Efficient Nonlinear Filtering Methods for Detection of Dim Targets by Passive Systems

In this paper, we outline an efficient system for detection of dim small targets that consists of three main parts: clutter rejection, track-before-detect (TBD), and detection subsystems. Detection and tracking of dim targets requires accumulation of sensor signal in time in order to resolve the target. There exist a number of powerful methods for detecting small low observable targets with stationary dynamics in image sequences provided by infrared (IR) and other imaging sensors. However, these methods need to be extended to handle maneuvering targets. We use a multiple switching model along with an optimal nonlinear filtering method to build an efficient TBD algorithm. We demonstrate that banks of interacting (nonlinear) Bayesian Matched filters (BIBMF) can be utilized for this purpose. The target’s dynamics is modeled by jump-linear systems. We present a computationally efficient (real time) TBD algorithm. The performance of the developed algorithm is compared to a Dynamic Programming (Viterbi) scheme and a bank of 3D Matched Filters (3DMF). The analysis shows that BIBMF outperforms the 3DMF algorithm and the sequential modification of Viterbi scheme. In addition, we develop a multi-target adaptive sequential detection algorithm that utilizes the estimates of targets’ positions obtained at the output of the TBD/BIBMF block. The performance of the algorithms is evaluated using real IR backgrounds obtained from SPAWAR Systems Center, San Diego (staring shipboard IR sensors). The results of this experimentation show that both developed algorithms, the TBD algorithm and the adaptive CFAR detection algorithm, perform very well for quite low effective signal-to-noise ratios.