Asif Iqbal Baba will defend his thesis:
Cleansing Indoor RFID Tracking Data
The defence takes place on Friday 18th of November 2016, 13:00, in room 0.2.13 at Selma Lagerlöfs Vej 300
Radio Frequency Identification (RFID) is a most flexible auto identification technology that uses radio frequencies to identify and track objects. RFID is taking over the traditional barcode systems and providing the ability to identify individual objects uniquely and without line-of-sight. An RFID system has two main components: readers and tags. RFID is a proximity-based technology in that a reader detects an object with a tag only when the object is within the reader’s detection range.
The commonly used passive RFID tags do not require battery to store information and exchange data with readers. Due to the flexibility and small size of RFID tags, a vast number of applications are using RFID tags for identifying and tracking objects. Such RFID applications generate massive amounts of raw RFID data. However, the generated data is not reliable enough for high-level analysis and processing. The raw RFID data contains errors such as redundant readings, cross readings and missing readings. These errors are mainly caused by the hardware limits, inconsistent radio signals, and the dynamic indoor environments. Therefore, it is paramount to make the raw RFID data reliable, which is aim of this PhD thesis. The PhD thesis is done as a part of projects NILTEK and BagTrack, which aim to develop an IT solution to improve aviation baggage handling quality. The projects introduce the auto-identification tagging for check-in bags at airports using RFID tags. The projects have a number of data management research challenges, one of which is cleansing indoor RFID data.
The PhD thesis provides several novel data cleansing solutions applicable for indoor RFID data. First, the thesis provides a cleansing solution to reduce the redundant and cross readings in raw RFID data. The solution is based on a graph model that captures the information about indoor constraints and the deployed reader characteristics. Second, the graph model is further enhanced with the information such as transition probabilities from reader to reader such that it can be used to handle missing readings in raw RFID data. Third, the thesis presents a learning based solution that is able to reduce cross readings and recover missing readings in the raw RFID data.
The approach introduces a hidden Markov model (HMM) that models the uncertainties inindoor RFID data and a number of state space designs that can be used to learn the parameters from raw RFID
data. After computing the most probable hidden state sequence, the learned model is able to determine the most probable observation sequence and use it as the cleansed data. Fourth, the thesis presents an approach to map raw RFID data to most probable indoor paths. The approach uses regular expressions to model RFID data and possible indoor paths, and constructs an automaton to capture all possible indoor paths in regular expressions. A probabilistic error model is designed to represent the errors between raw RFID data and semantic symbols in indoor paths. The automaton is used to find the most probable indoor paths according to the error model. All the proposed solutions have been validated through experiments using real and synthetic datasets. The experimental results show that our proposals are efficient and effective. The techniques proposed in the PhD thesis can be employed in real systems to improve the aviation baggage handling quality; they also make advancements in the research area of spatio-temporal data
Members of the assessment committee are
Prof. Mario Nascimento, University of Alberta, Canada, Spain
Prof. Yan Huang, University of North Texas, USA
Associate Prof. Bin Yang, Aalborg University.
Associate Professor Hua Lu, Aalborg University and Professor and Torben Bach Pedersen, Aalborg University, are Asif’s supervisors.
Moderator is Associate Professor
All interested parties are welcome. After the defense Department of Computer Science will host a small reception in cluster 3.