoder, r＝1,2,...,R, R represents the number of steps in the denoising process, when r＝1, the noisy future trajectory after the first step of denoising is converted to as the final future trajectory of the target vehicle. 2.根据权利要求1所述的轨迹预测方法，其特征在于，所述对所述目标车辆的历史轨迹进行编码，得到所述目标车辆的历史轨迹隐藏编码，包括：2. The trajectory prediction method according to claim 1, wherein encoding the historical trajectory of the target vehicle to obtain the hidden historical trajectory code of the target vehicle comprises: 对所述目标车辆的历史轨迹进行编码，得到所述目标车辆的历史轨迹编码；Encoding the historical trajectory of the target vehicle to obtain the historical trajectory code of the target vehicle; 对所述历史轨迹编码进行二次编码，得到历史轨迹隐藏编码。The historical trajectory code is re-encoded to obtain a historical trajectory hidden code. 3.根据权利要求1所述的轨迹预测方法，其特征在于，所述根据所有邻居车辆的历史轨迹隐藏编码，计算所述目标车辆的最终社会编码，包括：3. The trajectory prediction method according to claim 1, wherein the step of calculating the final social code of the target vehicle based on the hidden codes of the historical trajectories of all neighboring vehicles comprises: 分别针对每个所述邻居车辆，基于所述邻居车辆的历史轨迹隐藏编码，生成所述邻居车辆对应的社交编码，并利用所述社交编码计算所述邻居车辆的振幅嵌入和相位嵌入；For each of the neighbor vehicles, generating a social code corresponding to the neighbor vehicle based on the hidden code of the neighbor vehicle's historical trajectory, and using the social code to calculate the amplitude embedding and phase embedding of the neighbor vehicle; 基于所有振幅嵌入和所有相位嵌入计算每个邻居车辆的空间表征；Compute the spatial representation of each neighbor vehicle based on all amplitude embeddings and all phase embeddings; 根据所有空间表征构建所述目标车辆的初始社会编码；constructing an initial social encoding of the target vehicle based on all spatial representations; 利用注意力机制对所述初始社会编码进行计算，得到所述目标车辆的最终社会编码。The initial social code is calculated using an attention mechanism to obtain a final social code of the target vehicle. 4.根据权利要求3所述的轨迹预测方法，其特征在于，所述利用所述社交编码计算所述邻居车辆的振幅嵌入和相位嵌入，包括：4. The trajectory prediction method according to claim 3, wherein the step of calculating the amplitude embedding and phase embedding of the neighboring vehicle using the social coding comprises: 通过公式：By formula: zj＝Plain-FC(hj,Wz)z j =Plain-FC(h j ,W z ) θj＝Plain-FC(hj,Wθ)θ j =Plain-FC(h j ,W θ ) 计算第j个邻居车辆的振幅嵌入zj和相位嵌入θj；Calculate the amplitude embedding zj and phase embedding θj of the jth neighbor vehicle; 其中，hj表示所述第j个邻居车辆的社交编码，Plain-FC()表示朴素全连接，Wz表示计算振幅嵌入的可学习权重矩阵，Wθ表示计算相位嵌入的可学习权重矩阵，j＝1,2,…,n，n表示邻居车辆的数量。Wherein, hj represents the social encoding of the j-th neighbor vehicle, Plain-FC() represents naive full connection, Wz represents the learnable weight matrix for calculating amplitude embedding, Wθ represents the learnable weight matrix for calculating phase embedding, j = 1, 2,…, n, and n represents the number of neighbor vehicles. 5.根据权利要求4所述的轨迹预测方法，其特征在于，所述基于所有振幅嵌入和所有相位嵌入计算每个邻居车辆的空间表征，包括：5. The trajectory prediction method according to claim 4, wherein the step of calculating the spatial representation of each neighboring vehicle based on all amplitude embeddings and all phase embeddings comprises: 通过公式：By formula: 计算第j个邻居车辆的空间表征oj；Calculate the spatial representation o j of the jth neighbor vehicle; 其中，表示复值， 表示所有邻居车辆的振幅嵌入和相位嵌入，表示第1个邻居车辆的振幅嵌入和相位嵌入的组合，表示第2个邻居车辆的振幅嵌入和相位嵌入的组合，表示第n个邻居车辆的振幅嵌入和相位嵌入的组合，Mpos表示邻居车辆位置掩码矩阵，Wt、和均表示可学习权重矩阵，zk表示第k个邻居车辆的振幅嵌入，θk表示所述第k个邻居车辆的相位嵌入，zk⊙cosθk表示复值的真值，zk⊙sinθk表示复值的虚值，k∈(1,2,…,n)，k≠j。in, represents a complex value, represents the amplitude embedding and phase embedding of all neighbor vehicles, represents the combination of amplitude embedding and phase embedding of the first neighbor vehicle, represents the combination of amplitude embedding and phase embedding of the second neighbor vehicle, represents the combination of amplitude embedding and phase embedding of the nth neighbor vehicle, M pos represents the neighbor vehicle position mask matrix, W t , and denotes a learnable weight matrix, zk denotes the amplitude embedding of the kth neighbor vehicle, θk denotes the phase embedding of the kth neighbor vehicle, and zk⊙cosθk denotes a complex value The truth value of z k ⊙sinθ k represents the complex value imaginary value, k∈(1,2,…,n), k≠j. 6.