Nonparametric change detection in 2D random sensor field

The problem of detecting changes from data collected from a large-scale randomly deployed 2D sensor field is considered. Under a nonparametric change detection framework, we propose detection algorithms using two measures of change. The theoretical performance guarantee is derived from the Vapnik-Chervonenkis theory. By exploiting the structures of the search domain, we design a suboptimal recursive algorithm to detect the area of largest change which, for M sample points, runs in time O(M/sup 2/logM) (compared to an O(M/sup 4/) required for a straightforward exhaustive search). The lost of performance diminishes as M increases.