Adaptive Contour Estimation Using Collaborating Mobile Sensors

Many real life applications such as tracking of pollutant flows, studying plankton populations in lakes, monitoring ocean currents etc. require real-time detection and tracking of level sets in a dynamic(spatio-temporal) field. In this paper, we examine the problem of estimating a level set or a contour of a particular value within a bounded region of varying field values using a network of mobile sensors. Given the task of estimating a contour, the sensors need to move towards the contour (converge phase) and then trace the contour (coverage phase). In order to minimize the error in estimation, the sensors need to trace all the points on the contour faithfully and to minimize latency, the maximum number of steps taken by the sensors during converge and coverage phases needs to be minimized. In our algorithm called ACE (Adaptive Contour Estimation), we overlap converge and coverage phases to minimize latency. ACE uses the history of movement of sensors to estimate the distance from the contour and strikes a tradeoff between arriving at the contour as quickly as possible and spreading out, compromising on latency initially, but reducing the overall latency by distributing the task of tracing the contour. ACE incorporates a collaborative wall moving algorithm during coverage phase for covering the contour thereby guaranteeing minimization of error in estimation for closed and continuous contours. We demonstrate that, irrespective of the type of initial deployment of sensors, ACE performs well (> 50% reduction in latency) when compared to a previous approach. We present results using both simulated and experimentally measured fields obtained by performing actual measurements of light intensity.