一种水平面估计的水域环境视觉SLAM算法  

A visual SLAM for waterways algorithm environment with horizontal plane estimation

作　　者：高于科 章伟 胡陟 江鹏伟 谢贵亮 GAO Yuke;ZHANG Wei;HUZhi;JIANG Pengwei;XIEGuiliang(School of Mechanical and Automotive Engineering,Shanghai University of Engineering Science,Shanghai 201620,China)

机构地区：[1]上海工程技术大学机械与汽车工程学院,上海201620

出　　处：《测绘科学》2023年第5期104-112,128,共10页Science of Surveying and Mapping

基　　金：国家自然科学基金项目(62003207)。

摘　　要：针对内河水域环境水面镜像极易受扰动产生扭曲形变,从而导致视觉SLAM位姿估计不准确的问题,提出匹配地图点和最大梯度估计的方法用于估计水平面位置,区分了水域环境和真实环境的地图点。提出了一种判断水域环境可靠性方法,用于决定追踪侧重于真实环境或全局环境;建图环节中保留水域环境的地图点,用于估计当前水面范围。基于USVInland数据集提供的内河水域场景测试表明,相比原ORB-SLAM3算法,该文算法在仿真平台运行速度比原算法多花费约10%的时间,但绝对轨迹误差平均降低了47%。For the inland waterways environment,the surface mirror image is vulnerable to disturbance,which causes distortions and the inaccurateness of the visual simultaneous localization and mapping(SLAM)position,This article proposes the method of matching map points and maximum gradient estimates to estimate the position of the horizontal plane,distinguish the map point of the inland waterways environment and the real environment;propose a method of judging the reliability of the water environment to determine the focus on the real environment or the global environment.;The map point of the inland waterways environment is retained during the construction process,which is used to estimate the current water surface.Based on the inland waterways provided by the USVInland data set,compared with the original ORB-SLAM3 algorithm,The algorithm in this paper reduces the absolute trajectory error by an average of 47%and only takes about 10%more time to run on the simulation platform than the original algorithm.

关 键 词：即时定位与建图 内河水域 水面识别 ORB-SLAM3 USV 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]