Evaluation of Methods for Robust, Automatic Detection of Net Tear with Remotely Operated Vehicle and Remote Sensing

Abstract

Accompanying the continuous growth of the aquaculture fish farming industry in the recent years, the usage of Remotely Operated Vehicles (ROV) for regular inspections of net integrity has become increasingly common. For a human ROV operator, routine inspections can be repetitious and time consuming, and improving the regularity and efficiency of these operations are of interest. The aim of this study was therefore be to develop a robust technique for automatic detection of net damage with an ROV mounted camera and computer vision, which later can be employed either as an aid for a human operator or be embedded into an automatic solution in the future. Information from temporal background segmentation, edge detection, motion estimation and multiple image channels was incorporated into a high-redundancy combinatorial system design for background segmentation. Assessment of net damage was made from the resulting binary foreground image by employing a detection scheme based on morphological operations. The background segmentation performance, detection accuracy and robustness of the developed system was evaluated on previously recorded video material from real ROV operations and a simulated test setup. Results showed that the background segmentation process provided a stable and comprehensive binary foreground image, but with reduced ability to segment certain foreground objects. The damage assessment methodology, on the other hand, displayed a rigorous evaluation capability. With some additional measures, the developed procedure seems promising for achieving robust net damage detection in a practical implementation.