| Description |
Location of an object or person in in-door environments is a vital piece of in-formation. Traditionally, global positioning system-based devices do an excellent job in providing location information but are limited in in-door environments due to lack of an unobstructed line of sight. Wireless environments, with their extreme sensitivity to the positioning of objects inside them, provide excellent opportunities for obtaining location information of subjects. Received signal strength (RSS) based localization methods attract special attention as they can be readily implemented with "off-the-shelf" hardware and software. Device-free localization (DFL) presents a new and promising dimension in RSS-based localization research by providing a non-intrusive method of localization. However, existing RSS-based localization schemes assume a fixed or known transmit power. Any unexpected change in transmit power, not known to the receivers in the wireless network, can introduce errors in location estimate. Previous work has shown that meticulously planned power attacks can result in expected errors, in location of a transmitting sensor, in excess of 18 meters for an area of 75 X 50 m2. We find that the localization error in DFL can increase by four-fold when under power attack of 15 dB amplitude by multiple adversaries. Certain nonadversarial circumstances can also lead to unexpected changes in transmit power which would result in increased localization error. In this thesis, we focus on detection and isolation of wireless sensor nodes in a network which vary their transmit power to cause unexpected changes in RSS measurements and lead to increased localization errors in DFL. In the detection methods presented in this thesis, we do not require a training phase and hence, our methods are robust for use in dynamic environments where the training data may get obsolete frequently. We present our work with special focus on DFL methods using wireless sensor networks. However, the methods developed are generic and can be easily extended to active localization methods using both wireless sensor networks (WSN) and IEEE 802.11 protocols. To evaluate the effectiveness of our detection method, we perform extensive experiments in indoor settings using a network of 802.15.4 (Zigbee) compliant wireless sensor nodes and present evaluation results in the form of average detection rate, ROC curves, probability of missed detection and false alarm. |
