Perimeter detection in wireless sensor networks

Abstract

For a mobile robotic agent to bridge the gaps between disconnected networks, it is beneficial for the robot to first determine the network coverage boundary. Several techniques have been introduced to determine the boundary nodes of a network, but the correctness of these techniques is often ill-defined. We present a technique for obtaining boundary node ground truth from region adjacency analysis of a model-based image created from a network graph. The resulting ground truth baseline is then used for quantitative comparison of several boundary detection methods including a local application of the image region adjacency analysis and the computation of the local convex hull with the addition of a perturbation value to overcome small boundary concavities in the node location point set. Given our proposed metrics of the techniques evaluated, the perturbed convex hull technique demonstrates a high success rate for boundary node identification, particularly when the convex hull is formed using two-hop neighborhoods. This technique was successfully implemented on a physical 25-node network, and the performance of this network implementation is evaluated.