CN114312831B - Vehicle track prediction method based on spatial attention mechanism - Google Patents
Vehicle track prediction method based on spatial attention mechanism Download PDFInfo
- Publication number
- CN114312831B CN114312831B CN202111546023.9A CN202111546023A CN114312831B CN 114312831 B CN114312831 B CN 114312831B CN 202111546023 A CN202111546023 A CN 202111546023A CN 114312831 B CN114312831 B CN 114312831B
- Authority
- CN
- China
- Prior art keywords
- vehicle
- target vehicle
- time sequence
- track
- dimensional
- 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
Classifications
-
- 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
Landscapes
- Traffic Control Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a vehicle track prediction method based on a spatial attention mechanism, which comprises the following steps: s1, acquiring time sequence characteristics f of a target vehicle to be predicted a The method comprises the steps of carrying out a first treatment on the surface of the S2, acquiring interaction characteristics f of the target vehicle to be predicted and surrounding obstacles int The method comprises the steps of carrying out a first treatment on the surface of the S3, acquiring attention characteristics l of the target vehicle to be predicted and surrounding lane environments att The method comprises the steps of carrying out a first treatment on the surface of the S4, predicting the track of the vehicle according to the characteristics obtained in S1, S2 and S3. According to the method, the historical motion characteristics of the vehicle and the attention characteristics of surrounding environmental characteristics are utilized, meanwhile, the track interaction characteristics of the surrounding vehicles are generated, the distance influence weight factors are set, the motion track and the attention characteristics of the surrounding vehicles are added into the track prediction module, track prediction of the target vehicle is carried out, and the accuracy of a vehicle track prediction algorithm is improved.
Description
Technical Field
The invention relates to the technical field of track prediction, in particular to a vehicle track prediction method based on a spatial attention mechanism.
Background
The automatic driving vehicle runs on the road, and the running track of the surrounding obstacle vehicles at the current moment needs to be considered, so that the future running track of the surrounding vehicles is predicted, and the future running track of the automatic driving vehicle is planned, so that safe running is performed, and collision accidents with the surrounding obstacle vehicles are avoided. The future running track of the surrounding obstacles is accurately predicted, so that the automatic driving vehicle can run more stably, and comfortable and safe riding experience is brought. Traditional methods based on physical models or traditional machine learning algorithms, such as a hidden Markov model, a support vector machine, a dynamic Bayesian network and the like, are earliest in track prediction, but the traditional models need to set a large number of parameters for modeling, and the established models are not strong in applicability. Predicting the motion of surrounding vehicles is a critical capability for autonomous vehicles deployed in complex traffic. Most of the current track prediction methods are performed by adopting a neural network method, and models suitable for most environments can be trained through a large number of data sets. The track of the vehicle is influenced by surrounding vehicles, so that influence factors of the surrounding vehicles on the target vehicle are considered in track prediction, and the accuracy of the neural network on the track prediction of the vehicle can be greatly improved.
The method for predicting the multi-mode vehicle track based on the long-short-time memory network is disclosed in China patent literature, the publication number is CN113076599A, the publication date is 2021-07-06, the track information of each vehicle is coded into a coding sequence through LSTM, the motion state of each vehicle is coded, social pooling is proposed, the LSTM states of all vehicles around the predicted vehicle are pooled into one social tensor, the social tensor applies two convolution layers and one pooling layer to obtain the position relation information of the social pool, in addition, the LSTM states of the predicted vehicles are connected in series through a full-communication layer to obtain vehicle dynamic codes, the two codes are connected in series to form a complete track code, then the LSTM decoder and a mixed density network are used for performing multi-mode prediction on the future track information, and the active safety capability of the unmanned vehicle is improved. However, static scene information such as environmental characteristics of roads, lanes and the like and environmental characteristics of vehicles are not fully considered in the technology, so that the method plays an important role in a track prediction algorithm, the characteristics around the target vehicle are accurately extracted, and the accuracy of the vehicle track prediction algorithm can be improved.
