Description
Continuous observation of an indoor or semi-indoor venue to maintain situational awareness over a crowd: how many people are in each zone, how density evolves, where bottlenecks are forming, when intervention is required. Crowd monitoring is the operational counterpart of crowd-counting and occupancy-estimation — the same sensor stack, but a sustained dashboard / alerting use case rather than a static estimate. It is the application context that motivates the BLE-calibrated CSI thesis core.
Why it's hard
- Sensor coverage gaps create dead zones in dashboards even when individual sensors are accurate.
- Drift across hours/days requires online recalibration; static models degrade silently.
- Privacy and data-protection regulations restrict identifier persistence and trajectory storage.
- Cross-venue deployment usually means a new training data collection campaign per site.
- Operationalizing alerts has a high false-positive cost — operators stop trusting the system after a few bad warnings.
Common approaches
- Multi-modal fusion (CSI + BLE + camera + CO2) for robust per-zone estimates.
- Adaptive beacon deployment optimization to maximize coverage with fixed hardware budget.
- Privacy-preserving WiFi-fingerprint pipelines that hash identifiers at the sensor.
- Dashboard / SCADA-style visualization for venue operators.
Source Papers
- davies1995_b3cd ↗ — crowd monitoring using image processing (foundational).
- zhen2022_bb0b ↗ — adaptive beacon deployment for indoor crowd monitoring.
- darsena2023_50b7 ↗ — sensing technologies for crowd management in public transport.
- bendalibraham2021_476e ↗ — recent trends in crowd analysis (review).
- rusca2024_ccca ↗ — privacy-preserving WiFi-fingerprint counting for management.