New real-time resampling algorithm for particle filters

Two kinds of real-time resampling algorithms, local selection (LS) and Markov Chain Monte Carlo (MCMC), are briefly reviewed, and a new resampling algorithm suitable for real-time implementation of particle filters is proposed. The new resampling algorithm incorporates the concept of interleaving to overcome the drawback that when particles with large weights gather together, the performance will be degraded. The proposed resampling algorithm allows pipelined processing and parallelization implementation, which makes the particle filters more efficient. Simulations of a 2D target tracking application are conducted and the results indicate that the tracking accuracy of the proposed algorithm is superior to LS and MCMC. The hardware architecture of the algorithm is given. The hardware complexity including execution time and parallelization is analyzed, and it proves that the proposed method is efficient for real-time application.