Vision and ADS-B Integration: A Multimodal Pretraining Framework for Advanced Airport Surface Surveillance

Airport surface monitoring is increasingly essential for ensuring safety and optimizing operations in modern air traffic management. Although intelligent technologies have improved airport surface monitoring systems, key tasks, such as aircraft detection and trajectory prediction still encounter challenges due to the limitations of single-modality data, reducing the accuracy and robustness of existing methods. To address these issues, this article presents Vision–ADS-B Pretraining (VAP), a contrastive- learning-based multimodal pretraining framework that fuses visual data with automatic dependent surveillance-broadcast (ADS-B) information to generate a more comprehensive representation of surface activities. Crucially, VAP aligns these distinct data sources in a shared latent space by learning their underlying correlations through contrastive learning, thereby enhancing fusion performance and downstream tasks. We demonstrate the effectiveness of VAP through aircraft detection and trajectory prediction experiments using real-world data. Experimental results confirm that VAP outperforms existing methods, even with limited datasets, highlighting its potential to advance airport surface surveillance.