Abstract:This work introduces the multidimensional Graph Fourier Transformation Neural Network (GFTNN) for long-term trajectory predictions on highways. Similar to Graph Neural Networks (GNNs), the GFTNN is a novel network architecture that operates on graph structures. While several GNNs lack discriminative power due to suboptimal aggregation schemes, the proposed model aggregates scenario properties through a powerful operation: the multidimensional Graph Fourier Transformation (GFT). The spatio-temporal vehicle interaction graph of a scenario is converted into a spectral scenario representation using the GFT. This beneficial representation is input to the prediction framework composed of a neural network and a descriptive decoder. Even though the proposed GFTNN does not include any recurrent element, it outperforms state-of-the-art models in the task of highway trajectory prediction. For experiments and evaluation, the publicly available datasets highD and NGSIM are used

What problem does this paper attempt to address?

The problem that this paper attempts to solve is **long - term prediction of vehicle trajectories on highways**. Specifically, the author aims to improve the prediction accuracy of vehicle trajectories in traffic scenes by introducing a new neural network architecture - the multi - dimensional Graph Fourier Transform Neural Network (GFTNN). ### Problem Background In an autonomous driving system, predicting the movement intentions of surrounding road users is crucial for ensuring safe operation. However, vehicle trajectory prediction is a challenging task because it highly depends on the cooperative context of the scene. The cooperative context describes the spatio - temporal interaction relationships among traffic participants, and these interactions influence the behavior of each individual. Existing methods usually attempt to model or statistically learn these interaction relationships, but most deep - learning - based methods use Recurrent Neural Networks (RNNs) or other non - graph - structured models, which have limited performance in dealing with complex spatio - temporal dependencies. ### Main Contributions of the Paper 1. **Introduction of GFTNN**: This paper proposes a new neural network architecture - the multi - dimensional Graph Fourier Transform Neural Network (GFTNN) - for long - term prediction of vehicle trajectories on highways. 2. **Performance Evaluation**: Performance evaluations were carried out on publicly available highD and NGSIM datasets. 3. **Explanation of Prediction Performance**: Analyzed the performance of the proposed GFTNN in prediction. ### Working Principle of GFTNN The core idea of GFTNN is to transform the spatio - temporal dependencies in traffic scenes into frequency - domain representations through multi - dimensional Graph Fourier Transform (GFT), thereby extracting more expressive features. The specific steps are as follows: 1. **Graph Definition**: Construct a two - dimensional graph structure, where nodes represent traffic participants and edges represent the spatio - temporal dependencies between them. 2. **Multi - dimensional Graph Fourier Transform**: Apply multi - dimensional GFT to convert the graph signal to the frequency domain, preserving the characteristics of the spatio - temporal dimensions. 3. **Feature Generation**: Extract spectral features from the frequency - domain representation and input them into a Multi - Layer Perceptron (MLP) for processing. 4. **Encoder and Decoder**: Generate the final trajectory prediction through a lightweight encoder and a descriptive decoder. ### Mathematical Formulas - The Laplacian matrix of the graph \( L = D - W \), where \( D \) is the degree matrix and \( W \) is the weight matrix. - Multi - dimensional Graph Fourier Transform formula: \[ \hat{f}(\lambda^{(1)}_{l_1}, \lambda^{(2)}_{l_2})=\sum_{i_1 = 0}^{N_1-1}\sum_{i_2 = 0}^{N_2-1}f(i_1, i_2)u^{(1)}_{l_1}(i_1)u^{(2)}_{l_2}(i_2) \] where \( l_1=0,\ldots,N_1 - 1 \) and \( l_2 = 0,\ldots,N_2 - 1 \). - Feature selection information in the encoder: \[ h_s=s\odot w_s \] \[ h^k_{\text{norm}}=\frac{h^k_s - E[h^k_s]}{\sqrt{\text{VAR}[h^k_s]}+\epsilon} \] \[ h^k_l = W^k_l\Phi(W^k_n h^k_{\text{norm}}+b^k_n)+b^k_l \] ### Experimental Results The paper conducted experiments on the highD and NGSIM datasets. The results show that GFTNN outperforms the existing state - of - the - art models in the vehicle trajectory prediction task. Especially when dealing with highway scenes, GFTNN can better capture spatio - temporal dependencies, thus improving the prediction accuracy. Through the above methods, GFTNN successfully overcomes the limitations of existing methods in dealing with complex traffic scenes and provides a new solution for trajectory prediction in autonomous driving systems.

A Multidimensional Graph Fourier Transformation Neural Network for Vehicle Trajectory Prediction

Graph and Recurrent Neural Network-based Vehicle Trajectory Prediction For Highway Driving

Graph-Based Spatial-Temporal Convolutional Network for Vehicle Trajectory Prediction in Autonomous Driving

Graph-Based Interaction-Aware Multimodal 2D Vehicle Trajectory Prediction using Diffusion Graph Convolutional Networks

STS-DGNN: Vehicle Trajectory Prediction via Dynamic Graph Neural Network With Spatial–Temporal Synchronization

FourierGNN: Rethinking Multivariate Time Series Forecasting from a Pure Graph Perspective

Interaction-Aware Personalized Vehicle Trajectory Prediction Using Temporal Graph Neural Networks

Trajectory Forecasting with Neural Networks: an Empirical Evaluation and A New Hybrid Model

Representation learning and Graph Convolutional Networks for short-term vehicle trajectory prediction

Vehicle Interactive Dynamic Graph Neural Network Based Trajectory Prediction for Internet of Vehicles

DGInet: Dynamic Graph and Interaction-Aware Convolutional Network for Vehicle Trajectory Prediction.

Trajectory Prediction with Attention-Based Spatial–Temporal Graph Convolutional Networks for Autonomous Driving

VIF-GNN: A Novel Agent Trajectory Prediction Model Based on Virtual Interaction Force and GNN.

An LSTM Network for Highway Trajectory Prediction

GNN-RM: A Trajectory Completion Algorithm Based on Graph Neural Networks and Regeneration Modules

Connecting the Dots: Multivariate Time Series Forecasting with Graph Neural Networks

Trajectory prediction method based on a graph model for autonomous driving

MST-GNN: graph neural network with multi-granularity in space and time for traffic prediction

A novel hybrid time-varying graph neural network for traffic flow forecasting

Group Vehicle Trajectory Prediction With Global Spatio-Temporal Graph

Graph Neural Networks for Modelling Traffic Participant Interaction