Figure 3. The map matching system
Figure 4. Flowchart of the map matching algorithm used
(Adapted from Zhao (1997))
Many different approaches may be used in the actual
computations of Figure 4. Within the first step, the filtered
position and variance-covariance information is a by-product of
the Kalman filter computations (Figure 2). Once the filtered
position is known, it is possible to extract all the possible track
segment scenarios within a given range. This is achieved by using
a grid cell system. The map database file contains both grid and
vector based referencing. The track itself consists of a series of
vectors, linking three dimensional map nodes, which also have
grid cell attributes. The grid attributes are calculated by
overlaying a grid on the track and computing a grid cell reference
for each point using:
^Grid
E - E
^ Position Origin
Grid Interval
(1)
J Grid
N -N
Position Origin
Grid Interval
(2)
Given that each node (Figure 3) within the database has a known
grid reference, and the computed location of the train is known, it
is possible to compute a grid reference for the filtered position. It
is then possible to extract all track nodes within a certain number
of grid cells of the filtered position. Once the relevant track
segments are extracted, the Kalman gain, (Equation 3), and
the three dimensional perpendicular offset, or predicted residual,
F _ T v k (Equation 4), to each parallel track is computed
(Figure 3). The predicted residual is computed by assuming a
straight line between any two nodes (Figure 3).
fQ A t + 7’Qjc Jt * t ] ( 3 )
F-T y k = F X k~T X k ( 4 )
where,
fQ,.,. rQr x are the covariance matrix of the filtered
r x k x k . I x k x k
state and track respectively,
A k ,A k are the design matrix, and the transpose of the
design matrix respectively, and
F x k , T \ k are the filtered and track coordinate matrices
respectively.
Once, the gain and the perpendicular offset are calculated, the
map matched position, MM x k , is computed using:
X k F X k
(5)
Having constrained the filtered position to each track, three
statistical tests are performed in an attempt to determine on which
track the train is travelling. The first two tests developed by
Teunissen and Salzmann (1989) analyse the predicted residuals to
see if they match their expected stochastic properties. These tests
are known as the Local Overall Model (LOM) and Global Overall
Model (GOM). The local test only analyses one epoch of data at a
time and highlights solutions containing noise or large biases,
1-5-4