Integrating and Assessing an Indoor Positioning System for Emergency Evacuations

Sammanfattning

This thesis presents the development and performance assessment of an enhanced indoor positioning system for the EvacAware mobile application, designed for emergency evacuation guidance. The sys-tem architecture was refactored from a monolithic design to a modular, provider-based approach usinga factory pattern and unified interfaces. This facilitated the integration of multiple positioning meth-ods: an internal Bluetooth Low Energy (BLE) system, a Global Navigation Satellite System (GNSS)module, and an external BLE-based API (Eli). An ensemble selection system was implemented toorchestrate these methods, aiming to improve overall accuracy and reliability.The research methodology involved a mixed-methods approach, combining experimental designwith correlation analysis. Scenario-based tests (Static Positioning, Cross-Floor Movement, Out-of-Coverage) were conducted in Malmö University’s Niagara building to evaluate accuracy, latency, andreliability. Quantitative data, including position, floor, error distance, and beacon counts, were col-lected and analyzed.Performance assessment revealed significant challenges. GNSS, while accurate outdoors, sufferedfrom signal loss and substantial floor-level inaccuracies indoors. BLE-based methods (internal andAPI) exhibited considerable error margins (median error distances often exceeding 5-10 meters) andoperational unreliability, with long startup times (91-115 seconds for sub-10m accuracy) deemed im-practical for emergency scenarios. The ensemble strategy increased system flexibility but did notconsistently surpass the accuracy of the best single internal BLE method. BLE performance wasinfluenced by beacon density and environmental factors, but a higher beacon count did not alwaysyield proportionally better accuracy, highlighting the impact of building layout and potential radiointerference.Key limitations identified include the system’s critical dependency on WiFi for backend communi-cation, potential inaccuracies in manual ground truth collection (up to 3m), and inherent challengesof BLE technology such as inconsistent advertising rates and RSSI fluctuations. The study partiallyvalidated the hypothesis of correlation between accuracy, latency, and reliability metrics. However,the hypothesis that an ensemble method would significantly enhance positioning accuracy was onlypartially supported, as reliability improved but accuracy did not consistently outperform the bestindividual internal method.The findings underscore that while the integration of multiple positioning technologies was success-ful, the current system, particularly its indoor BLE components, does not meet the stringent accuracyand reliability requirements for mission-critical emergency navigation. Future considerations includeexploring BLE 5.1, BLE fingerprinting, optimized beacon placement, refined ensemble/voting logic,and leveraging WiFi-based positioning (802.11mc/az, fingerprinting) and mmWave radar to addresscurrent limitations and enhance overall system performance. Addressing the WiFi dependency iscrucial for robust deployment.