Information Technology Reference
In-Depth Information
3.2.1 Segmentation and Filtering
Most tracking systems adopting the Lagrangian perspective of movement (such as
GPS or VHF receivers) produce streams of location fixes, irrespective of the current
movement behavior of the tracked object. Here, segmentation is a useful approach
for structuring large volumes of raw movement data. Similar to image segmenta-
tion in image processing (Shapiro and Stockman 2001 ), segmentation then refers
to the process of partitioning movement data into multiple segments, with the goal
of simplifying or changing the representation of the trajectory into something that
is more meaningful or easier to analyze. Buchin et al. ( 2011b ) specifically define
segmentation as partitioning a trajectory into a (typically small) number of pieces,
where the obtained segments have uniform characteristics.
In the movement mining process segmentation and filtering can take the func-
tion of preprocessing steps, aiming at reducing noise and condensing the signal
for a given analytical task (Fig. 3.1 ). First the trajectory is segmented into coherent
segments, then only those segments relevant to the analysis task are selected for
subsequent processing. In the most basic case this involves separating stops from
moves. Many movement data sets contain large periods when the object is not mov-
ing. For instance, the GPS trackers producing the data in Laube and Purves ( P13 .
2011 ) used a movement model based on a Kalman filter, where even when the object
is immobile, a smooth curve is fitted to the location fixes resulting in trajectory coils
(see Fig. 3.4 ). Since such “pseudo movement” interfered with the multi-scale study
i
i
i
(1)
(2)
(3)
y
x
Fig. 3.4 Trajectory segmentation and filtering. ( 1 ) Application of an interval operator for computing
the average Euclidean distance to other fixes inside a temporal window i .( 2 ) Removal of all points
where average distance is less than a given threshold, i.e., filtering of static points, ( 3 ) filtering of
subtrajectories with less than a threshold temporal length. Adapted from Laube and Purves ( P13 .
2011 ) (Republished from Laube, P. and Purves, R., How fast is a cow? Cross-scale Analysis of
Movement Data, Transactions in GIS , 15(3), pp. 401-418, 2011, John Wiley & Sons Ltd, DOI:
10.1111/j.1467-9671.2011.01256.x)
 
 
Search WWH ::




Custom Search