Blind Tracking using Sparsity Penalized Multidimensional Scaling

In this paper, we consider the problem of target tracking using sensor network measurements. We assume no prior knowledge of the sensor locations and so we refer to this tracking as `blind´. Since any sensor localization algorithm can only find the sensor location estimates up to a rotation and translation, we propose a novel sparsity penalized multidimensional scaling (MDS) algorithm to align the current time sensor location estimates to those of the previous time-frames. In the presence of a target, only location estimates of those sensors in the vicinity of a target vary from their initially estimated values. Based on the differences in the sensor location estimates between two time-frames, we design a perturbation based algorithm naturally rising from the sparsity penalized MDS for tracking multiple targets relative to the initial sensor location estimates. Through a detailed numerical analysis, we show that the tracking algorithm based on sparsity penalized MDS outperforms the conventional likelihood ratio test (LRT) based tracking.