Lightweight clustering of spatio-temporal data in resource constrained mobile sensing

Abstract

The technological development of inexpensive GPS receivers has enabled a new realm of applications for embedded sensing systems. The availability of location information allows these sensing system to study the motion trajectories of humans, animals, and objects. The storage of the collected trajectory data, however, represents a challenge for constrained devices with limited memory. In fact, external memory is often required, which incurs an additional cost for the storage component, enlarges the physical dimensions of the device, and also results in a measurable increase of the node's energy expenditure. In this paper, we present a clustering approach for GPS location information that is specifically tailored to resource-constrained sensing platforms. While our approach can be generalised to wide variety of applications, we focus on wireless animal tracking as an illustrative example. Our two-stage clustering process only records areas in which the animal has spent an extended period of time, in order to reduce the storage requirement while ensuring a low memory foot-print and processing requirements. We evaluate our solution using real-world animal GPS traces and show that our scheme achieves 90% improvement in location accuracy while also reducing the memory footprint by up to 99% in comparison with the state-of-the-art.