A Particle Filter is a sequential Monte Carlo method that estimates the state of a dynamic system by representing the probability distribution over possible states as a weighted set of discrete samples, or "particles," which are iteratively propagated and reweighted based on observed data. In the context of WiFi/CSI-based crowd and pedestrian sensing, particle filters matter because they enable robust, real-time tracking and localization of individuals or groups even under nonlinear dynamics and non-Gaussian noise conditions, making them well-suited for integrating noisy sensor measurements with agent-based motion models. Key variants include the Sequential Importance Resampling (SIR) filter, which addresses particle degeneracy through periodic resampling, and Rao-Blackwellized particle filters, which improve computational efficiency by analytically marginalizing out portions of the state space.
Source Papers
- A Survey on Fusion-Based Indoor Positioning ↗ — A Survey on Fusion-Based Indoor Positioning
- Channel State Information from Pure Communication to Sense and Track Human Motion: A Survey ↗ — Channel State Information from Pure Communication to Sense a
- Crowd flow forecasting via agent-based simulations with sequential latent parameter estimation from aggregate observation ↗ — Crowd flow forecasting via agent-based simulations with sequ