2018/09/18 by Jahidul Islam, Michael Fulton, Islam, Md Jahidul +5 · 2 citations
Earth and Planetary Sciences · Engineering · Environmental Science · #FOS: Computer and information sciences #Marine animal studies overview #Robotics (cs.RO) #Robotics and Sensor-Based Localization #Underwater Acoustics Research #Underwater Vehicles and Communication Systems
paper · pdf · doi:10.48550/arxiv.1809.06849
openalex publication_date 2018/09/18 · openalex created_date 2022/08/03 · openalex updated_date 2026/07/28
This paper explores the design and development of a class of robust\ndiver-following algorithms for autonomous underwater robots. By considering the\noperational challenges for underwater visual tracking in diverse real-world\nsettings, we formulate a set of desired features of a generic diver following\nalgorithm. We attempt to accommodate these features and maximize general\ntracking performance by exploiting the state-of-the-art deep object detection\nmodels. We fine-tune the building blocks of these models with a goal of\nbalancing the trade-off between robustness and efficiency in an onboard setting\nunder real-time constraints. Subsequently, we design an architecturally simple\nConvolutional Neural Network (CNN)-based diver-detection model that is much\nfaster than the state-of-the-art deep models yet provides comparable detection\nperformances. In addition, we validate the performance and effectiveness of the\nproposed diver-following modules through a number of field experiments in\nclosed-water and open-water environments.\n