Description
A density field is a spatial map ρ(x) (or ρ(x, t)) giving local pedestrian density per unit area. It is the key state variable of continuum-crowd-model, the output of people-counting-as-regression pipelines, and the regulariser for physics-informed sensing inference. Discrete versions live on a grid; continuous versions are kernel-density estimates or learned heatmaps. The thesis uses density fields as the canonical bridge between BLE-derived ground truth and CSI-derived inference.
When it's used
- Output of crowd-counting CNN/Transformer heads (heatmap regression)
- State variable in continuum / fluid crowd models
- Supervisory signal for
csi-fingerprinting-style spatial inference - Visualisation surface for evaluation against BLE trajectories
Limitations
- Resolution traded against noise — fine grids overfit, coarse grids blur events
- Boundary effects at room edges need explicit handling
- Hard to validate without dense ground truth