根据权利要求1所述的轨迹预测方法，其特征在于，所述基于所述目标车辆的历史轨迹隐藏编码和最终社会编码，对所述目标车辆进行轨迹预测，得到未来轨迹分布，包括：6. The trajectory prediction method according to claim 1, wherein the step of performing trajectory prediction on the target vehicle based on the target vehicle's historical trajectory hidden code and final social code to obtain a future trajectory distribution comprises: 将所述目标车辆的历史轨迹隐藏编码和最终社会编码相加并进行规范化处理，得到交互信息向量；Adding the hidden code of the target vehicle's historical trajectory and the final social code and performing normalization processing to obtain an interactive information vector; 将所述交互信息向量和所有机动模态的映射矩阵进行自适应融合，得到融合向量；Adaptively fusing the interaction information vector and the mapping matrices of all maneuver modes to obtain a fusion vector; 将所述融合向量和所述交互信息向量相乘，得到最终向量，并对所述最终向量进行计算，得到未来轨迹分布其中，表示所述目标车辆在未来时刻位置的均值，表示所述目标车辆在未来时刻位置的方差，表示相关系数。Multiply the fusion vector and the mutual information vector to obtain the final vector, and calculate the final vector to obtain the future trajectory distribution in, represents the mean value of the target vehicle’s position at the future time, represents the variance of the target vehicle's position at the future moment, Represents the correlation coefficient. 7.根据权利要求6所述的轨迹预测方法，其特征在于，所述将所述交互信息向量和所有机动模态的映射矩阵进行自适应融合，得到融合向量，包括：7. The trajectory prediction method according to claim 6, wherein the step of adaptively fusing the interaction information vector with the mapping matrices of all maneuvering modes to obtain a fused vector comprises: 通过公式：By formula: 计算融合向量 Calculate the fusion vector 其中，ct-t'表示交互信息中第t-t'个历史时刻对应的元素， 表示所有机动模态的映射矩阵的组合向量， 表示包含所有机动模态的映射矩阵中第-Th个历史时刻对应的元素，表示包含所有机动模态的映射矩阵中第-2个历史时刻对应的元素，表示包含所有机动模态的映射矩阵中第-1个历史时刻对应的元素， 表示交互信息向量，表示第t-Th个历史时刻对应的交互信息，ct-2表示第t-2个历史时刻对应的交互信息，ct-1表示第t-1个历史时刻对应的交互信息，t表示当前时刻。Among them, c t-t' represents the element corresponding to the t-t'th historical moment in the interaction information, Represents the combined vector of the mapping matrices of all maneuver modes, represents the element corresponding to the -Th historical moment in the mapping matrix containing all maneuver modes, represents the element corresponding to the -2th historical moment in the mapping matrix containing all maneuver modes, represents the element corresponding to the -1th historical moment in the mapping matrix containing all maneuver modes, represents the mutual information vector, represents the interaction information corresponding to the tT h -th historical moment, c t-2 represents the interaction information corresponding to the t-2-th historical moment, c t-1 represents the interaction information corresponding to the t-1-th historical moment, and t represents the current moment. 8.一种终端设备，包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机程序，其特征在于，所述处理器执行所述计算机程序时实现如权利要求1至7任一项所述的基于去噪的轨迹预测方法。8. A terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the denoising-based trajectory prediction method according to any one of claims 1 to 7 is implemented.
CN202411444348.XA 2024-10-16 2024-10-16 Track prediction method based on denoising and related equipment Active CN119415828B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202411444348.XA CN119415828B (en) 2024-10-16 2024-10-16 Track prediction method based on denoising and related equipment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202411444348.XA CN119415828B (en) 2024-10-16 2024-10-16 Track prediction method based on denoising and related equipment

