Distributed particle filters for sensor networks

This paper describes two methodologies for performing distributed particle filtering in a sensor network. It considers the scenario in which a set of sensor nodes make multiple, noisy measurements of an underlying, time-varying state that describes the monitored system. The goal of the proposed algorithms is to perform on-line, distributed estimation of the current state at multiple sensor nodes, whilst attempting to minimize communication overhead. The first algorithm relies on likelihood factorization and the training of parametric models to approximate the likelihood factors. The second algorithm adds a predictive scalar quantizer training step into the more standard particle filtering framework, allowing adaptive encoding of the measurements. As its primary example, the paper describes the application of the quantization-based algorithm to tracking a maneuvering object. The paper concludes with a discussion of the limitations of the presented technique and an indication of future avenues for enhancement.