Modified Scoring in Multiple-Hypothesis Tracking

Track-oriented multiple hypothesis tracking (MHT) is well-established as a paradigm for multi-sensor multitarget tracking. The fundamental approach includes many variants. Hypothesis-oriented MHT was first proposed by Reid [10]. The initial integer-programming formulation of the problem is due to Morefield [8]. The hybrid-state decomposition that allows for computationally-efficient track-oriented MHT is due to Kurien [7]. An efficient solution to the optimization problem required for nscan hypothesis pruning via Lagrangian relaxation is due to Poore and Rijavec [9]. A linearprogramming based relaxation approach to the same optimization problem was proposed independently by Coraluppi et al [3] and by Storms and Spieksma [12]. In practice, MHT implementations must limit the number of local (or track) hypotheses. This can be achieved by measurement gating, by limiting hypothesis generation, and by pruning or merging existing hypotheses. Additionally, sequential track extraction schemes are adopted in lieu of optimal (batch) track extraction [1]. These techniques, while necessary for computationally-realizable and real-time MHT processing, lead to suboptimal data association decisions and track extraction. In this paper, we show that the suboptimality can be mitigated by favoring nearly-confirmed and confirmed tracks over tentative ones in the dataassociation process with suitable modification to the MHT track scoring equations. This paper is organized as follows. In Section 2, we summarize the hybrid-state derivation of the trackoriented MHT, with some modifications with respect to the original derivation [7]. In Section 3, we address briefly the target-death problem that arises in probability hypothesis density (PHD) filtering as discussed in Erdinc et al [6], and show that it does not arise in track-oriented MHT. In Section 4, we introduce the modified-scoring MHT equations and considering a limiting case of the general tracking problem that we call cardinality tracking. Section 5 provides simulation results that demonstrate the improved performance of modified-scoring MHT over standard MHT. Concluding remarks are in Section 6.