Publications (2)

Publication Number Publication Date
CN119415828A CN119415828A (en) 2025-02-11
CN119415828B true CN119415828B (en) 2025-10-17

Family

ID=94464698

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202411444348.XA Active CN119415828B (en) 2024-10-16 2024-10-16 Track prediction method based on denoising and related equipment

Country Status (1)

Country Link
CN (1) CN119415828B (en)

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117473032A (en) * 2023-11-20 2024-01-30 上海人工智能创新中心 Scene-level multi-agent trajectory generation method and device based on consistent diffusion

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11189171B2 (en) * 2018-03-13 2021-11-30 Nec Corporation Traffic prediction with reparameterized pushforward policy for autonomous vehicles

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117473032A (en) * 2023-11-20 2024-01-30 上海人工智能创新中心 Scene-level multi-agent trajectory generation method and device based on consistent diffusion

Also Published As

Publication number Publication date
CN119415828A (en) 2025-02-11

Similar Documents

Publication Publication Date Title
CN114022847B (en) A method, system, device and storage medium for predicting trajectory of intelligent agent
KR102770344B1 (en) System and method for geolocation prediction
Sadeghian et al. Sophie: An attentive gan for predicting paths compliant to social and physical constraints
CN112734808B (en) Trajectory prediction method for vulnerable road users in vehicle driving environment
CN112651374B (en) Future trajectory prediction method based on social information and automatic driving system
CN115690153B (en) Intelligent body track prediction method and system
CN116654022B (en) Pedestrian track prediction method, system, equipment and medium based on multiple interactions
CN116893409A (en) Methods and systems for object tracking
CN119149905B (en) Vehicle track prediction method and device suitable for different driving styles
CN116168062B (en) A 3D target tracking method and device
CN115577511B (en) Short-term track prediction method, device and system based on unmanned aerial vehicle motion state
CN113160279A (en) Method and device for detecting abnormal behaviors of pedestrians in subway environment
CN120633467B (en) Pedestrian track prediction method, system, equipment and medium based on continuous learning
US12573056B2 (en) Device and computer-implemented method for object tracking
CN114787880A (en) Method and system for training and validating a perception system
CN113658218B (en) A dual-template dense twin network tracking method, device and storage medium
CN119415828B (en) Track prediction method based on denoising and related equipment
Liu et al. Egocentric vulnerable road users trajectory prediction with incomplete observation
CN114283463A (en) Image processing method, image processing device, electronic equipment and storage medium
CN117009787B (en) Cross-modal semantic information supervision track prediction method
US12608073B2 (en) Predicting body motion
US20250259274A1 (en) System and method for deep equilibirum approach to adversarial attack of diffusion models
WO2025154553A1 (en) Rendering two-dimensional image of a dynamic three-dimensional scene
CN116245913B (en) Hierarchical Context-Guided Multi-Target Tracking Method
CN117291952A (en) A multi-target tracking method and device based on speed prediction and image reconstruction

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
GR01 Patent grant