This study introduces an information fusion (street view images and social media data "Weibo") method that leverages spatial dependencies and interactions to enhance urban traffic analysis. Our approach proposed a Global-Local Graph Neural Network (GL-GCN) to capture both close-range and long-range spatial relationships, leading to a significant improvement in traffic prediction. Achieving accuracies of 57.72% and 60.13% in speed and flow predictions, our method outperforms traditional models by 2.4% and 6.4% respectively, thus validating the efficacy of our method. A spatial interpretability analysis of the model predictions reveals that incorporating spatial interaction priors notably advances urban traffic tasks. Our research confirmed that traditional traffic forecasting methods may be inadequate for assessing congestion levels in high-density urban areas. Our findings not only find that there is abundant traffic environment information that could be inferred from street view imagery, but also underscore the sophisticated and dynamic nature of urban traffic flow and offer a more precise and dependable framework for urban traffic prediction.