Parallel particle filters for tracking in wireless sensor networks

An important application of wireless sensor networks is the tracking of objects moving through a monitored area. In some circumstances, a particle filter can perform substantially better than other tracking algorithms. A simple implementation is to transmit measurement data gathered at distributed sensors to a fusion centre and apply a single particle filter. The filter estimates the current position and predicts future locations so that appropriate sensors can be activated. This centralized approach can be energy-expensive and prone to failure: uncompressed data must be transmitted across multiple hops and there is a concentration of data transmission around the fusion site, which constitutes a single point of failure. This paper addresses these issues by proposing a distributed particle filter implementation in which parallel filters run at multiple nodes. These shared filters are used to quantize vectors of measurements. Simulations indicate that the scheme significantly reduces the energy expenditure of communication.