Description
A particle filter approximates the posterior of a nonlinear / non-Gaussian state-space model by a weighted set of samples ("particles"), advancing them with a process model and re-weighting them with each observation. It is the standard online estimator for indoor tracking and the dominant data-assimilation engine for fitting agent-based-model parameters to noisy real-world crowd observations.
When it's used
- Indoor tracking on noisy RSSI / CSI / IMU streams
- Data assimilation for agent-based crowd models
- BLE-CSI fusion under nonlinear measurement equations
- Sensor-fusion pipelines with non-Gaussian noise
Limitations
- Particle degeneracy without resampling
- Computational cost grows with state dimension
- Process / measurement models still need careful design
Source Papers
- malleson2020_7b38 ↗ — particle filter for ABM data assimilation
- ghorbani2023_c065 ↗ — particle filter for crowd-state inference
- makinoshima2022_7e21 ↗ — particle filter for evacuation modelling
- guo2020_267f ↗ — particle filter in indoor positioning
- chen2018_97e0 ↗ — particle-filter occupancy estimator