Disclosure of Invention
The invention provides a vehicle track prediction method based on a spatial attention mechanism, which aims to solve the problem that in the prior art, the interactive characteristics of a target vehicle and surrounding obstacle vehicles are emphasized to predict the vehicle track and the surrounding environment characteristics of the target vehicle are not considered, and the vehicle track prediction method based on the spatial attention mechanism utilizes the historical motion characteristics of the vehicle and the surrounding environment characteristic attention characteristics to generate the track interactive characteristics of the surrounding vehicles at the same time, sets a distance influence weight factor, adds the motion track and the attention characteristics of the surrounding vehicles into a track prediction module to predict the track of the target vehicle, and improves the accuracy of a vehicle track prediction algorithm.
In order to achieve the above purpose, the present invention adopts the following technical scheme:
a vehicle trajectory prediction method based on a spatial attention mechanism, comprising:
s1, acquiring time sequence characteristics f of a target vehicle to be predicted a ；
S2, acquiring interaction characteristics f of the target vehicle to be predicted and surrounding obstacles int ；
S3, acquiring attention characteristics l of the environment of the target vehicle and surrounding lanes with prediction att ；
S4, predicting the track of the vehicle according to the characteristics obtained in S1, S2 and S3.
In the invention, a self-attention mechanism mode is adopted, and the target vehicle track and surrounding lane characteristics are combined to form the attention characteristics, so that the network can pay attention to the characteristics of surrounding lanes during learning, and the vehicle track prediction result is more in accordance with the actual driving rule; meanwhile, vehicle interaction characteristics are introduced, weighting factors of different distances are set by utilizing distance information of the target vehicle and surrounding vehicles, and the characteristics of the surrounding vehicles are multiplied by the weighting factors, so that the target vehicle pays attention to the surrounding vehicles in different degrees; track prediction is performed by combining the time sequence characteristics, the attention characteristics and the interaction characteristics of the target vehicle, and accuracy of track prediction is improved.
Preferably, the step S1 includes the following steps:
s11, collecting motion characteristics of a target vehicle at T moments;
s12, coding and mapping the motion characteristics of T moments into high-dimensional characteristics;
s13, extracting time sequence features f of target vehicle through high-dimensional features a 。
The motion characteristics of the target vehicle comprise, but are not limited to, coordinate characteristics of the vehicle, speed characteristics and acceleration characteristics, and in addition, the operation of coding and mapping the motion characteristics to high-dimensional characteristics can adopt convolution operation or full-connection operators, and data is more easily linearly separable in a high-dimensional space than in a low-dimensional space; and then the high-dimensional characteristics are sent to an LSTM long-term memory neural network coding module to extract the time sequence characteristics.
Preferably, the step S2 includes the following steps:
s21, acquiring time sequence characteristics f of each obstacle vehicle around the target vehicle s S is the surrounding obstacle vehicles;
s22, acquiring the distance weight m between each obstacle vehicle and the target vehicle s ；
S23, weighting distance m s Normalization is carried out to obtain weight factors
S24, summing products of time sequence characteristics of each obstacle vehicle and corresponding weight factors to obtain interaction characteristics f int 。
According to the method, the distances from each obstacle vehicle to the target vehicle are different, the distances from each obstacle vehicle to the target vehicle are calculated, the reciprocal is taken as the distance weight of the obstacle vehicle, the distance weight is normalized to obtain a weight factor, and when the obstacle vehicle is closer to the target vehicle, the influence on the target vehicle is larger, and the corresponding weight factor is larger; the interactive features then reflect the different degrees of influence of the obstacle vehicles around the target vehicle.
Preferably, the step S3 includes the following steps:
s31, obtaining central line characteristics of a plurality of lanes around a target vehicle and mapping the central line characteristics into overall high-dimensional characteristics;
s32, combining the overall high-dimensional characteristic obtained in the step S31 with the time sequence characteristic f of the target vehicle a Performing dimension splicing to obtain fusion characteristics f al ；
S33, time sequence characteristic f a Dimension mapping is carried out to obtain a matrix Q, and features f are fused al Mapping two different dimensionsTo matrices K and V;
s34, multiplying Q by the transposed matrix of K and dividing by the dimension d k Is normalized and multiplied by a matrix V to obtain a attention characteristic l att ，d k Representing the characteristic dimension of K.
