<abstract xmlns="http://www.w3.org/1999/xhtml">

<p>Vehicle trajectory prediction is one of the key technologies to realize autonomous driving, which provides an important guarantee for the safety of vehicles in the process of autonomous driving. In this paper, with this as the starting point, a graph convolutional neural network is introduced through a graph attention mechanism to obtain scene features by modeling the temporal Transformer model of surrounding information. Based on the temporal convolutional model to obtain scene features, new feature vectors are calculated by aggregating the weights for the features of nodes and neighboring nodes. Then the input feature dimensions are transformed into the weight matrix of the output feature dimensions, and the output feature vector corresponding to the attention coefficients is calculated by using weighted summation. Then the effect of multiple training of the model is evaluated by taking the mean value and defining its structural relationship. The experimental results show that the prediction error of the proposed method is significantly smaller than that of the comparison method in scenarios with speeds less than or equal to 5m/s and greater than 5m/s. The prediction error based on target detection is reduced by 58.95%, indicating that the proposed method is more consistent with the operation scenarios of autonomous driving.</p>
</abstract>

Vehicle trajectory prediction is one of the key technologies to realize autonomous driving, which provides an important guarantee for the safety of vehicles in the process of autonomous driving. In this paper, with this as the starting point, a graph convolutional neural network is introduced through a graph attention mechanism to obtain scene features by modeling the temporal Transformer model of surrounding information. Based on the temporal convolutional model to obtain scene features, new feature vectors are calculated by aggregating the weights for the features of nodes and neighboring nodes. Then the input feature dimensions are transformed into the weight matrix of the output feature dimensions, and the output feature vector corresponding to the attention coefficients is calculated by using weighted summation. Then the effect of multiple training of the model is evaluated by taking the mean value and defining its structural relationship. The experimental results show that the prediction error of the proposed method is significantly smaller than that of the comparison method in scenarios with speeds less than or equal to 5m/s and greater than 5m/s. The prediction error based on target detection is reduced by 58.95%, indicating that the proposed method is more consistent with the operation scenarios of autonomous driving.

A trajectory prediction method based on graph attention mechanism

Department of Electromechanical and Information Engineering

Chongqing University of Posts and Telecommunications

Applied Mathematics and Nonlinear Sciences

This work is licensed under the Creative Commons Attribution 4.0 International License.

{"article-title":"A trajectory prediction method based on graph attention mechanism"}

Vehicle trajectory prediction is one of the key technologies to realize autonomous driving, which provides an important guarantee for the safety of vehicles in...

A trajectory prediction method based on graph attention mechanism

Online veröffentlicht: 30. Juni 2023

Seitenbereich: -

Eingereicht: 15. Sept. 2022

Akzeptiert: 09. Jan. 2023

DOI: https://doi.org/10.2478/amns.2023.1.00481

Schlüsselwörter
Trajectory prediction, Graph attention mechanism, Temporal Transformer model, Convolutional neural network, Feature vector

© 2023 Hejun Zhou et al., published by Sciendo

This work is licensed under the Creative Commons Attribution 4.0 International License.

A trajectory prediction method based on graph attention mechanism

Online veröffentlicht: 30. Juni 2023

Seitenbereich: -

Eingereicht: 15. Sept. 2022

Akzeptiert: 09. Jan. 2023

DOI: https://doi.org/10.2478/amns.2023.1.00481

SchlüsselwörterTrajectory prediction, Graph attention mechanism, Temporal Transformer model, Convolutional neural network, Feature vector

© 2023 Hejun Zhou et al., published by Sciendo

This work is licensed under the Creative Commons Attribution 4.0 International License.

Schlüsselwörter
Trajectory prediction, Graph attention mechanism, Temporal Transformer model, Convolutional neural network, Feature vector