Tracking Mobile Devices Using Crowdsourcing Technique in IoT Infrastructures for Outdoors and Indoors Positioning

In the Internet of Things (IoT) era, tracking humans’ daily life activities has faced a remarkable transformation, especially in terms of indoor positioning. Similarly, wireless positioning technologies, including Wi-Fi and LoRa, have been employed as an alternative to Global Navigation Satellite System (GNSS) technologies for indoor tracking. Wi-Fi and LoRa positioning frequently employs the received signal strength indicators (RSSI) of the Wi-Fi and LoRa signals. However, the RSSI-based approachs suffer from Multipath, Non-Line-Of-Sight (NLOS), and fluctuating RSSI measurements via Wi-Fi and LoRa chipsets. When these issues have the direct impact on the accuracy and reliability of the positioning techniques. In addition, the fingerprinting procedure is one of the most widely known positioning methods for RSSI- based techniques. Due to the absence of a stable matching algorithm, the fingerprinting-based method has an additional issue.

There are a number of matching algorithms, for example, weighted k- nearest neighbour (WkNN), k-mean clustering, decision tree, and deep learning algorithms such as Long-Short-Term Memory (LSTM). Two algorithms are proposed in this study to provide adequate positioning services.

The first algorithm is a novel integrated matching algorithm for Wi-Fi fingerprint-positioning technique, which is known as Norm_MSATE_LSTM, as a means of mitigating the drawbacks of the RSSI-based fingerprinting method. It is based on the Wi-Fi fingerprinting and proposed augmentation techniques with considering LSTM as a matching technique. To address the problem of a large number of RPs/classes in the LSTM, we first conduct the augmentation process to boost the RSSI data records using the Mean Standerd deviation Augmentation TEchnique (MSATE). The RSSI data are normalised (Norm), and the long short-term memory (LSTM) method is used to estimate the accurate positions. Finally, the recommended matching algorithm is compared with the stand-alone matching algorithms, including weighted k- nearest neighbours (WkNN) and LSTM.

The second algorithm is the hybrid positioning technique using existing Wi-Fi and LoRa technologies, which is known as Wi-Lo, and it is aimed at improving the outcomes of the first suggested algorithm. This one is based on the combination of the Wi-Fi and LoRa technologies and considering MM and trilateration techniques to provide seamless positioning from outdoor to indoor via building identifications. The approach is divided into two phases. The LoRa RSSI is used to identify buildings in the first phase. The second phase is known as Wi-Lo, and it combines LoRa and Wi-Fi technologies to improve Norm_MSATE_LSTM positioning accuracy outcomes.

Experiments and simulated investigations indicate that the proposed matching algorithm, Norm_MSATE_LSTM, may increase positioning accuracy of the LSTM by 45.83% and 72% when only augmentation and augmentation with normalisation are applied, respectively. On the other hand, the proposed Wi-Lo can increase the first algorithm’s accurate by 39.74% in terms of the positioning improvement.