Collaborative Techniques for Indoor Positioning Systems

The demand for Indoor Positioning Systems (IPSs) developed specifically for mobile and wearable devices is continuously growing as a consequence of the expansion of the global market of Location-based Services (LBS), increasing adoption of mobile LBS applications, and ubiquity of mobile/wearable devices in our daily life. Nevertheless, the design of mobile/wearable devices-based IPSs requires to fulfill additional design requirements, namely low power consumption, reuse of devices’ built-in technologies, and inexpensive and straightforward implementation. Within the available indoor positioning technologies, embedded in mobile/wearable devices, IEEE 802.11 Wireless LAN (Wi-Fi) and Bluetooth Low Energy (BLE) in combination with lateration and fingerprinting have received extensive attention from research communities to meet the requirements. Although these technologies are straightforward to implement in positioning approaches based on Received Signal Strength Indicator (RSSI), the positioning accuracy decreases mainly due to propagation signal fluctuations in Line-of-sight (LOS) and Non-line-of-sight (NLOS), and the heterogeneity of the devices’ hardware. Therefore, providing a solution to achieve the target accuracy within the given constraints remains an open issue. The motivation behind this doctoral thesis is to address the limitations of traditional IPSs for human positioning based on RSSI, which suffer from low accuracy due to signal fluctuations and hardware heterogeneity, and deployment cost constraints, considering the advantages provided by the ubiquity of mobile devices and collaborative and machine learning-based techniques. Therefore, the research undertaken in this doctoral thesis focuses on developing and evaluating mobile device-based collaborative indoor techniques, using Multilayer Perceptron (MLP) Artificial Neural Networks (ANNs), for human positioning to enhance the position accuracy of traditional indoor positioning systems based on RSSI (i.e., lateration and fingerprinting) in real-world conditions.

The methodology followed during the research consists of four phases. In the first phase, a comprehensive systematic review of Collaborative Indoor Positioning Systems (CIPSs) was conducted to identify the key design aspects and evaluations used in/for CIPSs and the main concerns, limitations, and gaps reported in the literature. In the second phase, extensive experimental data collections using mobile devices and considering collaborative scenarios were performed. The data collected was used to create a mobile device-based BLE database for testing ranging collaborative indoor positioning approaches, and BLE and Wi-Fi radio maps to estimate devices’ position in the non-collaborative phase. Moreover, a detailed description of the methodology used for collecting and processing data and creating the database, as well as its structure, was provided to guarantee the reproducibility, use, and expansion of the database. In the third phase, the traditional methods to estimate distance (i.e., based on Logarithmic Distance Path Loss (LDPL) and fuzzy logic) and position (i.e., RSSI-lateration and fingerprinting–9-Nearest Neighbors (9-NN)) were described and evaluated in order to present their limitations and challenges. Also, two novel approaches to improve distance and positioning accuracy were proposed. In the last phase, our two proposed variants of collaborative indoor positioning system using MLP ANNs were developed to enhance the accuracy of the traditional indoor positioning approaches (BLE–RSSI lateration-based and fingerprinting) and evaluated them under real-world conditions to demonstrate their feasibility and benefits, and to present their limitations and future research avenues.

The findings obtained in each of the aforementioned research phases correspond to the main contributions of this doctoral thesis. Specifically, the results of evaluating our CIPSs demonstrated that the first proposed variant of mobile device-based CIPS outperforms the positioning accuracy of the traditional lateration-based IPSs. Considering the distances among collaborating devices, our CIPS significantly outperforms the lateration baseline in short distances (≤ 4m), medium distances (>4m and ≤ 8m), and large distances (> 8m) with a maximum error reduction of 49.15 %, 19.24 %, and 21.48 % for the “median” metric, respectively. Regarding the second variant, the results demonstrated that for short distances between collaborating devices, our collaborative approach outperforms the traditional IPSs based on BLE–fingerprinting and Wi-Fi–fingerprinting with a maximum error reduction of 23.41% and 19.49% for the “75th percentile” and “90th percentile” metric, respectively. For medium distances, our proposed approach outperforms the traditional IPSs based on BLE–fingerprinting in the first 60% and after the 90% of cases in the Empirical Cumulative Distribution Function (ECDF) and only partially (20% of cases in the ECDF) the traditional IPSs based on Wi-Fi–fingerprinting. For larger distances, the performance of our proposed approach is worse than the traditional IPSs based on fingerprinting.

Overall, the results demonstrate the usefulness and usability of our CIPSs to improve the positioning accuracy of traditional IPSs, namely IPSs based on BLE– lateration, BLE–fingerprinting, and Wi-Fi–fingerprinting under specific conditions. Mainly, conditions where the collaborative devices have short and medium distances between them. Moreover, the integration of MLP ANNs model in CIPSs allows us to use our approach under different scenarios and technologies, showing its level of generalizability, usefulness, and feasibility.