Detection of Steady State in Pedestrian Experiments

Abstract

Initial conditions could have strong influences on the dynamics of pedestrian experiments. Thus, a careful differentiation between transient state and steady state is important and necessary for a thorough study. In this contribution a modified CUSUM algorithm is proposed to automatically detect steady state from time series of pedestrian experiments. Major modifications on the statistics include introducing a step function to enhance the sensitivity, adding a boundary to limit the increase, and simplifying the calculation to improve the computational efficiency. Furthermore, the threshold of the detection parameter is calibrated using an autoregressive process. By testing the robustness, the modified CUSUM algorithm is able to reproduce identical steady state with different references. Its application well contributes to accurate analysis and reliable comparison of experimental results.