Sparse multidimensional scaling for blind tracking in sensor networks

Abstract

In this chapter, we consider the problem of tracking a moving target using sensor network measurements. We assume no prior knowledge of the sensor locations and so we refer to this tracking as 'blind'. We use the distributed weighted multidimensional scaling (dwMDS) algorithm to obtain estimates of the sensor positions. Since dwMDS can only find sensor position estimates up to rotation and translation, there is a need for alignment of sensor positions from one time frame to another. We introduce a sparsity constraint to dwMDS to align current time sensor positions estimates with those of the previous time frame. In the presence of a target, location estimates of sensors in the vicinity of the target will vary from their initial values. We use this phenomenon to perform link level tracking relative to the initially estimated sensor locations. © 2008 Springer-Verlag US.