In the invention, the influence of a plurality of lane environments around the target vehicle on the prediction of the vehicle track is considered, and the static environment characteristics and the time sequence characteristics of the target vehicle are spliced into the fusion characteristics to reflect the relationship between the target vehicle and the surrounding lane environments; meanwhile, the dimension of the matrix Q obtained by dimension mapping of the time sequence features is unchanged or increased relative to the time sequence features, the dimension of the same matrix K and the dimension of the same matrix V are unchanged or increased relative to the fusion features, and the feature dimensions of the three matrices are the same, so that the attention features can be obtained by operation.
Preferably, the timing characteristic f of each obstacle vehicle around the target vehicle s Acquisition method of (a) and acquisition of time sequence characteristic f of target vehicle a The same method as in (a).
According to the invention, the LSTM long-term memory neural network can be used for extracting the time sequence characteristics of both the obstacle vehicle and the target vehicle, and the RNN circulating neural network or the GRU gating circulating unit and other time sequence network models can be used for extracting the time sequence characteristics.
Preferably, in S31, N lane centerlines around the target vehicle are selected, feature extraction is performed on each lane centerline, and then the feature extraction is mapped to high-dimensional features, and the high-dimensional features of the N lane centerlines are subjected to dimensional stitching to obtain overall high-dimensional features. According to the invention, different number of lanes are selected as considered environmental features according to the actual condition of the target vehicle, and the feature of the lane center line is extracted by sampling the center line of each lane at equal intervals.
Preferably, in the step S4, the time sequence characteristic f a And attention feature l att Interaction features f int After splicing, the signal is sent to a decoder to obtain LSTM decoding characteristics, full-connection operation is carried out on the decoding characteristics, and track prediction results of vehicles are output.
The invention has the following beneficial effects: the method comprises the steps of generating track interaction characteristics of surrounding vehicles by utilizing historical motion characteristics of the vehicles and attention characteristics of surrounding environmental characteristics, setting distance influence weight factors, adding motion tracks and attention characteristics of the surrounding vehicles into a track prediction module, predicting the track of a target vehicle, and improving the accuracy of a vehicle track prediction algorithm; the vehicle track is combined with the central line characteristics of surrounding lanes by adopting a mode of an attention mechanism, so that the neural network can pay attention to the characteristics of the surrounding lanes during learning, and the prediction result of the vehicle track is more in accordance with the actual driving rules; and introducing vehicle interaction characteristics, setting different weight factors by utilizing distance information of the target vehicle and surrounding vehicles, and multiplying the characteristics of the surrounding vehicles by the weight factors to enable the target vehicle to pay attention to the surrounding vehicles to different degrees.
Drawings
FIG. 1 is a flow chart of a vehicle trajectory prediction method of the present invention;
fig. 2 is a block diagram of a vehicle trajectory prediction process in an embodiment of the invention.
Detailed Description
The invention is further described with reference to the drawings and detailed description.
As shown in fig. 1, a vehicle track prediction method based on a spatial attention mechanism includes the following steps:
s1, acquiring historical motion characteristics of a target vehicle to be predicted and extracting time sequence characteristics f a ；
S1 comprises the following steps:
s11, collecting motion characteristics of a target vehicle at T different moments;
s12, coding and mapping the T motion features into high-dimensional features;
s13, sending the high-dimensional characteristics into a time sequence network model to extract time sequence characteristics f of the target vehicle a 。
S2, acquiring interaction characteristics f of the target vehicle to be predicted and surrounding obstacles int ；
S2 comprises the following steps:
s21, acquiring time sequence characteristics f of each obstacle vehicle around the target vehicle s S represents surrounding obstacle vehicles; timing characteristics f of each obstacle vehicle around the target vehicle s Acquisition method of (a) and acquisition of time sequence characteristic f of target vehicle a The same method as in (2);
s22, obtaining the distance between each obstacle vehicle and the target vehicle and taking the reciprocal to obtain a distance weight m s ；
S23, weighting distance m s Normalization processing is carried out to obtain weight factors
S24, time sequence characteristic f of each obstacle vehicle s And corresponding weight factorSummation of products of (2) to obtain interaction characteristic f int 。
S3, acquiring attention characteristic matrixes of the target vehicle to be predicted and surrounding lane environments, and calculating to obtain attention characteristics l att ；
S3, the following steps are included:
s31, obtaining central line characteristics of a plurality of lanes around a target vehicle and mapping the central line characteristics into overall high-dimensional characteristics; selecting N lane central lines around the target vehicle as the center, extracting features of each lane central line, mapping the extracted features to high-dimensional features, and performing dimensional splicing on the high-dimensional features of the N lane central lines to obtain overall high-dimensional features of the lane central lines;
s32, combining the overall high-dimensional characteristic of the lane center line with the time sequence characteristic f of the target vehicle a Performing dimension splicing to obtain fusion characteristics f al ；
S33, time sequence feature f a Dimension mapping is carried out to obtain a matrix Q, and fusion characteristics f are obtained al Respectively carrying out two different dimensional mappings to obtain matrixes K and V;
s34, multiplying the matrix Q by the transposed matrix of K and dividingIn dimension d k The square root of the obtained result is normalized to obtain an attention feature matrix, and the attention feature matrix is multiplied by the matrix V to obtain attention feature l att ，d k Representing the characteristic dimension of K.
