Graph-based Compression of Particle Clouds for Gossip-Based Decentralized Particle Filters


  1. Rabbat, M.G.
  2. Coates, M.J.
  3. Yu, J.Y.
  4. Ustebay, D.
  5. Blouin, S.
Corporate Authors
Defence Research and Development Canada, Atlantic Research Centre, Halifax NS (CAN);McGill Univ, Montreal QUE (CAN) Dept of Electrical Engineering
In gossip-based decentralized particle filtering, a network of sensors collaborate to track one or more tar- gets. Each node in the network maintains and updates a copy of the particle filter, and gossip algorithms are used to simultaneously synchronize the filters and fuse measurements after each measurement round. Particle filtering methods are attractive for tracking when the state dynamics and/or the measurement model are nonlinear and/or non-Gaussian. The accuracy of particle filters is generally related to the number of particles used to approximate the posterior density. A significant challenge in the decentralized setting is to accurately fuse and synchronize the weighted particle approximations while reducing the communication overhead. This is especially important when nodes communicate over unreliable wireless channels, in which case bandwidth-limitations restrict the amount of information that can be communicated in each measurement round. Also, when nodes are battery-powered, the energy consumed for each wireless transmission is significantly higher than that consumed for computation, and so reducing the amount of information communication can extend the lifetime of the system. We propose a graph-based method to synchronize the particle weights to values which are approximately equal to those that would be computed if every node had access to the measurements from all other sensors. Our approach leverages the fact that, in the decentralized particle filtering set
Report Number
DRDC-RDDC-2015-N037 — External Literature
Date of publication
30 May 2014
Number of Pages
Electronic Document(PDF)

Permanent link

Document 1 of 1

Date modified: