Non-Parametric Approach to Change Detection and Estimation in Large Scale Sensor Networks
Shai Ben-David, Ting He, Lang Tong · 2004
We consider a non-parametric, spatial sample-based scheme for the detection and estimation of changes of a random field by collecting packets from randomly distributed sensors. We assume that each sensor has a fixed probability of successfully sending to a mobile access point a packet containing its local state---either "excited" or "baseline" and its location. The task we are concerned with here is as follows: Given two sets of packets collected over two nonoverlapping time-windows, construct a test to determine if the distribution generating the sensor's states has changed between these two time windows. Furthermore, if a change of distribution has occurred, we wish to estimate the distribution of the change.