S4, predicting the track of the vehicle according to the characteristics obtained in the S1, the S2 and the S3; will time sequence feature f a And attention feature l att Interaction features f int After splicing, the signal is sent to a decoder to obtain LSTM decoding characteristics, full-connection operation is carried out on the decoding characteristics, and track prediction results of vehicles are output.
According to the invention, different number of lanes are selected as considered environmental features according to the actual condition of the target vehicle, and the feature of the lane center line is extracted by sampling the center line of each lane at equal intervals.
In the invention, a self-attention mechanism mode is adopted, and the target vehicle track and surrounding lane characteristics are combined to form the attention characteristics, so that the network can pay attention to the characteristics of surrounding lanes during learning, and the vehicle track prediction result is more in accordance with the actual driving rule; meanwhile, vehicle interaction characteristics are introduced, weighting factors of different distances are set by utilizing distance information of the target vehicle and surrounding vehicles, and the characteristics of the surrounding vehicles are multiplied by the weighting factors, so that the target vehicle pays attention to the surrounding vehicles in different degrees; track prediction is performed by combining the time sequence characteristics, the attention characteristics and the interaction characteristics of the target vehicle, so that track prediction accuracy is improved.
The motion characteristics of the target vehicle comprise, but are not limited to, coordinate characteristics of the vehicle, speed characteristics and acceleration characteristics, and in addition, the operation of coding and mapping the motion characteristics to high-dimensional characteristics can adopt convolution operation or full-connection operators, and data is more easily linearly separable in a high-dimensional space than in a low-dimensional space; and then the high-dimensional characteristics are sent to an LSTM long-term memory neural network coding module to extract the time sequence characteristics.
According to the method, the distances from each obstacle vehicle to the target vehicle are different, the distances from each obstacle vehicle to the target vehicle are calculated, the reciprocal is taken as the distance weight of the obstacle vehicle, the distance weight is normalized to obtain a weight factor, and when the obstacle vehicle is closer to the target vehicle, the influence on the target vehicle is larger, and the corresponding weight factor is larger; the interactive features then reflect the different degrees of influence of the obstacle vehicles around the target vehicle.
In the invention, the influence of a plurality of lane environments around the target vehicle on the prediction of the vehicle track is considered, and the static environment characteristics and the time sequence characteristics of the target vehicle are spliced into the fusion characteristics to reflect the relationship between the target vehicle and the surrounding lane environments; meanwhile, the dimension of the matrix Q obtained by dimension mapping of the time sequence features is unchanged or increased relative to the time sequence features, the dimension of the same matrix K and the dimension of the same matrix V are unchanged or increased relative to the fusion features, and the feature dimensions of the three matrices are the same, so that the attention features can be obtained by operation.
According to the invention, the LSTM long-term memory neural network can be used for extracting the time sequence characteristics of both the obstacle vehicle and the target vehicle, and the RNN circulating neural network or the GRU gating circulating unit and other time sequence network models can be used for extracting the time sequence characteristics.
In the embodiment of the present invention, the track prediction method of the vehicle track shown in fig. 2 is adopted to predict the track of the target vehicle to be predicted. The feature extraction needs to be performed on three groups of objects, namely, the target vehicle, the obstacle vehicles around the target vehicle and the lanes around the target vehicle.
