A Method for Regional Crowd Flow Prediction Based on Crowd Number Estimation Network

This paper presents a novel method for predicting crowd flow in various scenarios, such as workplaces, corporate campuses, and train stations, using a multi-task attention-based network, CNE-CBAM. The method first constructs a topological map of the scene using an adjacency matrix to model connections between different regions. The crowd number estimation network predicts the number of people in each area, leveraging attention mechanisms to enhance accuracy. To estimate crowd flow between regions, a linear programming model is applied, optimizing the flow of people based on movement data. Experiments conducted demonstrate the high accuracy of the model. This method offers valuable insights for optimizing space planning, safety measures, and operational efficiency in crowded public areas.