A tracking approach for low observable target using plot-sequences of multi-frame detection

This paper addresses the dim target tracking problem with respect to the multi-frame detection (MFD). Generally, the MFD strategy implicitly merges the tracking stage into its framework, by which the estimated target trajectory can be simultaneously returned when a target is declared. However, due to the batch processing manner of MFD, such products, essentially, are short state sequences in time series (or short track segments), rather than the wanted continuous long target trajectories. In this paper, a novel target detection and tracking scheme, i.e., the multi-frame detection and tracking (MFDT) is proposed, wherein a specially designed tracker is assigned after the MFD as a supplement, which can combine and fuse such short track segments into continuous, accurate target trajectory. By using some reasonable assumptions, an optimal filtering algorithm for this fusion problem is also derived in this paper. Numerical results show that the proposed MFDT method can effectively improve the tracking performance both compared with the classical tracking and the only MFD method.

I. Introduction

Unlike the classical target detection and tracking approaches [1] which declare target and update tracks for each single frame, the batch multi-frame detection (MFD) methods [2]–[4] can achieve superior performance by integrating target energy over several consecutive frames, especially for dim or fluctuating targets. This enhancement benefits from the difference of space-time correlation between the target and noise (or clutter). For dynamic targets to be detected, such integrating process along unknown target trajectory can be accomplished by resorting to some Track-Before-Detect (TB-D) techniques. It is worth mentioning that, since the MFD implicitly merges the tracking stage (may be rough or simple) during the integrating procedure, once a target is declared, the corresponding integrating path is simultaneously returned serving as the estimation of the target trajectory. In recent years, the efficiency and robustness of MFD have been widely demonstrated [5]–[10], and its extension to some real scenarios application have been researched in [11]–[17] as well.