Experimental analysis of adaptive clutter removal techniques in IR target detection systems

In many civilian and military applications, early warning IR detection systems have been developed over the years to detect long-range targets in scenarios characterized by highly structured background clutter. In this framework, a well-established detection scheme is realized with two cascaded stages: (i) background clutter removal, (ii) detection over the residual clutter. The performance of the whole detection system is especially determined by the choice and setting of the background estimation algorithm (BEA). In this paper, a novel procedure to automatically select the best performing BEA is proposed which relies on a selection criterion (BEA-SC) where the performances of the detection system are investigated via-simulation for the available BEAs and for different values of their parameters setting. The robustness of the BEA-SC is investigated to examine the performance of the detection system when the characteristics of the targets in the scene sensibly differ from the synthetic ones used in the BEA-SC, i.e. when the BEA is not perfectly tuned to the targets of interest in the scene. We consider target detection schemes that include BEAs based on well-established two-dimensional (2-D) filters. BEA-SC is applied to sequences of IR images acquired on scenarios typical of surveillance applications. Performance comparison is carried out in terms of experimental receiver operating characteristics (EX-ROC). The results show that the recently introduced BEA-SC is robust in the detection of targets whose characteristics are those expected in typical early warning systems.