Description
Estimating the position of a person or device inside a building, where GNSS is unavailable or unreliable. The problem comes in two flavors that share most of the literature: device-based (a phone or tag reports RSSI/CSI/UWB readings used to recover its own coordinates) and device-free (the environment is monitored and bodies are localized by the perturbation they cause to a wireless link). For the thesis, BLE provides device-based ground-truth trajectories during calibration campaigns; CSI handles device-free continuous monitoring between them.
Why it's hard
- Multipath propagation distorts the distance-RSS relationship in ways specific to floor geometry.
- Per-site fingerprint calibration is expensive and degrades with environmental change.
- Anchor density and placement dominate accuracy more than algorithmic choice in real deployments.
- NLOS conditions (walls, partitions) make ToF/AoA estimates biased rather than just noisy.
- Cross-floor disambiguation in multi-storey buildings is a persistent failure mode.
Common approaches
- Fingerprinting: kNN / SVM / DNN over RSSI or CSI feature vectors.
- Trilateration with ToF (UWB), AoA (WiFi 802.11mc), or RSS path-loss models.
- Particle filters / Kalman filters fusing inertial + RF measurements.
- Multi-modal sensor fusion (BLE + WiFi + IMU + magnetic).
- Deep-learning-based position regressors with site-adapted training.
Source Papers
- guo2020_267f ↗ — survey on fusion-based indoor positioning.
- sadowski2020_f2ba ↗ — memoryless techniques + wireless tech for IoT indoor localization.
- zhang2025_a250 ↗ — online-learning domain adaptation for WiFi device-free localization.
- choi2022_17c2 ↗ — Wi-CaL device-free counting and localization.