Optimizing Postprocessing of Range-Gated Viewing Data for Maritime Search and Rescue Operations at Night and in Bad Weather Conditions

Abstract

At night and in bad weather conditions the detection of persons and objects floating in the sea represents a major challenge for search and rescue operations (SAR). If conventional searchlights are used, backscattering from rain, fog and snow decreases detection range. Therefore, a compact and inexpensive range-gated viewing system which significantly reduces atmospheric backscattering was developed. The instrument was designed for detection ranges of several hundred meters. In this study, different image processing techniques were analyzed in terms of improved object detectability for a human observer and for a machine learning-based object detector, based on a real-world image dataset. On the one hand, noise of the camera is reduced by performing a non-uniformity correction (NUC) and on the other, the dynamic range of the images is adjusted and dark objects are accentuated by equalizing (EQ). The aim of this field study with the subsequent post processing steps was to improve visibility for both human observers and machine learning-based object detectors with low computing power, based on real-world image datasets. The results show that processing requirements are different in both cases, mainly due to human eye perception, which an automated detector does not rely on and therefore the performance of the object detector before the equalizing step is slightly better. However, the NUC improves the image quality in any case.