Texture-Based Detection of WellDefined Benthic Monoculture Boundaries From ROV Pilot Camera Images

The paper deals with an image processing method for extracting the direction of the propagation of the upper Neptune grass (Posidonia oceanica ) bed boundary along the sea-bottom by the use of a monocular camera. The ultimate goal of research is to integrate this classifier into a feedback loop allowing an ROV to navigate the upper sea-grass bed border autonomously. This facilitates geo- referenced mapping of the border. The algorithm for extraction features four distinct phases: multi-resolution analysis using wavelets, vector quantization, post-processing of the obtained binary image and the extraction of the line parameters. The classification and line-fitting procedure are computationally optimized and made more robust by using weights in the Least Squares fitting procedure, and using nonlinear binary-image domain processing.