Firstly, extracting historical motion characteristics and time sequence characteristics f of a target vehicle to be predicted a Namely, the motion characteristics of the target vehicle at T different moments are extracted, and specific coordinate information (x, y) is included. And carrying out coding mapping on the motion characteristics of the T moments by adopting convolution operation, and mapping the motion characteristics to high-dimensional characteristics. Then, the high-dimensional characteristics of the target vehicle are sent to a long-short-term memory neural network coding module, and the time sequence characteristics f of the target vehicle are extracted a 。
Secondly, extracting interaction characteristics f of the target vehicle to be predicted and surrounding obstacle vehicles int Selecting obstacle vehicles around the target vehicle, and acquiring historical transportation of each obstacle vehicleThe dynamic characteristics are coordinate information at T different moments, and the convolution operation is adopted for coding mapping to obtain the high-dimensional characteristics. The high-dimensional characteristics of each obstacle vehicle are sent to an LSTM coding module, and the time sequence characteristics f of each obstacle vehicle are obtained s Where s represents the surrounding obstacle vehicles, and the total obstacle vehicles have Z vehicles. Calculating the distance between each obstacle vehicle and the target vehicle, and taking the reciprocal of the distance to obtain a distance weight m s . And taking the Euclidean distance between the coordinate point of the current moment of the target vehicle and the coordinate point of the current moment of the surrounding obstacle vehicles as the distance between the coordinate point of the current moment of the target vehicle and the Euclidean distance. The distance weight calculation formula is as follows:
(x t ，y t ) Coordinate point (x) representing current time of target vehicle s ，y s ) Coordinate point representing current moment of surrounding obstacle vehicle, lambda represents super parameter, m s The distance weight is represented, wherein the super parameter lambda is a numerical value set in advance, the selection of specific data is determined by the road where the target vehicle is located, and the selection of different super parameters of the roads where the vehicles such as expressways, urban roads, viaducts and the like run is different.
In addition, the closer the obstacle vehicle is, the greater the influence on the target vehicle is, and the greater the distance weight is. Weighting the distance m of each obstacle vehicle s Normalization operation is carried out to obtain weight factors
Wherein Z represents a total of Z obstacle vehicles around the target vehicle, and the normalized weighting factor is usedMultiplying the corresponding time sequence characteristic f of the obstacle vehicle s Summarizing the characteristics of all obstacle vehicles to obtain an interaction characteristic vector f int 。
The self-attention mechanism is then used to obtain the attention characteristics l of the target vehicle to be predicted and the surrounding lane environment att And acquiring a vehicle coordinate point corresponding to the current moment in a historical track of the target vehicle, taking the coordinate point as a center, selecting N surrounding lane center lines, sampling each lane center line at equal intervals, dividing each lane center line into M points, and after dividing, obtaining N M equally-spaced points. And extracting features of M equidistant points of each lane center line in the N lane center lines, performing convolution operation, and mapping the features of the lane center lines to higher dimensionality. Performing LSTM coding on the high-dimensional features of each lane center line, performing dimensional splicing on the LSTM features of the N lane center lines, and then performing time sequence feature f with the target vehicle a Performing dimension splicing to obtain fusion characteristics f of the target vehicle and the center lines of all lanes al . Time sequence characteristic f of target vehicle a And performing dimension mapping to obtain a matrix Q. Fusion feature f for target vehicle and all lane centerlines al And respectively carrying out two different dimensional mappings to obtain a matrix K and a matrix V. The characteristic dimensions of the matrix Q, the matrix K and the matrix V are the same, and the row numbers of the matrix K and the matrix V are the same. Multiplying the transposed matrix of Q and K and dividing by the dimension d k Is further normalized by softmax to obtain a attention feature matrix, wherein d k Representing the characteristic dimension of the matrix K. Multiplying the attention characteristic matrix with the matrix V to obtain the attention characteristic l of the central line of the target vehicle and the lane att 。
Timing characteristics f of the target vehicle a Attention feature of the center line of the lane, target vehicle att Interaction features f of obstacle vehicles around the target vehicle int Splicing to obtain [ f ] a ；l att ；f int ]And then sending the information to an LSTM encoding module in a decoder to obtain LSTM decoding characteristics, performing full-connection operation on the decoding characteristics, and finally outputting a track prediction result of the target vehicle.
