Quantitative comparison of estimations for the density within pedestrian streams

In this work, the precision of estimators for the density within unidirectional pedestrian streams is evaluated. The analysis is done in controllable systems where the density is homogeneous and all the characteristics are known. The objectives are to estimate the global density with local measurements or density profile at high spatial resolution with no bias and low fluctuations. The classical estimation using discrete numbers of observed pedestrians is compared to continuous estimators using spacing distance, Voronoi diagram, Gaussian kernel as well as maximum likelihood. Mean squared error and bias of the estimators are calculated from empirical data and Monte Carlo experiments. The results show quantitatively how continuous approaches improve the precision of the estimations