Empirical analysis on external factors affecting pedestrian dynamics in high-density situations
Résumé
Predicting pedestrian dynamics in crowded environments is a complex task as pedestrian speed is influenced by multiple external factors. In this study, we aim to examine external factors that affect pedestrian walking speed in dense situations. To this purpose, we propose a set of prospective variables, including Mean Distance (MD), Environmental Effect (EE), Frontal Effect (FE), and Time-to-Collision (TTC), and evaluated their impact on pedestrian speed using a machine learning approach on the highdensity Jülich dataset [1]. The data for each variable, combined with fundamental inputs (FI) such as current speed, relative distance, and relative velocity of neighbors at time t, is fed into a neural network for training and predicting pedestrian walking speed at time t + ∆t. Furthermore, a sensitivityanalysis is conducted to determine the relative importance of these variables and identify the key factors that have greatest impact on pedestrian speed. Finally, the results are compared to those in low-density datasets [2, 3] to evaluate the difference between low-density and high-density circumstances. Our findings on the Mean Absolute Error (MAE) results of the prediction of pedestrian speed (as seen in Fig. 1) demonstrate that the Frontal Effect and Mean Distance are effective indicators as they exhibit a noticeable improvement in accuracy in both unidirectional and bidirectional scenarios in the Jülich datasets, whereas better results could not be observed from other factors. These insights can be utilized to improve the accuracy of pedestrian dynamics predictions in high-density situations by incorporating these factors as additional features in the model.
Origine | Fichiers produits par l'(les) auteur(s) |
---|