Tracking Dynamic Boundary Fronts using Range Sensors Subhasri

Tracking Dynamic Boundary Fronts using Range Sensors Subhasri Duttagupta (Ph. D student), Prof. Krithi Ramamritham Dept of Computer Sc. & Engg, Indian Institute of Technology, Bombay, India IIT Bombay 19 th Dec 2008

Early Warning System For Landslide Prediction using Sensor Networks Traffic Management on Highways IIT Bombay 19 th Dec 2008

Tracking Boundary Fronts • Compute confidence band with high accuracy. δ Width of the band • Estimate band with minimum communication overheads When is the tornado going to hit the city? [Manfredi et al. 2005] n, δ n = number of observations IIT Bombay Boundary Front Tracking k, loss of coverage 19 th Dec 2008 3

Combining Spatial and Temporal Estimation at a location no Observation change > threshold Temporal Estimation Spatial Estimation How to estimate Temporal Estimation When to update yes Spatial Estimation Feedback from Spatial IIT Bombay Multiple Observations Feedback improves the accuracy of Temporal Estimation 19 th Dec 2008 4

Placement of Estimation Points regions with high variance • • • Goal: Minimize LOC of interpolated band Start with a small set of equidistant points and perform spatial estimation at these points Add more estimation points in the region of high variance (variance implies spatial variation) Prediction Error Function can represent LOC without the knowledge of actual boundary IIT Bombay 19 th Dec 2008 5

Comparison of DBTR, SE, TE • DBTR performs • • better by 2 -4 % DBTR utilizes benefits of both the techniques Difference in accuracy does not change with δ. • Spatial Estimation provides more accuracy for lower δ • Temporal Estimation has better accuracy for larger δ IIT Bombay 19 th Dec 2008 6

Conclusions Tracking dynamic boundary fronts using range sensors • DBTR tracks both spatial and temporal variations with low communication overheads • Spatial estimation technique uses kernel smoothing to reduce the effect of noise • Temporal estimation technique uses Kalman filter modelbased approach updates estimate before the boundary moves out of confidence band IIT Bombay 19 th Dec 2008 7

Location of Spatial Estimation (SE) and Temporal Estimation (TE) h neighborhood actual boundary SE(xp 1, xp 2 ) TE(xp 2 ) SE(xp 1 ) TE(xp 1 ) Sensing nodes xp 1 IIT Bombay xp 2 19 th Dec 2008 Cluster heads 9