Analysis of sparse dataset for multi-city indoor localization

Abstract

The increasing use of smart technologies in the 21st century is changing the way we live, communicate and work. Outdoor solutions such as GPS (Global Positioning System) are almost 99% accurate for locating an object or person, but are not as effective underground, in a parking lot, inside a shopping center or similar indoor locations. In the era of automation, location-based services (LBS) have become more and more important for indoor navigation of a person or object. A wide range of context-aware applications, a growing number of location-based technologies, and the challenge of achieving greater indoor localization accuracy have greatly increased the volume and scope of research in this area using new fields such as ML, artificial intelligence, and so on and so forth. Indoor localization technologies such as RSSI, fingerprinting, Dead Reckoning, channel state information, scene analysis, and indoor localization technologies such as Wi-Fi, Bluetooth (BLE), ultra-wideband (UWB), and ultrasound are being studied by the researchers to discover the holy grail of indoor localization. High accuracy at lowest cost is a key requirement of LBS. This thesis would focus on indoor localization where complete “Sparse Indoor Loc” data was collected from a multi-city experimental region with 4852 samples, 49 classes and 187 APs. Proposed “Sparse Indoor Positioning” algorithm will be explored to predict the class label with 99% accuracy for selected threshold values of APs in this thesis.