Description
Sensor fusion combines observations from multiple modalities or anchors into a single state estimate, typically through a Bayesian filter (kalman-filter, particle-filter) or a learned multi-modal network. In the thesis it is the architectural pattern that lets BLE (sparse but trustworthy), CSI (dense but noisy), and IMU (where present) cooperate within one inference layer.
When it's used
- BLE + CSI + IMU joint indoor positioning
- Multi-anchor crowd-density estimation
- Combining radar with WiFi sensing for HAR
Limitations
- Modality-weight choice is sensitive to noise model accuracy
- Synchronisation across modalities is a deployment headache
- Late vs early fusion design space is large