EAGER - Integrating machine learning on autonomous platforms for target-tracking operations using stereo imagery

EAGER - 将机器学习集成到自主平台上，使用立体图像进行目标跟踪操作

• 批准号: 1812535
• 负责人: Kakani Young
• 金额: $ 26.92万
• 依托单位: Monterey Bay Aquarium Research Institute
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2018
• 资助国家: 美国
• 起止时间: 2018-01-01 至 2022-12-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1812535&HistoricalAwards=false
• 关键词: EAGER; Integrating; machine; learning; autonomous

The ocean's midwaters (depths from 200 to 1000 meters where sunlight is dim) are increasingly becoming an area of interest for scientific discovery and study. Efforts to further explore this vast and incredibly important region in the ocean involves the development of small, nimble, autonomous underwater vehicles (AUVs) that can be used for a variety of missions. This proposal will use a large database in video images collected over 25 years to train the vehicle to identify and track targets in real-time using a pair of stereo cameras. This project will involve a Postdoctoral Researcher who will be mentored by collaborators at MBARI and Stanford, who are pioneers in applying machine learning algorithms to underwater imagery. Results of this effort will be disseminated via conferences, publications, and outreach through industry and media partners. Media programs at MBARI and National Geographic Society will produce YouTube videos and social media posts detailing the efforts, the project's personnel, methods, and discoveries.The ocean's midwaters represent the largest ecosystem on earth with unique inhabitants and processes that link the surface waters to the seafloor. Efforts to further explore this vast and incredibly important region in the ocean involves development of AUVs that can be used for a variety of missions (e.g., transecting, tracking, fluid sampling). One of the key vehicle missions for these autonomous vehicles is to track targets in real-time. The tracking missions can be used for science questions as diverse as rates of marine snow sinking and its impact on biogeochemical cycling, the fate of rising methane from the benthos, and direct observations of organismal behavior to address their ecology and biomechanics. In order to conduct these tracking missions, robust algorithms are needed to identify and track targets as they change shape and state in realtime.

海洋的中层水域（深度从200米到1000米，阳光暗淡）越来越成为科学发现和研究的兴趣领域。为了进一步探索这一巨大而极其重要的海洋区域，需要开发可用于各种任务的小型、灵活的自主水下航行器（AUV）。该提案将使用25年来收集的视频图像中的大型数据库来训练车辆使用一对立体摄像机实时识别和跟踪目标。该项目将涉及一名博士后研究员，他将由MBARI和斯坦福大学的合作者指导，他们是将机器学习算法应用于水下图像的先驱。这一努力的成果将通过会议、出版物以及行业和媒体伙伴的外联活动加以传播。MBARI和国家地理学会的媒体计划将制作YouTube视频和社交媒体帖子，详细介绍项目的工作，人员，方法和发现。海洋的中间沃茨代表了地球上最大的生态系统，拥有独特的居民和连接表面沃茨和海底的过程。进一步探索海洋中这一广阔而极其重要的区域的努力涉及到可用于各种任务的AUV的开发（例如，横切、跟踪、流体取样）。这些自动驾驶车辆的关键任务之一是实时跟踪目标。跟踪任务可用于各种科学问题，如海洋积雪沉降率及其对生物地球化学循环的影响，底栖生物中甲烷上升的命运，以及直接观察生物行为以解决其生态学和生物力学问题。为了执行这些跟踪任务，需要鲁棒的算法来实时识别和跟踪改变形状和状态的目标。

项目成果

Visual tracking of deepwater animals using machine learning-controlled robotic underwater vehicles

• DOI: 10.1109/wacv48630.2021.00090
• 发表时间: 2021-01
• 期刊: 2021 IEEE Winter Conference on Applications of Computer Vision (WACV)
• 作者: K. Katija;P. Roberts;Joost Daniels;Alexander M. Lapides;K. Barnard;M. Risi;Ben Y Ranaan;Benjamin Woodward;Jonathan Takahashi