Database Reference
In-Depth Information
This technique determines the most effective window size automatically,
and continuously changes it over the course of the RFID stream, depend-
ing upon the underlying readings. One characteristic of this approach is
that it does not expose the smoothing window parameter to the partic-
ular application at hand. This makes the approach much more flexible
in different scenarios.
The approach proposed in [36] views RFID readings as
unequal prob-
ability random sample
of tags in the physical world. Therefore, the
tradeoff between reader unreliability and tag dynamics can be explored
in a principled and statistical manner. Furthermore, the approach can
be used to clean both “single-tag” and “multiple-tag” readings. In the
multiple tag case, it is assumed that single readings do not need to be
tracked. For example, a store may only need to track when the number
of items of a particular type falls below a given threshold. For the single
tag case, binomial sampling methods are used for the cleaning process.
For the multi-tag case, the aggregate signal over a tag population is
cleaned with the use of Horvitz-Thompson estimators.
One characteristic of effective cleaning methods is to use
declarative
methods
in the cleaning process [38, 39, 36]. The broad idea is to specify
cleaning stages with the use of high-level declarative queries over rela-
tional data streams. Once this is done, the system can translate the
queries into the required low level operations. Such an approach is use-
ful in helping programmers avoid writing low level interaction code, by
specifying the queries at the high level. Furthermore, such an approach
makes the system data- and device-independent, and the code does not
need to be changed if the underlying device fails, or is upgraded.
We note that the middleware approach to RFID data cleaning per-
forms all the processing on the data upfront, before applying any of the
data querying or analytical methods on it. However, different appli-
cations may define the anamolies or corrections on the same data set
in a different way. Therefore, the method in [59] introduces a deferred
approach for detecting and correcting RFID anamolies. Each applica-
tion uses declarative sequence-based rules in order specify, detect, and
correct relevant anamolies. We note that this approach is generally dif-
ferent from the methods proposed in [22, 36], which make the cleansing
process application-independent. Clearly, both approaches have their
own advantages in different scenarios. The generally accepted principle
[22] is that the separation of the middleware from the applications is
a desirable goal, because of the diversity of the applications in which
such data could be used, and the network limitations of the underlying
readers.
