3-dimension pedestrian route choice models with cross entropy methods using multi sensor observations

by Binchang Shen, Takumi Suga, Eiji Hato



This paper constructs a three-dimensional (3D) route choice model based on an adjusted RL model and observation model. In order to reproduce the pedestrian trajectory in the 3D space with precision, we designed an observation model using machine learning algorithms and proposed a Data Fusion (DF) based location strategy by integrating classifiers trained from multi-sensor data set to address with signal attenuation issue caused by the environment and access point (AP) distribution. In addition, the Recursive Logit (RL) model is utilized to build a route choice model on a choice-stage network (CSN). Since the output of the observation model is a link set in the form of probability, we introduce cross-entropy instead of the likelihood as the objective function for parameter estimation. A case study at Shibuya Station shows that our model is practical in complex 3D spaces. Pedestrians route preferences reflected in the model can provide a reference for future research on pedestrian behavior modeling and network optimization.