A FRAMEWORK FOR MINING UNIFYING TRAJECTORY PATTERNS USING SPATIOTEMPORAL DATASETS BASED ON VARYING TEMPORAL TIGHTNESS

Abstract

Trajectory patterns discovery is useful in learning interactions among moving objects. Different types of trajectory
patterns such as flock patterns, convoy patterns and swarm patterns have been proposed earlier, but methods were developed
for mining only a particular type of trajectory patterns. The pattern discovery becomes difficult and inefficient as users typically
may not know which types of trajectory patterns are present hidden in their data sets. One main observation is that trajectory
patterns can be arranged based on the strength of temporal tightness. In this paper, a framework of mining unifying trajectory
patterns also known as UT-patterns based on varying temporal tightness is proposed. The framework consists of three phases:
initial pattern identification, granularity adjustment, classification and visualization. The preprocessing is done by using
trajectory clustering algorithm and a set of initial UT-patterns identification are done in the first phase by using the spatiotemporal datasets. The granularity adjustments i.e., levels of detail are adjusted by drill down and roll up to detect other types of
UT-patterns in the second phase. Classifications of the UT-patterns are done using the Trajectory classification algorithm in the
third phase. Visualization of the classified UT-patterns is done in the final phase according to the patterns obtained as the result
of classification