Abstract:
This thesis presents a novel approach to locate victims in a post-disaster situation. Locating victims in disaster-affected regions is an important task, particularly because efficient search and localization can save human lives and resources. With an increase in natural and man-made disasters, localization in disaster-affected environment is becoming increasingly challenging. From a high level point of view, such localization closely resembles to indoor localization. Indoor localization in general is a well-studied research problem. This domain has flourished from simple signal fingerprinting approaches to complex channel state estimation, fusion of fingerprinting and inertial sensor systems, etc. However, indoor localization in disaster-affected environments is a much more challenging research problem. It often introduces unknown and previously unseen dynamics. For example, fire hazards lead to smoke and high temperature in a disaster-affected area, which is hostile to a working environment. Besides, structural collapses often lead to non-deterministic conditions where previously deployed sensor systems may fail and produce erroneous results.
To this extend, in this study, we present a new ad hoc WiFi based victim localization technique in post-disaster settings demanding no pre-installed WiFi infrastructures. Our proposed technique namely Victim Localization (VLoc) incorporates two separate stages of search and rescue mission operated by first-responders or rescuers. In the first stage, rescuers actively probe WiFi Received Signal Strength (RSS) from victims' phones, which is now ubiquitous in most parts of the world, to identify potential search location. This is done by tracking RSSI from victims' phones and observing changes in the RSSI levels by first-responders while moving across the disaster-affected sites. In the second stage, rescuers localize the victims through exploiting trilateration estimation techniques. VLoc leverages on an ad hoc setup that facilitates quick deployment and portability. We emulated several disaster-affected scenarios by setting up ten testbeds in four categories namely normal office environment, disaster-like scenarios with undamaged infrastructures, disaster-like scenario with fire, and disaster-like scenarios with damaged or collapsed infrastructure. Our experimental results reveal that we can estimate victims where localization errors range between 0.62m to 3.97m.
