Spatial Field Reconstruction with Distributed Kernel Least Squares in Mobile Sensor Networks

Inhalt:
Reconstructing spatial fields by sensor networks is a common problem in environmental monitoring applications. Usually, this task requires nonlinear techniques due to the underlying physical process. The so-called KDiCE algorithm is able to estimate such a spatial field in a distributed fashion by a nonlinear regression using kernel methods. To further enhance its reconstruction performance we consider a mobile sensor network in this paper. We utilize an iterative distributed scheme based on centroidal Voronoi tessellation where the sensors move to the center of mass of their Voronoi region. We include this sensor movement into the KDiCE algorithm and provide performance results regarding a distributed reconstruction of diffusion fields. Our evaluations show a significant gain in the performance by including sensor movement.