Hidden Markov Model Filter Banks for Dim Target Detection from Image Sequences

Abstract

The track-before-detect processing technique has been employed in numerous computer vision based algorithms to address the dim target detection problem. In this processing approach, target information (as often provided by an image processing stage that has emphasised target features or suppressed unwanted noise) is integrated over a period of time before the detection decision is made. In this paper, we compare two Hidden Markov Model (HMM) based track-before-detect temporal filtering approaches for dim target detection that use image data pre-processed with a Preserved-Sign morphological filter. The two compared temporal filtering approaches are: a standard HMM filter (recent studies have shown this to be close to the state-of-the-art) and a novel HMM filter bank approach. Results from our simulation study involving various combinations of target speeds and signal-to-noise ratios show that the proposed novel HMM filter bank approach achieves a higher detection rate than the standard HMM approach.