Cost-Efficient RSSI-Based Indoor Proximity Positioning, for Large/Complex Museum Exhibition Spaces

Area of Science:

• Indoor localization
• Wireless sensor networks
• Machine learning applications

Background:

• Received Signal Strength Indication (RSSI) offers simple, cost-effective indoor positioning but struggles with accuracy due to non-line-of-sight (NLOS) conditions, noise, and multipath fading.
• Bluetooth Low Energy (BLE) technology presents a scalable, energy-efficient, and market-available solution for indoor localization, particularly suitable for large, complex environments like museums with high visitor traffic.

Purpose of the Study:

• To evaluate the accuracy and feasibility of a BLE/RSSI-based indoor localization system in a real-world museum setting.
• To develop and test a prototype tool (VTT) integrating various data processing and prediction methods for visitor cell-level positioning.
• To compare the performance of traditional signal processing techniques against machine learning algorithms for improving localization accuracy.

Main Methods:

• A pilot study was conducted at the Museum of Modern Greek Culture using wearable BLE beacons for visitor tracking across 47 halls.
• The VTT prototype incorporated Kalman filters for RSSI smoothing, hybrid positioning, temporal/spatial filtering, and machine learning classifiers.
• Visitor movement was modeled, with the "ant" behavioral model selected for experimentation, and 15 methods/algorithms were evaluated across 20 RSSI datasets.

Main Results:

• Simple data smoothing and management methods achieved an average prediction accuracy of 81.53% across diverse datasets.
• Machine learning algorithms, specifically Random Forest, demonstrated superior performance, reaching an average prediction accuracy of 87.24%.
• The implemented infrastructure achieved a cost-efficiency of 8 Euro per square meter.

Conclusions:

• BLE/RSSI-based localization is a viable and cost-effective solution for indoor positioning in large public spaces like museums.
• Machine learning algorithms significantly enhance localization accuracy compared to traditional signal processing methods.
• The VTT prototype provides a scalable and adaptable architecture for improving indoor localization in complex environments.