The foregoing embodiments are further illustrative and explanatory of the invention, as is not restrictive of the invention, and any modifications, equivalents, and improvements made within the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (6)
1. A vehicle trajectory prediction method based on a spatial attention mechanism, comprising:
s1, acquiring time sequence characteristics f of a target vehicle to be predicted a ；
S2, acquiring interaction characteristics f of the target vehicle to be predicted and surrounding obstacles int ；
S3, acquiring attention characteristics l of the target vehicle to be predicted and surrounding lane environments att ；
S31, obtaining central line characteristics of a plurality of lanes around a target vehicle and mapping the central line characteristics into overall high-dimensional characteristics;
s32, combining the overall high-dimensional characteristic obtained in the step S31 with the time sequence characteristic f of the target vehicle a Performing dimension splicing to obtain fusion characteristics f al ；
S33, time sequence characteristic f a Dimension mapping is carried out to obtain a matrix Q, and features f are fused al Performing two different dimensional mappings to obtain matrixes K and V;
s34, multiplying Q by the transposed matrix of K and dividing by d k Is normalized and multiplied by a matrix V to obtain a attention characteristic l att ，d k Representing the feature dimension of K;
s4, predicting the track of the vehicle according to the characteristics obtained in S1, S2 and S3.
2. The method for predicting vehicle trajectories based on spatial attention mechanisms as set forth in claim 1, wherein said step S1 includes the steps of:
s11, collecting motion characteristics of a target vehicle at T moments;
s12, coding and mapping the motion characteristics of T moments into high-dimensional characteristics;
s13, extracting time sequence features f of target vehicle through high-dimensional features a 。
3. A vehicle trajectory prediction method based on a spatial attention mechanism according to claim 1 or 2, characterized in that said S2 comprises the steps of:
s21, acquiring time sequence characteristics f of each obstacle vehicle around the target vehicle s S is the surrounding obstacle vehicles;
s22, acquiring the distance weight m between each obstacle vehicle and the target vehicle s ；
S23, weighting distance m s Normalization is carried out to obtain weight factors
S24, summing products of time sequence characteristics of each obstacle vehicle and corresponding weight factors to obtain interaction characteristics f int 。
4. A vehicle trajectory prediction method based on a spatial attention mechanism as claimed in claim 3, characterized in that the timing characteristic f of each obstacle vehicle around the target vehicle s Acquisition method of (a) and acquisition of time sequence characteristic f of target vehicle a The same method as in (a).
5. The vehicle track prediction method based on the spatial attention mechanism according to claim 1, wherein in S31, N lane centerlines around the target vehicle are selected, feature extraction is performed on each lane centerline, and then the feature extraction is mapped to high-dimensional features, and the high-dimensional features of the N lane centerlines are subjected to dimensional stitching to obtain overall high-dimensional features.
6. A method for predicting vehicle trajectories based on spatial attention mechanisms as set forth in claim 1, 2, 4 or 5 wherein in S4, the timing characteristic f is set a And attention feature l att Interaction features f int After splicing, the signal is sent to a decoder to obtain LSTM decoding characteristics, full-connection operation is carried out on the decoding characteristics, and track prediction results of vehicles are output.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111546023.9A CN114312831B (en) | 2021-12-16 | 2021-12-16 | Vehicle track prediction method based on spatial attention mechanism |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111546023.9A CN114312831B (en) | 2021-12-16 | 2021-12-16 | Vehicle track prediction method based on spatial attention mechanism |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114312831A CN114312831A (en) | 2022-04-12 |
