Development of Program for Detection of Unspecified Floating Obstacles from Camera Images

In recent years there has got much interest in Ocean Surveillance, especially in the domain of Maritime Autonomous Surface Ships(MASS). Obstacle Detection System for MASS needs to detect automatically not only other ships but unspecified floating obstacles such as rubber boats or culturing rafts as well. The system consists of AIS, Radar and Camera-based Detection System as usual. The detection of obstacles through Camera Images mainly utilizes Detection Frameworks that use Deep-Learning Image Detection and Recognition Network Models such as YOLO. The Network Models are well-retrained by a large quantity of Sea photos with images of ships in the photos. However, each floating obstacle has various shapes that can’t be categorized properly. That makes it difficult to detect obstacles for the Network Model.