Multi-scale wavelet transform enhanced graph neural network for pedestrian trajectory prediction

Abstract

The pedestrian trajectory prediction forecasts future positions by analyzing historical data and environmental context. With the rapid advancement of artificial intelligence and data processing technologies, this technique has become increasingly significant in areas such as autonomous driving, video surveillance, and intelligent transportation systems. Traditional deep learning methods have primarily focused on time-domain modeling and have made great success. However, they struggle to capture multi-scale features and frequency-domain information in trajectories, making it challenging to effectively handle noise and uncertainty in trajectory data. To address these limitations, this paper proposes a Multi-Scale Wavelet Transform Enhanced Graph Neural Network (MSWTE-GNN) based on wavelet transform and multi-scale learning. The model processes trajectory sequences in the frequency domain using wavelet transform, extracting multi-scale features, and integrates multi-scale graph neural networks with cross-scale fusion to learn interaction information among pedestrians. Experimental results demonstrate that the proposed method significantly improves the accuracy and reliability of pedestrian trajectory prediction.