CLSTERS: A general system for reducing errors of trajectories under challenging localization situations

• Main Authors: WU, Hao, SUN, Weiwei, ZHENG, Baihua, YANG, Li, ZHOU, Wei
• Format: text
• Language: English
• Published: Institutional Knowledge at Singapore Management University 2017
• Subjects: Localization; error reduction; cellular-based trajectory; map matching; Databases and Information Systems; Software Engineering
• Online Access: https://ink.library.smu.edu.sg/sis_research/3869; https://ink.library.smu.edu.sg/context/sis_research/article/4871/viewcontent/UbiComp_2017_Final.pdf
• Institution: Singapore Management University

Trajectory data generated by outdoor activities have great potential for location based services. However, depending on the localization technique used, certain trajectory data could contain large errors. For example, the error of trajectories generated by cellular-based localization techniques is around 100m which is ten times larger than that of GPS-based trajectories. Hence, enhancing the utility of those large-error trajectories becomes a challenge. In this paper we show how to improve the quality of trajectory data having large errors. Some existing works reduce the error through hardware which requires information such as the time of arrival (TOA), received signal strength indication (RSSI), the position of cell towers, etc. Moreover, different positioning techniques will result in different hardware-based solutions and different data formats, which limit the generalizablity. Other works study a related but different problem, i.e., map matching, with the aid of road network information, to reduce the uncertainty and the noise of trajectory data. However, most of these approaches are designed for the GPS-sampled data, and hence they might not be able to achieve a similar performance when applied directly to trajectories with large errors. Motivated by this, we propose a general error reduction system namely CLSTERS for trajectories with large scale of errors. Our system is hardware independent and only requires the coordinates and the time stamp of each sample point which makes it general and ubiquitous. We present results from experiments using three real-world datasets in three different cities generated by two different localization techniques and the results show that our approach outperforms existing solutions.