Database Reference
In-Depth Information
Unusual vessel trajectory could be a sign of smuggling; outlying taking-
off/landing patterns could be a dangerous signal for aviation; and detection of
suspicious human movements could help prevent crimes and terrorism.
Spatiotemporal interactions of human may tell the semantic relationships among
them such as colleague, family or friend relationships. Different from cyber so-
cial network such as Facebook friends, spatiotemporal relationships reveal more
complicated physical social network.
This topic chapter discusses the state-of-art data mining methods to discover un-
derlying patterns in movements. Various patterns, characteristics, anomalies, and
actionable knowledge can be mined from massive moving object data. We will focus
on following three categories of movement patterns:
Individual periodic pattern.
One most basic pattern in moving objects is the
periodicity. Human repeat daily or weekly movement patterns. Animals have
seasonal migration patterns. We will discuss how to automatically
detect the
periods
in a trajectory and how to
mine frequent periodic patterns
after periods
are detected. We will also describe the methods of using periodic patterns for
future movement prediction
.
Pairwise movement pattern.
Focusing on two moving objects only, we will discuss
different trajectory
similarity measures
and the methods to mine generic, behav-
ioral and semantic patterns.
Generic patterns
include the attraction or avoidance
relationships between two moving objects. In
behavioral patterns
, we will mainly
discuss how to detect the following and leadership patterns. To mine
semantic re-
lationships
, such as colleague or friends, we will discuss the supervised learning
frameworks with various spatiotemporal features.
Aggregate patterns over multiple trajectories.
The aggregate patterns describe a
group of moving objects share similar movement patterns.
Frequent trajectory
patterns
can find the frequent sequential transitions among spatial regions.
Mov-
ing object clusters
, such as flock, convoy and swarm, will detect a group of moving
objects being spatially close for a relatively long period of time.
Trajectory clus-
tering
groups similar (sub-)trajectories and reveals the popular paths shared by
trajectories.
The rest of the chapter is organized as follows. Section
2
introduces the basic def-
initions and concepts in spatiotemporal data mining. We then study the individual
periodic patterns in Sect.
3
. Section
4
covers pairwise movement patterns. And we
present aggregate patterns in Sect.
5
. Finally, we summarize the chapter in Sect.
6
.
2
Basic Concept
2.1
Spatiotemporal Data Collection
Spatiotemporal data is a broad concept. As long as the data is related to spatial
and temporal information, we call it spatiotemporal data. Two most frequently seen
spatiotemporal data are (1) ID-based spatiotemporal data collected from GPS and
(2) location-based data collected from sensors.
