Database Reference
In-Depth Information
ture more frequently. Unless the index is specially designed for such a
query workload, frequent updates can be very costly and even outweigh
any benefit the index structure provides for query processing.
Tracking is critical to managing both the spatial and temporal un-
certainty in an object's position. Accurate tracking is challenging, es-
pecially at the database scale (i.e. tracking thousands to hundreds of
thousands of objects), due to the computational constraints. Inferences
and predictions about an object's position must be made quickly, and
should use all of the data that has been observed thus far.
The diculties in mining for patterns in spatiotemporal data are sim-
ilar to those mentioned for querying. Core mining problems, such as
that of identifying groups or clusters, is made significantly more dicult
when the data change positions over time. New definitions and objec-
tives must be defined which take into account not only the current data
configuration, but also the past (or predicted) configurations.
Additionally, the problem of data uncertainty is inherent in all areas
of managing spatiotemporal data. Despite technological improvements,
the ability to localize mobile objects is still only available up to a degree
of error. Due to the nature of dealing with inexact data, new approaches
to indexing, querying, and mining are necessary to effectively account
for ambiguities in the data [32, 15, 82, 31]. Although data uncertainty
spawns from a variety of sources, it can be broadly categorized as one
of two types:
spatial
and
temporal
uncertainty.
Spatial Uncertainty
is uncertainty in the location of an object at
the instance an observation is made. That is, spatial uncertainty
describes the limitations of a sensor to provide an accurate reading
of an object's position. For example, high quality GPS sensors
typically provide measurement accuracy in the range of 1
−
10
meters, lower quality hardware is in the range of 10
−
50 meters, and
localization from cellular tower triangulation may resolve position
to only within 100
−
2
,
000 meters [6, 79].
Temporal Uncertainty
is the uncertainty in an object's position
since the previously received update. Temporal uncertainty arises
due to the update schedule of how frequently an object will send
information about its position to a database. Since objects may
move continuously but only report their positions intermittently,
there is a time-lag in which the database contains stale information.
In several datasets, GPS traces have shown incredible variance in
the frequency with which measurements are provided. The tem-
poral resolution ranges from very high (1 second intervals) to very
low (
>
2 min. intervals) [97].
