Movement Mining - Computational Movement Analysis - page 48

first focus on patterns of individuals (i.e., changes in position and other movement
characteristics of an entity over time, individual movement behaviors , IMB) that may
then be found in groups of entities to form dynamic collective behaviors (DCB).
Finally, Wood and Galton contribute a much needed ontological analysis of col-
lective motion, aiming for the development and formalization of a comprehensive
classification of collectives and their motions (Wood and Galton 2009a , b ).
Segmentation and filtering . Many techniques for segmenting trajectories emerge
from transdisciplinary collaborations between GIScience, databases, and especially
computational geometry. Some approaches focus on the shape of a trajectory, search-
ing for characteristic points where the geometric structure of the trajectory changes
substantially (Yoon and Shahabi 2008 ). Other approaches rather focus on derived
movement parameters such as speed, heading, or sinuosity, and search for subtra-
jectories expressing uniform movement parameters (Buchin et al. 2010b , 2011b ).
Yet other work tries to understand the semantics of moves segmenting trajectories
according to travel mode changes or moves between points of interest (Sester et al.
2012 ). Most of that work is rather methods-driven, its applications are hence mostly
illustrative. In some CMA application areas, however, trajectory segmentation is
closely related to applied research questions. For example in movement ecology,
trajectory segmentation is sometimes referred to as classification such that the task
is to semantically annotate segments of a trajectory with the most likely behavior
of the observed animal. For instance, Shamoun-Baranes et al. ( 2012 ) use trajectory
data in conjunction with additional sensory data from an accelerometer aiming at
segmenting and labeling fixes according to a set of predefined behavior classes (here
fly, forage, body care, stand, and sit) using supervised classification trees.
In the database community segmentation concepts have been suggested for struc-
turing raw streams of position data into semantically meaningful units, aiming at
supporting a meaningful interpretation of trajectories. In their conceptual view on
trajectories Spaccapietra et al. ( 2008 ) base the semantic enrichment of raw move-
ment data on an initial segmentation of trajectories into stops and moves. Subsequent
semantic annotation could then label stops and moves with, for example, the type
of activity (e.g., commute) or the type of the visited location (e.g., home vs. work).
Baglioni and Fernandes de Macedo ( 2009 ) employ formal ontologies to enable such
semantic enrichment, aiming at augmenting both the semantics of the trajectory data
and also patterns mined from the trajectories. The authors argue that most pattern
mining approaches produce patterns, one could say in a mechanistic way, that are
then difficult to link to the actual movement behavior, i.e., the semantics (Baglioni
et al. 2009 , p. 272). By contrast, their semantic enrichment bases the interpretation
of mined patterns in a domain ontology representing the geographical knowledge of
the relevant application domain.
Similarity and clustering . There is ample relatedwork on the similarity of trajectories.
Some of these focus on the spatial or temporal characteristics of trajectories, or
specifically aimat their spatio-temporal nature. Fromwithin computational geometry
emerged a family of similarity measures based on the Fréchet distance between
two curves (Buchin et al. 2010a ). When a person on one curve walks a dog on