How to determine people's location via Wi-Fi: BIM and semantics improve accuracy

The project aimed to collect room occupancy and movement statistics without expensive infrastructure like UWB. The team developed an Android app for collecting RSSI measurements from access points and a desktop app for building coverage maps and calculating coordinates with RMSE. Field experiments were conducted in two types of premises: a combined office with rooms and a corridor where each room had its own access point, and an openspace with central access points and meeting rooms along the wall. Two positioning approaches were implemented: classical kNN and a compact neural network with one hidden layer using Keras. Additional improvements included inertial sensors of the phone, two calibration methods for radio wave propagation (brute-force and ray tracing), and the key feature: using BIM data (floor plans, wall materials) and semantic probabilities of a person's location based on schedule. Two semantic approaches were tested: positioning with probability weights and per-room polygons whose sizes depend on maximum calibration error. Results showed that simple kNN methods work but benefit significantly from calibration; adding BIM and semantic probabilities reduces mispositioning into adjacent rooms at boundary RSSI values; combinations of approaches give better stability than any single method. Accuracy strongly depends on Wi-Fi density and layout: rooms with one AP each performed better than openspace with several central APs. The proof-of-concept confirmed that for occupancy analytics (statistics, heatmaps), Wi-Fi plus BIM is usually sufficient, while ultra-precise localization (0.5 m accuracy) still requires specialized equipment. Potential business applications include quick evaluation of premises usage without heavy investment, optimizing layouts and schedules, and improving energy efficiency. Privacy considerations are addressed by collecting anonymized aggregated statistics rather than identifying individuals, requiring employee consent and compliance with local data protection laws. Future work includes scaling the PoC to larger floors, integrating with booking systems and HVAC automation, adding BLE and UWB for high-density zones, and automating calibration with adaptive learning.