Abstract: Data streams are more and more commonly generated in a large number of scenarios by audio and video devices, Global Positioning System (GPS), Radio Frequency Identification (RFID) and other types of sensors. In particular, RFID technology has recently gained significant popularity, especially for real-time people and goods tracking, however the noisy, redundant and unreliable nature of RFID streams, coupled with their huge size, can make their exploitation and management difficult. In this paper, we present a realtime system for RFID Probabilistic Data Management (RPDM). The system manages unreliable and noisy raw RFID data and transforms them into reliable meaningful probabilistic data streams by means of a newly proposed method based on a probabilistic Hidden Markov Model (HMM). Moreover, to handle the huge data volume generated by RFID deployments, RPDM proposes and implements a simple on-line summarization mechanism, which is able to provide small space representation for the massive RFID probabilistic data streams while preserving the meaningful information. The results are promptly stored in a probabilistic database, in such a way that a wide range of probabilistic queries can be submitted and answered effectively. The experimental evaluation proves the feasibility of the approach in real-world object tracking scenarios.
Keywords: RFID data streams, Hidden Markov Model, probabilistic data management, data reduction, object tracking
