Greedy and $K$-Greedy Algorithms for Multidimensional Data Association

IEEE Transactions on Aerospace and Electronic Systems (Impact Factor: 1.76). 08/2011; 47(3):1915 - 1925. DOI: 10.1109/TAES.2011.5937273
Source: IEEE Xplore


The multidimensional assignment (MDA) problem is a combinatorial optimization problem arising in many applications, for instance multitarget tracking (MTT). The objective of an MDA problem of dimension d ∈ N is to match groups of d objects in such a way that each measurement is associated with at most one track and each track is associated with at most one measurement from each list, optimizing a certain objective function. It is well known that the MDA problem is NP-hard for d ≥ 3. In this paper five new polynomial time heuristics to solve the MDA problem arising in MTT are presented. They are all based on the semi-greedy approach introduced in earlier research. Experimental results on the accuracy and speed of the proposed algorithms in MTT problems are provided.

1 Follower
41 Reads
  • [Show abstract] [Hide abstract]
    ABSTRACT: This paper shows the results of two different methods of implementing the semi-greedy auction algorithm for hypothesis selection in the multiple hypothesis radar data association problem. The goal is to compare the Semi-Greedy Track Selection (SGTS) technique proposed by Waard, Capponi et. al. to a traditional semi-greedy approach [2], [3], [4], [5], [6], [12]. This study uses detection data generated by a medium-fidelity digital simulation of targets and sensors passed through the developed multiple hypothesis system. The results show that there is a crossover point at 8 solution sets for simplistic scenarios and a crossover point of 3 solution sets for more complex scenarios. This result would suggest that implementations where more than 8 solution sets in the semi-greedy approach are to be considered, the traditional semi-greedy approach is favorable. In problems where less than 3 solution sets are to be considered, the SGTS method provides better performance.
    No preview · Conference Paper · Mar 2014