Human centric sensing by Android phones - WOLoc
Tan, Nicholas Yan Ming
Date of Issue2016
With the increasing distribution of WiFi deployment in urban areas, outdoor localization without the aid of GPS is made possible by relying on the WiFi framework of mobile devices and existing network infrastructures. Despite the presence of existing outdoor localization solutions, it provides an unsatisfactory accuracy. Furthermore, there has been much research on indoor localization but the outdoor aspect has been largely overlooked. To address these issues, this paper proposes WOLoc (WiFi-only Outdoor Localization) as a solution which returns meter-level accuracy achieved by comprehensively processing the WiFi hotspot labels gathered by crowdsensing. Comparing against existing solutions, WOLoc avoids fingerprinting metropolitan areas with the labels due to the complexity of networks outdoor. WOLoc also does not use over-simplified data synthesis methods (e.g., centroid) which omits crucial information in the labels. Alternatively, using a semi-supervised manifold learning technique, labeled and unlabeled data is processed. The output of the unlabeled part will contain the estimated locations for both users and WiFi hotspots. Through conducting extensive experiments with WOLoc in several outdoor zones with varying density of known access points, the results offer higher accuracy over other contemporary methods.
Final Year Project (FYP)
Nanyang Technological University