Abstract
This paper shows a new approach to estimate parameters of pedestrian behavior model. The approach applies general state space model. In this modelling, parameter estimation is formulated as a sequential Bayesian filtering. Thus we can estimate time varying parameters which are adaptive to sequentially acquired data. Also we can deal with observation errors. Firstly we include parameters into state vectors of a general state space model and formulate an estimation equation. Secondly we define components of that model: state vector, observation vector, system model, observation model and initial distributions. Then we apply this settings to data acquired at a railway station and estimate time varying parameters. We compared estimation results with estimation settings; we discuss effects of initial distributions, variance of random walk parameters and observation accuracies. In addition, we compared estimated parameters with real situations and confirm that parameters are adaptively changed to congestions.
