In recent years, the development of advanced smart campus systems, accurately predicting and managing pedestrian volumes in public areas has become a crucial task for enhancing campus safety and information security. This critical task forms the foundation for essential decisions in areas like campus safety, event coordination, and resource allocation and more. University administrations have set higher expectations for accurately forecasting and controlling pedestrian volume in these public areas. However, the distinct spatiotemporal patterns and inherent intricacies of pedestrian volume on campuses, combined with the frequent collection of data as sparse trajectories, make traditional forecasting techniques struggle with both precision and computational efficiency in these areas. To address these challenges, this study introduces a two‐stage algorithmic framework. Furthermore, this research underscores the potential of accurate pedestrian volume forecasts in bolstering campus information security measures. By enabling more strategic deployment of security resources and facilitating informed decision‐making, our model contributes to the creation of safer campus environments. In the first step, we use the geospatial encoding algorithm “Geohash” to transform the sparse trajectory data from the campus into pedestrian volume information for public areas. Subsequently, we introduce the GPVP‐transformer (generalized pedestrian volume prediction transformer), a modified algorithm derived from the transformer’s encoder–decoder structure. In parallel, we compare our approach with traditional statistical methods, machine learning algorithms, and state‐of‐the‐art (SOTA) techniques in time series forecasting as our baseline comparisons. The findings demonstrate the robustness of our model in all evaluation results.
ZHUANG et al. (Thu,) studied this question.