MMOSPA estimation with unknown number of objects

Document Type

Conference Proceeding

Publication Date

8-31-2015

Publication Title

2015 IEEE China Summit and International Conference on Signal and Information Processing, ChinaSIP 2015 - Proceedings

First Page

706

Keywords

Minimum mean OSPA (MMOSPA) estimate, Multi-object filtering, Multi-object systems, Multitarget tracking, Optimal sub-pattern assignment (OSPA), Point processes, Random finite sets (RFS), Wasserstein distance

Last Page

710

Abstract

We consider the problem of estimating unordered sets of objects, which arises when the object labels are irrelevant. The widely used minimum mean square error (MMSE) estimators are not applicable for the estimation of unordered objects. Recently, a new type of estimator, known as the minimum mean OSPA (MMOSPA) estimator, which minimizes the optimal sub-pattern assignment (OSPA) metric, was proposed. Unfortunately, the MMOSPA estimator is unable to deliver a closed form solution when the objects, represented as a random finite set (RFS), are multidimensional or when the underlying posterior density is non-Gaussian; also, the existing MMOSPA estimators have not bee used to estimate unknown numbers of objects. In this paper, we derive a particle-based algorithm for the estimation of unknown number of objects which is optimal in the MMOSPA sense; also, the proposed algorithm is not limited by the dimension of the RFS or the requirement of Gaussian posterior density.

DOI

10.1109/ChinaSIP.2015.7230496

ISBN

9781479919482

Link to Full Text

Share

COinS