Description
Modeling and estimating the macroscopic motion of people through space as a continuous flux field, in close analogy to fluid dynamics. The state variables are density and velocity per location-time; the observables are typically aggregated trajectory counts or per-zone occupancy time series. Pedestrian flow estimation is the bridge between sparse sensor measurements and a continuum description compatible with the thesis's BLE-calibrated CSI proposal.
Why it's hard
- The continuum assumption breaks below ~0.5 ped/m² — most indoor settings sit in this regime most of the time.
- Boundary conditions (doors, stairs, signage) dominate flow patterns and are hard to model from data.
- Counter-flows and intersecting streams are notoriously difficult: simple gradient-descent route models fail.
- Recovering velocity (not just density) from coarse sensor readings is under-constrained without additional priors.
- Conservation-of-mass enforcement during inference often fights data-driven fits in non-stationary regimes.
Common approaches
- Hughes' continuum model and its successors (second-order PDEs).
- Lattice Boltzmann pedestrian flow.
- Mass-conserving neural surrogates with continuity-equation regularization.
- Fundamental diagram parameter fitting from trajectory or density-time data.
- Mesoscopic kinetic models from kinetic-theory analogies.
Source Papers
- hughes2002_57b4 ↗ — continuum theory for the flow of pedestrians.
- huang2009_292f ↗ — efficient solution algorithm for Hughes' dynamic continuum model.
- maity2024_4dd4 ↗ — high-resolution meshfree particle method for second-order pedestrian flow.
- helbing2005_94a7 ↗ — self-organized pedestrian flow phenomena.