Detecting Marine Animals in Underwater Video: Let's Start with Salmon

Abstract

Environmental Decision Support Systems (EDSS) constitute an important emerging technology for the management of marine environments and resources world-wide. A requirement for such systems is the provision of accurate and comprehensive data on marine animal populations and habitats. Underwater video is increasing in importance as a technology for collecting this data, but manual analysis of the video is time-consuming and subjective, and so automated techniques need to be developed before video technology becomes effective as a major data collection method for EDSS. A central requirement in many applications is the automated detection of specific marine animals in the video footage. Our aim is to develop a range of techniques for detecting marine animals in underwater video scenes and, as a starting point, we have developed a segmentation technique to automatically detect salmon depicted in images taken from video cameras placed in fish farm cages.