Advancing pedestrian trajectory prediction with interaction-aware 3D-dual contextualized modeling

In the contemporary landscape, numerous models have emerged to predict pedestrian trajectories. However, a significant limitation of many existing models is their exclusive reliance on historical motion data, which may lead to undesirable outcomes such as pedestrians intersecting with roadside features. Furthermore, current research predominantly relies on spatial assumptions, making it challenging to adjust the graph arrangement for un-specified environments in online systems, and there is a notable absence of an evaluation methodology to assess the impact of relational modeling on prediction execution. This study addresses these limitations by developing a trajectory prediction model incorporating environmental factors affecting pedestrians. The proposed 3D-dual contextualized model (DCM) utilizes adaptive relational aggregation to capture the intricate relationships between pedestrians and their contextual data. Moreover, integrating a Graph Convolutional Network (GCN) with the Pedestrian Visual Acuity Module (PVAM) aims to replicate pedestrians’ perception of their surroundings, eliminating extraneous data and reducing computational complexity. Supplementary environmental data was introduced to enrich the information set. Evaluation of the dataset demonstrates that the proposed model, incorporating dual-contextualized information such as background and vision information, outperforms the prediction accuracy of cutting-edge baseline models. Experimental results demonstrate that the 3D-DCM outperforms state-of-the-art models, achieving significant improvements in prediction accuracy, particularly in scenarios with dynamic crowd behavior and environmental influences. This work contributes to the advancement of trajectory prediction by providing a robust framework that incorporates both environmental and visual data, setting the stage for more accurate and scalable applications in intelligent transportation systems and autonomous driving.