A distributed algorithm for object tracking in wireless sensor networks using data mining based predicition

Object tracking is an important application of Wireless Sensor Networks (WSNs). One of the main challenges for designing an object tracking technique is energy conservation. Prediction based strategies are used to save energy. We consider an application where a WSN is installed in a forest for tracking the movement of animals (objects). Since, animals move in certain paths rather moving entirely randomly, we attempt to use this property. The movement patterns of the objects are extracted using a data mining technique. This information is used to predict the object movement with the aim to reduce the energy consumption in tracking. In situations where the prediction is erroneous, we propose a systematic recovery mechanism. The recovery mechanism has three levels. The number of additional nodes woken up progressively increases with the increase in the recovery level. If an object has been detected at a lower level, normal operations are resumed. It is observed that there are significant energy savings due to this systematic recovery mechanism. The proposed technique, called DESPOT, is fully distributed and energy efficient. We have carried out extensive simulation of DESPOT and have compared its performance with an existing technique, called PTSP. Simulation results show that DESPOT conserves significant energy and it has a better tracking efficiency than PTSP.