A Novel Outlier Detection Method for Spatio-Tempral Trajectory Data

Abstract

The development of mobile device technology and localization technology makes the collection of spatio-temporal information from moving objects much easier than before, and outlier detection for spatio-temporal trajectory is becoming increasingly attractive to data mining community. However, there is a lack of serious studies in this area. Several existing trajectory outlier methods such as the partition-and-detect framework can only deal with the trajectory data which only includes spatial attributes. It cannot be applied to the spatio-temporal trajectory data which includes both spatial and temporal attributes. In this paper, we propose an enhanced partition-and-detect framework to detect the outliers of spatio-temporal trajectory data. In this framework, we mainly introduce an outlier detection method which uses trajectory MBBs(Minimum Boundary Boxs). Based on this enhanced framework, we propose a congestion outlier detection method. Finally, the efficiency and accuracy are evaluated through experiments which use a real traffic dataset called US Highway 101 Dataset.