Description
Crowd flow estimation produces a vector field v(x, t) describing the bulk motion of the crowd: speed and direction averaged over local neighbourhoods. It complements density-field (a scalar) and together with it satisfies a continuity equation, which is the conservation law the thesis uses as a physics-informed inductive bias. Flow estimates can come from optical flow on cameras, from particle tracking on BLE, or from Doppler-bin extraction on CSI.
When it's used
- Velocity field for
continuum-crowd-modelcalibration - BLE-derived ground-truth flows for CSI training labels
- Anomaly detection (flow reversal, congestion onset)
- Indoor flow-routing dashboards
Limitations
- Bidirectional flow on overlapping paths is hard to disentangle
- CSI-derived flow only resolves radial velocity components, not tangential
- Requires temporal averaging that smooths fast events
Source Papers
- duives2013_3924 ↗ — flow estimation in crowd-modelling review
- maury2018_d24a ↗ — flow estimation in continuum/microscopic models
- jebrane2026_3e34 ↗ — flow estimation in agent-based crowd simulation
- makinoshima2022_7e21 ↗ — flow estimation under data assimilation
- bendalibraham2021_476e ↗ — flow / density / velocity coupling