Analysis and modeling of detours in pedestrian operational navigation
Qiancheng Xu,Zhilu Yuan,Renzhong Guo,Biao He,Mohcine Chraibi
DOI: https://doi.org/10.1016/j.trc.2024.104584
IF: 9.022
2024-04-01
Transportation Research Part C Emerging Technologies
Abstract:Traditional operational navigation models for pedestrian dynamics demonstrate limitations in reproducing the circle antipode experiment, an artificially designed multi-directional flow scenario. In the experiment, pedestrians take detours to avoid the congestion caused by others taking straight paths. Although the pedestrian detour action is commonly incorporated into the route choice model, it frequently goes unaddressed in the operational navigation model, resulting in a disparity between simulation outcomes and empirical observations. To reveal the mechanism underlying pedestrian detours in the circle antipode experiment, this study employed the K-means clustering method rather than using a threshold approach to categorize the experimental participants into groups taking direct or detour routes. Following this, a heuristic function is formulated to determine the desired direction of agents in a collision-free velocity model, reflecting the trade-off between shorter routes and faster speeds. The parameter of the proposed function is calibrated using the proportion of agents choosing detours, where the route types of agents are identified by a classifier based on the random forest. Compared to two traditional models that do not consider detours, the proposed model can more realistically reproduce the trajectory distributions, the travel time, the route length, and the time series of relevant variables in the circle antipode experiment. The study offers insights into employing machine learning methodologies for analyzing pedestrian flow, validating pedestrian dynamics models, and providing an accurate simulation tool for designing transportation facilities and crowd management at large events.
transportation science & technology