Database Reference
In-Depth Information
Figure 10.6.
The (general) architecture of a STDB or a MOD. Both systems require
an algorithm to turn the raw location estimates from the moving objects into a usable
trajectory. In the case of the MOD, this conversion happens in real-time as new
locations are made available (filtering) and in the the case of STDB, the complete
sequence of locations and time stamps are available (smoothing). The quality of the
inferred locations directly affects query accuracy, thus the tracking algorithm is a
vital component of the data management system.
filtering task, then we review the Kalman filter model (KFM) [39, 88]
in detail. Lastly, we discuss some methods which address the track-
ing problem when objects are constrained to move on a road network.
Table 10.1
contains a list of notation used throughout this section.
3.1 The Tracking Problem
The general task of tracking can be formulated as a Bayesian filtering
problem where we would like to estimate the value of an unobserved ran-
dom variable
x
, given an observation
z
. Because
x
defines the state of
a mobile object (position, velocity, altitude, etc.), this value will change
over time and we would like to re-estimate it each time a new observa-
tion becomes available. The general problem of Bayesian filtering is then
to update our beliefs about
x
t
, incorporating all available information
(i.e.
z
1:
t
) by computing the posterior distribution
p
(
x
t
|
z
1:
t
). To keep our
presentation clear, we will assume that the state of a mobile object,
x
,
is described by a vector containing the object's current position and ve-