| CN114312831B true CN114312831B (en) | 2023-10-03 |
Family
ID=81051879
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111546023.9A Active CN114312831B (en) | 2021-12-16 | 2021-12-16 | Vehicle track prediction method based on spatial attention mechanism |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114312831B (en) |
Citations (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104298236A (en) * | 2014-04-15 | 2015-01-21 | 明日交通公司 | Travel route planning and managing method of intelligent bus system |
| CN109885643A (en) * | 2019-03-04 | 2019-06-14 | 上海图趣信息科技有限公司 | A kind of position prediction algorithm and storage medium based on semantic track |
| CN111091591A (en) * | 2019-12-23 | 2020-05-01 | 百度国际科技（深圳）有限公司 | Collision detection method and device, electronic equipment and storage medium |
| US10671076B1 (en) * | 2017-03-01 | 2020-06-02 | Zoox, Inc. | Trajectory prediction of third-party objects using temporal logic and tree search |
| CN111316286A (en) * | 2019-03-27 | 2020-06-19 | 深圳市大疆创新科技有限公司 | Trajectory prediction method and device, storage medium, driving system and vehicle |
| CN111523643A (en) * | 2020-04-10 | 2020-08-11 | 商汤集团有限公司 | Trajectory prediction method, apparatus, device and storage medium |
| CN111942407A (en) * | 2020-07-31 | 2020-11-17 | 商汤集团有限公司 | Trajectory prediction method, apparatus, device and storage medium |
| CN111982143A (en) * | 2020-08-11 | 2020-11-24 | 北京汽车研究总院有限公司 | Vehicle and vehicle path planning method and device |
| CN112215337A (en) * | 2020-09-30 | 2021-01-12 | 江苏大学 | Vehicle trajectory prediction method based on environment attention neural network model |
| CN112347567A (en) * | 2020-11-27 | 2021-02-09 | 青岛莱吉传动系统科技有限公司 | Vehicle intention and track prediction method |
| CN112805730A (en) * | 2019-12-30 | 2021-05-14 | 华为技术有限公司 | Trajectory prediction method and related equipment |
| CN113076599A (en) * | 2021-04-15 | 2021-07-06 | 河南大学 | Multimode vehicle trajectory prediction method based on long-time and short-time memory network |
| WO2021134172A1 (en) * | 2019-12-30 | 2021-07-08 | 华为技术有限公司 | Trajectory prediction method and related device |
| CN113401143A (en) * | 2021-07-19 | 2021-09-17 | 电子科技大学长三角研究院(衢州) | Individualized self-adaptive trajectory prediction method based on driving style and intention |
| WO2021238303A1 (en) * | 2020-05-29 | 2021-12-02 | 华为技术有限公司 | Motion planning method and apparatus |
Family Cites Families (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR102138979B1 (en) * | 2018-11-29 | 2020-07-29 | 한국과학기술원 | Lane-based Probabilistic Surrounding Vehicle Motion Prediction and its Application for Longitudinal Control |
| CN109583151B (en) * | 2019-02-20 | 2023-07-21 | 阿波罗智能技术(北京)有限公司 | Method and device for predicting running track of vehicle |
| EP3767541A1 (en) * | 2019-07-17 | 2021-01-20 | Robert Bosch GmbH | A machine learnable system with conditional normalizing flow |
| US11161502B2 (en) * | 2019-08-13 | 2021-11-02 | Zoox, Inc. | Cost-based path determination |
| US11989952B2 (en) * | 2019-12-10 | 2024-05-21 | Samsung Electronics Co., Ltd. | Systems and methods for trajectory prediction |
| US11860634B2 (en) * | 2019-12-12 | 2024-01-02 | Baidu Usa Llc | Lane-attention: predicting vehicles' moving trajectories by learning their attention over lanes |
| US11891087B2 (en) * | 2019-12-20 | 2024-02-06 | Uatc, Llc | Systems and methods for generating behavioral predictions in reaction to autonomous vehicle movement |
| US11127142B2 (en) * | 2019-12-31 | 2021-09-21 | Baidu Usa Llc | Vehicle trajectory prediction model with semantic map and LSTM |
-
2021
- 2021-12-16 CN CN202111546023.9A patent/CN114312831B/en active Active
Patent Citations (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104298236A (en) * | 2014-04-15 | 2015-01-21 | 明日交通公司 | Travel route planning and managing method of intelligent bus system |
| US10671076B1 (en) * | 2017-03-01 | 2020-06-02 | Zoox, Inc. | Trajectory prediction of third-party objects using temporal logic and tree search |
| CN109885643A (en) * | 2019-03-04 | 2019-06-14 | 上海图趣信息科技有限公司 | A kind of position prediction algorithm and storage medium based on semantic track |
| CN111316286A (en) * | 2019-03-27 | 2020-06-19 | 深圳市大疆创新科技有限公司 | Trajectory prediction method and device, storage medium, driving system and vehicle |
