标题： 立体声侧视和前视声纳的特征几何学 显示英文标题

标题： Feature Geometry for Stereo Sidescan and Forward-looking Sonar

摘要： 在本文中，我们研究了海洋机器人中的立体声学数据融合，并提出了一种基于几何的方法，用于将一个声纳观测到的特征投影到另一个声纳上，适用于包含前视声纳和侧扫声纳的跨模态立体声纳设置。 我们的侧扫声纳和前视声纳的声学几何结构受到立体相机的极线几何的启发，我们利用相对位姿信息来投影在一个声纳图像中观测到的特征在另一个声纳图像中的位置。 此外，我们分析了特征相对于声纳的位置以及两个声纳之间的相对位姿如何影响投影。 从仿真结果中，我们确定了适用于现场机器人应用（如特征对应和特征三维信息恢复）的理想的立体配置。 显示英文摘要

摘要： In this paper, we address stereo acoustic data fusion for marine robotics and propose a geometry-based method for projecting observed features from one sonar to another for a cross-modal stereo sonar setup that consists of both a forward-looking and a sidescan sonar. Our acoustic geometry for sidescan and forward-looking sonar is inspired by the epipolar geometry for stereo cameras, and we leverage relative pose information to project where an observed feature in one sonar image will be found in the image of another sonar. Additionally, we analyze how both the feature location relative to the sonar and the relative pose between the two sonars impact the projection. From simulated results, we identify desirable stereo configurations for applications in field robotics like feature correspondence and recovery of the 3D information of the feature.