Computer Vision Methods for Marine Ecology

点击翻译按钮获取中文摘要

Computer Vision is an applied field of research, which means that various application domains have shaped its progress in distinct ways; ecology and environmental monitoring are likely to be the next drivers for major innovations in the computer vision field. I will thus explore and define new roles for computer vision in marine ecology; the proposed research program is built upon my demonstrated expertise and recent contributions to underwater image enhancement, habitat mapping and analysis, detection and counting of animals, and animal behaviour analysis. My overarching vision is to champion a paradigm shift in the methodology for monitoring core dimensions of marine ecology, such as abundance, biodiversity, and animal behaviour. More specifically, the research program will significantly contribute to a successful transition from time-consuming, resource-intensive, small-scale observational studies towards scalable, automatic, data-driven methods for marine ecology. During the next five years, I will focus on three short-term objectives, as follows: A. Develop automated methods for the visual quantification of marine species abundance and biodiversity B. Develop automated methods for the visual detection and analysis of marine animal behaviour C. Develop automated methods for efficient and sustainable underwater visual data management The short-term objectives support the proposed paradigm shift in environmental monitoring methodologies, which is my long-term focus. Biodiversity and abundance are currently estimated manually within limited spatiotemporal contexts: objectives A and B provide a much needed up-scaling of the current protocols. Objective C addresses the efficient use of the vast amount of data acquired when monitoring large-scale ecosystems over extended periods of time. From a societal standpoint, my research addresses the stringent need for more accurate, sustainable (from a computational viewpoint) and comprehensive environmental monitoring. Due to maturing hardware technologies allowing for cost-efficient deployment and operation of cameras in remote environments, long-term visual data collection has recently become prevalent in many key areas of ecological and biological research. However, the vast amount of collected imagery clearly exceeds human analysis capabilities. Computer vision can efficiently address this Big Data problem, and thus can contribute to monitoring and improving the health of ecosystems, which is intricately linked to the well-being of our society. We hope that this contribution will inform the development of new Canadian environmental policies, and thus will enable Canada to play a leading role in promoting and implementing a climate resilient global economy.