Database Reference
In-Depth Information
Information about the constraints in the underlying application,
such as the maximum capacity of a room or shelf, can be used in
order to improve accuracy.
The work in [15] proposes a Bayesian inference framework, which takes
full advantage of the duplicate readings, and the additional background
information in order to maximize the accuracy of RFID data collection.
A different method, proposed in [52], is to use a
KerneL dEnsity-bAsed
Probability cleaning method (KLEAP)
to remove cross-reads within a
sliding window. The method estimates the density of each tag using
a kernel-based function. The idea is that the most relevant reader will
have a much larger number of objects in a similar position as this object.
Therefore, the reader corresponding to the micro-cluster with the largest
density will be regarded as the relevant position of the tagged object in
the current window. The reads which are derived from the other readers
will be treated as cross-reads.
3. RFID Data Management and Warehousing
Some of the earliest work on temporal management of RFID data
was proposed in [68]. This work develops a
Dynamic Relationship ER
(DRER) Model
for temporal management of RFID data. This system
is built on top of the ER model with relatively few extensions. The
technique maintains the history of events and state changes, so that
complex queries can be supported. A rules-based framework is used to
transform business logic data into user configured rules. In addition
to location, another concept which is introduced is that of
containment
.
Containment implies a hierarchical relationship between a set of objects.
For example, a pallet may be loaded with cases, and both the pallet and
the cases would have their own separate EPCs.
The RFID data contains two basic categories of data, corresponding
to
static
and
dynamic
data. The static data is related to commercial en-
tities such as location information, product level information, and serial
information. There are two kinds of dynamic data: (a) The first corre-
sponds to
instance data
such as serial number and the date of manufac-
ture; and (b) The second corresponds to
temporal data
such as location
observations and temporal changes in the containment of objects.
The second kind of temporal data are captured through EPC tag
readings, and is related to the movement of products. The four primary
kinds of entities which interact with one another in such a system are
EPC-tagged objects
,
readers
,
locations
,and
transactions
. These entities
interact with one another, as object locations change, and entity con-
tainment relationships change as well. We note that even sensor (reader)
