MBSIN: Multi-Branch Spatiotemporal Interaction Network for Pedestrian Trajectory Prediction

Abstract

Pedestrian trajectory prediction plays a crucial role in intelligent transportation and autonomous driving applications. This task remains challenging due to the complex spatiotemporal correlations involved. Existing methods typically extract spatial and temporal information independently, overlooking their coupling relationship. To address this, we propose a novel multi-branch spatiotemporal interaction network (MBSIN) that learns the spatial, temporal, and spatiotemporal features through three separate branches. Each branch employs the self-attention mechanism to dynamically generate a graph structure and applies the global sparsification strategy to eliminate redundant connections. Additionally, the feature enhancement network with information sharing is integrated to enhance the feature representation ability. Furthermore, we introduce gated network to dynamically fuse the features from different branches, and a Temporal Convolution Aggregation Network (TACN) is utilized to further improve trajectory prediction accuracy. Extensive experiments on the ETH, UCY, SDD, and NBA datasets demonstrate that MBSIN consistently surpasses existing state-of-the-art methods. Note to Practitioners—This study presents MBSIN as a valuable tool for practitioners in the field of intelligent surveillance and monitoring scenarios. MBSIN effectively models pedestrian interactions using a multi-branch spatiotemporal approach. It integrates an information-sharing feature enhancement network to strengthen feature representation, a gated fusion network to dynamically combine spatial and temporal information, and a Temporal Convolution Aggregation Network (TACN) to refine trajectory predictions. These components work together to improve predictive accuracy. Extensive experiments on the ETH, UCY, SDD, and NBA datasets demonstrate that MBSIN significantly outperforms existing baseline models, making it a reliable solution for real-world applications.