Point Target Detection in IR Image Sequences using Spatio-Temporal Hypotheses Testing

摘要

This paper addresses the problem of detecting weak, moving point targets in infrared (IR) image sequences that also contain evolving cloud clutter. The problem is initially attacked in the temporal domain, where there is a clear distinction between targets and cloud clutter. We formulate the temporal detection problem in the context of a hypothesis testing procedure on individual pixel temporal profiles, leading to a theoretically sound and computationally efficient statistical test. The technique assumes we have deterministic and statistical models for the temporal behavior of the background noise, target and clutter, on a single pixel basis. The target temporal profile can be modeled by scaled versions of the point spread function (PSF) of the imager, while the clutter can be well described using a first order Markov model. Based on these models, which are experimentally verified using real data, we develop a generalized likelihood ratio test and perfect measurement performance analysis, and present the resulting decision rule. We demonstrate the effectiveness of the technique by applying the resulting algorithm to real world infrared image sequences containing targets of opportunity. For severe clutter situations which result in false alarms, we suggest an additional spatial hypothesis testing procedure, designed to exploit the difference in the spatial signature of point targets and cloud clutter. As for the temporal case, we propose models for the spatial signatures of targets and cloud clutter and derive the resulting decision rule. Application to real IR image sequences shows that the composite spatio-temporal algorithm results in reduced false alarm rates and increased probability of detection compared to the purely temporal approach.