Efficient Spatio-Temporal Information Fusion in Sensor Networks

Making the sensor data look more meaningful in its representation of an observed entity is the primary goal of sensor data fusion. Due to the energy constraint on sensors, there exists a need for algorithms that minimize the fusion cost while maintaining the validity of the data sent to the base station. Maintaining validity is even more difficult when we have a limited knowledge of the factors that govern an observed sensor entity. To achieve this goal, we modeled the uncertainties in sensor data and fed them into the system, employing recursive data estimation. By doing so, we considered the dynamically changing environmental parameters affecting the network to produce the most accurate representation of the observed system state. We propose here a spatio-temporal, correlation-based estimation procedure to corroborate the detection of an event in a sensor field. The number of in-network communications plays a great role from the networking perspective. This is because the power consumption during communication is several times greater than the power consumption during computation. To achieve this, our algorithm ensures that communication is done only during the time of an event. At all other times, the sensor motes maintain an updated global estimate, without communicating, by using a prediction algorithm. This reduces the need for frequent sensor synchronization. We conducted experiments using our distributed fusion architecture to show our algorithm's effectiveness; by a reduction in the power consumption,