Jae Sung Choi

Graduation Semester and Year




Document Type


Degree Name

Doctor of Philosophy in Computer Science


Computer Science and Engineering

First Advisor

Ramez Elmasri


In a smart environment, accurate locations of objects are a fundamental and critical issue. To achieve this goal, we present several methods based on passive far-field UHF RFID technologies, which can satisfy accuracy, robustness and reliability, cost efficiency, simplicity, compatibility, and scalability. Our research overcomes several negative characteristics of the use of cost efficient passive UHF RFID. Our research has several important contributions.First, we study the causes of the problems of using passive UHF RFID in localization, with detailed empirical results, and then, present the impacts of the causes on existing localization techniques such as KNN. Second, we present a new model of backscattered signal strength for passive far-field UHF RFID system under tag-to-tag interference. We propose a method to estimate power variations due to tag interference, based on a tag-to-tag distance and angle using a second order under-damped system. We present a novel localization algorithm to estimate target object location using our Tag-to-tag Interference Model (LMTI). According to the empirical results, LMTI improves accuracy be over 200% compared with RSSI based KNN algorithm when objects are empty boxes, and 127% improvement when objects are the print cartridges contained in aluminum foil bags. Third, we present another approach to achieve accurate localization. Localization using Detection of Tag Interference (LDTI) algorithm, which detects the tag interference on a map of reference tags to estimate target location. To avoid selection of spatially non-adjacent reference tags, we also present the most interfered reference group finding algorithm, which considers spatial relations between reference tags. LDTI based smart shelf performs on average 0.0948m estimation error for 9 empty cardboard boxes, and average 0.1831m estimation error for 9 print cartridge containers, which is a 71% accuracy improvement compared to the KNN algorithm. Finally, we present a novel Vision and passive UHF RFID integrated Localization (VRL) system on smart shelf application to improve performance under harsh conditions. VRL performs on average 0.079m estimation error for 10 print cartridge containers, which is a 61% accuracy improvement compared to the LDTI algorithm under low False Negative Reference (FNR) interrogation conditions. Moreover, it shows 555% computation overhead reduction compared a homogeneous vision system. In high FNR conditions, VRL system achieves over 620% increased accuracy compared to LDTI, and 437% reduced computation time compared to a pure vision based localization system.


Computer Sciences | Physical Sciences and Mathematics


Degree granted by The University of Texas at Arlington