Pedestrian intention prediction aims to infer future crossing decisions and actions by modeling temporal dynamics, social interactions, and environmental context. However, it remains challenging due to the inherent uncertainty of human behavior and the complexity of contextual factors. Existing approaches typically rely on historical trajectories or pedestrian body features, while neglecting the intricate interdependencies that characterize human-environment and human-human interactions. To bridge these limitations, we propose a Adaptive Spatiotemporal Interaction-aware (ASINet) framework inspired by Conditional Random Field (CRF) for pedestrian intention prediction. Specifically, we extract individual pedestrian spatiotemporal features and encode the global environmental context to capture pedestrian-pedestrian and pedestrian-environment interactions. These interactions are represented as node and edge potentials within a CRF. Finally, we reformulate the optimization process as an energy minimization problem, achieving accurate pedestrian prediction while ensuring the consistency of individual interactions with the global environment as well as with other pedestrians. Extensive experiments on public datasets validate the effectiveness of ASINet and show its superior performance over other baselines.