On Localization and Tracking for Wireless Capsule Endoscopy

Abstract

Capsule endoscopy is a new and painless technique which has revolutionizedthe gastrointestinal imaging. However, one of the main shortcomingsof the new method is its inability to tag a precise location to each image. Inthis thesis we address various challenges associated with localizing a capsuleendoscope. First, we derive a pathloss model for the wireless channelinside the human body, this model is based on simulating the electromagneticwave propagation in human tissues. Based on the pathloss model wecalculate the mathematical lower bound on the localization accuracy usingthe Received Signal Strength (RSS) method. We show that RSS based methodscan potentially reach 1 cm in localization accuracy. Then we derivethe same bound for the case that magnetic sensors are utilized to locate thecapsule. For this type of sensors the localization accuracy bound is shownto be in the order of 2-3 millimeter. In the second part of the thesis, weshift to a Bayesian paradigm, which assumes prior information on the capsulemovements. We model the movements of the capsule and we invokea multi model tracking scheme for localization. Through simulations, weshow that a significant gain in the performance is achieved by using a correctmodel for capsule movements. It is also shown that accurately detectingthe maneuvers in capsule movements, is crucial in order to reduce theoverall mean square error. Therefore, we investigate further the detectionproblem associated with the multi model tracking and we show that utilizingthe available information on the maneuver and the state of the Kalmanfilter it is possible to improve the tracking accuracy.