Database Reference
In-Depth Information
data may be interpreted as the absence of a particular item, rather
than a system failure.
The chapter will discuss the processes involved in the storage, manage-
ment and cleaning of RFID data. The chapter is organized as follows. In
the next section, we will discuss the process of RFID Data Cleaning and
compression. In section 3, we will discuss issues in the data management
and warehousing of RFID data. An important goal of tracking RFID
data is to use it for detecting interesting semantic events in the data,
especially in real-time streaming scenarios. Section 4 discusses methods
for event detection from RFID data streams. Section 5 discusses issues
related to privacy and security of RFID data. Section 6 contains the
conclusions and summary.
2. Raw RFID Data Cleaning and Compression
A variety of middleware architectures are used in order to collect and
process RFID data [11, 22, 33, 37, 80, 68]. RFID data, by its very nature,
is extremely noisy, incomplete and redundant because of cross-reads from
multiple sensor readers. For example, it has been shown [38, 65] that a
large fraction of the readings in RFID streams are essentially dropped.
It has been estimated in [38, 65], that as many as 30% of the sensor
readings are lost (i.e. the tag identifiers do not appear at all), because of
the reader unreliability. Therefore, RFID middleware systems are used
in order to correct for dropped readings. A commonly used method in
many data cleaning systems [33, 80] is to use a temporal smoothing filter,
in which a sliding window over the reader's data stream interpolates for
lost readings from each tag within the time window. This approach
provides each tag more opportunities to be read within the smoothing
window. This reduces the number of distinct tags which are lost, because
they will show up in one or more tag readings, when the window size
is increased. Typically, the window size is fixed, as in [22], and the
smoothing is performed on the basis of the readings which are received
within this fixed window.
It has been observed in [36], that the choice of window size can be a
critical parameter, which leads to different tradeoffs between false posi-
tives and false negatives. Using a window size which is too small will lead
to missed readings (or
false negatives
), because the tag has fewer oppor-
tunities to be scanned by the reader. On the other hand, a larger window
size will cause
false positives
, because it will lead to scanned readings,
even after the tag has moved out of the reader's detection range. The
work in [36] proposes
SMURF (Statistical sMoothing for Unreliable RFid
data)
, which is an adaptive smoothing filter for raw RFID data streams.