| CN111091591A (en) * | 2019-12-23 | 2020-05-01 | 百度国际科技（深圳）有限公司 | Collision detection method and device, electronic equipment and storage medium |
| CN112805730A (en) * | 2019-12-30 | 2021-05-14 | 华为技术有限公司 | Trajectory prediction method and related equipment |
| WO2021134172A1 (en) * | 2019-12-30 | 2021-07-08 | 华为技术有限公司 | Trajectory prediction method and related device |
| CN111523643A (en) * | 2020-04-10 | 2020-08-11 | 商汤集团有限公司 | Trajectory prediction method, apparatus, device and storage medium |
| WO2021238303A1 (en) * | 2020-05-29 | 2021-12-02 | 华为技术有限公司 | Motion planning method and apparatus |
| CN111942407A (en) * | 2020-07-31 | 2020-11-17 | 商汤集团有限公司 | Trajectory prediction method, apparatus, device and storage medium |
| CN111982143A (en) * | 2020-08-11 | 2020-11-24 | 北京汽车研究总院有限公司 | Vehicle and vehicle path planning method and device |
| CN112215337A (en) * | 2020-09-30 | 2021-01-12 | 江苏大学 | Vehicle trajectory prediction method based on environment attention neural network model |
| CN112347567A (en) * | 2020-11-27 | 2021-02-09 | 青岛莱吉传动系统科技有限公司 | Vehicle intention and track prediction method |
| CN113076599A (en) * | 2021-04-15 | 2021-07-06 | 河南大学 | Multimode vehicle trajectory prediction method based on long-time and short-time memory network |
| CN113401143A (en) * | 2021-07-19 | 2021-09-17 | 电子科技大学长三角研究院(衢州) | Individualized self-adaptive trajectory prediction method based on driving style and intention |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114312831A (en) | 2022-04-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110610271B (en) | Multi-vehicle track prediction method based on long and short memory network | |
| CN113954864B (en) | Intelligent automobile track prediction system and method integrating peripheral automobile interaction information | |
| CN111931902B (en) | Generating countermeasure network model and vehicle track prediction method using generating countermeasure network model | |
| CN111339867B (en) | Pedestrian trajectory prediction method based on generation of countermeasure network | |
| CN111930110A (en) | Intent track prediction method for generating confrontation network by combining society | |
| CN112734808B (en) | Trajectory prediction method for vulnerable road users in vehicle driving environment | |
| CN112257850A (en) | Vehicle track prediction method based on generation countermeasure network | |
| CN111311009B (en) | Pedestrian track prediction method based on long-term and short-term memory | |
| CN114399743B (en) | Method for generating future track of obstacle | |
| CN112651374B (en) | Future trajectory prediction method based on social information and automatic driving system | |
| CN114898550B (en) | Pedestrian track prediction method and system | |
| CN115049009A (en) | Track prediction method based on semantic fusion representation | |
| CN115293237A (en) | Vehicle track prediction method based on deep learning | |
| CN117141518A (en) | Vehicle track prediction method based on intention perception spatiotemporal attention network | |
| CN114368387B (en) | Attention mechanism-based driver intention recognition and vehicle track prediction method | |
| CN115071762A (en) | Pedestrian trajectory prediction method, model and storage medium oriented to urban scene | |
| CN111369053A (en) | Method for predicting tracks of pedestrians and vehicles in human-vehicle interaction scene | |
| CN116382267B (en) | Robot dynamic obstacle avoidance method based on multi-mode pulse neural network | |
| CN114312831B (en) | Vehicle track prediction method based on spatial attention mechanism | |
| CN111443701A (en) | Unmanned vehicle/robot behavior planning method based on heterogeneous deep learning | |
| CN115700509A (en) | Guideline generation method and system based on simulation feedback data | |
| CN113191539B (en) | High-density composite scene track prediction method based on heterogeneous graph aggregation network | |
| CN114905505B (en) | Navigation control method, system and storage medium of mobile robot | |
| CN117475615A (en) | Vehicle track prediction method based on spatial motion relation and long-short time memory network | |
| CN115797892A (en) | Target track prediction method |
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 |