Mining mobile object trajectories: frameworks and algorithms

Trajectory mining;Pattern mining;Trajectory clustering;Road networks;Mobile object trajectory

Han, Binh Thi ; Liu, Ling Computer Science Omiecinski, Edward Pu, Calton Navathe, Shamkant Lofstead, Gerald Krishnamurthy, Vikram ; Liu, Ling

University:Georgia Institute of Technology

Department:Computer Science

关键词: Trajectory mining;    Pattern mining;    Trajectory clustering;    Road networks;    Mobile object trajectory;   

Others  :  https://smartech.gatech.edu/bitstream/1853/52968/1/HAN-DISSERTATION-2013.pdf

美国|英语

来源: SMARTech Repository

【 摘 要 】

The proliferation of mobile devices and advances in geo-positioning technologies has fueled the growth of location-based applications, systems and services. Many location-based applications have now gained high popularity and permeated the daily activities of mobile users. This has led to a huge amount of geo-location data generated on a daily basis, which draws significant interests in analyzing and mining ubiquitous location data, especially trajectories of mobile objects moving in road networks (MO trajectories). Mobile trajectories are complex spatio-temporal sequences of location points with varying sample sizes and varying lengths. Mining interesting patterns from large collection of complex MO trajectories presents interesting challenges and opportunities which can reveal valuable insights to the studies of human mobility in many perspectives. This dissertation research contributes original ideas and innovative techniques for mining complex trajectories from whole trajectories, from subtrajectories of significant characteristics, and from semantic location sequences within large-scale datasets of MO trajectories. Concretely, the first unique contribution of this dissertation is the development of NEAT, a three-phase road-network aware trajectory clustering framework to organize MO subtrajectories into spatial clusters representing highly dense and highly continuous traffic flows in a road network. Compared with existing trajectory clustering approaches, NEAT yields highly accurate clustering results and runs orders of magnitude faster by smartly utilizing traffic locality with respect to physical constraints of the road network, traffic flows among consecutive road segments and flow-based density of mobile traffic as well as road network based distances. The second original contribution of this dissertation is the design and development of TraceMob, a methodical and high performance framework for clustering whole trajectories of mobile objects. To our best knowledge, this is the first whole trajectory clustering system for MO trajectories in road networks. The core idea of TraceMob is to develop a road-network aware transformation algorithm that can map complex trajectories of varying lengths from a road network space into a multidimensional data space while preserving the relative distances between complex trajectories in the transformed metric space. The third novel contribution is the design and implementation of a fast and effective trajectory pattern mining algorithm TrajPod. TrajPod can extract the complete set of frequent trajectory patterns from large-scale trajectory datasets by utilizing space-efficient data structures and locality-aware spatial and temporal correlations for computational efficiency. A comprehensive performance study shows that TrajPod outperforms existing sequential pattern mining algorithms by an order of magnitude.