Algorithms for the multi-sensor assignment problem in the d-generalized labeled multi-Bernoulli filter


  1. Yu, J.Y.
  2. Saucan, A-A.
  3. Coates, M.
  4. Rabbat, M.
Corporate Authors
Defence Research and Development Canada, Atlantic Research Centre, Halifax NS (CAN);McGill Univ, Montreal Que (CAN) Dept of Electrical and Computer Engineering
Practical implementations of the multi-sensor d-generalized labeled multi-Bernoulli filter require solving the multi-sensor assignment problem which is NP-hard. In this paper, we present two different algorithms, the combination method and the cross-entropy method, that find T highly likely target-measurement associations without exhaustive enumeration of all possible multi-sensor assignments. Numerical simulations are conducted to evaluate the aforementioned multi-sensor assignment methods together with the standard sequential processing method and a stochastic optimization algorithm based on Gibbs sampling. The combination method is based on an approximate assignment score function which leads to a lower running time, and it also explores a greater portion of the space of assignments compared to other methods. The cross-entropy method does not rely on the approximation and achieves better tracking performance than the sequential method, albeit at a higher computational overhead. The impact of the approximate score function on the algorithms’ performance is also studied via simulations and it is shown that the cross-entropy method consistently yields the best tracking performance whereas the combination method has the shortest runtime at high measurement noise level or high clutter rate.
target tracking;multi-sensor;filter
Report Number
DRDC-RDDC-2017-C244 — Contract Report
Date of publication
01 May 2018
Number of Pages
Electronic Document(PDF)

Permanent link

Document 1 of 1

Date modified: