Modern Intelligent Traffic Systems (ITS) rely on diverse data sources such as vehicle mobility, wireless communication signals, and environmental configurations to support real-time decisions. However, in practice, this data is often incomplete, unsynchronized, or noisy, which poses challenges for accurate traffic analysis. This paper presents a cross-modal alignment framework that brings together these heterogeneous data sources into a shared representation space without requiring labeled or perfectly synchronized inputs. The method leverages mutual information maximization using a contrastive InfoNCE loss to align different modalities. Evaluation on the TiHanX_V2X dataset shows that the proposed approach achieves retrieval accuracies of 91.7% between mobility and communication modalities, 75.0% between mobility and scenario data, and 58.3% between communication and scenario inputs. Alternate training strategies such as regression loss and intra-modal contrastive learning, both implemented for comparative analysis in this study, showed limited effectiveness in achieving cross-modal alignment, with retrieval accuracies not exceeding 3.12%. The performance of a random selection strategy, used as a lower bound baseline, was measured at 1.56%, aligning with expectations for untrained models. Visualizations further confirm that the learned latent representations are semantically consistent across modalities. The empirical evidence highlights that mutual information maximization offers a scalable and effective approach for aligning asynchronous and incomplete modalities in Intelligent Traffic Systems, substantially outperforming baseline and alternative methods under realistic constraints.