Online filtering and uncertainty management techniques for rfid data processing

RFID is one of the emerging technologies for a wide-range of applications, including supply chain and asset management, healthcare and intruder localization. However, the nature of an RFID data stream is noisy, redundant and unreliable, making it unsuitable for direct use in applications. In this paper, we propose specific RFID Online Filtering and Uncertainty Management techniques that operate on unreliable and imprecise data streams in order to transform them into reliable probabilistic data that can be meaningful to the applications. Our proposal makes use of an Hidden Markov Model (HMM) that continuously infers hidden variables (locations, in case of above example) based on sensor readings. The resulting data can be directly stored in a probabilistic database table for further analysis. All the techniques presented in this paper are implemented in a complete framework and succesfully evaluated in real-world object tracking scenarios.