<p>Accurate prediction of pedestrian trajectories in complex urban environments is crucial for autonomous driving systems and smart city applications, yet remains challenging due to the complex interplay of spatial interactions and temporal dependencies. This paper introduces DualStream-TGCN, a novel parallel framework for pedestrian trajectory prediction that synergistically combines Transformer networks and Graph Convolutional Networks (GCNs). While existing models often prioritize either temporal or spatial relationships independently, our approach processes both dimensions simultaneously through a dual-stream architecture. The GCN module constructs a dynamic graph structure to model spatial relationships between pedestrians and surrounding agents, capturing complex interaction patterns. Concurrently, the Transformer network processes historical trajectory sequences to model long-range temporal dependencies. These parallel streams are integrated through a spatio-temporal attention mechanism that dynamically constructs adjacency matrices, enabling the model to adapt to evolving pedestrian relationships over time. This design allows DualStream-TGCN to comprehensively capture both social interactions and temporal movement patterns, resulting in more accurate trajectory predictions. Extensive evaluations on the ETH/UCY benchmarks demonstrate that our approach significantly outperforms state-of-the-art methods, achieving superior accuracy in both Average Displacement Error (ADE) and Final Displacement Error (FDE) metrics across diverse urban environments with varying crowd densities.</p>