Abstract
The performance of outdoor positioning has become excellent with the emergence of Global Positioning System (GPS), but GPS is not reliable indoors. The ability of a system to perform indoor positioning without GPS is still challenging and has gained a lot of attention in recent years.
Indoor positioning has become a focus of research during the past decade. Despite a lot of research efforts, existing indoor positioning systems based on different technologies are still limited because most of them either require expensive infrastructure (ultrasound), offer limited coverage (Wi-Fi, Bluetooth) or provide low accuracy (audible sound). On the other hand, machine vision offers the potential for a cheap and effective solution for robust indoor positioning.
This thesis describes the research, experiments and analysis conducted to develop a machine vision based system, known as "Indoor Positioning System (iPoS)", which can provide reliable positioning in indoor environments. iPoS is based on a client server model where the client is a smartphone application and the server uses the proposed "BoWLocator" algorithm to match the incoming query image from the application. The key approach to the system is the use of minimum information i.e. a single image of a location from the phone camera for localisation. The main purpose of iPoS is to use it as a navigation aid for blind people and guide them while they move in unfamiliar indoor environments because they often feel lost in the absence of current location information.
To create a reliable indoor positioning, iPoS uses three proposed components (1) voting module, (2) homography verification method, and (3) post-verification method. iPoS has been demonstrated to localise on four realistic datasets covering a total of 50 indoor self-similar locations with a correct acceptance rate of 72-93% with few wrong matches depending on the test set and queries typically require 5-14 seconds on average to return a result. iPoS gives a very low localisation error with an average wrong match rate of 5.5%.