An adaptive refresh distributed model for estimation and efficient tracking of dynamic boundaries

We propose a distributed algorithm for tracking dynamic boundaries in a ranging sensor network. The main aim here is to minimize the number of data pushes to the sink by the observing sensors. Contour is modeled as correlated Brownian motion with drift. Sensors continuously sample the data. Neighboring sensors communicate and exploit the spatio-temporal correlation and using the parameters of the contour the time to push the sample data to the sink is predicted. A multihop path is established between sensor to sink to route the data. To get the global view of the contour sink apply non parametric regression on the sensor data. Along with the sample data sensors push the mean value so that the sink can estimate the sample point till the next push of data from sensor. The sensors push the data when the confidence in the estimate by the sink is below a specified thereshold. The performance of this model is compared with centralized model with respect to energy consumption for routing samples to sink.